diff --git a/.gitattributes b/.gitattributes index c3cd56d9a7cd4cef19c97bcde5beb2baa6c62f30..804b6222e4a723f774c7af20deb21508e5da698a 100644 --- a/.gitattributes +++ b/.gitattributes @@ -7759,3 +7759,233 @@ parse/train/gDcaUj4Myhn/gDcaUj4Myhn_origin.pdf filter=lfs diff=lfs merge=lfs -te parse/train/ByxY8CNtvr/ByxY8CNtvr_span.pdf filter=lfs diff=lfs merge=lfs -text parse/train/ByxY8CNtvr/ByxY8CNtvr_layout.pdf filter=lfs diff=lfs merge=lfs -text parse/train/ByxY8CNtvr/ByxY8CNtvr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1xQNlBYPS/r1xQNlBYPS_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1xQNlBYPS/r1xQNlBYPS_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1xQNlBYPS/r1xQNlBYPS_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/o81ZyBCojoA/o81ZyBCojoA_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/o81ZyBCojoA/o81ZyBCojoA_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/o81ZyBCojoA/o81ZyBCojoA_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByxBFsRqYm/ByxBFsRqYm_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByxBFsRqYm/ByxBFsRqYm_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByxBFsRqYm/ByxBFsRqYm_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJYoqzbC-/HJYoqzbC-_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJYoqzbC-/HJYoqzbC-_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJYoqzbC-/HJYoqzbC-_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/umIdUL8rMH/umIdUL8rMH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/umIdUL8rMH/umIdUL8rMH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/umIdUL8rMH/umIdUL8rMH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/sUgpxb9QD/sUgpxb9QD_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/sUgpxb9QD/sUgpxb9QD_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/sUgpxb9QD/sUgpxb9QD_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJeq43AqF7/HJeq43AqF7_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJeq43AqF7/HJeq43AqF7_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJeq43AqF7/HJeq43AqF7_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bk8ZcAxR-/Bk8ZcAxR-_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bk8ZcAxR-/Bk8ZcAxR-_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bk8ZcAxR-/Bk8ZcAxR-_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJzMATlAZ/SJzMATlAZ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJzMATlAZ/SJzMATlAZ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJzMATlAZ/SJzMATlAZ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rylWVnR5YQ/rylWVnR5YQ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rylWVnR5YQ/rylWVnR5YQ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rylWVnR5YQ/rylWVnR5YQ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Q32U7dzWXpc/Q32U7dzWXpc_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Q32U7dzWXpc/Q32U7dzWXpc_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Q32U7dzWXpc/Q32U7dzWXpc_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkMW1hRqKX/rkMW1hRqKX_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkMW1hRqKX/rkMW1hRqKX_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkMW1hRqKX/rkMW1hRqKX_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygh9j09KX/Bygh9j09KX_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygh9j09KX/Bygh9j09KX_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygh9j09KX/Bygh9j09KX_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/R-616EWWKF5/R-616EWWKF5_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/R-616EWWKF5/R-616EWWKF5_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/R-616EWWKF5/R-616EWWKF5_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SybqeKgA-/SybqeKgA-_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SybqeKgA-/SybqeKgA-_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SybqeKgA-/SybqeKgA-_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Arn2E4IRjEB/Arn2E4IRjEB_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Arn2E4IRjEB/Arn2E4IRjEB_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Arn2E4IRjEB/Arn2E4IRjEB_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/H4e7mBnC9f0/H4e7mBnC9f0_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/H4e7mBnC9f0/H4e7mBnC9f0_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/H4e7mBnC9f0/H4e7mBnC9f0_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkhQHMW0W/SkhQHMW0W_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkhQHMW0W/SkhQHMW0W_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkhQHMW0W/SkhQHMW0W_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1erHoR5t7/S1erHoR5t7_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1erHoR5t7/S1erHoR5t7_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1erHoR5t7/S1erHoR5t7_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/7J-fKoXiReA/7J-fKoXiReA_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/7J-fKoXiReA/7J-fKoXiReA_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/7J-fKoXiReA/7J-fKoXiReA_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HygrAR4tPS/HygrAR4tPS_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HygrAR4tPS/HygrAR4tPS_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HygrAR4tPS/HygrAR4tPS_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SylVJTNKDr/SylVJTNKDr_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SylVJTNKDr/SylVJTNKDr_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SylVJTNKDr/SylVJTNKDr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ryQu7f-RZ/ryQu7f-RZ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ryQu7f-RZ/ryQu7f-RZ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ryQu7f-RZ/ryQu7f-RZ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkgHY0NYwr/rkgHY0NYwr_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkgHY0NYwr/rkgHY0NYwr_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rkgHY0NYwr/rkgHY0NYwr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/6MaBrlQ5JM/6MaBrlQ5JM_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/6MaBrlQ5JM/6MaBrlQ5JM_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/6MaBrlQ5JM/6MaBrlQ5JM_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJgsskrFwH/rJgsskrFwH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJgsskrFwH/rJgsskrFwH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJgsskrFwH/rJgsskrFwH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BkfbpsAcF7/BkfbpsAcF7_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BkfbpsAcF7/BkfbpsAcF7_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BkfbpsAcF7/BkfbpsAcF7_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Cnon5ezMHtu/Cnon5ezMHtu_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Cnon5ezMHtu/Cnon5ezMHtu_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Cnon5ezMHtu/Cnon5ezMHtu_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/qZzy5urZw9/qZzy5urZw9_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/qZzy5urZw9/qZzy5urZw9_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/qZzy5urZw9/qZzy5urZw9_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/jnkE5c5f9m/jnkE5c5f9m_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/jnkE5c5f9m/jnkE5c5f9m_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/jnkE5c5f9m/jnkE5c5f9m_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TV9INIrmtWN/TV9INIrmtWN_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TV9INIrmtWN/TV9INIrmtWN_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TV9INIrmtWN/TV9INIrmtWN_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkekl0NFPr/Hkekl0NFPr_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkekl0NFPr/Hkekl0NFPr_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkekl0NFPr/Hkekl0NFPr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/zdrls6LIX4W/zdrls6LIX4W_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/zdrls6LIX4W/zdrls6LIX4W_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/zdrls6LIX4W/zdrls6LIX4W_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TMUR2ovJfjE/TMUR2ovJfjE_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TMUR2ovJfjE/TMUR2ovJfjE_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TMUR2ovJfjE/TMUR2ovJfjE_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJgExaVtwr/HJgExaVtwr_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJgExaVtwr/HJgExaVtwr_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJgExaVtwr/HJgExaVtwr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/DAaaaqPv9-q/DAaaaqPv9-q_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/DAaaaqPv9-q/DAaaaqPv9-q_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/DAaaaqPv9-q/DAaaaqPv9-q_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlxmAKlg/BJlxmAKlg_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlxmAKlg/BJlxmAKlg_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlxmAKlg/BJlxmAKlg_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/dUEpGV2mhf/dUEpGV2mhf_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/dUEpGV2mhf/dUEpGV2mhf_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/dUEpGV2mhf/dUEpGV2mhf_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygq-H9eg/Bygq-H9eg_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygq-H9eg/Bygq-H9eg_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bygq-H9eg/Bygq-H9eg_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/UFWnZn2v0bV/UFWnZn2v0bV_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/UFWnZn2v0bV/UFWnZn2v0bV_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/UFWnZn2v0bV/UFWnZn2v0bV_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hy6b4Pqee/Hy6b4Pqee_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hy6b4Pqee/Hy6b4Pqee_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hy6b4Pqee/Hy6b4Pqee_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HOFxeCutxZR/HOFxeCutxZR_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HOFxeCutxZR/HOFxeCutxZR_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HOFxeCutxZR/HOFxeCutxZR_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bys_NzbC-/Bys_NzbC-_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bys_NzbC-/Bys_NzbC-_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Bys_NzbC-/Bys_NzbC-_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1eXygBFPH/B1eXygBFPH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1eXygBFPH/B1eXygBFPH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1eXygBFPH/B1eXygBFPH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/bYi_2708mKK/bYi_2708mKK_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/bYi_2708mKK/bYi_2708mKK_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/bYi_2708mKK/bYi_2708mKK_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJRpRfKxx/SJRpRfKxx_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJRpRfKxx/SJRpRfKxx_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SJRpRfKxx/SJRpRfKxx_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SygwwGbRW/SygwwGbRW_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SygwwGbRW/SygwwGbRW_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SygwwGbRW/SygwwGbRW_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BygWRaVYwH/BygWRaVYwH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BygWRaVYwH/BygWRaVYwH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BygWRaVYwH/BygWRaVYwH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gE6TEYDB/S1gE6TEYDB_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gE6TEYDB/S1gE6TEYDB_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gE6TEYDB/S1gE6TEYDB_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skk3Jm96W/Skk3Jm96W_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skk3Jm96W/Skk3Jm96W_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skk3Jm96W/Skk3Jm96W_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyxvSiCcFQ/SyxvSiCcFQ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyxvSiCcFQ/SyxvSiCcFQ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyxvSiCcFQ/SyxvSiCcFQ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/YDGJ5YExiw6/YDGJ5YExiw6_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/YDGJ5YExiw6/YDGJ5YExiw6_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/YDGJ5YExiw6/YDGJ5YExiw6_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByZvfijeg/ByZvfijeg_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByZvfijeg/ByZvfijeg_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/ByZvfijeg/ByZvfijeg_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJTzHtqee/HJTzHtqee_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJTzHtqee/HJTzHtqee_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJTzHtqee/HJTzHtqee_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skeke3C5Fm/Skeke3C5Fm_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skeke3C5Fm/Skeke3C5Fm_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skeke3C5Fm/Skeke3C5Fm_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/O3bqkf_Puys/O3bqkf_Puys_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/O3bqkf_Puys/O3bqkf_Puys_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/O3bqkf_Puys/O3bqkf_Puys_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkgEaj05t7/SkgEaj05t7_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkgEaj05t7/SkgEaj05t7_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkgEaj05t7/SkgEaj05t7_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkpSlKIel/SkpSlKIel_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkpSlKIel/SkpSlKIel_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkpSlKIel/SkpSlKIel_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Twf_XYunk5j/Twf_XYunk5j_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Twf_XYunk5j/Twf_XYunk5j_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Twf_XYunk5j/Twf_XYunk5j_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyqShMZRb/SyqShMZRb_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyqShMZRb/SyqShMZRb_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SyqShMZRb/SyqShMZRb_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lfF2NYvH/r1lfF2NYvH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lfF2NYvH/r1lfF2NYvH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lfF2NYvH/r1lfF2NYvH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0z1HScLBEpb/0z1HScLBEpb_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0z1HScLBEpb/0z1HScLBEpb_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0z1HScLBEpb/0z1HScLBEpb_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkex2a4FPr/Hkex2a4FPr_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkex2a4FPr/Hkex2a4FPr_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Hkex2a4FPr/Hkex2a4FPr_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gmrxHFvB/S1gmrxHFvB_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gmrxHFvB/S1gmrxHFvB_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1gmrxHFvB/S1gmrxHFvB_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlzm64tDH/BJlzm64tDH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlzm64tDH/BJlzm64tDH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJlzm64tDH/BJlzm64tDH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/mSAKhLYLSsl/mSAKhLYLSsl_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/hQGRD1Zael7/hQGRD1Zael7_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/hQGRD1Zael7/hQGRD1Zael7_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/WWRBHhH158K/WWRBHhH158K_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/WWRBHhH158K/WWRBHhH158K_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/WWRBHhH158K/WWRBHhH158K_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/5CGPY2VeEGb/5CGPY2VeEGb_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/5CGPY2VeEGb/5CGPY2VeEGb_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/5CGPY2VeEGb/5CGPY2VeEGb_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/xWq1MVj7YrE/xWq1MVj7YrE_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/xWq1MVj7YrE/xWq1MVj7YrE_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/xWq1MVj7YrE/xWq1MVj7YrE_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/EbIDjBynYJ8/EbIDjBynYJ8_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/EbIDjBynYJ8/EbIDjBynYJ8_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/EbIDjBynYJ8/EbIDjBynYJ8_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1AG8zYeg/S1AG8zYeg_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1AG8zYeg/S1AG8zYeg_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1AG8zYeg/S1AG8zYeg_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S18Su--CW/S18Su--CW_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S18Su--CW/S18Su--CW_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S18Su--CW/S18Su--CW_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1nmx5l0W/r1nmx5l0W_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1nmx5l0W/r1nmx5l0W_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1nmx5l0W/r1nmx5l0W_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJg7x1HFvB/BJg7x1HFvB_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJg7x1HFvB/BJg7x1HFvB_span.pdf filter=lfs diff=lfs merge=lfs -text diff --git a/parse/train/0z1HScLBEpb/0z1HScLBEpb_layout.pdf b/parse/train/0z1HScLBEpb/0z1HScLBEpb_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7da932b8e621af986147b71fbe6bb9a5a23ff9fc --- /dev/null +++ b/parse/train/0z1HScLBEpb/0z1HScLBEpb_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffd52b98c1013bcf5559ad3345f07584838e2837a51c829fd663486bdcdad426 +size 3929436 diff --git a/parse/train/0z1HScLBEpb/0z1HScLBEpb_origin.pdf b/parse/train/0z1HScLBEpb/0z1HScLBEpb_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c77879fb70e38ba2f5f2ce42204949986eea440d --- /dev/null +++ b/parse/train/0z1HScLBEpb/0z1HScLBEpb_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09db2bbdaebeafc084fafb0a4baaab0a995d2435785a1a73ba828e81a0fbebcb +size 3734672 diff --git a/parse/train/0z1HScLBEpb/0z1HScLBEpb_span.pdf b/parse/train/0z1HScLBEpb/0z1HScLBEpb_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..891af8668229fc8e095227e9e8ecd65d6999d974 --- /dev/null +++ b/parse/train/0z1HScLBEpb/0z1HScLBEpb_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60e96013f07b9a7b8e3d087e53a1b1d909fee2ce0c1fbed88166eec0d06692a3 +size 3944673 diff --git a/parse/train/2zCRcTafea/2zCRcTafea.md b/parse/train/2zCRcTafea/2zCRcTafea.md new file mode 100644 index 0000000000000000000000000000000000000000..07156364eb92e8ca2797bbe9e08705a2132b5445 --- /dev/null +++ b/parse/train/2zCRcTafea/2zCRcTafea.md @@ -0,0 +1,267 @@ +# Focal Attention for Long-Range Interactions in Vision Transformers + +Jianwei Yang1 Chunyuan $\mathbf { L i } ^ { 1 }$ Pengchuan Zhang1 Xiyang Dai2 Bin Xiao2 Lu Yuan2 Jianfeng Gao1 1Microsoft Research at Redmond, 2Microsoft Cloud $^ +$ AI +{jianwyan,chunyl,penzhan,xidai,bixi,luyuan,jfgao}@microsoft.com + +# Abstract + +Recently, Vision Transformer and its variants have shown great promise on various computer vision tasks. The ability of capturing local and global visual dependencies through self-attention is the key to its success. However, this also brings challenges due to quadratic computational overhead, especially for the high-resolution vision tasks (e.g., object detection). Many recent works have attempted to reduce the cost and improve model performance by applying either coarse-grained global attention or fine-grained local attention. However, both approaches cripple the modeling power of the original self-attention mechanism of multi-layer Transformers, leading to sub-optimal solutions. In this paper, we present focal attention, a new attention mechanism that incorporates both fine-grained local and coarse-grained global interactions. In this new mechanism, each token attends its closest surrounding tokens at fine granularity and the tokens far away at coarse granularity, and thus can capture both short- and long-range visual dependencies efficiently and effectively. With focal attention, we build a new variant of Vision Transformer models, called Focal Transformers, which achieve superior performance over the state-of-theart (SoTA) Vision Transformers on a range of public image classification and object detection benchmarks. In particular, our Focal Transformer models with a moderate size of 51.1M and a large size of $8 9 . 8 \mathbf { M }$ achieve $\mathbf { 8 3 . 6 \% }$ and $\mathbf { 8 4 . 0 \% }$ Top-1 accuracy, respectively, on ImageNet classification at $2 2 4 \times 2 2 4$ . When employed as the backbones, Focal Transformers achieve consistent and substantial improvements over the current SoTA Swin Transformers [43] across 6 different object detection methods. Our largest Focal Transformer yields 58.7/59.0 box mAPs and 50.9/51.3 mask mAPs on COCO mini-val/test-dev, and 55.4 mIoU on ADE20K for semantic segmentation, creating new SoTA on three of the most challenging computer vision tasks. Our code is available at: https://github. com/microsoft/Focal-Transformer. + +# 1 Introduction + +Nowadays, Transformer [57] has become a prevalent model architecture in natural language processing (NLP) [20, 6]. In the light of its success in NLP, there is an increasing effort on adapting it to computer vision (CV) [47, 50]. Since its promise firstly demonstrated in Vision Transformer (ViT) [21], we have witnessed a flourish of full-Transformer models for image classification [55, 60, 64, 43, 76, 56], object detection [8, 85, 79, 18] and semantic segmentation [58, 62]. Beyond these static image tasks, it has also been applied on various temporal understanding tasks, such as action recognition [40, 78, 10], object tracking [13, 59], scene flow estimation [38]. + +The self-attention mechanism is arguably the key component that differentiates Transformers from the widely used convolutional neural networks (CNNs) [37] in computer vision. At each Transformer layer, self-attention enables global content-dependent interactions among different image regions for modeling short- and long-range dependencies, respectively. Through the visualization of full selfattention results1, we indeed observe that self-attention learns to attend local surroundings (like CNNs) and the global contexts at the same time, as illustrated in Fig. 1 (Left). Nevertheless, when dealing with high-resolution vision tasks such as object detection or segmentation, an efficient implementation of a global and fine-grained self-attention becomes non-trivial due to the quadratic computational cost with respect to the number of tokens in feature maps. Recent works have alternatively exploited either a coarse-grained global self-attention [60, 64] or a fine-grained local self-attention [43, 76, 56], for the sake of reducing the computational cost. However, both approaches cripple the power of the original full self-attention i.e., the ability to simultaneously capture local and global visual dependencies. + +![](images/621f200345a7ebe619be7e8a0aedd96409fbf6fca08314c5e32fce4ac604fb01.jpg) +Figure 1: Left: Visualization of the attention maps of the three heads at the given query patch (blue) in the first layer of the DeiT-Tiny model [55]. Right: An illustrative depiction of focal attention mechanism. Three granularity levels are used to compose the attention region for the blue query. + +In this paper, we present a new attention mechanism to capture both short- and long-range interactions in Transformer layers for high-resolution input images. Considering that the visual dependencies between the nearby (local) regions are usually much stronger than the dependencies between the regions that are far away, we perform the fine-grained attention only in local regions while the coarse-grained attention globally. As depicted in Fig. 1 (Right), a query token in the feature map attends its closest local surroundings at the finest granularity as itself. However, when it goes to the regions far away, it attends to summarized tokens to capture coarse-grained visual dependencies. We call this new mechanism focal attention, as each token attends the others in a focal manner. We will show in this study that focal attention allows to effectively model visual dependencies among all regions covering the whole high-resolution feature maps while introducing much less number of tokens in the computation than that in the standard self-attention mechanism. + +Equipped with focal attention, a series of Focal Transformers are developed and validated via a comprehensive empirical study across three core vision tasks, including image classification, object detection and segmentation. Results show that Focal Transformers consistently outperform the SoTA Vision Transformers across various settings (i.e., in model sizes and complexities). Notably, the small Focal Transformer with 51.1M parameters achieves $8 3 . 6 \%$ top-1 accuracy on ImageNet-1K, and the base model with 89.8M parameters obtains $8 4 . 0 \%$ top-1 accuracy. In the fine-tuning experiments for object detection, Focal Transformers consistently outperform the SoTA Swin Transformers [43] across six popular object detection methods. Our largest Focal Transformer model achieves 59.0 box mAP and 51.3 mask mAP on COCO test-dev for object detection and instance segmentation, respectively, and 55.4 mIoU on ADE20K for semantic segmentation. These results demonstrate that focal attention is highly effective in modeling the global interactions in Vision Transformers. + +# 2 Related work + +Vision Transformers. Vision Transformer (ViT) is first introduced in [21]. It applies a standard Transformer, originally developed for NLP [57], to encode an image by analogously splitting the image into a sequence of visual tokens. It has demonstrated superior performance to CNNs such as ResNet [33] on multiple image classification benchmarks, when trained with sufficient data [21] and carefully designed data augmentation and regularization methods [55]. The results thus inspire researchers to explore the applications of ViT on various vision tasks beyond image classification, such as self-supervised learning [14, 9, 39], object detection [8, 85, 79, 18] and semantic segmentation [58, 62, 81]. There are also increasing number of studies for improving ViT via data-efficient training [55], improved patch embedding/encoding [16, 71, 31], integrating convolutional projections into transformers [64, 70], and using multi-scale architectures and efficient self-attention mechanisms for high-resolution vision tasks [60, 64, 43, 76, 15]. Recent surveys include [36, 30, 36]. This paper focuses on improving the self-attention mechanism of ViT for encoding high-resolution images. + +![](images/895ff52491908f44661414bd57eca7fc36f34ebd4f7c17cd1d28f3b83da587af.jpg) +Figure 2: Model architecture for our Focal Transformers. As highlighted in light blue boxes, our main innovation is the proposed focal attention in each Transformer layer. + +Efficient global and local self-attention. In many real-world tasks, Transformers need to cope with a large number of input tokens, such as long documents in NLP and high-resolution images in computer vision (CV). Recently, many efficient self-attention mechanisms have been proposed to deal with the quadratic computational and memory cost incurred by the standard self-attention mechanism. On one hand, a number of works in both NLP and CV resort to coarse-grained global self-attention (i.e., attending the down-sampled or summarized tokens) to capture the long-range interactions [49, 46, 60, 64, 31, 23]. Although this approach improves the model efficiency, it loses the detailed context information surrounding the query tokens. On the other hand, to make the computational cost manageable, various local fine-grained attention mechanism (i.e., attending neighboring tokens within a pre-set window size) are used for both NLP [3, 74, 1] and CV [56, 43, 76]. In this paper, we argue that both global and local attentions are important for model performance. This is also validated by some recent studies that aim to improve CNNs by incorporating ways of modeling global attentions [35, 63, 61, 68, 2, 7, 51]. The standard self-attention mechanism used by ViT can indeed learned both types of attentions, as shown in Fig. 1 (Left). But it often incurs a prohibitively high cost for high-resolution images. To the best of our knowledge, the proposed focal attention provides the first mechanism to incorporate local and global attention in a single Transformer layer 2. It can capture both short- and long-range interactions as standard self-attention but in a much more efficient and effective way, especially for high-resolution images. + +# 3 Method + +# 3.1 Model architecture + +To accommodate high-resolution dense prediction tasks, we employ a multi-scale model architecture as in [60, 76, 43]. As shown in Fig. 2, an image $I \in \mathcal { R } ^ { H \times W \times 3 }$ is first partitioned into patches of size $4 \times 4$ , resulting in ${ \frac { H } { 4 } } \times { \frac { W } { 4 } }$ visual tokens with dimension $4 \times 4 \times 3$ . Then, we use a patch embedding layer, consisting of a convolutional layer with filter size and stride both equal to 4, to project these patches into hidden features with dimension $d$ . We then pass this spatial feature map to the four stages of Focal Transformer blocks. In each stage $i \in \{ 1 , 2 , 3 , 4 \}$ , the Focal Transformer block consists of $N _ { i }$ Focal Transformer layers. After each stage, we use a patch embedding layer to reduce the spatial size of feature map by factor 2 and increase the feature dimension by 2. For image classification tasks, we take the average of the output from the last stage and send it to a classification layer. For object detection, the feature maps from the last 3 or all 4 stages are fed to a particular object detector head, depending on the specific detection method we choose to use. The model capacity can be customized by varying the input feature dimension $d$ and the number of Focal Transformer layers. + +Standard self-attention can capture both short- and long-range interactions at fine-grain, but suffers from high computational cost when it performs attention on high-resolution feature maps as noted in [76]. Take stage 1 in Fig. 2 as an example. For a feature map of size ${ \frac { H } { 4 } } \times { \frac { W } { 4 } } \times d$ , the complexity of self-attention is $\begin{array} { r } { \mathcal { O } ( ( \frac { H } { 4 } \times \frac { W } { 4 } ) ^ { 2 } d ) } \end{array}$ , resulting in an explosion of time and memory cost, considering that $\operatorname* { m i n } ( H , W )$ could be 800 or even larger for object detection. In the next section, we describe how we address this issue with the proposed focal attention mechanism. + +# 3.2 Token-wise focal attention + +Focal attention is proposed to make the Transformer layers suitable for encoding high-resolution input images. Instead of attending all tokens at fine-grain, we attend the fine-grain tokens only locally, but the summarized ones (i.e., the coarse-grained tokens generated by sub-window pooling, which is illustrated in Fig. 4 and will be described later) globally. As such, focal attention can cover the same amount of image regions as standard self-attention but with much less cost. In Fig. 3, we show the size of the receptive field for standard self-attention and our focal attention as a function of the number of attended tokens. For a given query position, by reducing the granularity of its surroundings based on their distance to the query, focal attention can have significantly larger receptive fields at the same cost measured by the number of visual tokens, compared to the standard self-attention mechanism. + +![](images/bf6a6de3b259a05fb0ac374d04a980cb6b72063e837f8ca7329f069632ea03a7.jpg) +Figure 3: The size of receptive field (yaxis) as a function of the number of used visual tokens $\mathbf { \bar { x } }$ -axis) in regular (standard) self-attention and focal attention. When plotting the curve for focal attention, we increase the focal window size by 2 for each focal level up to the maximal window size of 8. + +Theoretically, the focal attention mechanism enables global interaction with much less time and memory cost, because it attends a much smaller number of surrounding (summarized) tokens. In practice, however, extracting the surrounding tokens for each query position could incur high time cost since we need to duplicate the extraction of each token for all queries that the token surrounds. This issue had been extensively discussed in [56, 76, 43] and a common solution is to partition the input feature map into windows. Thus, in our Focal Transformers, we resort to performing focal attention at the window level. We elaborate the window-wise focal attention in the following. + +# 3.2.1 Window-wise focal attention + +Given a feature map of $\boldsymbol { x } \in \mathcal { R } ^ { M \times N \times d }$ with spatial size $M \times N$ , we first partition it into a grid of windows of size $s _ { p } \times s _ { p }$ . Then, we extract the surroundings for each window rather than each individual token. The proposed window-wise focal attention is illustrated in Fig. 4. To clarify, we first define three terms: + +• Focal level $L$ refers to the granularity level at which we extract the tokens for focal attention. +• Focal window size $s _ { w } ^ { l }$ is the size of sub-window on which the summarized tokens are formed via +sub-window pooling at granularity level of $l \in \{ 1 , . . . , L \}$ . +• Focal region size $s _ { r } ^ { l }$ denotes the number of sub-windows that are filled up horizontally (or vertically) in an attended region at level $l$ . + +Now, we detail how window-wise focal attention works in the following two steps, sub-window pooling and attention computing. + +Sub-window pooling. Consider input feature map $\boldsymbol { x } \in \mathcal { R } ^ { M \times N \times d }$ , where $M \times N$ is the spatial dimension and $d$ the feature dimension. We perform sub-window pooling for all $L$ levels. At focal level $l$ , we first split the input feature map $x$ into a grid of sub-windows with size $s _ { w } ^ { l } \times s _ { w } ^ { l }$ . Then we use a linear projection layer $f _ { p } ^ { l }$ to pool the sub-windows spatially by + +$$ +\begin{array} { r } { x ^ { l } = f _ { p } ^ { l } ( \hat { x } ) \in \mathcal { R } ^ { \frac { M } { s _ { w } ^ { l } } \times \frac { N } { s _ { w } ^ { l } } \times d } , \quad \hat { x } = \mathrm { R e s h a p e } ( x ) \in \mathcal { R } ^ { ( \frac { M } { s _ { w } ^ { l } } \times \frac { N } { s _ { w } ^ { l } } \times d ) \times ( s _ { w } ^ { l } \times s _ { w } ^ { l } ) } . } \end{array} +$$ + +The pooled feature maps $\{ x ^ { l } \} _ { 1 } ^ { L }$ at different levels $l$ provide rich information at both fine-grain and coarse-grain. Since we set $s _ { w } ^ { l } = 1$ for the first focal level which has the same granularity as the input + +![](images/8d7c5bc6d5803f11463c3ab8656b31fec8dbb82ddb891bec59961a9ad98dfe04.jpg) +Figure 4: An illustration of focal attention at window level. Each of the square cells represents a visual token that is either from the original feature map or a summarized token formed by sub-window pooling. Suppose we have an input feature map of size $2 0 \times 2 0$ . We first partition it into $5 \times 5$ windows of size $4 \times 4$ . Take the $4 \times 4$ blue window in the middle as the query set, we extract its surrounding tokens at three granularity levels as its keys and values. For the first level, we extract the $8 \times 8$ tokens which are closest to the blue window at the finest grain. At the second level, we expand the attention region and pool the surrounding $2 \times 2$ sub-windows to form summarized tokens, which results in $6 \times 6$ summarized tokens. At the third level, we attend a larger region covering the whole feature map and pool $4 \times 4$ sub-windows, which leads to $5 \times 5$ summarized tokens. Finally, these three levels of tokens are concatenated to compute the keys and values for the $4 \times 4 = 1 6$ tokens (queries) in the blue window. + +feature map, there is no need to perform any sub-window pooling. Considering that the focal window size is usually very small (7 maximally in our settings), the number of extra parameters introduced by sub-window pooling is negligible. + +Attention computing. Once we obtain the pooled feature maps $\{ x ^ { l } \} _ { 1 } ^ { L }$ at all $L$ levels, we compute the query at the first level, and key and value for all levels using three linear projection layers $f _ { q } , f _ { k }$ and $f _ { v }$ , respectively, as + +$$ +Q = f _ { q } ( x ^ { 1 } ) , \quad K = \{ K ^ { l } \} _ { 1 } ^ { L } = f _ { k } ( \{ x ^ { 1 } , . . . , x ^ { L } \} ) , \quad V = \{ V ^ { l } \} _ { 1 } ^ { L } = f _ { v } ( \{ x ^ { 1 } , . . . , x ^ { L } \} ) . +$$ + +To perform focal attention, we need to first extract the surrounding tokens for each query token in the feature map. As mentioned earlier, tokens inside a window partition $s _ { p } \times s _ { p }$ share the same set of surroundings. For the queries inside the $i$ -th window $Q _ { i } \in \mathcal { R } ^ { s _ { p } \times s _ { p } \times d }$ , we extract the $s _ { r } ^ { l } \times s _ { r } ^ { l }$ keys and values from $K ^ { l }$ and $V ^ { l }$ surrounding the window which the query lies in, and then gather the keys and values from all $L$ levels to obtain $\breve { K } _ { i } = \{ K _ { i } ^ { 1 } , . . . , K _ { i } ^ { L } \} \in \mathscr { R } ^ { \bar { s } \times d }$ and $V _ { i } = \{ V _ { i } ^ { 1 } , . . . , \mathbf { \bar { V } } _ { i } ^ { L } \} \in \mathcal { R } ^ { s \times d }$ , where imple $s$ is the sum of focal regions from all levels, i.e., ntation of focal attention following Fig. 1 requires $\begin{array} { r } { s = \sum _ { l = 1 } ^ { L } ( s _ { r } ^ { l } ) ^ { 2 } } \end{array}$ . Note that a canonicalverlapped regions across different levels. In our implementation, we intentionally keep them in order to capture the pyramid information for the overlapped regions. Finally, we follow [43] to include a relative position bias and compute the focal attention for $Q _ { i }$ by + +$$ +\mathrm { A t t e n t i o n } ( Q _ { i } , K _ { i } , V _ { i } ) = \mathrm { S o f t m a x } ( \frac { Q _ { i } K _ { i } ^ { T } } { \sqrt { d } } + B ) V _ { i } , +$$ + +where $B = \{ B ^ { l } \} _ { 1 } ^ { L }$ is the learnable relative position bias. It consists of $L$ subsets for $L$ focal levels. Similar to [43], for the first level, we parameterize it as $B ^ { 1 } \in \mathcal { R } ^ { ( 2 s _ { p } - 1 ) \times ( 2 s _ { p } - 1 ) }$ , considering that the horizontal and vertical position ranges are both in $[ - s _ { p } + 1 , s _ { p } - 1 ]$ . For the other focal levels, considering that they have different granularity with respect to the queries, we treat all the queries inside a window equally and use $B ^ { l } \in \mathcal { R } ^ { s _ { r } ^ { l } \times s _ { r } ^ { l } }$ to represent the relative position bias between the query window and each of $s _ { r } ^ { l } \times s _ { r } ^ { l }$ summarized tokens. Since the focal attention for each window can be performed independent of the others, we can compute Eq. (3) in parallel. Once we obtain attention scores for the whole input feature map, we send them to LayerNorm and the MLP block. + +
Output SizeLayer NameFocal-TinyFocal-SmallFocal-Base
stage 156×56Patch Embeddingp1= 4;c1 = 96p1=4;c1= 96p1 = 4;c1 = 128
56×56Transformer Block{1,13} 三 s={7,7×2二 {1,13} ={7,7}×2{1,13} ={7,7}×2
stage 228×28Patch EmbeddingP2=2;c=192P2=2;C=192P2=2;C= 256
28×28Transformer Block{1,13} 三 swr={7,5} 1×2={1,13} sw,r={7,5} 1×2{1,13} 二 su,r={7,5}×2
stage 314 × 14Patch Embeddingp3=2;c3=384p3=2; c3= 384p3=2; c3= 512
14 × 14Transformer Block={1,13} s={7,3}×6={1,13} ={7,3}×18={1,13} ={7,3}×18
stage 47×7Patch EmbeddingP4=2;C4=768P4=2;C4=768P4= 2;C4=1024
7×7Transformer Block二 {1,7}×2{1,7} s 三 ={7,1}×2二 {1,7} su,r {7,1} 二×2
+ +Table 1: Model configurations for Focal Transformers. We use three configurations with different model capacities: Focal-Tiny, Focal-Small and Focal-Base. + +# 3.2.2 Complexity analysis + +We analyze the computational complexity for the two steps of focal attention described above. For the input feature map $\dot { \boldsymbol { x } } \in \mathcal { R } ^ { M \times N \times d }$ , we have $\begin{array} { r } { \frac { M } { s _ { w } ^ { l } } \times \frac { N } { s _ { w } ^ { l } } } \end{array}$ sub-windows at focal level l. For each sub-window, the pooling operation in Eq.1 has the complexity of $\mathcal { O } ( ( s _ { w } ^ { l } ) ^ { 2 } d )$ . Aggregating all sub-windows brings us $O ( ( M N ) d )$ . Then for all focal levels, we have the complexity of $\mathcal { O } ( L ( M N ) d )$ in total, which is independent of the sub-window size at each focal level. Regarding the attention computation in Eq. 3, the computational cost for a query window $s _ { p } \times s _ { p }$ is $\mathcal { O } ( ( s _ { p } ) ^ { 2 } \textstyle \sum _ { l } ( s _ { r } ^ { l } ) ^ { 2 } d )$ , and $\mathcal { O } ( \dot { \sum } _ { l } ( s _ { r } ^ { l } ) ^ { 2 } ( M \dot { N } ) d )$ for the whole input feature map. To sum up, the overall computational cost for focal attention is $\begin{array} { r } { \mathcal { O } ( ( L + \sum _ { l } ( s _ { r } ^ { l } ) ^ { \bar { 2 } } ) ( M N ) d ) } \end{array}$ . In an extreme case, one can set $s _ { r } ^ { \hat { L } } = 2 \times \operatorname* { m a x } ( M , N ) / s _ { w } ^ { L }$ to ensure a global receptive field for all queries (including both corner and middle queries) in this layer. + +# 3.3 Model configurations + +For fair comparison, we consider three network configurations for Focal Transformers, following [60, 64, 43]. Specifically, we follow the design of the Tiny, Small and Base models in Swin Transformer [43], as shown in Table 1. Our models take $2 2 4 \times 2 2 4$ images as inputs and the window partition size is set to 7 to make our models comparable to Swin Transformers. For the focal attention layer, we introduce two levels, one for fine-grained local attention and the other for coarse-grained global attention. Except for the last stage, the focal region size is set to 13 for the window partition size of 7, which means that we expand 3 tokens for each window partition. For the last stage, since the whole feature map is $7 \times 7$ , the focal region size at level 0 is set to 7, which is sufficient to cover the entire feature map. For the coarse-grained global attention, we set its focal window size the same as the window partition size 7, but gradually decrease the focal region size to get $\{ 7 , 5 , 3 , 1 \}$ for the four stages, respectively. For the patch embedding layer, the spatial reduction ratio $p _ { i }$ for the four stages are all $\{ 4 , 2 , 2 , 2 \}$ . Note that Focal-Base has a higher hidden dimension $c _ { i }$ , compared to Focal-Tiny and Focal-Small. + +# 4 Experiments + +# 4.1 Image classification on ImageNet-1K + +We compare different methods on ImageNet-1K [19]. For fair comparison, we follow the training recipes in [55, 60]. All models are trained for 300 epochs with batch size 1024. The initial learning rate is set to $1 0 ^ { - 3 }$ with 20 epochs of linear warm-up starting from $1 0 ^ { - 5 }$ . For optimization, we use AdamW [44] as the optimizer with a cosine learning rate scheduler. The weight decay is set to 0.05 and the maximal gradient norm is clipped to 5.0. We use the same set of data augmentation and regularization strategies used in [55] after excluding random erasing [82], repeated augmentation [4, 34] and exponential moving average (EMA) [48]. The stochastic depth drop rates are set to 0.2, 0.2 and 0.3 for our tiny, small and base models, respectively. During training, we crop images randomly to $2 2 4 \times 2 2 4$ , while a center crop is used during evaluation on the validation set. + +
Model#Params. FLOPsTop-1 (%)
ResNet-50 [33]25.0 4.176.2
DeiT-Small/16 [55]22.1 4.679.9
PVT-Small [60]24.5 3.879.8
ViL-Small [76]24.6 5.182.0
CvT-13 [64]20.0 4.581.6
Swin-Tiny [43]28.3 4.581.2
Focal-Tiny (Ours)28.9 4.982.2
ResNet-101[33]45.0 7.977.4
PVT-Medium [60]44.2 6.781.2
CvT-21 [64]32.0 7.182.5
ViL-Medium [76]39.7 9.183.3
Swin-Small [43]49.6 8.783.1
Focal-Small (Ours)51.1 9.483.6
ResNet-152[33]60.0 11.078.3
ViT-Base/16 [21]86.6 17.677.9
DeiT-Base/16 [55]17.581.8
86.6
PVT-Large [60]61.4 9.881.7
ViL-Base[76]55.7 13.483.2
Swin-Base [43]87.8 15.483.4
Focal-Base (Ours)89.816.4 84.0
+ +Table 2: Comparison of image classification on ImageNet-1K for different models. Except for ViT-Base/16, all other models are trained and evaluated on $2 2 4 \times 2 2 4$ resolution. + +Table 3: Comparisons with CNN and Transformer baselines and SoTA methods on COCO object detection. The box mAP $( A P ^ { b } )$ and mask mAP $( A P ^ { m } )$ are reported for RetinaNet and Mask R-CNN trained with $1 \times$ schedule. More detailed comparisons with $3 \times$ schedule are in Table 4. + +
BackboneRetinaNetMask R-CNN
APbApbAPm
ResNet-50 [33]36.338.034.4
PVT-Small40.440.437.8
ViL-Small [76]41.641.838.5
Swin-Tiny [43]42.043.739.8
Focal-Tiny (Ours)43.7 (+1.7)44.8 (+1.1) 41.0 (+1.3)
ResNet-101[33]38.540.436.4
ResNeXt101-32x4d [67]39.941.937.5
PVT-Medium [60]41.942.039.0
ViL-Medium [76]42.943.439.7
Swin-Small [43]45.046.542.1
Focal-Small (Ours)45.6 (+0.6)47.4 (+0.9) 42.8 (+0.7)
ResNeXt101-64x4d[67] 41.042.838.4
PVT-Large [60]42.642.939.5
ViL-Base[76]44.345.141.0
Swin-Base [43]45.046.942.3
Focal-Base (Ours)46.3 (+1.3)47.8 (+0.9)43.2 (+0.9)
+ +In Table 2, we summarize the results for baseline models and the state-of-the-art models on image classification task. We can see that Focal Transformers consistently outperform other methods with similar model sizes (#Params.) and computational complexities (GFLOPs). Specifically, Focal-Tiny improves over the Transformer baseline DeiT-Small/16 by $2 . 3 \%$ . Meanwhile, using the same model configuration (2-2-6-2) and a few extra parameters and computations, Focal-Tiny improves over Swin-Tiny by 1.0 point. For small and base models, Focal-Small with 51.1M parameters can reach $8 3 . 6 \%$ which is better than all the counterpart small and base models using much less parameters. By increasing the model size, Focal-Base model achieves $8 4 . 0 \%$ , surpassing all the other models with comparable parameters and FLOPs. + +To compare with the large-scale models, we further build Focal-Large Transformer by increasing the hidden dimension in Focal-Base from 128 to 196 while keeping all the other hyperparameters the same. We follow the common practice to pretrain our Focal-Large Transformer on ImageNet-22K and transfer it to detection and segmentation tasks [64, 43]. + +# 4.2 Object detection and instance segmentation + +We benchmark our models on object detection with COCO 2017 [42]. The pretrained models are used as visual backbones and then plugged into two representative pipelines, RetinaNet [41] and Mask R-CNN [32]. All models are trained on the $1 1 8 \mathrm { k }$ training images and the results are reported on 5K validation set. We use the two standard training schedules, $1 \times$ with 12 epochs and $3 \times$ with 36 epochs. For the $1 \times$ schedule, we resize image’s shorter side to 800 while keeping its longer side no more than 1,333. For the $3 \times$ schedule, we use the multi-scale training strategy by randomly resizing its shorter side to the range of [480, 800]. Considering this higher input resolution, we adaptively increase the focal sizes at four stages to (15, 13, 9, 7), to ensures that the focal attention covers more than half of the image region at the first two stages, and the whole image at the last two stages. With the focal size increased, the relative position biases are accordingly up-sampled to the corresponding sizes using bilinear interpolation. During training, we use AdamW [44] for optimization with initial learning rate $1 0 ^ { - 4 }$ and weight decay 0.05. Similarly, we use 0.2, 0.3 and 0.5 stochastic depth drop rates to regularize the training for our Tiny, Small and Base models, respectively. Since Swin Transformer does not report the results on RetinaNet, we obtain the results by ourselves using their official code with the same hyper-parameters as that of Focal Transformers. + +In Table 3, we show the performance for both CNN-based models and the current Transformerbased state-of-the-art models. The bbox mAP $( A P ^ { b } )$ and mask mAP $( A P ^ { m } )$ are reported. We see that Focal Transformers outperform the CNN-based models consistently with the gap of 4.8-7.1 points. Compared with the other methods which also use multi-scale Transformer architectures, + +Table 4: COCO object detection and segmentation results with RetinaNet [41] and Mask R-CNN [33]. All models are trained with $3 \times$ schedule and multi-scale inputs (MS). The numbers before and after “/” at column 2 and 3 are the model size and complexity for RetinaNet and Mask R-CNN, respectively. + +
Backbone#Params (M)FLOPs (G)RetinaNet 3x schedule + MSMask R-CNN 3x schedule + MS
APAP5APAPsAPMAPLAP6APAPApmAPAP
ResNet50 [33]37.7/44.2239/26039.058.441.822.442.851.641.061.744.937.158.440.1
PVT-Small[60]34.2/44.1226/24542.262.745.026.245.257.243.065.346.939.962.542.8
ViL-Small [76]35.7/45.0252/174 42.963.845.627.846.456.343.464.947.039.662.142.4
Swin-Tiny [43]38.5/47.8245/264 45.065.948.429.748.958.146.068.150.341.665.144.9
Focal-Tiny (Ours)39.4/48.8265/291 45.566.348.831.249.258.747.269.451.942.766.5 45.9
ResNet101 [33]56.7/63.2315/33640.960.144.023.745.053.842.863.247.138.560.141.3
ResNeXt101-32x4d [67]56.4/62.8319/34041.461.044.323.945.553.744.064.448.039.261.441.9
PVT-Medium [60]53.9/63.9283/30243.263.846.127.346.358.944.266.048.240.563.143.5
ViL-Medium [76]50.8/60.1339/26143.764.646.427.947.156.944.666.348.540.763.843.7
Swin-Small [43]59.8/69.1335/354 46.467.050.131.050.160.348.570.253.543.367.346.6
Focal-Small (Ours)61.7/71.2367/40147.367.851.031.650.961.148.870.553.643.867.747.2
ResNeXt101-64x4d [67]95.5/102473/49341.861.544.425.245.454.644.464.948.839.761.942.6
PVT-Large[60]71.1/81.0345/364 43.463.646.126.146.059.544.566.048.340.763.443.7
ViL-Base [76]66.7/76.1443/365 44.765.547.629.948.058.145.767.249.941.364.444.5
Swin-Base 43]98.4/107477/496 45.866.449.129.949.460.348.569.853.243.466.846.9
Focal-Base (Ours)100.8/110.0 514/533 46.967.850.331.950.361.549.070.153.643.767.647.0
+ +Focal Transformers show substantial gains across all settings and metrics. Particularly, Focal Transformers brings 0.7-1.7 points of mAP against the current best approach Swin Transformer [43] at comparable settings. Different from the other multi-scale Transformer models, Focal Transformers can simultaneously enable short-range fine-grain and long-range coarse-grain interactions for each visual token, and thus capture richer visual contexts at each layer for better dense predictions. To have more comprehensive comparisons, we train all models using the $3 \times$ schedule and show the detailed numbers for RetinaNet and Mask R-CNN in Table 4. As we can see, even with the $3 \times$ schedule, Focal Transformers can still achieve 0.3-1.1 gain over Swin Transformer models in comparable settings. + +Comparison with large SoTA detection models. We follow Swin Transformers to use HTC [11] as the detection method in that it reported SoTA performance on COCO detection when using Swin Transformer as the backbone. For fair comparison, we also use soft-NMS [5], instaboost [24] and a multi-scale training strategy with the shorter side in range [400, 1400] and the longer side no more than 1600. We train the model using AdamW [44] with base learning rate 1e-4 and weight decay 0.1. The model is trained using the standard $3 \times$ schedule. The box and mask mAPs on COCO validation set and test-dev are reported in Table 5, where both single-scale evaluation and multi-scale evaluation results are presented. Our Focal-Large model with multi-scale test achieves 58.1 box mAP and 50.9 mask mAP on mini-val set, which is better than the reported numbers for Swin-Large in [43]. When evaluating our model on the test-dev set, it achieves 58.4 box mAP and 51.3 mask mAP, which is slightly better than Swin Transformer. Note that because our model does not include the global self-attention layer used in Swin Transformer at the last stage, it has a smaller model size and fewer FLOPs. More recently, DyHead [17] achieves new SoTA on COCO, when combined with Swin-Large. We replace the Swin-Large model with the Focal-Large model, and use the same $2 \times$ training schedule as in [17]. We report the box mAPs for both mini-val and test-dev. Focal-Large achieves 58.7 and 59.0 on mini-val and test-dev, respectively. + +# 4.3 Semantic Segmentation + +In addition to the instance segmentation results, we also evaluate our models on the semantic segmentation task which usually takes high-resolution input images and requires capturing long-range interactions. We benchmark our methods on ADE20K [83]. We use UperNet [65] as the segmentation method and Focal Transformers as the backbones. We train three models as Focal-Tiny, Focal-Small, Focal-Base, respectively. For all the models, we use a standard recipe that sets the input size to $5 1 2 \times 5 1 2$ and trains the model for 160k iterations with batch size 16. Table 6 shows the comparison results. We see that Focal-Tiny, Focal-Small and Focal-Base models consistently outperform Swin Transformers of the similar size in single-scale and multi-scale mIoUs. + +Comparison with large SoTA semantic segmentation models. We use the pretrained Focal-Large model as the backbone for semantic segmentation. Follow the setting in [43], we use input image size $6 4 0 \times 6 4 0$ and train the model for 160k iterations with a batch size of 16. We set the initial learning to 6e-5 and use a polynomial learning rate decay. The weight decay is set to 0.01. For + +
Method#Param FLOPsmini-valtest-dev
ApbAPmApbApm
X101-64x4d [67]155M1033G 52.346.0
EfficientNet-D7 [54]77M410G54.4-55.1=
GCNet*[7]-1041G51.844.752.345.4
ResNeSt-200 [75]=52.5=53.347.1
Copy-paste [28]185M1440G 55.947.256.047.4
BoTNet-200 [51]-49.7-
SpineNet-190 [22]164M1885G 52.652.8
CenterNet2 [84]-=--56.4=
Swin-L (HTC++) [43]284M1470G 57.149.557.750.2
Swin-L (DyHead)[17]213M965G56.2---
Swin-L† (HTC++) [43]284M58.050.458.751.1
Swin-L† (DyHead) [17]213M58.4-58.7
Swin-L† (QueryInst) [25]-56.1156.1
Focal-L (HTC++) (Ours)265M1165G57.049.9-=
Focal-L (DyHead) (Ours)229M1081G56.4--=
Focal-L† (HTC++) (Ours)265M-58.150.958.451.3
Focal-L† (DyHead) (Ours)229M-58.7-59.0-
+ +Table 5: Comparison with state-of-the-art methods on COCO object detection and instance segmentation. The numbers are reported on 5K val set and test-dev. Augmented HTC [11] (denoted by $\mathrm { H T C + + }$ ) and DyHead [17] are used as the detection methods. † means multi-scale evaluation. + +
BackboneMethod#Param FLOPs mIoU +MS
ResNet-101DANet [45]69M1119G45.3
ResNet-101ACNet [26]==45.9
ResNet-101DNL [69]69M1249G46.0
ResNet-101UperNet [65]86M1029G44.9
HRNet-w48 [53]OCRNet [73]71M664G45.7=
ResNeSt-200 [75]DLab.v3+ [12]88M1381G48.4=
Swin-T[43]UperNet [65]60M945G44.545.8
Swin-S [43]UperNet [65]81M1038G47.649.5
Swin-B [43]UperNet [65]121M1188G48.149.7
Twins-SVT-L[15]UperNet [65]133M48.850.2
MiT-B5 [66]SegFormer [66]85M51.051.8
ViT-L/16+ [21]SETR[80]308M50.3=
Swin-L* [43]UperNet [65]234M3230G52.153.5
ViT-L/16 [21]Segmenter [52]334M=51.853.6
Swin-L‡ [43]K-Net [77]==54.3
Swin-L‡ [43]PatchDiverse [29]234M53.154.4
VOLO-D5 [72]UperNet [65]=-54.3
Focal-T (Ours)UperNet [65]62M998G45.847.0
Focal-S (Ours)UperNet [65]85M1130G48.050.0
Focal-B (Ours)UperNet [65]126M1354G49.050.5
Focal-L‡ (Ours)UperNet [65]240M3376G54.055.4
+ +Table 6: Comparison with SoTA methods for semantic segmentation on ADE20K [83] val set. Single- and multi-scale evaluations are reported in the last two columns. $^ \ddag$ means ImageNet-22K is used as the pretraining dataset. + +
Model W-Size FLOPs Top-1(%) APb APm
Swin-Tiny74.581.2 43.739.8
144.982.1 44.0 40.5
Focal-Tiny74.982.2 44.941.1
145.282.3 45.5 41.5
+ +Table 7: Impact of different window sizes (WSize). We alter the default size 7 to 14 and observe consistent improvements for both methods. + +
Model W-Shift Top-1(%) APb APm
Swin-Tiny80.2 81.238.8 43.736.4 39.8
Focal-Tiny82.244.841.0
81.944.941.1
+ +Table 8: Impact of window shift (W-Shift) on Swin Transformer and Focal Transformer. Tiny models are used. + +multi-scale evaluation, we use the same scaling ratios [0.5, 0.75, 1.0, 1.25, 1.5, 1.75] as in previous works. The results in Table 6 show that Focal-Large achieves significantly better performance than Swin-Large. In both single-scale and multi-scale evaluations, Focal-Large leads to more than 1 point mIoU improvement, creating new SoTA for semantic segmentation on ADE20K. + +# 4.4 Ablation studies + +We conduct a series of ablation studies to inspect the model’s capacity from different aspects. We use Focal-Tiny and the image classification and object detection tasks. + +Effect of varying the window size. We have demonstrated that it is crucial to model both short- and long-range interactions. Thus, a related question is whether increasing the window size helps as it leads to a larger receptive field. Table 7 shows the performance of Swin-Tiny and Focal-Tiny with window sizes 7 and 14. Clearly, a larger window size is beneficial for both methods measured in all three metrics, and Focal-Tiny consistently outperforms Swin-Tiny in both window sizes. Comparing the second and third row, we find that Focal-Tiny outperforms Swin-Tiny even with a smaller window size $( 7 \nu . s . \ 1 4 )$ . We suspect that the gain is attributed to our focal attention’s superior capability of capturing long-range dependencies among visual tokens. + +The necessity of window shift. In Swin Transformer [43], window shift is proposed to capture crosswindow interactions between two successive layers. In contrast, visual tokens in Focal Transformers can always communicate with each other across windows at both fine- and coarse-grain. Thus, it is interesting to investigate whether adding window shift to Focal Transformers can lead to any improvement. To answer the question, we remove window shift from Swin Transformer while adding it to Focal Transformers. As shown in Table 8, Swin Transformer shows a severe degradation after removing the window shift. However, adding window shift to Focal Transformer hurts classification performance. The result indicates that window shift is unnecessary for Focal Transformers. While in + +![](images/bba4db737a9b7677713c20afacad8a3124328743b54298c0a2ae2792bcf81178.jpg) +Figure 5: Ablating Focal-Tiny model by adding local, global and both interactions, respectively. Blue bars are image classification results and orange bars object detection results. This figure is better viewed in color. + +
Depths Model #Params. FLOPs Top-1(%) APb Apm
2-2-2-2Swin21.23.178.738.2 35.7
Focal21.73.479.940.5 37.6
2-2-4-2Swin24.73.880.241.2 38.1
Focal25.44.181.443.3 39.8
2-2-6-2Swin28.34.581.243.7 39.8
Focal29.14.982.244.8 41.0
+ +Table 9: Impact of the change of model depth. We gradually reduce the number of transformer layers at the third stage from original 6 to 4 and further 2. Our Focal Transformers has much slower drop rate than Swin Transformer. + +Swin Transformers, there should always be an even number of layers in each stage for the alternative window shift operation, Focal Transformers do not have such a constraint. + +Contributions of local and global interactions. To investigate the relative contributions of capturing local fine-grain and global coarse-grain interactions in Focal Transformers, we have developed several variants of Focal-Tiny: a) Focal-Tiny-Window merely performs attention inside each window; b) Focal-Tiny-Local attends the additional fine-grain surrounding tokens and c) Focal-Tiny-Global attends the extra coarse-grain summarized tokens. We train these models using the same setting as Focal-Tiny and report their performance on image classification and object detection using Mask R-CNN $1 \times$ schedule. As shown in Fig. 5, Focal-Tiny-Window suffers from a significant performance drop on both image classification $8 2 . 2 \substack { 8 0 . 1 }$ ) and object detection $\cdot 4 4 . 8 \mathrm { \ - } \to 3 8 . 3$ ). This is expected since the communication across windows is completely cut off at each Transformer layer. After we enable either the local fine-grain or global coarse-grain interactions (middle two columns), we observe significant performance boost. When we combine short- and long-range interactions, we observe additional improvements on both tasks. This implies that these two type of interactions are complementary and both are beneficial to model performance. + +Model capacity against model depth. Focal attention allows a Transformer model to capture shortand long-range interactions at each Transformer layer. An interesting question is whether Focal Transformers need fewer layers to obtain a similar modeling capacity as the Transformer models that does not use focal attention, such as Swin Transformer. To answer this question, we conduct an experiment by training a series of Swin-Tiny and Focal-Tiny models by varying the number of Transformer layers at stage 3. As shown in Table 9, Focal-Tiny outperforms Swin-Tiny consistently with the same depth. More importantly, using fewer layers, Focal-Tiny can sometimes achieve comparable or even better performance than Swin Transformer. For example, Focal-Tiny with (2-2-4-2) achieves 81.4 on image classification which is better than Swin-Tiny with (2-2-6-2). + +# 5 Conclusion + +In this paper, we have presented a new focal attention mechanism that enables efficient long-range interactions in Vision Transformers. Different from previous works, it performs the local attention at fine-grain and global attention at coarse-grain, providing an effective way of capturing both shortand long-range context with a manageable computational cost. By applying focal attention into a multi-scale Transformer architecture, we propose Focal Transformers as general-purpose backbones for a wide range of dense vision tasks. A comprehensive empirical study shows that our Focal Transformers outperform the SoTA Vision Transformers on a range of vision tasks including image classification, object detection and segmentation. + +Limitations and future work. Although our experiments show that focal attention can significantly boost the performance on image classification and dense prediction tasks, focal attention does introduce extra computational and memory cost, since each query token needs to attend more (summarized) tokens in addition to tokens inside a window. A cost-effective implementation of Focal Transformer is necessary to make it more applicable to many real-world scenarios. This study focuses on incorporating focal attention into multi-scale Vision Transformers for CV tasks. However, we notice that focal attention is an effective sparse attention mechanism that is widely applicable to all attention-based neural network models that are developed for processing natural language, images, videos etc. This is an exciting future direction. + +References +[1] Joshua Ainslie, Santiago Ontanon, Chris Alberti, Philip Pham, Anirudh Ravula, and Sumit Sanghai. Etc: Encoding long and structured data in transformers. arXiv preprint arXiv:2004.08483, 2020. +[2] Irwan Bello, Barret Zoph, Ashish Vaswani, Jonathon Shlens, and Quoc V Le. Attention augmented convolutional networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3286–3295, 2019. +[3] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150, 2020. +[4] Maxim Berman, Hervé Jégou, Andrea Vedaldi, Iasonas Kokkinos, and Matthijs Douze. Multigrain: a unified image embedding for classes and instances. arXiv preprint arXiv:1902.05509, 2019. +[5] Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S Davis. Soft-nms–improving object detection with one line of code. In Proceedings of the IEEE international conference on computer vision, pages 5561–5569, 2017. +[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020. +[7] Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. Gcnet: Non-local networks meet squeezeexcitation networks and beyond. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, pages 0–0, 2019. +[8] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In European Conference on Computer Vision, pages 213–229. Springer, 2020. +[9] Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. arXiv preprint arXiv:2104.14294, 2021. +[10] Shuning Chang, Pichao Wang, Fan Wang, Hao Li, and Jiashi Feng. Augmented transformer with adaptive graph for temporal action proposal generation. arXiv preprint arXiv:2103.16024, 2021. +[11] Kai Chen, Jiangmiao Pang, Jiaqi Wang, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jianping Shi, Wanli Ouyang, et al. Hybrid task cascade for instance segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4974–4983, 2019. +[12] Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European conference on computer vision (ECCV), pages 801–818, 2018. +[13] Xin Chen, Bin Yan, Jiawen Zhu, Dong Wang, Xiaoyun Yang, and Huchuan Lu. Transformer tracking. arXiv preprint arXiv:2103.15436, 2021. +[14] Xinlei Chen, Saining Xie, and Kaiming He. An empirical study of training self-supervised visual transformers. arXiv preprint arXiv:2104.02057, 2021. +[15] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting spatial attention design in vision transformers. arXiv preprint arXiv:2104.13840, 2021. +[16] Xiangxiang Chu, Zhi Tian, Bo Zhang, Xinlong Wang, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Conditional positional encodings for vision transformers. Arxiv preprint 2102.10882, 2021. +[17] Xiyang Dai, Yinpeng Chen, Bin Xiao, Dongdong Chen, Mengchen Liu, Lu Yuan, and Lei Zhang. Dynamic head: Unifying object detection heads with attentions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7373–7382, 2021. +[18] Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object detection with transformers. arXiv preprint arXiv:2011.09094, 2020. +[19] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. +[20] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. NAACL, 2019. +[21] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. +[22] Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Golnaz Ghiasi, Mingxing Tan, Yin Cui, Quoc V Le, and Xiaodan Song. Spinenet: Learning scale-permuted backbone for recognition and localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11592–11601, 2020. +[23] Haoqi Fan, Bo Xiong, Karttikeya Mangalam, Yanghao Li, Zhicheng Yan, Jitendra Malik, and Christoph Feichtenhofer. Multiscale vision transformers. arXiv preprint arXiv:2104.11227, 2021. +[24] Hao-Shu Fang, Jianhua Sun, Runzhong Wang, Minghao Gou, Yong-Lu Li, and Cewu Lu. Instaboost: Boosting instance segmentation via probability map guided copy-pasting. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 682–691, 2019. +[25] Yuxin Fang, Shusheng Yang, Xinggang Wang, Yu Li, Chen Fang, Ying Shan, Bin Feng, and Wenyu Liu. Instances as queries, 2021. +[26] Jun Fu, Jing Liu, Yuhang Wang, Yong Li, Yongjun Bao, Jinhui Tang, and Hanqing Lu. Adaptive context network for scene parsing. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6748–6757, 2019. +[27] Jianfeng Gao, Patrick Pantel, Michael Gamon, Xiaodong He, and Li Deng. Modeling interestingness with deep neural networks. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2–13, Doha, Qatar, October 2014. Association for Computational Linguistics. +[28] Golnaz Ghiasi, Yin Cui, Aravind Srinivas, Rui Qian, Tsung-Yi Lin, Ekin D Cubuk, Quoc V Le, and Barret Zoph. Simple copy-paste is a strong data augmentation method for instance segmentation. arXiv preprint arXiv:2012.07177, 2020. +[29] Chengyue Gong, Dilin Wang, Meng Li, Vikas Chandra, and Qiang Liu. Vision transformers with patch diversification, 2021. +[30] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020. +[31] Kai Han, An Xiao, Enhua Wu, Jianyuan Guo, Chunjing Xu, and Yunhe Wang. Transformer in transformer, 2021. +[32] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. +[33] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. +[34] Elad Hoffer, Tal Ben-Nun, Itay Hubara, Niv Giladi, Torsten Hoefler, and Daniel Soudry. Augment your batch: Improving generalization through instance repetition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8129–8138, 2020. +[35] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018. +[36] Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, and Mubarak Shah. Transformers in vision: A survey. arXiv preprint arXiv:2101.01169, 2021. +[37] Yann LeCun, Yoshua Bengio, et al. Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks, 3361(10):1995, 1995. +[38] Bing Li, Cheng Zheng, Silvio Giancola, and Bernard Ghanem. Sctn: Sparse convolution-transformer network for scene flow estimation. arXiv preprint arXiv:2105.04447, 2021. +[39] Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Mei Gao, Bin Xiao, Xiyang Dai, Lu Yuan, and Jianfeng Gao. Efficient self-supervised vision transformers for representation learning. arXiv preprint arXiv:2106.09785, 2021. +[40] Xiangyu Li, Yonghong Hou, Pichao Wang, Zhimin Gao, Mingliang Xu, and Wanqing Li. Trear: Transformer-based rgb-d egocentric action recognition. arXiv preprint arXiv:2101.03904, 2021. +[41] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017. +[42] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, 2014. +[43] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. +[44] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. +[45] Hyeonseob Nam, Jung-Woo Ha, and Jeonghee Kim. Dual attention networks for multimodal reasoning and matching. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 299–307, 2017. +[46] R. Pappagari, P. Zelasko, J. Villalba, Y. Carmiel, and N. Dehak. Hierarchical transformers for long document classification. In 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 838–844, 2019. +[47] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR, 2018. +[48] Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM journal on control and optimization, 30(4):838–855, 1992. +[49] Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, and Timothy P Lillicrap. Compressive transformers for long-range sequence modelling. arXiv preprint arXiv:1911.05507, 2019. +[50] Prajit Ramachandran, Niki Parmar, Ashish Vaswani, Irwan Bello, Anselm Levskaya, and Jonathon Shlens. Stand-alone self-attention in vision models. arXiv preprint arXiv:1906.05909, 2019. +[51] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani. Bottleneck transformers for visual recognition, 2021. +[52] Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. Segmenter: Transformer for semantic segmentation. arXiv preprint arXiv:2105.05633, 2021. +[53] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, 2019. +[54] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International Conference on Machine Learning, pages 6105–6114. PMLR, 2019. +[55] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. +[56] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12894–12904, 2021. +[57] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NeurIPS, 2017. +[58] Huiyu Wang, Yukun Zhu, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen. Max-deeplab: End-to-end panoptic segmentation with mask transformers. arXiv preprint arXiv:2012.00759, 2020. +[59] Ning Wang, Wengang Zhou, Jie Wang, and Houqaing Li. Transformer meets tracker: Exploiting temporal context for robust visual tracking. arXiv preprint arXiv:2103.11681, 2021. +[60] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. arXiv preprint arXiv:2102.12122, 2021. +[61] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018. +[62] Yuqing Wang, Zhaoliang Xu, Xinlong Wang, Chunhua Shen, Baoshan Cheng, Hao Shen, and Huaxia Xia. End-to-end video instance segmentation with transformers. arXiv preprint arXiv:2011.14503, 2020. +[63] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Convolutional block attention module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19, 2018. +[64] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021. +[65] Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene understanding. In Proceedings of the European Conference on Computer Vision (ECCV), pages 418–434, 2018. +[66] Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. Segformer: Simple and efficient design for semantic segmentation with transformers, 2021. +[67] Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregated residual transformations for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1492–1500, 2017. +[68] Jianwei Yang, Zhile Ren, Chuang Gan, Hongyuan Zhu, and Devi Parikh. Cross-channel communication networks. In Proceedings of the 33rd International Conference on Neural Information Processing Systems, pages 1297–1306, 2019. +[69] Minghao Yin, Zhuliang Yao, Yue Cao, Xiu Li, Zheng Zhang, Stephen Lin, and Han Hu. Disentangled non-local neural networks. In European Conference on Computer Vision, pages 191–207. Springer, 2020. +[70] Kun Yuan, Shaopeng Guo, Ziwei Liu, Aojun Zhou, Fengwei Yu, and Wei Wu. Incorporating convolution designs into visual transformers. arXiv preprint arXiv:2103.11816, 2021. +[71] Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi, Francis EH Tay, Jiashi Feng, and Shuicheng Yan. Tokens-to-token vit: Training vision transformers from scratch on imagenet. arXiv preprint arXiv:2101.11986, 2021. +[72] Li Yuan, Qibin Hou, Zihang Jiang, Jiashi Feng, and Shuicheng Yan. Volo: Vision outlooker for visual recognition. arXiv preprint arXiv:2106.13112, 2021. +[73] Yuhui Yuan, Xilin Chen, and Jingdong Wang. Object-contextual representations for semantic segmentation. arXiv preprint arXiv:1909.11065, 2019. +[74] Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, et al. Big bird: Transformers for longer sequences. arXiv preprint arXiv:2007.14062, 2020. +[75] Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Haibin Lin, Zhi Zhang, Yue Sun, Tong He, Jonas Mueller, R Manmatha, et al. Resnest: Split-attention networks. arXiv preprint arXiv:2004.08955, 2020. +[76] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multiscale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint arXiv:2103.15358, 2021. +[77] Wenwei Zhang, Jiangmiao Pang, Kai Chen, and Chen Change Loy. K-net: Towards unified image segmentation, 2021. +[78] Jiaojiao Zhao, Xinyu Li, Chunhui Liu, Shuai Bing, Hao Chen, Cees GM Snoek, and Joseph Tighe. Tuber: Tube-transformer for action detection. arXiv preprint arXiv:2104.00969, 2021. +[79] Minghang Zheng, Peng Gao, Xiaogang Wang, Hongsheng Li, and Hao Dong. End-to-end object detection with adaptive clustering transformer. arXiv preprint arXiv:2011.09315, 2020. +[80] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. arXiv preprint arXiv:2012.15840, 2020. +[81] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6881–6890, 2021. +[82] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 13001–13008, 2020. +[83] Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing through ade20k dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 633–641, 2017. +[84] Xingyi Zhou, Vladlen Koltun, and Philipp Krähenbühl. Probabilistic two-stage detection. arXiv preprint arXiv:2103.07461, 2021. +[85] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. \ No newline at end of file diff --git a/parse/train/2zCRcTafea/2zCRcTafea_content_list.json b/parse/train/2zCRcTafea/2zCRcTafea_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..cad69af2e12083a44fc83bf714e013b134ef17bb --- /dev/null +++ b/parse/train/2zCRcTafea/2zCRcTafea_content_list.json @@ -0,0 +1,1009 @@ +[ + { + "type": "text", + "text": "Focal Attention for Long-Range Interactions in Vision Transformers ", + "text_level": 1, + "bbox": [ + 215, + 122, + 784, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Jianwei Yang1 Chunyuan $\\mathbf { L i } ^ { 1 }$ Pengchuan Zhang1 Xiyang Dai2 Bin Xiao2 Lu Yuan2 Jianfeng Gao1 1Microsoft Research at Redmond, 2Microsoft Cloud $^ +$ AI \n{jianwyan,chunyl,penzhan,xidai,bixi,luyuan,jfgao}@microsoft.com ", + "bbox": [ + 227, + 219, + 772, + 280 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 314, + 535, + 330 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Recently, Vision Transformer and its variants have shown great promise on various computer vision tasks. The ability of capturing local and global visual dependencies through self-attention is the key to its success. However, this also brings challenges due to quadratic computational overhead, especially for the high-resolution vision tasks (e.g., object detection). Many recent works have attempted to reduce the cost and improve model performance by applying either coarse-grained global attention or fine-grained local attention. However, both approaches cripple the modeling power of the original self-attention mechanism of multi-layer Transformers, leading to sub-optimal solutions. In this paper, we present focal attention, a new attention mechanism that incorporates both fine-grained local and coarse-grained global interactions. In this new mechanism, each token attends its closest surrounding tokens at fine granularity and the tokens far away at coarse granularity, and thus can capture both short- and long-range visual dependencies efficiently and effectively. With focal attention, we build a new variant of Vision Transformer models, called Focal Transformers, which achieve superior performance over the state-of-theart (SoTA) Vision Transformers on a range of public image classification and object detection benchmarks. In particular, our Focal Transformer models with a moderate size of 51.1M and a large size of $8 9 . 8 \\mathbf { M }$ achieve $\\mathbf { 8 3 . 6 \\% }$ and $\\mathbf { 8 4 . 0 \\% }$ Top-1 accuracy, respectively, on ImageNet classification at $2 2 4 \\times 2 2 4$ . When employed as the backbones, Focal Transformers achieve consistent and substantial improvements over the current SoTA Swin Transformers [43] across 6 different object detection methods. Our largest Focal Transformer yields 58.7/59.0 box mAPs and 50.9/51.3 mask mAPs on COCO mini-val/test-dev, and 55.4 mIoU on ADE20K for semantic segmentation, creating new SoTA on three of the most challenging computer vision tasks. Our code is available at: https://github. com/microsoft/Focal-Transformer. ", + "bbox": [ + 232, + 347, + 764, + 705 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 731, + 312, + 747 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Nowadays, Transformer [57] has become a prevalent model architecture in natural language processing (NLP) [20, 6]. In the light of its success in NLP, there is an increasing effort on adapting it to computer vision (CV) [47, 50]. Since its promise firstly demonstrated in Vision Transformer (ViT) [21], we have witnessed a flourish of full-Transformer models for image classification [55, 60, 64, 43, 76, 56], object detection [8, 85, 79, 18] and semantic segmentation [58, 62]. Beyond these static image tasks, it has also been applied on various temporal understanding tasks, such as action recognition [40, 78, 10], object tracking [13, 59], scene flow estimation [38]. ", + "bbox": [ + 174, + 756, + 826, + 852 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The self-attention mechanism is arguably the key component that differentiates Transformers from the widely used convolutional neural networks (CNNs) [37] in computer vision. At each Transformer layer, self-attention enables global content-dependent interactions among different image regions for modeling short- and long-range dependencies, respectively. Through the visualization of full selfattention results1, we indeed observe that self-attention learns to attend local surroundings (like CNNs) and the global contexts at the same time, as illustrated in Fig. 1 (Left). Nevertheless, when dealing with high-resolution vision tasks such as object detection or segmentation, an efficient implementation of a global and fine-grained self-attention becomes non-trivial due to the quadratic computational cost with respect to the number of tokens in feature maps. Recent works have alternatively exploited either a coarse-grained global self-attention [60, 64] or a fine-grained local self-attention [43, 76, 56], for the sake of reducing the computational cost. However, both approaches cripple the power of the original full self-attention i.e., the ability to simultaneously capture local and global visual dependencies. ", + "bbox": [ + 176, + 858, + 825, + 901 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/621f200345a7ebe619be7e8a0aedd96409fbf6fca08314c5e32fce4ac604fb01.jpg", + "image_caption": [ + "Figure 1: Left: Visualization of the attention maps of the three heads at the given query patch (blue) in the first layer of the DeiT-Tiny model [55]. Right: An illustrative depiction of focal attention mechanism. Three granularity levels are used to compose the attention region for the blue query. " + ], + "image_footnote": [], + "bbox": [ + 183, + 95, + 816, + 223 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 284, + 825, + 409 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this paper, we present a new attention mechanism to capture both short- and long-range interactions in Transformer layers for high-resolution input images. Considering that the visual dependencies between the nearby (local) regions are usually much stronger than the dependencies between the regions that are far away, we perform the fine-grained attention only in local regions while the coarse-grained attention globally. As depicted in Fig. 1 (Right), a query token in the feature map attends its closest local surroundings at the finest granularity as itself. However, when it goes to the regions far away, it attends to summarized tokens to capture coarse-grained visual dependencies. We call this new mechanism focal attention, as each token attends the others in a focal manner. We will show in this study that focal attention allows to effectively model visual dependencies among all regions covering the whole high-resolution feature maps while introducing much less number of tokens in the computation than that in the standard self-attention mechanism. ", + "bbox": [ + 174, + 415, + 825, + 566 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Equipped with focal attention, a series of Focal Transformers are developed and validated via a comprehensive empirical study across three core vision tasks, including image classification, object detection and segmentation. Results show that Focal Transformers consistently outperform the SoTA Vision Transformers across various settings (i.e., in model sizes and complexities). Notably, the small Focal Transformer with 51.1M parameters achieves $8 3 . 6 \\%$ top-1 accuracy on ImageNet-1K, and the base model with 89.8M parameters obtains $8 4 . 0 \\%$ top-1 accuracy. In the fine-tuning experiments for object detection, Focal Transformers consistently outperform the SoTA Swin Transformers [43] across six popular object detection methods. Our largest Focal Transformer model achieves 59.0 box mAP and 51.3 mask mAP on COCO test-dev for object detection and instance segmentation, respectively, and 55.4 mIoU on ADE20K for semantic segmentation. These results demonstrate that focal attention is highly effective in modeling the global interactions in Vision Transformers. ", + "bbox": [ + 174, + 573, + 825, + 724 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 Related work ", + "text_level": 1, + "bbox": [ + 174, + 746, + 316, + 763 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Vision Transformers. Vision Transformer (ViT) is first introduced in [21]. It applies a standard Transformer, originally developed for NLP [57], to encode an image by analogously splitting the image into a sequence of visual tokens. It has demonstrated superior performance to CNNs such as ResNet [33] on multiple image classification benchmarks, when trained with sufficient data [21] and carefully designed data augmentation and regularization methods [55]. The results thus inspire researchers to explore the applications of ViT on various vision tasks beyond image classification, such as self-supervised learning [14, 9, 39], object detection [8, 85, 79, 18] and semantic segmentation [58, 62, 81]. There are also increasing number of studies for improving ViT via data-efficient training [55], improved patch embedding/encoding [16, 71, 31], integrating convolutional projections into transformers [64, 70], and using multi-scale architectures and efficient self-attention mechanisms for high-resolution vision tasks [60, 64, 43, 76, 15]. Recent surveys include [36, 30, 36]. This paper focuses on improving the self-attention mechanism of ViT for encoding high-resolution images. ", + "bbox": [ + 174, + 772, + 825, + 883 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/895ff52491908f44661414bd57eca7fc36f34ebd4f7c17cd1d28f3b83da587af.jpg", + "image_caption": [ + "Figure 2: Model architecture for our Focal Transformers. As highlighted in light blue boxes, our main innovation is the proposed focal attention in each Transformer layer. " + ], + "image_footnote": [], + "bbox": [ + 179, + 90, + 816, + 273 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 351, + 825, + 406 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Efficient global and local self-attention. In many real-world tasks, Transformers need to cope with a large number of input tokens, such as long documents in NLP and high-resolution images in computer vision (CV). Recently, many efficient self-attention mechanisms have been proposed to deal with the quadratic computational and memory cost incurred by the standard self-attention mechanism. On one hand, a number of works in both NLP and CV resort to coarse-grained global self-attention (i.e., attending the down-sampled or summarized tokens) to capture the long-range interactions [49, 46, 60, 64, 31, 23]. Although this approach improves the model efficiency, it loses the detailed context information surrounding the query tokens. On the other hand, to make the computational cost manageable, various local fine-grained attention mechanism (i.e., attending neighboring tokens within a pre-set window size) are used for both NLP [3, 74, 1] and CV [56, 43, 76]. In this paper, we argue that both global and local attentions are important for model performance. This is also validated by some recent studies that aim to improve CNNs by incorporating ways of modeling global attentions [35, 63, 61, 68, 2, 7, 51]. The standard self-attention mechanism used by ViT can indeed learned both types of attentions, as shown in Fig. 1 (Left). But it often incurs a prohibitively high cost for high-resolution images. To the best of our knowledge, the proposed focal attention provides the first mechanism to incorporate local and global attention in a single Transformer layer 2. It can capture both short- and long-range interactions as standard self-attention but in a much more efficient and effective way, especially for high-resolution images. ", + "bbox": [ + 173, + 411, + 825, + 661 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 Method ", + "text_level": 1, + "bbox": [ + 174, + 685, + 269, + 703 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 Model architecture ", + "text_level": 1, + "bbox": [ + 174, + 713, + 346, + 728 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "To accommodate high-resolution dense prediction tasks, we employ a multi-scale model architecture as in [60, 76, 43]. As shown in Fig. 2, an image $I \\in \\mathcal { R } ^ { H \\times W \\times 3 }$ is first partitioned into patches of size $4 \\times 4$ , resulting in ${ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } }$ visual tokens with dimension $4 \\times 4 \\times 3$ . Then, we use a patch embedding layer, consisting of a convolutional layer with filter size and stride both equal to 4, to project these patches into hidden features with dimension $d$ . We then pass this spatial feature map to the four stages of Focal Transformer blocks. In each stage $i \\in \\{ 1 , 2 , 3 , 4 \\}$ , the Focal Transformer block consists of $N _ { i }$ Focal Transformer layers. After each stage, we use a patch embedding layer to reduce the spatial size of feature map by factor 2 and increase the feature dimension by 2. For image classification tasks, we take the average of the output from the last stage and send it to a classification layer. For object detection, the feature maps from the last 3 or all 4 stages are fed to a particular object detector head, depending on the specific detection method we choose to use. The model capacity can be customized by varying the input feature dimension $d$ and the number of Focal Transformer layers. ", + "bbox": [ + 173, + 739, + 825, + 878 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 90, + 823, + 119 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Standard self-attention can capture both short- and long-range interactions at fine-grain, but suffers from high computational cost when it performs attention on high-resolution feature maps as noted in [76]. Take stage 1 in Fig. 2 as an example. For a feature map of size ${ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } } \\times d$ , the complexity of self-attention is $\\begin{array} { r } { \\mathcal { O } ( ( \\frac { H } { 4 } \\times \\frac { W } { 4 } ) ^ { 2 } d ) } \\end{array}$ , resulting in an explosion of time and memory cost, considering that $\\operatorname* { m i n } ( H , W )$ could be 800 or even larger for object detection. In the next section, we describe how we address this issue with the proposed focal attention mechanism. ", + "bbox": [ + 174, + 126, + 825, + 212 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 Token-wise focal attention ", + "text_level": 1, + "bbox": [ + 174, + 227, + 395, + 242 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Focal attention is proposed to make the Transformer layers suitable for encoding high-resolution input images. Instead of attending all tokens at fine-grain, we attend the fine-grain tokens only locally, but the summarized ones (i.e., the coarse-grained tokens generated by sub-window pooling, which is illustrated in Fig. 4 and will be described later) globally. As such, focal attention can cover the same amount of image regions as standard self-attention but with much less cost. In Fig. 3, we show the size of the receptive field for standard self-attention and our focal attention as a function of the number of attended tokens. For a given query position, by reducing the granularity of its surroundings based on their distance to the query, focal attention can have significantly larger receptive fields at the same cost measured by the number of visual tokens, compared to the standard self-attention mechanism. ", + "bbox": [ + 174, + 253, + 549, + 473 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/bf6a6de3b259a05fb0ac374d04a980cb6b72063e837f8ca7329f069632ea03a7.jpg", + "image_caption": [ + "Figure 3: The size of receptive field (yaxis) as a function of the number of used visual tokens $\\mathbf { \\bar { x } }$ -axis) in regular (standard) self-attention and focal attention. When plotting the curve for focal attention, we increase the focal window size by 2 for each focal level up to the maximal window size of 8. " + ], + "image_footnote": [], + "bbox": [ + 560, + 253, + 812, + 343 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Theoretically, the focal attention mechanism enables global interaction with much less time and memory cost, because it attends a much smaller number of surrounding (summarized) tokens. In practice, however, extracting the surrounding tokens for each query position could incur high time cost since we need to duplicate the extraction of each token for all queries that the token surrounds. This issue had been extensively discussed in [56, 76, 43] and a common solution is to partition the input feature map into windows. Thus, in our Focal Transformers, we resort to performing focal attention at the window level. We elaborate the window-wise focal attention in the following. ", + "bbox": [ + 173, + 479, + 825, + 577 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2.1 Window-wise focal attention ", + "text_level": 1, + "bbox": [ + 174, + 590, + 423, + 604 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Given a feature map of $\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }$ with spatial size $M \\times N$ , we first partition it into a grid of windows of size $s _ { p } \\times s _ { p }$ . Then, we extract the surroundings for each window rather than each individual token. The proposed window-wise focal attention is illustrated in Fig. 4. To clarify, we first define three terms: ", + "bbox": [ + 174, + 613, + 825, + 670 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• Focal level $L$ refers to the granularity level at which we extract the tokens for focal attention. \n• Focal window size $s _ { w } ^ { l }$ is the size of sub-window on which the summarized tokens are formed via \nsub-window pooling at granularity level of $l \\in \\{ 1 , . . . , L \\}$ . \n• Focal region size $s _ { r } ^ { l }$ denotes the number of sub-windows that are filled up horizontally (or vertically) in an attended region at level $l$ . ", + "bbox": [ + 174, + 676, + 826, + 747 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Now, we detail how window-wise focal attention works in the following two steps, sub-window pooling and attention computing. ", + "bbox": [ + 171, + 752, + 823, + 782 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Sub-window pooling. Consider input feature map $\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }$ , where $M \\times N$ is the spatial dimension and $d$ the feature dimension. We perform sub-window pooling for all $L$ levels. At focal level $l$ , we first split the input feature map $x$ into a grid of sub-windows with size $s _ { w } ^ { l } \\times s _ { w } ^ { l }$ . Then we use a linear projection layer $f _ { p } ^ { l }$ to pool the sub-windows spatially by ", + "bbox": [ + 174, + 786, + 825, + 845 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/2a14e2c3c947dc9cd476c4ca7ba03d6c99cd76f5d3158e8561986ada02fbfb84.jpg", + "text": "$$\n\\begin{array} { r } { x ^ { l } = f _ { p } ^ { l } ( \\hat { x } ) \\in \\mathcal { R } ^ { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d } , \\quad \\hat { x } = \\mathrm { R e s h a p e } ( x ) \\in \\mathcal { R } ^ { ( \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d ) \\times ( s _ { w } ^ { l } \\times s _ { w } ^ { l } ) } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 272, + 848, + 723, + 871 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The pooled feature maps $\\{ x ^ { l } \\} _ { 1 } ^ { L }$ at different levels $l$ provide rich information at both fine-grain and coarse-grain. Since we set $s _ { w } ^ { l } = 1$ for the first focal level which has the same granularity as the input ", + "bbox": [ + 174, + 881, + 821, + 912 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/8d7c5bc6d5803f11463c3ab8656b31fec8dbb82ddb891bec59961a9ad98dfe04.jpg", + "image_caption": [ + "Figure 4: An illustration of focal attention at window level. Each of the square cells represents a visual token that is either from the original feature map or a summarized token formed by sub-window pooling. Suppose we have an input feature map of size $2 0 \\times 2 0$ . We first partition it into $5 \\times 5$ windows of size $4 \\times 4$ . Take the $4 \\times 4$ blue window in the middle as the query set, we extract its surrounding tokens at three granularity levels as its keys and values. For the first level, we extract the $8 \\times 8$ tokens which are closest to the blue window at the finest grain. At the second level, we expand the attention region and pool the surrounding $2 \\times 2$ sub-windows to form summarized tokens, which results in $6 \\times 6$ summarized tokens. At the third level, we attend a larger region covering the whole feature map and pool $4 \\times 4$ sub-windows, which leads to $5 \\times 5$ summarized tokens. Finally, these three levels of tokens are concatenated to compute the keys and values for the $4 \\times 4 = 1 6$ tokens (queries) in the blue window. " + ], + "image_footnote": [], + "bbox": [ + 194, + 92, + 808, + 292 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "feature map, there is no need to perform any sub-window pooling. Considering that the focal window size is usually very small (7 maximally in our settings), the number of extra parameters introduced by sub-window pooling is negligible. ", + "bbox": [ + 174, + 472, + 825, + 513 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Attention computing. Once we obtain the pooled feature maps $\\{ x ^ { l } \\} _ { 1 } ^ { L }$ at all $L$ levels, we compute the query at the first level, and key and value for all levels using three linear projection layers $f _ { q } , f _ { k }$ and $f _ { v }$ , respectively, as ", + "bbox": [ + 174, + 520, + 825, + 563 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/adc5271b876d4969c2420a5cdd389cd1720ece17ee5092a72df5a2920aba267f.jpg", + "text": "$$\nQ = f _ { q } ( x ^ { 1 } ) , \\quad K = \\{ K ^ { l } \\} _ { 1 } ^ { L } = f _ { k } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) , \\quad V = \\{ V ^ { l } \\} _ { 1 } ^ { L } = f _ { v } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) .\n$$", + "text_format": "latex", + "bbox": [ + 240, + 569, + 758, + 588 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To perform focal attention, we need to first extract the surrounding tokens for each query token in the feature map. As mentioned earlier, tokens inside a window partition $s _ { p } \\times s _ { p }$ share the same set of surroundings. For the queries inside the $i$ -th window $Q _ { i } \\in \\mathcal { R } ^ { s _ { p } \\times s _ { p } \\times d }$ , we extract the $s _ { r } ^ { l } \\times s _ { r } ^ { l }$ keys and values from $K ^ { l }$ and $V ^ { l }$ surrounding the window which the query lies in, and then gather the keys and values from all $L$ levels to obtain $\\breve { K } _ { i } = \\{ K _ { i } ^ { 1 } , . . . , K _ { i } ^ { L } \\} \\in \\mathscr { R } ^ { \\bar { s } \\times d }$ and $V _ { i } = \\{ V _ { i } ^ { 1 } , . . . , \\mathbf { \\bar { V } } _ { i } ^ { L } \\} \\in \\mathcal { R } ^ { s \\times d }$ , where imple $s$ is the sum of focal regions from all levels, i.e., ntation of focal attention following Fig. 1 requires $\\begin{array} { r } { s = \\sum _ { l = 1 } ^ { L } ( s _ { r } ^ { l } ) ^ { 2 } } \\end{array}$ . Note that a canonicalverlapped regions across different levels. In our implementation, we intentionally keep them in order to capture the pyramid information for the overlapped regions. Finally, we follow [43] to include a relative position bias and compute the focal attention for $Q _ { i }$ by ", + "bbox": [ + 173, + 602, + 826, + 747 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/a272e4bce5871346781b2ad83e37e4087f5507b82017abb37846d96c857a7475.jpg", + "text": "$$\n\\mathrm { A t t e n t i o n } ( Q _ { i } , K _ { i } , V _ { i } ) = \\mathrm { S o f t m a x } ( \\frac { Q _ { i } K _ { i } ^ { T } } { \\sqrt { d } } + B ) V _ { i } ,\n$$", + "text_format": "latex", + "bbox": [ + 343, + 755, + 651, + 786 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $B = \\{ B ^ { l } \\} _ { 1 } ^ { L }$ is the learnable relative position bias. It consists of $L$ subsets for $L$ focal levels. Similar to [43], for the first level, we parameterize it as $B ^ { 1 } \\in \\mathcal { R } ^ { ( 2 s _ { p } - 1 ) \\times ( 2 s _ { p } - 1 ) }$ , considering that the horizontal and vertical position ranges are both in $[ - s _ { p } + 1 , s _ { p } - 1 ]$ . For the other focal levels, considering that they have different granularity with respect to the queries, we treat all the queries inside a window equally and use $B ^ { l } \\in \\mathcal { R } ^ { s _ { r } ^ { l } \\times s _ { r } ^ { l } }$ to represent the relative position bias between the query window and each of $s _ { r } ^ { l } \\times s _ { r } ^ { l }$ summarized tokens. Since the focal attention for each window can be performed independent of the others, we can compute Eq. (3) in parallel. Once we obtain attention scores for the whole input feature map, we send them to LayerNorm and the MLP block. ", + "bbox": [ + 173, + 794, + 826, + 912 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/2e084ddc28027667351980bf367187c19e42a436814641051a239edbb9b8f4a0.jpg", + "table_caption": [], + "table_footnote": [ + "Table 1: Model configurations for Focal Transformers. We use three configurations with different model capacities: Focal-Tiny, Focal-Small and Focal-Base. " + ], + "table_body": "
Output SizeLayer NameFocal-TinyFocal-SmallFocal-Base
stage 156×56Patch Embeddingp1= 4;c1 = 96p1=4;c1= 96p1 = 4;c1 = 128
56×56Transformer Block{1,13} 三 s={7,7×2二 {1,13} ={7,7}×2{1,13} ={7,7}×2
stage 228×28Patch EmbeddingP2=2;c=192P2=2;C=192P2=2;C= 256
28×28Transformer Block{1,13} 三 swr={7,5} 1×2={1,13} sw,r={7,5} 1×2{1,13} 二 su,r={7,5}×2
stage 314 × 14Patch Embeddingp3=2;c3=384p3=2; c3= 384p3=2; c3= 512
14 × 14Transformer Block={1,13} s={7,3}×6={1,13} ={7,3}×18={1,13} ={7,3}×18
stage 47×7Patch EmbeddingP4=2;C4=768P4=2;C4=768P4= 2;C4=1024
7×7Transformer Block二 {1,7}×2{1,7} s 三 ={7,1}×2二 {1,7} su,r {7,1} 二×2
", + "bbox": [ + 178, + 88, + 810, + 311 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.2.2 Complexity analysis ", + "text_level": 1, + "bbox": [ + 174, + 358, + 366, + 372 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We analyze the computational complexity for the two steps of focal attention described above. For the input feature map $\\dot { \\boldsymbol { x } } \\in \\mathcal { R } ^ { M \\times N \\times d }$ , we have $\\begin{array} { r } { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } } \\end{array}$ sub-windows at focal level l. For each sub-window, the pooling operation in Eq.1 has the complexity of $\\mathcal { O } ( ( s _ { w } ^ { l } ) ^ { 2 } d )$ . Aggregating all sub-windows brings us $O ( ( M N ) d )$ . Then for all focal levels, we have the complexity of $\\mathcal { O } ( L ( M N ) d )$ in total, which is independent of the sub-window size at each focal level. Regarding the attention computation in Eq. 3, the computational cost for a query window $s _ { p } \\times s _ { p }$ is $\\mathcal { O } ( ( s _ { p } ) ^ { 2 } \\textstyle \\sum _ { l } ( s _ { r } ^ { l } ) ^ { 2 } d )$ , and $\\mathcal { O } ( \\dot { \\sum } _ { l } ( s _ { r } ^ { l } ) ^ { 2 } ( M \\dot { N } ) d )$ for the whole input feature map. To sum up, the overall computational cost for focal attention is $\\begin{array} { r } { \\mathcal { O } ( ( L + \\sum _ { l } ( s _ { r } ^ { l } ) ^ { \\bar { 2 } } ) ( M N ) d ) } \\end{array}$ . In an extreme case, one can set $s _ { r } ^ { \\hat { L } } = 2 \\times \\operatorname* { m a x } ( M , N ) / s _ { w } ^ { L }$ to ensure a global receptive field for all queries (including both corner and middle queries) in this layer. ", + "bbox": [ + 173, + 375, + 825, + 505 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.3 Model configurations ", + "text_level": 1, + "bbox": [ + 174, + 522, + 361, + 537 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "For fair comparison, we consider three network configurations for Focal Transformers, following [60, 64, 43]. Specifically, we follow the design of the Tiny, Small and Base models in Swin Transformer [43], as shown in Table 1. Our models take $2 2 4 \\times 2 2 4$ images as inputs and the window partition size is set to 7 to make our models comparable to Swin Transformers. For the focal attention layer, we introduce two levels, one for fine-grained local attention and the other for coarse-grained global attention. Except for the last stage, the focal region size is set to 13 for the window partition size of 7, which means that we expand 3 tokens for each window partition. For the last stage, since the whole feature map is $7 \\times 7$ , the focal region size at level 0 is set to 7, which is sufficient to cover the entire feature map. For the coarse-grained global attention, we set its focal window size the same as the window partition size 7, but gradually decrease the focal region size to get $\\{ 7 , 5 , 3 , 1 \\}$ for the four stages, respectively. For the patch embedding layer, the spatial reduction ratio $p _ { i }$ for the four stages are all $\\{ 4 , 2 , 2 , 2 \\}$ . Note that Focal-Base has a higher hidden dimension $c _ { i }$ , compared to Focal-Tiny and Focal-Small. ", + "bbox": [ + 173, + 541, + 825, + 720 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4 Experiments ", + "text_level": 1, + "bbox": [ + 174, + 742, + 312, + 760 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1 Image classification on ImageNet-1K ", + "text_level": 1, + "bbox": [ + 174, + 767, + 467, + 782 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We compare different methods on ImageNet-1K [19]. For fair comparison, we follow the training recipes in [55, 60]. All models are trained for 300 epochs with batch size 1024. The initial learning rate is set to $1 0 ^ { - 3 }$ with 20 epochs of linear warm-up starting from $1 0 ^ { - 5 }$ . For optimization, we use AdamW [44] as the optimizer with a cosine learning rate scheduler. The weight decay is set to 0.05 and the maximal gradient norm is clipped to 5.0. We use the same set of data augmentation and regularization strategies used in [55] after excluding random erasing [82], repeated augmentation [4, 34] and exponential moving average (EMA) [48]. The stochastic depth drop rates are set to 0.2, 0.2 and 0.3 for our tiny, small and base models, respectively. During training, we crop images randomly to $2 2 4 \\times 2 2 4$ , while a center crop is used during evaluation on the validation set. ", + "bbox": [ + 174, + 786, + 825, + 911 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/9f9cf0ebeecc16dfc400d5f622af2df25d867b1a4030e821ba17190b276a552a.jpg", + "table_caption": [], + "table_footnote": [ + "Table 2: Comparison of image classification on ImageNet-1K for different models. Except for ViT-Base/16, all other models are trained and evaluated on $2 2 4 \\times 2 2 4$ resolution. " + ], + "table_body": "
Model#Params. FLOPsTop-1 (%)
ResNet-50 [33]25.0 4.176.2
DeiT-Small/16 [55]22.1 4.679.9
PVT-Small [60]24.5 3.879.8
ViL-Small [76]24.6 5.182.0
CvT-13 [64]20.0 4.581.6
Swin-Tiny [43]28.3 4.581.2
Focal-Tiny (Ours)28.9 4.982.2
ResNet-101[33]45.0 7.977.4
PVT-Medium [60]44.2 6.781.2
CvT-21 [64]32.0 7.182.5
ViL-Medium [76]39.7 9.183.3
Swin-Small [43]49.6 8.783.1
Focal-Small (Ours)51.1 9.483.6
ResNet-152[33]60.0 11.078.3
ViT-Base/16 [21]86.6 17.677.9
DeiT-Base/16 [55]17.581.8
86.6
PVT-Large [60]61.4 9.881.7
ViL-Base[76]55.7 13.483.2
Swin-Base [43]87.8 15.483.4
Focal-Base (Ours)89.816.4 84.0
", + "bbox": [ + 179, + 88, + 457, + 347 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/2eccde7cb9cd36279d6f0e1059ce2743dcec73d9f7583e972b12720e375f7d35.jpg", + "table_caption": [ + "Table 3: Comparisons with CNN and Transformer baselines and SoTA methods on COCO object detection. The box mAP $( A P ^ { b } )$ and mask mAP $( A P ^ { m } )$ are reported for RetinaNet and Mask R-CNN trained with $1 \\times$ schedule. More detailed comparisons with $3 \\times$ schedule are in Table 4. " + ], + "table_footnote": [], + "table_body": "
BackboneRetinaNetMask R-CNN
APbApbAPm
ResNet-50 [33]36.338.034.4
PVT-Small40.440.437.8
ViL-Small [76]41.641.838.5
Swin-Tiny [43]42.043.739.8
Focal-Tiny (Ours)43.7 (+1.7)44.8 (+1.1) 41.0 (+1.3)
ResNet-101[33]38.540.436.4
ResNeXt101-32x4d [67]39.941.937.5
PVT-Medium [60]41.942.039.0
ViL-Medium [76]42.943.439.7
Swin-Small [43]45.046.542.1
Focal-Small (Ours)45.6 (+0.6)47.4 (+0.9) 42.8 (+0.7)
ResNeXt101-64x4d[67] 41.042.838.4
PVT-Large [60]42.642.939.5
ViL-Base[76]44.345.141.0
Swin-Base [43]45.046.942.3
Focal-Base (Ours)46.3 (+1.3)47.8 (+0.9)43.2 (+0.9)
", + "bbox": [ + 488, + 88, + 818, + 320 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In Table 2, we summarize the results for baseline models and the state-of-the-art models on image classification task. We can see that Focal Transformers consistently outperform other methods with similar model sizes (#Params.) and computational complexities (GFLOPs). Specifically, Focal-Tiny improves over the Transformer baseline DeiT-Small/16 by $2 . 3 \\%$ . Meanwhile, using the same model configuration (2-2-6-2) and a few extra parameters and computations, Focal-Tiny improves over Swin-Tiny by 1.0 point. For small and base models, Focal-Small with 51.1M parameters can reach $8 3 . 6 \\%$ which is better than all the counterpart small and base models using much less parameters. By increasing the model size, Focal-Base model achieves $8 4 . 0 \\%$ , surpassing all the other models with comparable parameters and FLOPs. ", + "bbox": [ + 173, + 420, + 825, + 545 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "To compare with the large-scale models, we further build Focal-Large Transformer by increasing the hidden dimension in Focal-Base from 128 to 196 while keeping all the other hyperparameters the same. We follow the common practice to pretrain our Focal-Large Transformer on ImageNet-22K and transfer it to detection and segmentation tasks [64, 43]. ", + "bbox": [ + 174, + 551, + 825, + 608 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2 Object detection and instance segmentation ", + "text_level": 1, + "bbox": [ + 174, + 625, + 516, + 638 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We benchmark our models on object detection with COCO 2017 [42]. The pretrained models are used as visual backbones and then plugged into two representative pipelines, RetinaNet [41] and Mask R-CNN [32]. All models are trained on the $1 1 8 \\mathrm { k }$ training images and the results are reported on 5K validation set. We use the two standard training schedules, $1 \\times$ with 12 epochs and $3 \\times$ with 36 epochs. For the $1 \\times$ schedule, we resize image’s shorter side to 800 while keeping its longer side no more than 1,333. For the $3 \\times$ schedule, we use the multi-scale training strategy by randomly resizing its shorter side to the range of [480, 800]. Considering this higher input resolution, we adaptively increase the focal sizes at four stages to (15, 13, 9, 7), to ensures that the focal attention covers more than half of the image region at the first two stages, and the whole image at the last two stages. With the focal size increased, the relative position biases are accordingly up-sampled to the corresponding sizes using bilinear interpolation. During training, we use AdamW [44] for optimization with initial learning rate $1 0 ^ { - 4 }$ and weight decay 0.05. Similarly, we use 0.2, 0.3 and 0.5 stochastic depth drop rates to regularize the training for our Tiny, Small and Base models, respectively. Since Swin Transformer does not report the results on RetinaNet, we obtain the results by ourselves using their official code with the same hyper-parameters as that of Focal Transformers. ", + "bbox": [ + 174, + 642, + 825, + 849 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In Table 3, we show the performance for both CNN-based models and the current Transformerbased state-of-the-art models. The bbox mAP $( A P ^ { b } )$ and mask mAP $( A P ^ { m } )$ are reported. We see that Focal Transformers outperform the CNN-based models consistently with the gap of 4.8-7.1 points. Compared with the other methods which also use multi-scale Transformer architectures, ", + "bbox": [ + 176, + 856, + 825, + 911 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/ddccdbf44fb137bfc4faa3d93f6ef7c7ed8810948684d7d2e48c12d371518931.jpg", + "table_caption": [ + "Table 4: COCO object detection and segmentation results with RetinaNet [41] and Mask R-CNN [33]. All models are trained with $3 \\times$ schedule and multi-scale inputs (MS). The numbers before and after “/” at column 2 and 3 are the model size and complexity for RetinaNet and Mask R-CNN, respectively. " + ], + "table_footnote": [], + "table_body": "
Backbone#Params (M)FLOPs (G)RetinaNet 3x schedule + MSMask R-CNN 3x schedule + MS
APAP5APAPsAPMAPLAP6APAPApmAPAP
ResNet50 [33]37.7/44.2239/26039.058.441.822.442.851.641.061.744.937.158.440.1
PVT-Small[60]34.2/44.1226/24542.262.745.026.245.257.243.065.346.939.962.542.8
ViL-Small [76]35.7/45.0252/174 42.963.845.627.846.456.343.464.947.039.662.142.4
Swin-Tiny [43]38.5/47.8245/264 45.065.948.429.748.958.146.068.150.341.665.144.9
Focal-Tiny (Ours)39.4/48.8265/291 45.566.348.831.249.258.747.269.451.942.766.5 45.9
ResNet101 [33]56.7/63.2315/33640.960.144.023.745.053.842.863.247.138.560.141.3
ResNeXt101-32x4d [67]56.4/62.8319/34041.461.044.323.945.553.744.064.448.039.261.441.9
PVT-Medium [60]53.9/63.9283/30243.263.846.127.346.358.944.266.048.240.563.143.5
ViL-Medium [76]50.8/60.1339/26143.764.646.427.947.156.944.666.348.540.763.843.7
Swin-Small [43]59.8/69.1335/354 46.467.050.131.050.160.348.570.253.543.367.346.6
Focal-Small (Ours)61.7/71.2367/40147.367.851.031.650.961.148.870.553.643.867.747.2
ResNeXt101-64x4d [67]95.5/102473/49341.861.544.425.245.454.644.464.948.839.761.942.6
PVT-Large[60]71.1/81.0345/364 43.463.646.126.146.059.544.566.048.340.763.443.7
ViL-Base [76]66.7/76.1443/365 44.765.547.629.948.058.145.767.249.941.364.444.5
Swin-Base 43]98.4/107477/496 45.866.449.129.949.460.348.569.853.243.466.846.9
Focal-Base (Ours)100.8/110.0 514/533 46.967.850.331.950.361.549.070.153.643.767.647.0
", + "bbox": [ + 179, + 88, + 825, + 313 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Focal Transformers show substantial gains across all settings and metrics. Particularly, Focal Transformers brings 0.7-1.7 points of mAP against the current best approach Swin Transformer [43] at comparable settings. Different from the other multi-scale Transformer models, Focal Transformers can simultaneously enable short-range fine-grain and long-range coarse-grain interactions for each visual token, and thus capture richer visual contexts at each layer for better dense predictions. To have more comprehensive comparisons, we train all models using the $3 \\times$ schedule and show the detailed numbers for RetinaNet and Mask R-CNN in Table 4. As we can see, even with the $3 \\times$ schedule, Focal Transformers can still achieve 0.3-1.1 gain over Swin Transformer models in comparable settings. ", + "bbox": [ + 173, + 372, + 825, + 483 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Comparison with large SoTA detection models. We follow Swin Transformers to use HTC [11] as the detection method in that it reported SoTA performance on COCO detection when using Swin Transformer as the backbone. For fair comparison, we also use soft-NMS [5], instaboost [24] and a multi-scale training strategy with the shorter side in range [400, 1400] and the longer side no more than 1600. We train the model using AdamW [44] with base learning rate 1e-4 and weight decay 0.1. The model is trained using the standard $3 \\times$ schedule. The box and mask mAPs on COCO validation set and test-dev are reported in Table 5, where both single-scale evaluation and multi-scale evaluation results are presented. Our Focal-Large model with multi-scale test achieves 58.1 box mAP and 50.9 mask mAP on mini-val set, which is better than the reported numbers for Swin-Large in [43]. When evaluating our model on the test-dev set, it achieves 58.4 box mAP and 51.3 mask mAP, which is slightly better than Swin Transformer. Note that because our model does not include the global self-attention layer used in Swin Transformer at the last stage, it has a smaller model size and fewer FLOPs. More recently, DyHead [17] achieves new SoTA on COCO, when combined with Swin-Large. We replace the Swin-Large model with the Focal-Large model, and use the same $2 \\times$ training schedule as in [17]. We report the box mAPs for both mini-val and test-dev. Focal-Large achieves 58.7 and 59.0 on mini-val and test-dev, respectively. ", + "bbox": [ + 173, + 489, + 825, + 710 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3 Semantic Segmentation ", + "text_level": 1, + "bbox": [ + 174, + 727, + 377, + 741 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In addition to the instance segmentation results, we also evaluate our models on the semantic segmentation task which usually takes high-resolution input images and requires capturing long-range interactions. We benchmark our methods on ADE20K [83]. We use UperNet [65] as the segmentation method and Focal Transformers as the backbones. We train three models as Focal-Tiny, Focal-Small, Focal-Base, respectively. For all the models, we use a standard recipe that sets the input size to $5 1 2 \\times 5 1 2$ and trains the model for 160k iterations with batch size 16. Table 6 shows the comparison results. We see that Focal-Tiny, Focal-Small and Focal-Base models consistently outperform Swin Transformers of the similar size in single-scale and multi-scale mIoUs. ", + "bbox": [ + 173, + 744, + 825, + 856 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Comparison with large SoTA semantic segmentation models. We use the pretrained Focal-Large model as the backbone for semantic segmentation. Follow the setting in [43], we use input image size $6 4 0 \\times 6 4 0$ and train the model for 160k iterations with a batch size of 16. We set the initial learning to 6e-5 and use a polynomial learning rate decay. The weight decay is set to 0.01. For ", + "bbox": [ + 176, + 856, + 823, + 911 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/fbecb97ab8c83eccb4496fc4e3c85a4b9127f6ec402c3956112b387672b7d462.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Method#Param FLOPsmini-valtest-dev
ApbAPmApbApm
X101-64x4d [67]155M1033G 52.346.0
EfficientNet-D7 [54]77M410G54.4-55.1=
GCNet*[7]-1041G51.844.752.345.4
ResNeSt-200 [75]=52.5=53.347.1
Copy-paste [28]185M1440G 55.947.256.047.4
BoTNet-200 [51]-49.7-
SpineNet-190 [22]164M1885G 52.652.8
CenterNet2 [84]-=--56.4=
Swin-L (HTC++) [43]284M1470G 57.149.557.750.2
Swin-L (DyHead)[17]213M965G56.2---
Swin-L† (HTC++) [43]284M58.050.458.751.1
Swin-L† (DyHead) [17]213M58.4-58.7
Swin-L† (QueryInst) [25]-56.1156.1
Focal-L (HTC++) (Ours)265M1165G57.049.9-=
Focal-L (DyHead) (Ours)229M1081G56.4--=
Focal-L† (HTC++) (Ours)265M-58.150.958.451.3
Focal-L† (DyHead) (Ours)229M-58.7-59.0-
", + "bbox": [ + 178, + 95, + 498, + 314 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Table 5: Comparison with state-of-the-art methods on COCO object detection and instance segmentation. The numbers are reported on 5K val set and test-dev. Augmented HTC [11] (denoted by $\\mathrm { H T C + + }$ ) and DyHead [17] are used as the detection methods. † means multi-scale evaluation. ", + "bbox": [ + 173, + 315, + 501, + 400 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/87bb36ea5cbafa44649b82024642f9740f5a663fda0b6f453aa106965b3c0739.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
BackboneMethod#Param FLOPs mIoU +MS
ResNet-101DANet [45]69M1119G45.3
ResNet-101ACNet [26]==45.9
ResNet-101DNL [69]69M1249G46.0
ResNet-101UperNet [65]86M1029G44.9
HRNet-w48 [53]OCRNet [73]71M664G45.7=
ResNeSt-200 [75]DLab.v3+ [12]88M1381G48.4=
Swin-T[43]UperNet [65]60M945G44.545.8
Swin-S [43]UperNet [65]81M1038G47.649.5
Swin-B [43]UperNet [65]121M1188G48.149.7
Twins-SVT-L[15]UperNet [65]133M48.850.2
MiT-B5 [66]SegFormer [66]85M51.051.8
ViT-L/16+ [21]SETR[80]308M50.3=
Swin-L* [43]UperNet [65]234M3230G52.153.5
ViT-L/16 [21]Segmenter [52]334M=51.853.6
Swin-L‡ [43]K-Net [77]==54.3
Swin-L‡ [43]PatchDiverse [29]234M53.154.4
VOLO-D5 [72]UperNet [65]=-54.3
Focal-T (Ours)UperNet [65]62M998G45.847.0
Focal-S (Ours)UperNet [65]85M1130G48.050.0
Focal-B (Ours)UperNet [65]126M1354G49.050.5
Focal-L‡ (Ours)UperNet [65]240M3376G54.055.4
", + "bbox": [ + 521, + 88, + 813, + 320 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Table 6: Comparison with SoTA methods for semantic segmentation on ADE20K [83] val set. Single- and multi-scale evaluations are reported in the last two columns. $^ \\ddag$ means ImageNet-22K is used as the pretraining dataset. ", + "bbox": [ + 517, + 323, + 818, + 406 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/71e84f7ae2932bbde0991a681df51b358ad5e6359c4edd0e3856bdf12c691fef.jpg", + "table_caption": [], + "table_footnote": [ + "Table 7: Impact of different window sizes (WSize). We alter the default size 7 to 14 and observe consistent improvements for both methods. " + ], + "table_body": "
Model W-Size FLOPs Top-1(%) APb APm
Swin-Tiny74.581.2 43.739.8
144.982.1 44.0 40.5
Focal-Tiny74.982.2 44.941.1
145.282.3 45.5 41.5
", + "bbox": [ + 178, + 417, + 485, + 503 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/f9b110f7022a0af916099cc762e9ee21cf07047e230774dd117201440a6f74e6.jpg", + "table_caption": [], + "table_footnote": [ + "Table 8: Impact of window shift (W-Shift) on Swin Transformer and Focal Transformer. Tiny models are used. " + ], + "table_body": "
Model W-Shift Top-1(%) APb APm
Swin-Tiny80.2 81.238.8 43.736.4 39.8
Focal-Tiny82.244.841.0
81.944.941.1
", + "bbox": [ + 531, + 417, + 800, + 503 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "multi-scale evaluation, we use the same scaling ratios [0.5, 0.75, 1.0, 1.25, 1.5, 1.75] as in previous works. The results in Table 6 show that Focal-Large achieves significantly better performance than Swin-Large. In both single-scale and multi-scale evaluations, Focal-Large leads to more than 1 point mIoU improvement, creating new SoTA for semantic segmentation on ADE20K. ", + "bbox": [ + 174, + 559, + 825, + 614 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4 Ablation studies ", + "text_level": 1, + "bbox": [ + 174, + 630, + 326, + 643 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We conduct a series of ablation studies to inspect the model’s capacity from different aspects. We use Focal-Tiny and the image classification and object detection tasks. ", + "bbox": [ + 174, + 648, + 821, + 676 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Effect of varying the window size. We have demonstrated that it is crucial to model both short- and long-range interactions. Thus, a related question is whether increasing the window size helps as it leads to a larger receptive field. Table 7 shows the performance of Swin-Tiny and Focal-Tiny with window sizes 7 and 14. Clearly, a larger window size is beneficial for both methods measured in all three metrics, and Focal-Tiny consistently outperforms Swin-Tiny in both window sizes. Comparing the second and third row, we find that Focal-Tiny outperforms Swin-Tiny even with a smaller window size $( 7 \\nu . s . \\ 1 4 )$ . We suspect that the gain is attributed to our focal attention’s superior capability of capturing long-range dependencies among visual tokens. ", + "bbox": [ + 174, + 683, + 825, + 794 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "The necessity of window shift. In Swin Transformer [43], window shift is proposed to capture crosswindow interactions between two successive layers. In contrast, visual tokens in Focal Transformers can always communicate with each other across windows at both fine- and coarse-grain. Thus, it is interesting to investigate whether adding window shift to Focal Transformers can lead to any improvement. To answer the question, we remove window shift from Swin Transformer while adding it to Focal Transformers. As shown in Table 8, Swin Transformer shows a severe degradation after removing the window shift. However, adding window shift to Focal Transformer hurts classification performance. The result indicates that window shift is unnecessary for Focal Transformers. While in ", + "bbox": [ + 174, + 800, + 825, + 911 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/bba4db737a9b7677713c20afacad8a3124328743b54298c0a2ae2792bcf81178.jpg", + "image_caption": [ + "Figure 5: Ablating Focal-Tiny model by adding local, global and both interactions, respectively. Blue bars are image classification results and orange bars object detection results. This figure is better viewed in color. " + ], + "image_footnote": [], + "bbox": [ + 176, + 92, + 483, + 205 + ], + "page_idx": 9 + }, + { + "type": "table", + "img_path": "images/f441a274a5f6aa85d84879890ce822e6335cb56e3b0d5e7463cbc04311f50dfa.jpg", + "table_caption": [], + "table_footnote": [ + "Table 9: Impact of the change of model depth. We gradually reduce the number of transformer layers at the third stage from original 6 to 4 and further 2. Our Focal Transformers has much slower drop rate than Swin Transformer. " + ], + "table_body": "
Depths Model #Params. FLOPs Top-1(%) APb Apm
2-2-2-2Swin21.23.178.738.2 35.7
Focal21.73.479.940.5 37.6
2-2-4-2Swin24.73.880.241.2 38.1
Focal25.44.181.443.3 39.8
2-2-6-2Swin28.34.581.243.7 39.8
Focal29.14.982.244.8 41.0
", + "bbox": [ + 509, + 92, + 816, + 198 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Swin Transformers, there should always be an even number of layers in each stage for the alternative window shift operation, Focal Transformers do not have such a constraint. ", + "bbox": [ + 173, + 291, + 821, + 319 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Contributions of local and global interactions. To investigate the relative contributions of capturing local fine-grain and global coarse-grain interactions in Focal Transformers, we have developed several variants of Focal-Tiny: a) Focal-Tiny-Window merely performs attention inside each window; b) Focal-Tiny-Local attends the additional fine-grain surrounding tokens and c) Focal-Tiny-Global attends the extra coarse-grain summarized tokens. We train these models using the same setting as Focal-Tiny and report their performance on image classification and object detection using Mask R-CNN $1 \\times$ schedule. As shown in Fig. 5, Focal-Tiny-Window suffers from a significant performance drop on both image classification $8 2 . 2 \\substack { 8 0 . 1 }$ ) and object detection $\\cdot 4 4 . 8 \\mathrm { \\ - } \\to 3 8 . 3$ ). This is expected since the communication across windows is completely cut off at each Transformer layer. After we enable either the local fine-grain or global coarse-grain interactions (middle two columns), we observe significant performance boost. When we combine short- and long-range interactions, we observe additional improvements on both tasks. This implies that these two type of interactions are complementary and both are beneficial to model performance. ", + "bbox": [ + 174, + 325, + 825, + 505 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Model capacity against model depth. Focal attention allows a Transformer model to capture shortand long-range interactions at each Transformer layer. An interesting question is whether Focal Transformers need fewer layers to obtain a similar modeling capacity as the Transformer models that does not use focal attention, such as Swin Transformer. To answer this question, we conduct an experiment by training a series of Swin-Tiny and Focal-Tiny models by varying the number of Transformer layers at stage 3. As shown in Table 9, Focal-Tiny outperforms Swin-Tiny consistently with the same depth. More importantly, using fewer layers, Focal-Tiny can sometimes achieve comparable or even better performance than Swin Transformer. For example, Focal-Tiny with (2-2-4-2) achieves 81.4 on image classification which is better than Swin-Tiny with (2-2-6-2). ", + "bbox": [ + 174, + 511, + 825, + 636 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "5 Conclusion ", + "text_level": 1, + "bbox": [ + 174, + 655, + 299, + 672 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this paper, we have presented a new focal attention mechanism that enables efficient long-range interactions in Vision Transformers. Different from previous works, it performs the local attention at fine-grain and global attention at coarse-grain, providing an effective way of capturing both shortand long-range context with a manageable computational cost. By applying focal attention into a multi-scale Transformer architecture, we propose Focal Transformers as general-purpose backbones for a wide range of dense vision tasks. A comprehensive empirical study shows that our Focal Transformers outperform the SoTA Vision Transformers on a range of vision tasks including image classification, object detection and segmentation. ", + "bbox": [ + 174, + 680, + 825, + 791 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Limitations and future work. Although our experiments show that focal attention can significantly boost the performance on image classification and dense prediction tasks, focal attention does introduce extra computational and memory cost, since each query token needs to attend more (summarized) tokens in addition to tokens inside a window. A cost-effective implementation of Focal Transformer is necessary to make it more applicable to many real-world scenarios. This study focuses on incorporating focal attention into multi-scale Vision Transformers for CV tasks. However, we notice that focal attention is an effective sparse attention mechanism that is widely applicable to all attention-based neural network models that are developed for processing natural language, images, videos etc. This is an exciting future direction. ", + "bbox": [ + 174, + 796, + 825, + 911 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References \n[1] Joshua Ainslie, Santiago Ontanon, Chris Alberti, Philip Pham, Anirudh Ravula, and Sumit Sanghai. Etc: Encoding long and structured data in transformers. arXiv preprint arXiv:2004.08483, 2020. \n[2] Irwan Bello, Barret Zoph, Ashish Vaswani, Jonathon Shlens, and Quoc V Le. Attention augmented convolutional networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3286–3295, 2019. \n[3] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150, 2020. \n[4] Maxim Berman, Hervé Jégou, Andrea Vedaldi, Iasonas Kokkinos, and Matthijs Douze. Multigrain: a unified image embedding for classes and instances. arXiv preprint arXiv:1902.05509, 2019. \n[5] Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S Davis. Soft-nms–improving object detection with one line of code. In Proceedings of the IEEE international conference on computer vision, pages 5561–5569, 2017. \n[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020. \n[7] Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. Gcnet: Non-local networks meet squeezeexcitation networks and beyond. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, pages 0–0, 2019. \n[8] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In European Conference on Computer Vision, pages 213–229. Springer, 2020. \n[9] Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. arXiv preprint arXiv:2104.14294, 2021. \n[10] Shuning Chang, Pichao Wang, Fan Wang, Hao Li, and Jiashi Feng. Augmented transformer with adaptive graph for temporal action proposal generation. arXiv preprint arXiv:2103.16024, 2021. \n[11] Kai Chen, Jiangmiao Pang, Jiaqi Wang, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jianping Shi, Wanli Ouyang, et al. Hybrid task cascade for instance segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4974–4983, 2019. \n[12] Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European conference on computer vision (ECCV), pages 801–818, 2018. \n[13] Xin Chen, Bin Yan, Jiawen Zhu, Dong Wang, Xiaoyun Yang, and Huchuan Lu. Transformer tracking. arXiv preprint arXiv:2103.15436, 2021. \n[14] Xinlei Chen, Saining Xie, and Kaiming He. An empirical study of training self-supervised visual transformers. arXiv preprint arXiv:2104.02057, 2021. \n[15] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting spatial attention design in vision transformers. arXiv preprint arXiv:2104.13840, 2021. \n[16] Xiangxiang Chu, Zhi Tian, Bo Zhang, Xinlong Wang, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Conditional positional encodings for vision transformers. Arxiv preprint 2102.10882, 2021. \n[17] Xiyang Dai, Yinpeng Chen, Bin Xiao, Dongdong Chen, Mengchen Liu, Lu Yuan, and Lei Zhang. Dynamic head: Unifying object detection heads with attentions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7373–7382, 2021. \n[18] Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object detection with transformers. arXiv preprint arXiv:2011.09094, 2020. \n[19] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. \n[20] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. NAACL, 2019. \n[21] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. \n[22] Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Golnaz Ghiasi, Mingxing Tan, Yin Cui, Quoc V Le, and Xiaodan Song. Spinenet: Learning scale-permuted backbone for recognition and localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11592–11601, 2020. \n[23] Haoqi Fan, Bo Xiong, Karttikeya Mangalam, Yanghao Li, Zhicheng Yan, Jitendra Malik, and Christoph Feichtenhofer. Multiscale vision transformers. arXiv preprint arXiv:2104.11227, 2021. \n[24] Hao-Shu Fang, Jianhua Sun, Runzhong Wang, Minghao Gou, Yong-Lu Li, and Cewu Lu. Instaboost: Boosting instance segmentation via probability map guided copy-pasting. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 682–691, 2019. \n[25] Yuxin Fang, Shusheng Yang, Xinggang Wang, Yu Li, Chen Fang, Ying Shan, Bin Feng, and Wenyu Liu. Instances as queries, 2021. \n[26] Jun Fu, Jing Liu, Yuhang Wang, Yong Li, Yongjun Bao, Jinhui Tang, and Hanqing Lu. Adaptive context network for scene parsing. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6748–6757, 2019. \n[27] Jianfeng Gao, Patrick Pantel, Michael Gamon, Xiaodong He, and Li Deng. Modeling interestingness with deep neural networks. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2–13, Doha, Qatar, October 2014. Association for Computational Linguistics. \n[28] Golnaz Ghiasi, Yin Cui, Aravind Srinivas, Rui Qian, Tsung-Yi Lin, Ekin D Cubuk, Quoc V Le, and Barret Zoph. Simple copy-paste is a strong data augmentation method for instance segmentation. arXiv preprint arXiv:2012.07177, 2020. \n[29] Chengyue Gong, Dilin Wang, Meng Li, Vikas Chandra, and Qiang Liu. Vision transformers with patch diversification, 2021. \n[30] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020. \n[31] Kai Han, An Xiao, Enhua Wu, Jianyuan Guo, Chunjing Xu, and Yunhe Wang. Transformer in transformer, 2021. \n[32] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. \n[33] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. \n[34] Elad Hoffer, Tal Ben-Nun, Itay Hubara, Niv Giladi, Torsten Hoefler, and Daniel Soudry. Augment your batch: Improving generalization through instance repetition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8129–8138, 2020. \n[35] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018. \n[36] Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, and Mubarak Shah. Transformers in vision: A survey. arXiv preprint arXiv:2101.01169, 2021. \n[37] Yann LeCun, Yoshua Bengio, et al. Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks, 3361(10):1995, 1995. \n[38] Bing Li, Cheng Zheng, Silvio Giancola, and Bernard Ghanem. Sctn: Sparse convolution-transformer network for scene flow estimation. arXiv preprint arXiv:2105.04447, 2021. \n[39] Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Mei Gao, Bin Xiao, Xiyang Dai, Lu Yuan, and Jianfeng Gao. Efficient self-supervised vision transformers for representation learning. arXiv preprint arXiv:2106.09785, 2021. \n[40] Xiangyu Li, Yonghong Hou, Pichao Wang, Zhimin Gao, Mingliang Xu, and Wanqing Li. Trear: Transformer-based rgb-d egocentric action recognition. arXiv preprint arXiv:2101.03904, 2021. \n[41] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017. \n[42] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, 2014. \n[43] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. \n[44] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. \n[45] Hyeonseob Nam, Jung-Woo Ha, and Jeonghee Kim. Dual attention networks for multimodal reasoning and matching. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 299–307, 2017. \n[46] R. Pappagari, P. Zelasko, J. Villalba, Y. Carmiel, and N. Dehak. Hierarchical transformers for long document classification. In 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 838–844, 2019. \n[47] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR, 2018. \n[48] Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM journal on control and optimization, 30(4):838–855, 1992. \n[49] Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, and Timothy P Lillicrap. Compressive transformers for long-range sequence modelling. arXiv preprint arXiv:1911.05507, 2019. \n[50] Prajit Ramachandran, Niki Parmar, Ashish Vaswani, Irwan Bello, Anselm Levskaya, and Jonathon Shlens. Stand-alone self-attention in vision models. arXiv preprint arXiv:1906.05909, 2019. \n[51] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani. Bottleneck transformers for visual recognition, 2021. \n[52] Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. Segmenter: Transformer for semantic segmentation. arXiv preprint arXiv:2105.05633, 2021. \n[53] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, 2019. \n[54] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International Conference on Machine Learning, pages 6105–6114. PMLR, 2019. \n[55] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. \n[56] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12894–12904, 2021. \n[57] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NeurIPS, 2017. \n[58] Huiyu Wang, Yukun Zhu, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen. Max-deeplab: End-to-end panoptic segmentation with mask transformers. arXiv preprint arXiv:2012.00759, 2020. \n[59] Ning Wang, Wengang Zhou, Jie Wang, and Houqaing Li. Transformer meets tracker: Exploiting temporal context for robust visual tracking. arXiv preprint arXiv:2103.11681, 2021. \n[60] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. arXiv preprint arXiv:2102.12122, 2021. \n[61] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018. \n[62] Yuqing Wang, Zhaoliang Xu, Xinlong Wang, Chunhua Shen, Baoshan Cheng, Hao Shen, and Huaxia Xia. End-to-end video instance segmentation with transformers. arXiv preprint arXiv:2011.14503, 2020. \n[63] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Convolutional block attention module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19, 2018. \n[64] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021. \n[65] Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene understanding. In Proceedings of the European Conference on Computer Vision (ECCV), pages 418–434, 2018. \n[66] Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. Segformer: Simple and efficient design for semantic segmentation with transformers, 2021. \n[67] Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregated residual transformations for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1492–1500, 2017. \n[68] Jianwei Yang, Zhile Ren, Chuang Gan, Hongyuan Zhu, and Devi Parikh. Cross-channel communication networks. In Proceedings of the 33rd International Conference on Neural Information Processing Systems, pages 1297–1306, 2019. \n[69] Minghao Yin, Zhuliang Yao, Yue Cao, Xiu Li, Zheng Zhang, Stephen Lin, and Han Hu. Disentangled non-local neural networks. In European Conference on Computer Vision, pages 191–207. Springer, 2020. \n[70] Kun Yuan, Shaopeng Guo, Ziwei Liu, Aojun Zhou, Fengwei Yu, and Wei Wu. Incorporating convolution designs into visual transformers. arXiv preprint arXiv:2103.11816, 2021. \n[71] Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi, Francis EH Tay, Jiashi Feng, and Shuicheng Yan. Tokens-to-token vit: Training vision transformers from scratch on imagenet. arXiv preprint arXiv:2101.11986, 2021. \n[72] Li Yuan, Qibin Hou, Zihang Jiang, Jiashi Feng, and Shuicheng Yan. Volo: Vision outlooker for visual recognition. arXiv preprint arXiv:2106.13112, 2021. \n[73] Yuhui Yuan, Xilin Chen, and Jingdong Wang. Object-contextual representations for semantic segmentation. arXiv preprint arXiv:1909.11065, 2019. \n[74] Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, et al. Big bird: Transformers for longer sequences. arXiv preprint arXiv:2007.14062, 2020. \n[75] Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Haibin Lin, Zhi Zhang, Yue Sun, Tong He, Jonas Mueller, R Manmatha, et al. Resnest: Split-attention networks. arXiv preprint arXiv:2004.08955, 2020. \n[76] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multiscale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint arXiv:2103.15358, 2021. \n[77] Wenwei Zhang, Jiangmiao Pang, Kai Chen, and Chen Change Loy. K-net: Towards unified image segmentation, 2021. \n[78] Jiaojiao Zhao, Xinyu Li, Chunhui Liu, Shuai Bing, Hao Chen, Cees GM Snoek, and Joseph Tighe. Tuber: Tube-transformer for action detection. arXiv preprint arXiv:2104.00969, 2021. \n[79] Minghang Zheng, Peng Gao, Xiaogang Wang, Hongsheng Li, and Hao Dong. End-to-end object detection with adaptive clustering transformer. arXiv preprint arXiv:2011.09315, 2020. \n[80] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. arXiv preprint arXiv:2012.15840, 2020. \n[81] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6881–6890, 2021. \n[82] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 13001–13008, 2020. \n[83] Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing through ade20k dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 633–641, 2017. \n[84] Xingyi Zhou, Vladlen Koltun, and Philipp Krähenbühl. Probabilistic two-stage detection. arXiv preprint arXiv:2103.07461, 2021. \n[85] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. ", + "bbox": [ + 173, + 80, + 828, + 915 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 51, + 828, + 915 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 60, + 828, + 919 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 50, + 828, + 919 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 92, + 826, + 237 + ], + "page_idx": 14 + } +] \ No newline at end of file diff --git a/parse/train/2zCRcTafea/2zCRcTafea_middle.json b/parse/train/2zCRcTafea/2zCRcTafea_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..97328524eb945a841255f17a92769ac2d82d301b --- /dev/null +++ b/parse/train/2zCRcTafea/2zCRcTafea_middle.json @@ -0,0 +1,39923 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 132, + 97, + 480, + 137 + ], + "lines": [ + { + "bbox": [ + 129, + 97, + 481, + 118 + ], + "spans": [ + { + "bbox": [ + 129, + 97, + 481, + 118 + ], + "score": 1.0, + "content": "Focal Attention for Long-Range Interactions in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 228, + 117, + 383, + 137 + ], + "spans": [ + { + "bbox": [ + 228, + 117, + 383, + 137 + ], + "score": 1.0, + "content": "Vision Transformers", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 139, + 174, + 473, + 222 + ], + "lines": [ + { + "bbox": [ + 137, + 175, + 473, + 189 + ], + "spans": [ + { + "bbox": [ + 137, + 175, + 255, + 189 + ], + "score": 1.0, + "content": "Jianwei Yang1 Chunyuan", + "type": "text" + }, + { + "bbox": [ + 255, + 175, + 271, + 186 + ], + "score": 0.32, + "content": "\\mathbf { L i } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 175, + 473, + 189 + ], + "score": 1.0, + "content": "Pengchuan Zhang1 Xiyang Dai2 Bin Xiao2", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 247, + 186, + 363, + 200 + ], + "spans": [ + { + "bbox": [ + 247, + 186, + 363, + 200 + ], + "score": 1.0, + "content": "Lu Yuan2 Jianfeng Gao1", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 188, + 196, + 424, + 212 + ], + "spans": [ + { + "bbox": [ + 188, + 196, + 400, + 212 + ], + "score": 1.0, + "content": "1Microsoft Research at Redmond, 2Microsoft Cloud", + "type": "text" + }, + { + "bbox": [ + 400, + 200, + 409, + 209 + ], + "score": 0.56, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 196, + 424, + 212 + ], + "score": 1.0, + "content": "AI", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 139, + 210, + 472, + 223 + ], + "spans": [ + { + "bbox": [ + 139, + 210, + 472, + 223 + ], + "score": 1.0, + "content": "{jianwyan,chunyl,penzhan,xidai,bixi,luyuan,jfgao}@microsoft.com", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 283, + 249, + 328, + 262 + ], + "lines": [ + { + "bbox": [ + 282, + 249, + 330, + 263 + ], + "spans": [ + { + "bbox": [ + 282, + 249, + 330, + 263 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 142, + 275, + 468, + 559 + ], + "lines": [ + { + "bbox": [ + 142, + 275, + 469, + 287 + ], + "spans": [ + { + "bbox": [ + 142, + 275, + 469, + 287 + ], + "score": 1.0, + "content": "Recently, Vision Transformer and its variants have shown great promise on various", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 285, + 470, + 298 + ], + "spans": [ + { + "bbox": [ + 141, + 285, + 470, + 298 + ], + "score": 1.0, + "content": "computer vision tasks. The ability of capturing local and global visual dependencies", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 296, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 470, + 310 + ], + "score": 1.0, + "content": "through self-attention is the key to its success. However, this also brings challenges", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 308, + 470, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 470, + 321 + ], + "score": 1.0, + "content": "due to quadratic computational overhead, especially for the high-resolution vision", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 319, + 470, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 319, + 470, + 331 + ], + "score": 1.0, + "content": "tasks (e.g., object detection). Many recent works have attempted to reduce the cost", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 330, + 469, + 342 + ], + "spans": [ + { + "bbox": [ + 142, + 330, + 469, + 342 + ], + "score": 1.0, + "content": "and improve model performance by applying either coarse-grained global attention", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 340, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 469, + 354 + ], + "score": 1.0, + "content": "or fine-grained local attention. However, both approaches cripple the modeling", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 351, + 470, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 365 + ], + "score": 1.0, + "content": "power of the original self-attention mechanism of multi-layer Transformers, leading", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "score": 1.0, + "content": "to sub-optimal solutions. In this paper, we present focal attention, a new attention", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 373, + 470, + 386 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 470, + 386 + ], + "score": 1.0, + "content": "mechanism that incorporates both fine-grained local and coarse-grained global", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 383, + 470, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 383, + 470, + 397 + ], + "score": 1.0, + "content": "interactions. In this new mechanism, each token attends its closest surrounding", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 395, + 470, + 407 + ], + "spans": [ + { + "bbox": [ + 141, + 395, + 470, + 407 + ], + "score": 1.0, + "content": "tokens at fine granularity and the tokens far away at coarse granularity, and thus can", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 406, + 470, + 418 + ], + "spans": [ + { + "bbox": [ + 141, + 406, + 470, + 418 + ], + "score": 1.0, + "content": "capture both short- and long-range visual dependencies efficiently and effectively.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 416, + 470, + 428 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 470, + 428 + ], + "score": 1.0, + "content": "With focal attention, we build a new variant of Vision Transformer models, called", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 428, + 470, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 428, + 470, + 439 + ], + "score": 1.0, + "content": "Focal Transformers, which achieve superior performance over the state-of-the-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "art (SoTA) Vision Transformers on a range of public image classification and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 450, + 469, + 461 + ], + "spans": [ + { + "bbox": [ + 142, + 450, + 469, + 461 + ], + "score": 1.0, + "content": "object detection benchmarks. In particular, our Focal Transformer models with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 460, + 469, + 472 + ], + "spans": [ + { + "bbox": [ + 141, + 460, + 331, + 472 + ], + "score": 1.0, + "content": "a moderate size of 51.1M and a large size of", + "type": "text" + }, + { + "bbox": [ + 331, + 460, + 360, + 471 + ], + "score": 0.28, + "content": "8 9 . 8 \\mathbf { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 460, + 394, + 472 + ], + "score": 1.0, + "content": "achieve", + "type": "text" + }, + { + "bbox": [ + 394, + 460, + 422, + 470 + ], + "score": 0.86, + "content": "\\mathbf { 8 3 . 6 \\% }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 460, + 441, + 472 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 460, + 469, + 471 + ], + "score": 0.86, + "content": "\\mathbf { 8 4 . 0 \\% }", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 471, + 470, + 483 + ], + "spans": [ + { + "bbox": [ + 141, + 471, + 390, + 483 + ], + "score": 1.0, + "content": "Top-1 accuracy, respectively, on ImageNet classification at", + "type": "text" + }, + { + "bbox": [ + 390, + 471, + 435, + 482 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 471, + 470, + 483 + ], + "score": 1.0, + "content": ". When", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 482, + 469, + 493 + ], + "spans": [ + { + "bbox": [ + 141, + 482, + 469, + 493 + ], + "score": 1.0, + "content": "employed as the backbones, Focal Transformers achieve consistent and substantial", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "spans": [ + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "score": 1.0, + "content": "improvements over the current SoTA Swin Transformers [43] across 6 different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 142, + 504, + 470, + 515 + ], + "spans": [ + { + "bbox": [ + 142, + 504, + 470, + 515 + ], + "score": 1.0, + "content": "object detection methods. Our largest Focal Transformer yields 58.7/59.0 box", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 515, + 469, + 526 + ], + "spans": [ + { + "bbox": [ + 142, + 515, + 469, + 526 + ], + "score": 1.0, + "content": "mAPs and 50.9/51.3 mask mAPs on COCO mini-val/test-dev, and 55.4 mIoU on", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 526, + 470, + 538 + ], + "spans": [ + { + "bbox": [ + 141, + 526, + 470, + 538 + ], + "score": 1.0, + "content": "ADE20K for semantic segmentation, creating new SoTA on three of the most", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 536, + 470, + 549 + ], + "spans": [ + { + "bbox": [ + 141, + 536, + 470, + 549 + ], + "score": 1.0, + "content": "challenging computer vision tasks. Our code is available at: https://github.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 142, + 548, + 309, + 560 + ], + "spans": [ + { + "bbox": [ + 142, + 548, + 309, + 560 + ], + "score": 1.0, + "content": "com/microsoft/Focal-Transformer.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 579, + 191, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 192, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 192, + 595 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 506, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 598, + 507, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 507, + 612 + ], + "score": 1.0, + "content": "Nowadays, Transformer [57] has become a prevalent model architecture in natural language pro-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 610, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 506, + 622 + ], + "score": 1.0, + "content": "cessing (NLP) [20, 6]. In the light of its success in NLP, there is an increasing effort on adapt-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 621, + 507, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 507, + 633 + ], + "score": 1.0, + "content": "ing it to computer vision (CV) [47, 50]. Since its promise firstly demonstrated in Vision Trans-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 632, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 506, + 643 + ], + "score": 1.0, + "content": "former (ViT) [21], we have witnessed a flourish of full-Transformer models for image classifica-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "score": 1.0, + "content": "tion [55, 60, 64, 43, 76, 56], object detection [8, 85, 79, 18] and semantic segmentation [58, 62].", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 653, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 507, + 666 + ], + "score": 1.0, + "content": "Beyond these static image tasks, it has also been applied on various temporal understanding tasks,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 664, + 473, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 473, + 677 + ], + "score": 1.0, + "content": "such as action recognition [40, 78, 10], object tracking [13, 59], scene flow estimation [38].", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 108, + 680, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 505, + 694 + ], + "score": 1.0, + "content": "The self-attention mechanism is arguably the key component that differentiates Transformers from", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "the widely used convolutional neural networks (CNNs) [37] in computer vision. At each Transformer", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 703, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 715 + ], + "score": 1.0, + "content": "layer, self-attention enables global content-dependent interactions among different image regions for", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 731, + 385, + 741 + ], + "lines": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "spans": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 132, + 97, + 480, + 137 + ], + "lines": [ + { + "bbox": [ + 129, + 97, + 481, + 118 + ], + "spans": [ + { + "bbox": [ + 129, + 97, + 481, + 118 + ], + "score": 1.0, + "content": "Focal Attention for Long-Range Interactions in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 228, + 117, + 383, + 137 + ], + "spans": [ + { + "bbox": [ + 228, + 117, + 383, + 137 + ], + "score": 1.0, + "content": "Vision Transformers", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 139, + 174, + 473, + 222 + ], + "lines": [ + { + "bbox": [ + 137, + 175, + 473, + 189 + ], + "spans": [ + { + "bbox": [ + 137, + 175, + 255, + 189 + ], + "score": 1.0, + "content": "Jianwei Yang1 Chunyuan", + "type": "text" + }, + { + "bbox": [ + 255, + 175, + 271, + 186 + ], + "score": 0.32, + "content": "\\mathbf { L i } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 175, + 473, + 189 + ], + "score": 1.0, + "content": "Pengchuan Zhang1 Xiyang Dai2 Bin Xiao2", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 247, + 186, + 363, + 200 + ], + "spans": [ + { + "bbox": [ + 247, + 186, + 363, + 200 + ], + "score": 1.0, + "content": "Lu Yuan2 Jianfeng Gao1", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 188, + 196, + 424, + 212 + ], + "spans": [ + { + "bbox": [ + 188, + 196, + 400, + 212 + ], + "score": 1.0, + "content": "1Microsoft Research at Redmond, 2Microsoft Cloud", + "type": "text" + }, + { + "bbox": [ + 400, + 200, + 409, + 209 + ], + "score": 0.56, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 196, + 424, + 212 + ], + "score": 1.0, + "content": "AI", + "type": "text" + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 139, + 210, + 472, + 223 + ], + "spans": [ + { + "bbox": [ + 139, + 210, + 472, + 223 + ], + "score": 1.0, + "content": "{jianwyan,chunyl,penzhan,xidai,bixi,luyuan,jfgao}@microsoft.com", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 3.5, + "bbox_fs": [ + 137, + 175, + 473, + 223 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 249, + 328, + 262 + ], + "lines": [ + { + "bbox": [ + 282, + 249, + 330, + 263 + ], + "spans": [ + { + "bbox": [ + 282, + 249, + 330, + 263 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 142, + 275, + 468, + 559 + ], + "lines": [ + { + "bbox": [ + 142, + 275, + 469, + 287 + ], + "spans": [ + { + "bbox": [ + 142, + 275, + 469, + 287 + ], + "score": 1.0, + "content": "Recently, Vision Transformer and its variants have shown great promise on various", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 285, + 470, + 298 + ], + "spans": [ + { + "bbox": [ + 141, + 285, + 470, + 298 + ], + "score": 1.0, + "content": "computer vision tasks. The ability of capturing local and global visual dependencies", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 296, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 470, + 310 + ], + "score": 1.0, + "content": "through self-attention is the key to its success. However, this also brings challenges", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 308, + 470, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 470, + 321 + ], + "score": 1.0, + "content": "due to quadratic computational overhead, especially for the high-resolution vision", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 319, + 470, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 319, + 470, + 331 + ], + "score": 1.0, + "content": "tasks (e.g., object detection). Many recent works have attempted to reduce the cost", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 330, + 469, + 342 + ], + "spans": [ + { + "bbox": [ + 142, + 330, + 469, + 342 + ], + "score": 1.0, + "content": "and improve model performance by applying either coarse-grained global attention", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 340, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 469, + 354 + ], + "score": 1.0, + "content": "or fine-grained local attention. However, both approaches cripple the modeling", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 351, + 470, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 365 + ], + "score": 1.0, + "content": "power of the original self-attention mechanism of multi-layer Transformers, leading", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "score": 1.0, + "content": "to sub-optimal solutions. In this paper, we present focal attention, a new attention", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 373, + 470, + 386 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 470, + 386 + ], + "score": 1.0, + "content": "mechanism that incorporates both fine-grained local and coarse-grained global", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 383, + 470, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 383, + 470, + 397 + ], + "score": 1.0, + "content": "interactions. In this new mechanism, each token attends its closest surrounding", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 395, + 470, + 407 + ], + "spans": [ + { + "bbox": [ + 141, + 395, + 470, + 407 + ], + "score": 1.0, + "content": "tokens at fine granularity and the tokens far away at coarse granularity, and thus can", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 406, + 470, + 418 + ], + "spans": [ + { + "bbox": [ + 141, + 406, + 470, + 418 + ], + "score": 1.0, + "content": "capture both short- and long-range visual dependencies efficiently and effectively.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 416, + 470, + 428 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 470, + 428 + ], + "score": 1.0, + "content": "With focal attention, we build a new variant of Vision Transformer models, called", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 428, + 470, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 428, + 470, + 439 + ], + "score": 1.0, + "content": "Focal Transformers, which achieve superior performance over the state-of-the-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "art (SoTA) Vision Transformers on a range of public image classification and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 450, + 469, + 461 + ], + "spans": [ + { + "bbox": [ + 142, + 450, + 469, + 461 + ], + "score": 1.0, + "content": "object detection benchmarks. In particular, our Focal Transformer models with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 460, + 469, + 472 + ], + "spans": [ + { + "bbox": [ + 141, + 460, + 331, + 472 + ], + "score": 1.0, + "content": "a moderate size of 51.1M and a large size of", + "type": "text" + }, + { + "bbox": [ + 331, + 460, + 360, + 471 + ], + "score": 0.28, + "content": "8 9 . 8 \\mathbf { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 460, + 394, + 472 + ], + "score": 1.0, + "content": "achieve", + "type": "text" + }, + { + "bbox": [ + 394, + 460, + 422, + 470 + ], + "score": 0.86, + "content": "\\mathbf { 8 3 . 6 \\% }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 460, + 441, + 472 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 460, + 469, + 471 + ], + "score": 0.86, + "content": "\\mathbf { 8 4 . 0 \\% }", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 471, + 470, + 483 + ], + "spans": [ + { + "bbox": [ + 141, + 471, + 390, + 483 + ], + "score": 1.0, + "content": "Top-1 accuracy, respectively, on ImageNet classification at", + "type": "text" + }, + { + "bbox": [ + 390, + 471, + 435, + 482 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 471, + 470, + 483 + ], + "score": 1.0, + "content": ". When", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 482, + 469, + 493 + ], + "spans": [ + { + "bbox": [ + 141, + 482, + 469, + 493 + ], + "score": 1.0, + "content": "employed as the backbones, Focal Transformers achieve consistent and substantial", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "spans": [ + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "score": 1.0, + "content": "improvements over the current SoTA Swin Transformers [43] across 6 different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 142, + 504, + 470, + 515 + ], + "spans": [ + { + "bbox": [ + 142, + 504, + 470, + 515 + ], + "score": 1.0, + "content": "object detection methods. Our largest Focal Transformer yields 58.7/59.0 box", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 515, + 469, + 526 + ], + "spans": [ + { + "bbox": [ + 142, + 515, + 469, + 526 + ], + "score": 1.0, + "content": "mAPs and 50.9/51.3 mask mAPs on COCO mini-val/test-dev, and 55.4 mIoU on", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 526, + 470, + 538 + ], + "spans": [ + { + "bbox": [ + 141, + 526, + 470, + 538 + ], + "score": 1.0, + "content": "ADE20K for semantic segmentation, creating new SoTA on three of the most", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 536, + 470, + 549 + ], + "spans": [ + { + "bbox": [ + 141, + 536, + 470, + 549 + ], + "score": 1.0, + "content": "challenging computer vision tasks. Our code is available at: https://github.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 142, + 548, + 309, + 560 + ], + "spans": [ + { + "bbox": [ + 142, + 548, + 309, + 560 + ], + "score": 1.0, + "content": "com/microsoft/Focal-Transformer.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 19.5, + "bbox_fs": [ + 141, + 275, + 470, + 560 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 579, + 191, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 192, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 192, + 595 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 506, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 598, + 507, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 507, + 612 + ], + "score": 1.0, + "content": "Nowadays, Transformer [57] has become a prevalent model architecture in natural language pro-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 610, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 506, + 622 + ], + "score": 1.0, + "content": "cessing (NLP) [20, 6]. In the light of its success in NLP, there is an increasing effort on adapt-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 621, + 507, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 507, + 633 + ], + "score": 1.0, + "content": "ing it to computer vision (CV) [47, 50]. Since its promise firstly demonstrated in Vision Trans-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 632, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 506, + 643 + ], + "score": 1.0, + "content": "former (ViT) [21], we have witnessed a flourish of full-Transformer models for image classifica-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "score": 1.0, + "content": "tion [55, 60, 64, 43, 76, 56], object detection [8, 85, 79, 18] and semantic segmentation [58, 62].", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 653, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 507, + 666 + ], + "score": 1.0, + "content": "Beyond these static image tasks, it has also been applied on various temporal understanding tasks,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 664, + 473, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 473, + 677 + ], + "score": 1.0, + "content": "such as action recognition [40, 78, 10], object tracking [13, 59], scene flow estimation [38].", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 598, + 507, + 677 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 680, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 505, + 694 + ], + "score": 1.0, + "content": "The self-attention mechanism is arguably the key component that differentiates Transformers from", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "the widely used convolutional neural networks (CNNs) [37] in computer vision. At each Transformer", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 703, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 715 + ], + "score": 1.0, + "content": "layer, self-attention enables global content-dependent interactions among different image regions for", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 225, + 507, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 507, + 238 + ], + "score": 1.0, + "content": "modeling short- and long-range dependencies, respectively. Through the visualization of full self-", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "attention results1, we indeed observe that self-attention learns to attend local surroundings (like CNNs)", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 245, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 261 + ], + "score": 1.0, + "content": "and the global contexts at the same time, as illustrated in Fig. 1 (Left). Nevertheless, when dealing", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 257, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 506, + 271 + ], + "score": 1.0, + "content": "with high-resolution vision tasks such as object detection or segmentation, an efficient implementation", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "score": 1.0, + "content": "of a global and fine-grained self-attention becomes non-trivial due to the quadratic computational cost", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "score": 1.0, + "content": "with respect to the number of tokens in feature maps. Recent works have alternatively exploited either", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 104, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "a coarse-grained global self-attention [60, 64] or a fine-grained local self-attention [43, 76, 56], for the", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 301, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 314 + ], + "score": 1.0, + "content": "sake of reducing the computational cost. However, both approaches cripple the power of the original", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 312, + 493, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 493, + 325 + ], + "score": 1.0, + "content": "full self-attention i.e., the ability to simultaneously capture local and global visual dependencies.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 680, + 505, + 715 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 76, + 500, + 177 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 76, + 500, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 76, + 500, + 177 + ], + "spans": [ + { + "bbox": [ + 112, + 76, + 500, + 177 + ], + "score": 0.97, + "type": "image", + "image_path": "621f200345a7ebe619be7e8a0aedd96409fbf6fca08314c5e32fce4ac604fb01.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 76, + 500, + 109.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 109.66666666666666, + 500, + 143.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 143.33333333333331, + 500, + 176.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 180, + 502, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 179, + 504, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 504, + 192 + ], + "score": 1.0, + "content": "Figure 1: Left: Visualization of the attention maps of the three heads at the given query patch (blue)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 191, + 504, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 504, + 203 + ], + "score": 1.0, + "content": "in the first layer of the DeiT-Tiny model [55]. Right: An illustrative depiction of focal attention", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 201, + 493, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 493, + 215 + ], + "score": 1.0, + "content": "mechanism. Three granularity levels are used to compose the attention region for the blue query.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 507, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 507, + 238 + ], + "score": 1.0, + "content": "modeling short- and long-range dependencies, respectively. Through the visualization of full self-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "attention results1, we indeed observe that self-attention learns to attend local surroundings (like CNNs)", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 245, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 261 + ], + "score": 1.0, + "content": "and the global contexts at the same time, as illustrated in Fig. 1 (Left). Nevertheless, when dealing", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 257, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 506, + 271 + ], + "score": 1.0, + "content": "with high-resolution vision tasks such as object detection or segmentation, an efficient implementation", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "score": 1.0, + "content": "of a global and fine-grained self-attention becomes non-trivial due to the quadratic computational cost", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "score": 1.0, + "content": "with respect to the number of tokens in feature maps. Recent works have alternatively exploited either", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 104, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "a coarse-grained global self-attention [60, 64] or a fine-grained local self-attention [43, 76, 56], for the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 301, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 314 + ], + "score": 1.0, + "content": "sake of reducing the computational cost. However, both approaches cripple the power of the original", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 312, + 493, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 493, + 325 + ], + "score": 1.0, + "content": "full self-attention i.e., the ability to simultaneously capture local and global visual dependencies.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 329, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 341 + ], + "score": 1.0, + "content": "In this paper, we present a new attention mechanism to capture both short- and long-range interactions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "in Transformer layers for high-resolution input images. Considering that the visual dependencies", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "between the nearby (local) regions are usually much stronger than the dependencies between the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "regions that are far away, we perform the fine-grained attention only in local regions while the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 371, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 505, + 386 + ], + "score": 1.0, + "content": "coarse-grained attention globally. As depicted in Fig. 1 (Right), a query token in the feature map", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "attends its closest local surroundings at the finest granularity as itself. However, when it goes to the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "regions far away, it attends to summarized tokens to capture coarse-grained visual dependencies. We", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 417 + ], + "score": 1.0, + "content": "call this new mechanism focal attention, as each token attends the others in a focal manner. We will", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "show in this study that focal attention allows to effectively model visual dependencies among all", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "regions covering the whole high-resolution feature maps while introducing much less number of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 438, + 415, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 415, + 450 + ], + "score": 1.0, + "content": "tokens in the computation than that in the standard self-attention mechanism.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "Equipped with focal attention, a series of Focal Transformers are developed and validated via a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 466, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 506, + 478 + ], + "score": 1.0, + "content": "comprehensive empirical study across three core vision tasks, including image classification, object", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "detection and segmentation. Results show that Focal Transformers consistently outperform the SoTA", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "score": 1.0, + "content": "Vision Transformers across various settings (i.e., in model sizes and complexities). Notably, the small", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 313, + 510 + ], + "score": 1.0, + "content": "Focal Transformer with 51.1M parameters achieves", + "type": "text" + }, + { + "bbox": [ + 313, + 498, + 340, + 509 + ], + "score": 0.87, + "content": "8 3 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "top-1 accuracy on ImageNet-1K, and the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 508, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 279, + 522 + ], + "score": 1.0, + "content": "base model with 89.8M parameters obtains", + "type": "text" + }, + { + "bbox": [ + 279, + 509, + 306, + 519 + ], + "score": 0.86, + "content": "8 4 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 508, + 506, + 522 + ], + "score": 1.0, + "content": "top-1 accuracy. In the fine-tuning experiments for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "object detection, Focal Transformers consistently outperform the SoTA Swin Transformers [43] across", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "score": 1.0, + "content": "six popular object detection methods. Our largest Focal Transformer model achieves 59.0 box mAP", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "and 51.3 mask mAP on COCO test-dev for object detection and instance segmentation, respectively,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "and 55.4 mIoU on ADE20K for semantic segmentation. These results demonstrate that focal attention", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 419, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 419, + 576 + ], + "score": 1.0, + "content": "is highly effective in modeling the global interactions in Vision Transformers.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 31 + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 194, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 196, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 196, + 606 + ], + "score": 1.0, + "content": "2 Related work", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 505, + 624 + ], + "score": 1.0, + "content": "Vision Transformers. Vision Transformer (ViT) is first introduced in [21]. It applies a standard", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 623, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 635 + ], + "score": 1.0, + "content": "Transformer, originally developed for NLP [57], to encode an image by analogously splitting the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "image into a sequence of visual tokens. It has demonstrated superior performance to CNNs such", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "as ResNet [33] on multiple image classification benchmarks, when trained with sufficient data [21]", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 656, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 505, + 668 + ], + "score": 1.0, + "content": "and carefully designed data augmentation and regularization methods [55]. The results thus inspire", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 667, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 679 + ], + "score": 1.0, + "content": "researchers to explore the applications of ViT on various vision tasks beyond image classification,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 678, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 506, + 689 + ], + "score": 1.0, + "content": "such as self-supervised learning [14, 9, 39], object detection [8, 85, 79, 18] and semantic segmenta-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "tion [58, 62, 81]. There are also increasing number of studies for improving ViT via data-efficient", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 115, + 712, + 487, + 722 + ], + "lines": [ + { + "bbox": [ + 120, + 710, + 489, + 723 + ], + "spans": [ + { + "bbox": [ + 120, + 710, + 489, + 723 + ], + "score": 1.0, + "content": "1DeiT-Tiny model, checkpoint downloaded from https://github.com/facebookresearch/deit.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 76, + 500, + 177 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 76, + 500, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 76, + 500, + 177 + ], + "spans": [ + { + "bbox": [ + 112, + 76, + 500, + 177 + ], + "score": 0.97, + "type": "image", + "image_path": "621f200345a7ebe619be7e8a0aedd96409fbf6fca08314c5e32fce4ac604fb01.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 76, + 500, + 109.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 109.66666666666666, + 500, + 143.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 143.33333333333331, + 500, + 176.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 180, + 502, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 179, + 504, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 504, + 192 + ], + "score": 1.0, + "content": "Figure 1: Left: Visualization of the attention maps of the three heads at the given query patch (blue)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 191, + 504, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 504, + 203 + ], + "score": 1.0, + "content": "in the first layer of the DeiT-Tiny model [55]. Right: An illustrative depiction of focal attention", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 201, + 493, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 493, + 215 + ], + "score": 1.0, + "content": "mechanism. Three granularity levels are used to compose the attention region for the blue query.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 324 + ], + "lines": [], + "index": 10, + "bbox_fs": [ + 104, + 225, + 507, + 325 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 329, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 341 + ], + "score": 1.0, + "content": "In this paper, we present a new attention mechanism to capture both short- and long-range interactions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "in Transformer layers for high-resolution input images. Considering that the visual dependencies", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "between the nearby (local) regions are usually much stronger than the dependencies between the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "regions that are far away, we perform the fine-grained attention only in local regions while the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 371, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 505, + 386 + ], + "score": 1.0, + "content": "coarse-grained attention globally. As depicted in Fig. 1 (Right), a query token in the feature map", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "attends its closest local surroundings at the finest granularity as itself. However, when it goes to the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "regions far away, it attends to summarized tokens to capture coarse-grained visual dependencies. We", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 417 + ], + "score": 1.0, + "content": "call this new mechanism focal attention, as each token attends the others in a focal manner. We will", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "show in this study that focal attention allows to effectively model visual dependencies among all", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "regions covering the whole high-resolution feature maps while introducing much less number of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 438, + 415, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 415, + 450 + ], + "score": 1.0, + "content": "tokens in the computation than that in the standard self-attention mechanism.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 329, + 506, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "Equipped with focal attention, a series of Focal Transformers are developed and validated via a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 466, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 506, + 478 + ], + "score": 1.0, + "content": "comprehensive empirical study across three core vision tasks, including image classification, object", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "detection and segmentation. Results show that Focal Transformers consistently outperform the SoTA", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "score": 1.0, + "content": "Vision Transformers across various settings (i.e., in model sizes and complexities). Notably, the small", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 313, + 510 + ], + "score": 1.0, + "content": "Focal Transformer with 51.1M parameters achieves", + "type": "text" + }, + { + "bbox": [ + 313, + 498, + 340, + 509 + ], + "score": 0.87, + "content": "8 3 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "top-1 accuracy on ImageNet-1K, and the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 508, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 279, + 522 + ], + "score": 1.0, + "content": "base model with 89.8M parameters obtains", + "type": "text" + }, + { + "bbox": [ + 279, + 509, + 306, + 519 + ], + "score": 0.86, + "content": "8 4 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 508, + 506, + 522 + ], + "score": 1.0, + "content": "top-1 accuracy. In the fine-tuning experiments for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "object detection, Focal Transformers consistently outperform the SoTA Swin Transformers [43] across", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "score": 1.0, + "content": "six popular object detection methods. Our largest Focal Transformer model achieves 59.0 box mAP", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "and 51.3 mask mAP on COCO test-dev for object detection and instance segmentation, respectively,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "and 55.4 mIoU on ADE20K for semantic segmentation. These results demonstrate that focal attention", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 419, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 419, + 576 + ], + "score": 1.0, + "content": "is highly effective in modeling the global interactions in Vision Transformers.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 454, + 506, + 576 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 194, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 196, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 196, + 606 + ], + "score": 1.0, + "content": "2 Related work", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 505, + 624 + ], + "score": 1.0, + "content": "Vision Transformers. Vision Transformer (ViT) is first introduced in [21]. It applies a standard", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 623, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 635 + ], + "score": 1.0, + "content": "Transformer, originally developed for NLP [57], to encode an image by analogously splitting the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "image into a sequence of visual tokens. It has demonstrated superior performance to CNNs such", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "as ResNet [33] on multiple image classification benchmarks, when trained with sufficient data [21]", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 656, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 505, + 668 + ], + "score": 1.0, + "content": "and carefully designed data augmentation and regularization methods [55]. The results thus inspire", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 667, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 679 + ], + "score": 1.0, + "content": "researchers to explore the applications of ViT on various vision tasks beyond image classification,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 678, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 506, + 689 + ], + "score": 1.0, + "content": "such as self-supervised learning [14, 9, 39], object detection [8, 85, 79, 18] and semantic segmenta-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "tion [58, 62, 81]. There are also increasing number of studies for improving ViT via data-efficient", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "training [55], improved patch embedding/encoding [16, 71, 31], integrating convolutional projections", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "into transformers [64, 70], and using multi-scale architectures and efficient self-attention mechanisms", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "for high-resolution vision tasks [60, 64, 43, 76, 15]. Recent surveys include [36, 30, 36]. This paper", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 311, + 491, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 491, + 324 + ], + "score": 1.0, + "content": "focuses on improving the self-attention mechanism of ViT for encoding high-resolution images.", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 612, + 506, + 701 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 72, + 500, + 217 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 72, + 500, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 72, + 500, + 217 + ], + "spans": [ + { + "bbox": [ + 110, + 72, + 500, + 217 + ], + "score": 0.969, + "type": "image", + "image_path": "895ff52491908f44661414bd57eca7fc36f34ebd4f7c17cd1d28f3b83da587af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 72, + 500, + 120.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 120.33333333333334, + 500, + 168.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 168.66666666666669, + 500, + 217.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 231, + 504, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "score": 1.0, + "content": "Figure 2: Model architecture for our Focal Transformers. As highlighted in light blue boxes, our", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 242, + 403, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 403, + 255 + ], + "score": 1.0, + "content": "main innovation is the proposed focal attention in each Transformer layer.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 505, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "training [55], improved patch embedding/encoding [16, 71, 31], integrating convolutional projections", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "into transformers [64, 70], and using multi-scale architectures and efficient self-attention mechanisms", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "for high-resolution vision tasks [60, 64, 43, 76, 15]. Recent surveys include [36, 30, 36]. This paper", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 311, + 491, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 491, + 324 + ], + "score": 1.0, + "content": "focuses on improving the self-attention mechanism of ViT for encoding high-resolution images.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "score": 1.0, + "content": "Efficient global and local self-attention. In many real-world tasks, Transformers need to cope", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "with a large number of input tokens, such as long documents in NLP and high-resolution images", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "in computer vision (CV). Recently, many efficient self-attention mechanisms have been proposed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "to deal with the quadratic computational and memory cost incurred by the standard self-attention", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "mechanism. On one hand, a number of works in both NLP and CV resort to coarse-grained global", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "score": 1.0, + "content": "self-attention (i.e., attending the down-sampled or summarized tokens) to capture the long-range", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "interactions [49, 46, 60, 64, 31, 23]. Although this approach improves the model efficiency, it", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "loses the detailed context information surrounding the query tokens. On the other hand, to make", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "the computational cost manageable, various local fine-grained attention mechanism (i.e., attending", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "neighboring tokens within a pre-set window size) are used for both NLP [3, 74, 1] and CV [56, 43, 76].", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "In this paper, we argue that both global and local attentions are important for model performance.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "This is also validated by some recent studies that aim to improve CNNs by incorporating ways of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 457, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 506, + 472 + ], + "score": 1.0, + "content": "modeling global attentions [35, 63, 61, 68, 2, 7, 51]. The standard self-attention mechanism used", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 470, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 506, + 483 + ], + "score": 1.0, + "content": "by ViT can indeed learned both types of attentions, as shown in Fig. 1 (Left). But it often incurs", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 480, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 480, + 506, + 493 + ], + "score": 1.0, + "content": "a prohibitively high cost for high-resolution images. To the best of our knowledge, the proposed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 492, + 504, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 504, + 504 + ], + "score": 1.0, + "content": "focal attention provides the first mechanism to incorporate local and global attention in a single", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 501, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 506, + 516 + ], + "score": 1.0, + "content": "Transformer layer 2. It can capture both short- and long-range interactions as standard self-attention", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 512, + 448, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 448, + 527 + ], + "score": 1.0, + "content": "but in a much more efficient and effective way, especially for high-resolution images.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 107, + 543, + 165, + 557 + ], + "lines": [ + { + "bbox": [ + 104, + 542, + 168, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 542, + 168, + 559 + ], + "score": 1.0, + "content": "3 Method", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 107, + 565, + 212, + 577 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 213, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 213, + 578 + ], + "score": 1.0, + "content": "3.1 Model architecture", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "To accommodate high-resolution dense prediction tasks, we employ a multi-scale model architecture", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 595, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 295, + 612 + ], + "score": 1.0, + "content": "as in [60, 76, 43]. As shown in Fig. 2, an image", + "type": "text" + }, + { + "bbox": [ + 295, + 597, + 355, + 608 + ], + "score": 0.93, + "content": "I \\in \\mathcal { R } ^ { H \\times W \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 595, + 506, + 612 + ], + "score": 1.0, + "content": "is first partitioned into patches of size", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 603, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 129, + 619 + ], + "score": 0.88, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 603, + 180, + 627 + ], + "score": 1.0, + "content": ", resulting in", + "type": "text" + }, + { + "bbox": [ + 180, + 608, + 213, + 622 + ], + "score": 0.94, + "content": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 603, + 330, + 627 + ], + "score": 1.0, + "content": "visual tokens with dimension", + "type": "text" + }, + { + "bbox": [ + 331, + 609, + 370, + 619 + ], + "score": 0.9, + "content": "4 \\times 4 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 603, + 505, + 627 + ], + "score": 1.0, + "content": ". Then, we use a patch embedding", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "layer, consisting of a convolutional layer with filter size and stride both equal to 4, to project these", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 629, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 280, + 643 + ], + "score": 1.0, + "content": "patches into hidden features with dimension", + "type": "text" + }, + { + "bbox": [ + 280, + 631, + 286, + 640 + ], + "score": 0.66, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 629, + 506, + 643 + ], + "score": 1.0, + "content": ". We then pass this spatial feature map to the four stages", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 280, + 654 + ], + "score": 1.0, + "content": "of Focal Transformer blocks. In each stage", + "type": "text" + }, + { + "bbox": [ + 281, + 641, + 340, + 653 + ], + "score": 0.93, + "content": "i \\in \\{ 1 , 2 , 3 , 4 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 641, + 506, + 654 + ], + "score": 1.0, + "content": ", the Focal Transformer block consists of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 651, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 107, + 653, + 119, + 663 + ], + "score": 0.86, + "content": "N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 651, + 506, + 665 + ], + "score": 1.0, + "content": "Focal Transformer layers. After each stage, we use a patch embedding layer to reduce the spatial", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "score": 1.0, + "content": "size of feature map by factor 2 and increase the feature dimension by 2. For image classification tasks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "we take the average of the output from the last stage and send it to a classification layer. For object", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 685, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 506, + 698 + ], + "score": 1.0, + "content": "detection, the feature maps from the last 3 or all 4 stages are fed to a particular object detector head,", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 711, + 358, + 722 + ], + "lines": [ + { + "bbox": [ + 119, + 709, + 359, + 723 + ], + "spans": [ + { + "bbox": [ + 119, + 709, + 359, + 723 + ], + "score": 1.0, + "content": "2A similar focal mechanism has been used in CNNs for NLP [27].", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 72, + 500, + 217 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 72, + 500, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 72, + 500, + 217 + ], + "spans": [ + { + "bbox": [ + 110, + 72, + 500, + 217 + ], + "score": 0.969, + "type": "image", + "image_path": "895ff52491908f44661414bd57eca7fc36f34ebd4f7c17cd1d28f3b83da587af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 72, + 500, + 120.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 120.33333333333334, + 500, + 168.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 168.66666666666669, + 500, + 217.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 231, + 504, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "score": 1.0, + "content": "Figure 2: Model architecture for our Focal Transformers. As highlighted in light blue boxes, our", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 242, + 403, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 403, + 255 + ], + "score": 1.0, + "content": "main innovation is the proposed focal attention in each Transformer layer.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 505, + 322 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 279, + 506, + 324 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "score": 1.0, + "content": "Efficient global and local self-attention. In many real-world tasks, Transformers need to cope", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "with a large number of input tokens, such as long documents in NLP and high-resolution images", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "in computer vision (CV). Recently, many efficient self-attention mechanisms have been proposed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "to deal with the quadratic computational and memory cost incurred by the standard self-attention", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "mechanism. On one hand, a number of works in both NLP and CV resort to coarse-grained global", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "score": 1.0, + "content": "self-attention (i.e., attending the down-sampled or summarized tokens) to capture the long-range", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "interactions [49, 46, 60, 64, 31, 23]. Although this approach improves the model efficiency, it", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "loses the detailed context information surrounding the query tokens. On the other hand, to make", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "the computational cost manageable, various local fine-grained attention mechanism (i.e., attending", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "neighboring tokens within a pre-set window size) are used for both NLP [3, 74, 1] and CV [56, 43, 76].", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "In this paper, we argue that both global and local attentions are important for model performance.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "This is also validated by some recent studies that aim to improve CNNs by incorporating ways of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 457, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 506, + 472 + ], + "score": 1.0, + "content": "modeling global attentions [35, 63, 61, 68, 2, 7, 51]. The standard self-attention mechanism used", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 470, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 506, + 483 + ], + "score": 1.0, + "content": "by ViT can indeed learned both types of attentions, as shown in Fig. 1 (Left). But it often incurs", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 480, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 480, + 506, + 493 + ], + "score": 1.0, + "content": "a prohibitively high cost for high-resolution images. To the best of our knowledge, the proposed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 492, + 504, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 504, + 504 + ], + "score": 1.0, + "content": "focal attention provides the first mechanism to incorporate local and global attention in a single", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 501, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 506, + 516 + ], + "score": 1.0, + "content": "Transformer layer 2. It can capture both short- and long-range interactions as standard self-attention", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 512, + 448, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 448, + 527 + ], + "score": 1.0, + "content": "but in a much more efficient and effective way, especially for high-resolution images.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 327, + 506, + 527 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 543, + 165, + 557 + ], + "lines": [ + { + "bbox": [ + 104, + 542, + 168, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 542, + 168, + 559 + ], + "score": 1.0, + "content": "3 Method", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 107, + 565, + 212, + 577 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 213, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 213, + 578 + ], + "score": 1.0, + "content": "3.1 Model architecture", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "To accommodate high-resolution dense prediction tasks, we employ a multi-scale model architecture", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 595, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 295, + 612 + ], + "score": 1.0, + "content": "as in [60, 76, 43]. As shown in Fig. 2, an image", + "type": "text" + }, + { + "bbox": [ + 295, + 597, + 355, + 608 + ], + "score": 0.93, + "content": "I \\in \\mathcal { R } ^ { H \\times W \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 595, + 506, + 612 + ], + "score": 1.0, + "content": "is first partitioned into patches of size", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 603, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 129, + 619 + ], + "score": 0.88, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 603, + 180, + 627 + ], + "score": 1.0, + "content": ", resulting in", + "type": "text" + }, + { + "bbox": [ + 180, + 608, + 213, + 622 + ], + "score": 0.94, + "content": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 603, + 330, + 627 + ], + "score": 1.0, + "content": "visual tokens with dimension", + "type": "text" + }, + { + "bbox": [ + 331, + 609, + 370, + 619 + ], + "score": 0.9, + "content": "4 \\times 4 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 603, + 505, + 627 + ], + "score": 1.0, + "content": ". Then, we use a patch embedding", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "layer, consisting of a convolutional layer with filter size and stride both equal to 4, to project these", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 629, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 280, + 643 + ], + "score": 1.0, + "content": "patches into hidden features with dimension", + "type": "text" + }, + { + "bbox": [ + 280, + 631, + 286, + 640 + ], + "score": 0.66, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 629, + 506, + 643 + ], + "score": 1.0, + "content": ". We then pass this spatial feature map to the four stages", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 280, + 654 + ], + "score": 1.0, + "content": "of Focal Transformer blocks. In each stage", + "type": "text" + }, + { + "bbox": [ + 281, + 641, + 340, + 653 + ], + "score": 0.93, + "content": "i \\in \\{ 1 , 2 , 3 , 4 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 641, + 506, + 654 + ], + "score": 1.0, + "content": ", the Focal Transformer block consists of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 651, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 107, + 653, + 119, + 663 + ], + "score": 0.86, + "content": "N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 651, + 506, + 665 + ], + "score": 1.0, + "content": "Focal Transformer layers. After each stage, we use a patch embedding layer to reduce the spatial", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "score": 1.0, + "content": "size of feature map by factor 2 and increase the feature dimension by 2. For image classification tasks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "we take the average of the output from the last stage and send it to a classification layer. For object", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 685, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 506, + 698 + ], + "score": 1.0, + "content": "detection, the feature maps from the last 3 or all 4 stages are fed to a particular object detector head,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "score": 1.0, + "content": "depending on the specific detection method we choose to use. The model capacity can be customized", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 452, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 263, + 97 + ], + "score": 1.0, + "content": "by varying the input feature dimension", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 264, + 84, + 270, + 93 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 271, + 83, + 452, + 97 + ], + "score": 1.0, + "content": "and the number of Focal Transformer layers.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 586, + 506, + 698 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 504, + 95 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "score": 1.0, + "content": "depending on the specific detection method we choose to use. The model capacity can be customized", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 452, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 263, + 97 + ], + "score": 1.0, + "content": "by varying the input feature dimension", + "type": "text" + }, + { + "bbox": [ + 264, + 84, + 270, + 93 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 83, + 452, + 97 + ], + "score": 1.0, + "content": "and the number of Focal Transformer layers.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 505, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "Standard self-attention can capture both short- and long-range interactions at fine-grain, but suffers", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 110, + 504, + 122 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 504, + 122 + ], + "score": 1.0, + "content": "from high computational cost when it performs attention on high-resolution feature maps as noted", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 102, + 115, + 507, + 138 + ], + "spans": [ + { + "bbox": [ + 102, + 115, + 389, + 138 + ], + "score": 1.0, + "content": "in [76]. Take stage 1 in Fig. 2 as an example. For a feature map of size", + "type": "text" + }, + { + "bbox": [ + 390, + 121, + 440, + 135 + ], + "score": 0.93, + "content": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } } \\times d", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 115, + 507, + 138 + ], + "score": 1.0, + "content": ", the complexity", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 132, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 104, + 132, + 181, + 149 + ], + "score": 1.0, + "content": "of self-attention is", + "type": "text" + }, + { + "bbox": [ + 182, + 133, + 248, + 147 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\mathcal { O } ( ( \\frac { H } { 4 } \\times \\frac { W } { 4 } ) ^ { 2 } d ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 132, + 506, + 149 + ], + "score": 1.0, + "content": ", resulting in an explosion of time and memory cost, considering", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 124, + 158 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 146, + 173, + 158 + ], + "score": 0.91, + "content": "\\operatorname* { m i n } ( H , W )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "could be 800 or even larger for object detection. In the next section, we describe", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 157, + 396, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 396, + 169 + ], + "score": 1.0, + "content": "how we address this issue with the proposed focal attention mechanism.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 107, + 180, + 242, + 192 + ], + "lines": [ + { + "bbox": [ + 105, + 180, + 242, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 242, + 193 + ], + "score": 1.0, + "content": "3.2 Token-wise focal attention", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 201, + 336, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 199, + 338, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 338, + 214 + ], + "score": 1.0, + "content": "Focal attention is proposed to make the Transformer lay-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 338, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 338, + 225 + ], + "score": 1.0, + "content": "ers suitable for encoding high-resolution input images.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 223, + 337, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 337, + 234 + ], + "score": 1.0, + "content": "Instead of attending all tokens at fine-grain, we attend the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 232, + 337, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 337, + 245 + ], + "score": 1.0, + "content": "fine-grain tokens only locally, but the summarized ones", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 244, + 337, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 337, + 257 + ], + "score": 1.0, + "content": "(i.e., the coarse-grained tokens generated by sub-window", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 255, + 337, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 337, + 267 + ], + "score": 1.0, + "content": "pooling, which is illustrated in Fig. 4 and will be described", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 266, + 338, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 338, + 278 + ], + "score": 1.0, + "content": "later) globally. As such, focal attention can cover the same", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 277, + 337, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 337, + 288 + ], + "score": 1.0, + "content": "amount of image regions as standard self-attention but", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 288, + 337, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 337, + 299 + ], + "score": 1.0, + "content": "with much less cost. In Fig. 3, we show the size of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 299, + 337, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 337, + 309 + ], + "score": 1.0, + "content": "receptive field for standard self-attention and our focal", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 310, + 338, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 338, + 321 + ], + "score": 1.0, + "content": "attention as a function of the number of attended tokens.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 321, + 338, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 338, + 333 + ], + "score": 1.0, + "content": "For a given query position, by reducing the granularity of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 331, + 337, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 337, + 343 + ], + "score": 1.0, + "content": "its surroundings based on their distance to the query, focal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 343, + 337, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 337, + 354 + ], + "score": 1.0, + "content": "attention can have significantly larger receptive fields at", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 353, + 338, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 338, + 366 + ], + "score": 1.0, + "content": "the same cost measured by the number of visual tokens,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 364, + 315, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 315, + 376 + ], + "score": 1.0, + "content": "compared to the standard self-attention mechanism.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 16.5 + }, + { + "type": "image", + "bbox": [ + 343, + 201, + 497, + 272 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 343, + 201, + 497, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 201, + 497, + 272 + ], + "spans": [ + { + "bbox": [ + 343, + 201, + 497, + 272 + ], + "score": 0.968, + "type": "image", + "image_path": "bf6a6de3b259a05fb0ac374d04a980cb6b72063e837f8ca7329f069632ea03a7.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 343, + 201, + 497, + 215.2 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 343, + 215.2, + 497, + 229.39999999999998 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 343, + 229.39999999999998, + 497, + 243.59999999999997 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 343, + 243.59999999999997, + 497, + 257.79999999999995 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 343, + 257.79999999999995, + 497, + 271.99999999999994 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 277, + 506, + 365 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 344, + 276, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 344, + 276, + 506, + 290 + ], + "score": 1.0, + "content": "Figure 3: The size of receptive field (y-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 343, + 289, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 343, + 289, + 505, + 299 + ], + "score": 1.0, + "content": "axis) as a function of the number of used", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 343, + 299, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 343, + 299, + 405, + 311 + ], + "score": 1.0, + "content": "visual tokens", + "type": "text" + }, + { + "bbox": [ + 406, + 301, + 411, + 309 + ], + "score": 0.47, + "content": "\\mathbf { \\bar { x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 299, + 506, + 311 + ], + "score": 1.0, + "content": "-axis) in regular (stan-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 344, + 309, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 344, + 309, + 506, + 321 + ], + "score": 1.0, + "content": "dard) self-attention and focal attention.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 343, + 321, + 507, + 333 + ], + "spans": [ + { + "bbox": [ + 343, + 321, + 507, + 333 + ], + "score": 1.0, + "content": "When plotting the curve for focal atten-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 343, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 343, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "tion, we increase the focal window size", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 343, + 343, + 507, + 354 + ], + "spans": [ + { + "bbox": [ + 343, + 343, + 507, + 354 + ], + "score": 1.0, + "content": "by 2 for each focal level up to the maxi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 343, + 354, + 435, + 364 + ], + "spans": [ + { + "bbox": [ + 343, + 354, + 435, + 364 + ], + "score": 1.0, + "content": "mal window size of 8.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5 + } + ], + "index": 30.25 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "Theoretically, the focal attention mechanism enables global interaction with much less time and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "memory cost, because it attends a much smaller number of surrounding (summarized) tokens. In", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "practice, however, extracting the surrounding tokens for each query position could incur high time", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "score": 1.0, + "content": "cost since we need to duplicate the extraction of each token for all queries that the token surrounds.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "This issue had been extensively discussed in [56, 76, 43] and a common solution is to partition the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "input feature map into windows. Thus, in our Focal Transformers, we resort to performing focal", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 445, + 480, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 480, + 459 + ], + "score": 1.0, + "content": "attention at the window level. We elaborate the window-wise focal attention in the following.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41 + }, + { + "type": "title", + "bbox": [ + 107, + 468, + 259, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 259, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 259, + 481 + ], + "score": 1.0, + "content": "3.2.1 Window-wise focal attention", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 505, + 531 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 205, + 500 + ], + "score": 1.0, + "content": "Given a feature map of", + "type": "text" + }, + { + "bbox": [ + 206, + 486, + 268, + 497 + ], + "score": 0.92, + "content": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 484, + 340, + 500 + ], + "score": 1.0, + "content": "with spatial size", + "type": "text" + }, + { + "bbox": [ + 340, + 487, + 374, + 497 + ], + "score": 0.9, + "content": "M \\times N", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 484, + 506, + 500 + ], + "score": 1.0, + "content": ", we first partition it into a grid", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 186, + 511 + ], + "score": 1.0, + "content": "of windows of size", + "type": "text" + }, + { + "bbox": [ + 187, + 499, + 219, + 510 + ], + "score": 0.9, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 497, + 506, + 511 + ], + "score": 1.0, + "content": ". Then, we extract the surroundings for each window rather than each", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "individual token. The proposed window-wise focal attention is illustrated in Fig. 4. To clarify, we", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 519, + 201, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 201, + 531 + ], + "score": 1.0, + "content": "first define three terms:", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47.5 + }, + { + "type": "text", + "bbox": [ + 107, + 536, + 506, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 488, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 162, + 549 + ], + "score": 1.0, + "content": "• Focal level", + "type": "text" + }, + { + "bbox": [ + 162, + 537, + 170, + 546 + ], + "score": 0.73, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 534, + 488, + 549 + ], + "score": 1.0, + "content": "refers to the granularity level at which we extract the tokens for focal attention.", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 194, + 559 + ], + "score": 1.0, + "content": "• Focal window size", + "type": "text" + }, + { + "bbox": [ + 194, + 546, + 206, + 559 + ], + "score": 0.9, + "content": "s _ { w } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "is the size of sub-window on which the summarized tokens are formed via", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 112, + 556, + 347, + 571 + ], + "spans": [ + { + "bbox": [ + 112, + 556, + 288, + 571 + ], + "score": 1.0, + "content": "sub-window pooling at granularity level of", + "type": "text" + }, + { + "bbox": [ + 288, + 558, + 343, + 570 + ], + "score": 0.93, + "content": "l \\in \\{ 1 , . . . , L \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 556, + 347, + 571 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 568, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 187, + 583 + ], + "score": 1.0, + "content": "• Focal region size", + "type": "text" + }, + { + "bbox": [ + 188, + 569, + 198, + 581 + ], + "score": 0.89, + "content": "s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 568, + 506, + 583 + ], + "score": 1.0, + "content": "denotes the number of sub-windows that are filled up horizontally (or verti-", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 114, + 581, + 266, + 593 + ], + "spans": [ + { + "bbox": [ + 114, + 581, + 257, + 593 + ], + "score": 1.0, + "content": "cally) in an attended region at level", + "type": "text" + }, + { + "bbox": [ + 258, + 581, + 262, + 591 + ], + "score": 0.73, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 581, + 266, + 593 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 52 + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 504, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "Now, we detail how window-wise focal attention works in the following two steps, sub-window", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 608, + 241, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 241, + 622 + ], + "score": 1.0, + "content": "pooling and attention computing.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 55.5 + }, + { + "type": "text", + "bbox": [ + 107, + 623, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 318, + 637 + ], + "score": 1.0, + "content": "Sub-window pooling. Consider input feature map", + "type": "text" + }, + { + "bbox": [ + 318, + 623, + 381, + 635 + ], + "score": 0.92, + "content": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 622, + 414, + 637 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 414, + 624, + 448, + 635 + ], + "score": 0.9, + "content": "M \\times N", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 622, + 506, + 637 + ], + "score": 1.0, + "content": "is the spatial", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 635, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 168, + 648 + ], + "score": 1.0, + "content": "dimension and", + "type": "text" + }, + { + "bbox": [ + 168, + 636, + 175, + 645 + ], + "score": 0.8, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 635, + 432, + 648 + ], + "score": 1.0, + "content": "the feature dimension. We perform sub-window pooling for all", + "type": "text" + }, + { + "bbox": [ + 432, + 637, + 440, + 645 + ], + "score": 0.85, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 635, + 506, + 648 + ], + "score": 1.0, + "content": "levels. At focal", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 104, + 645, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 128, + 661 + ], + "score": 1.0, + "content": "level", + "type": "text" + }, + { + "bbox": [ + 128, + 647, + 132, + 656 + ], + "score": 0.63, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 645, + 273, + 661 + ], + "score": 1.0, + "content": ", we first split the input feature map", + "type": "text" + }, + { + "bbox": [ + 274, + 649, + 281, + 656 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 645, + 428, + 661 + ], + "score": 1.0, + "content": "into a grid of sub-windows with size", + "type": "text" + }, + { + "bbox": [ + 429, + 646, + 464, + 658 + ], + "score": 0.92, + "content": "s _ { w } ^ { l } \\times s _ { w } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 645, + 506, + 661 + ], + "score": 1.0, + "content": ". Then we", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 658, + 380, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 220, + 671 + ], + "score": 1.0, + "content": "use a linear projection layer", + "type": "text" + }, + { + "bbox": [ + 220, + 659, + 230, + 672 + ], + "score": 0.9, + "content": "f _ { p } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 658, + 380, + 671 + ], + "score": 1.0, + "content": "to pool the sub-windows spatially by", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 58.5 + }, + { + "type": "interline_equation", + "bbox": [ + 167, + 672, + 443, + 690 + ], + "lines": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "spans": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "score": 0.9, + "content": "\\begin{array} { r } { x ^ { l } = f _ { p } ^ { l } ( \\hat { x } ) \\in \\mathcal { R } ^ { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d } , \\quad \\hat { x } = \\mathrm { R e s h a p e } ( x ) \\in \\mathcal { R } ^ { ( \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d ) \\times ( s _ { w } ^ { l } \\times s _ { w } ^ { l } ) } . } \\end{array}", + "type": "interline_equation", + "image_path": "2a14e2c3c947dc9cd476c4ca7ba03d6c99cd76f5d3158e8561986ada02fbfb84.jpg" + } + ] + } + ], + "index": 61, + "virtual_lines": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "spans": [], + "index": 61 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 503, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 209, + 713 + ], + "score": 1.0, + "content": "The pooled feature maps", + "type": "text" + }, + { + "bbox": [ + 209, + 698, + 234, + 711 + ], + "score": 0.89, + "content": "\\{ x ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 696, + 309, + 713 + ], + "score": 1.0, + "content": "at different levels", + "type": "text" + }, + { + "bbox": [ + 309, + 700, + 313, + 709 + ], + "score": 0.77, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 696, + 506, + 713 + ], + "score": 1.0, + "content": "provide rich information at both fine-grain and", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 709, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 212, + 724 + ], + "score": 1.0, + "content": "coarse-grain. Since we set", + "type": "text" + }, + { + "bbox": [ + 212, + 711, + 242, + 723 + ], + "score": 0.91, + "content": "s _ { w } ^ { l } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 709, + 506, + 724 + ], + "score": 1.0, + "content": "for the first focal level which has the same granularity as the input", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "4", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 504, + 95 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 73, + 505, + 97 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 505, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "Standard self-attention can capture both short- and long-range interactions at fine-grain, but suffers", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 110, + 504, + 122 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 504, + 122 + ], + "score": 1.0, + "content": "from high computational cost when it performs attention on high-resolution feature maps as noted", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 102, + 115, + 507, + 138 + ], + "spans": [ + { + "bbox": [ + 102, + 115, + 389, + 138 + ], + "score": 1.0, + "content": "in [76]. Take stage 1 in Fig. 2 as an example. For a feature map of size", + "type": "text" + }, + { + "bbox": [ + 390, + 121, + 440, + 135 + ], + "score": 0.93, + "content": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } } \\times d", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 115, + 507, + 138 + ], + "score": 1.0, + "content": ", the complexity", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 132, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 104, + 132, + 181, + 149 + ], + "score": 1.0, + "content": "of self-attention is", + "type": "text" + }, + { + "bbox": [ + 182, + 133, + 248, + 147 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\mathcal { O } ( ( \\frac { H } { 4 } \\times \\frac { W } { 4 } ) ^ { 2 } d ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 132, + 506, + 149 + ], + "score": 1.0, + "content": ", resulting in an explosion of time and memory cost, considering", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 124, + 158 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 146, + 173, + 158 + ], + "score": 0.91, + "content": "\\operatorname* { m i n } ( H , W )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "could be 800 or even larger for object detection. In the next section, we describe", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 157, + 396, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 396, + 169 + ], + "score": 1.0, + "content": "how we address this issue with the proposed focal attention mechanism.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 102, + 100, + 507, + 169 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 180, + 242, + 192 + ], + "lines": [ + { + "bbox": [ + 105, + 180, + 242, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 242, + 193 + ], + "score": 1.0, + "content": "3.2 Token-wise focal attention", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 201, + 336, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 199, + 338, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 338, + 214 + ], + "score": 1.0, + "content": "Focal attention is proposed to make the Transformer lay-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 338, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 338, + 225 + ], + "score": 1.0, + "content": "ers suitable for encoding high-resolution input images.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 223, + 337, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 337, + 234 + ], + "score": 1.0, + "content": "Instead of attending all tokens at fine-grain, we attend the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 232, + 337, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 337, + 245 + ], + "score": 1.0, + "content": "fine-grain tokens only locally, but the summarized ones", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 244, + 337, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 337, + 257 + ], + "score": 1.0, + "content": "(i.e., the coarse-grained tokens generated by sub-window", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 255, + 337, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 337, + 267 + ], + "score": 1.0, + "content": "pooling, which is illustrated in Fig. 4 and will be described", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 266, + 338, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 338, + 278 + ], + "score": 1.0, + "content": "later) globally. As such, focal attention can cover the same", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 277, + 337, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 337, + 288 + ], + "score": 1.0, + "content": "amount of image regions as standard self-attention but", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 288, + 337, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 337, + 299 + ], + "score": 1.0, + "content": "with much less cost. In Fig. 3, we show the size of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 299, + 337, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 337, + 309 + ], + "score": 1.0, + "content": "receptive field for standard self-attention and our focal", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 310, + 338, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 338, + 321 + ], + "score": 1.0, + "content": "attention as a function of the number of attended tokens.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 321, + 338, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 338, + 333 + ], + "score": 1.0, + "content": "For a given query position, by reducing the granularity of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 331, + 337, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 337, + 343 + ], + "score": 1.0, + "content": "its surroundings based on their distance to the query, focal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 343, + 337, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 337, + 354 + ], + "score": 1.0, + "content": "attention can have significantly larger receptive fields at", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 353, + 338, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 338, + 366 + ], + "score": 1.0, + "content": "the same cost measured by the number of visual tokens,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 364, + 315, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 315, + 376 + ], + "score": 1.0, + "content": "compared to the standard self-attention mechanism.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 199, + 338, + 376 + ] + }, + { + "type": "image", + "bbox": [ + 343, + 201, + 497, + 272 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 343, + 201, + 497, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 201, + 497, + 272 + ], + "spans": [ + { + "bbox": [ + 343, + 201, + 497, + 272 + ], + "score": 0.968, + "type": "image", + "image_path": "bf6a6de3b259a05fb0ac374d04a980cb6b72063e837f8ca7329f069632ea03a7.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 343, + 201, + 497, + 215.2 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 343, + 215.2, + 497, + 229.39999999999998 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 343, + 229.39999999999998, + 497, + 243.59999999999997 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 343, + 243.59999999999997, + 497, + 257.79999999999995 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 343, + 257.79999999999995, + 497, + 271.99999999999994 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 277, + 506, + 365 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 344, + 276, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 344, + 276, + 506, + 290 + ], + "score": 1.0, + "content": "Figure 3: The size of receptive field (y-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 343, + 289, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 343, + 289, + 505, + 299 + ], + "score": 1.0, + "content": "axis) as a function of the number of used", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 343, + 299, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 343, + 299, + 405, + 311 + ], + "score": 1.0, + "content": "visual tokens", + "type": "text" + }, + { + "bbox": [ + 406, + 301, + 411, + 309 + ], + "score": 0.47, + "content": "\\mathbf { \\bar { x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 299, + 506, + 311 + ], + "score": 1.0, + "content": "-axis) in regular (stan-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 344, + 309, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 344, + 309, + 506, + 321 + ], + "score": 1.0, + "content": "dard) self-attention and focal attention.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 343, + 321, + 507, + 333 + ], + "spans": [ + { + "bbox": [ + 343, + 321, + 507, + 333 + ], + "score": 1.0, + "content": "When plotting the curve for focal atten-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 343, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 343, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "tion, we increase the focal window size", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 343, + 343, + 507, + 354 + ], + "spans": [ + { + "bbox": [ + 343, + 343, + 507, + 354 + ], + "score": 1.0, + "content": "by 2 for each focal level up to the maxi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 343, + 354, + 435, + 364 + ], + "spans": [ + { + "bbox": [ + 343, + 354, + 435, + 364 + ], + "score": 1.0, + "content": "mal window size of 8.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5 + } + ], + "index": 30.25 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "Theoretically, the focal attention mechanism enables global interaction with much less time and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "memory cost, because it attends a much smaller number of surrounding (summarized) tokens. In", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "practice, however, extracting the surrounding tokens for each query position could incur high time", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "score": 1.0, + "content": "cost since we need to duplicate the extraction of each token for all queries that the token surrounds.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "This issue had been extensively discussed in [56, 76, 43] and a common solution is to partition the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "input feature map into windows. Thus, in our Focal Transformers, we resort to performing focal", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 445, + 480, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 480, + 459 + ], + "score": 1.0, + "content": "attention at the window level. We elaborate the window-wise focal attention in the following.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 380, + 506, + 459 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 468, + 259, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 259, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 259, + 481 + ], + "score": 1.0, + "content": "3.2.1 Window-wise focal attention", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 505, + 531 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 205, + 500 + ], + "score": 1.0, + "content": "Given a feature map of", + "type": "text" + }, + { + "bbox": [ + 206, + 486, + 268, + 497 + ], + "score": 0.92, + "content": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 484, + 340, + 500 + ], + "score": 1.0, + "content": "with spatial size", + "type": "text" + }, + { + "bbox": [ + 340, + 487, + 374, + 497 + ], + "score": 0.9, + "content": "M \\times N", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 484, + 506, + 500 + ], + "score": 1.0, + "content": ", we first partition it into a grid", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 186, + 511 + ], + "score": 1.0, + "content": "of windows of size", + "type": "text" + }, + { + "bbox": [ + 187, + 499, + 219, + 510 + ], + "score": 0.9, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 497, + 506, + 511 + ], + "score": 1.0, + "content": ". Then, we extract the surroundings for each window rather than each", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "individual token. The proposed window-wise focal attention is illustrated in Fig. 4. To clarify, we", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 519, + 201, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 201, + 531 + ], + "score": 1.0, + "content": "first define three terms:", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 484, + 506, + 531 + ] + }, + { + "type": "list", + "bbox": [ + 107, + 536, + 506, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 488, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 162, + 549 + ], + "score": 1.0, + "content": "• Focal level", + "type": "text" + }, + { + "bbox": [ + 162, + 537, + 170, + 546 + ], + "score": 0.73, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 534, + 488, + 549 + ], + "score": 1.0, + "content": "refers to the granularity level at which we extract the tokens for focal attention.", + "type": "text" + } + ], + "index": 50, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 194, + 559 + ], + "score": 1.0, + "content": "• Focal window size", + "type": "text" + }, + { + "bbox": [ + 194, + 546, + 206, + 559 + ], + "score": 0.9, + "content": "s _ { w } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "is the size of sub-window on which the summarized tokens are formed via", + "type": "text" + } + ], + "index": 51, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 556, + 347, + 571 + ], + "spans": [ + { + "bbox": [ + 112, + 556, + 288, + 571 + ], + "score": 1.0, + "content": "sub-window pooling at granularity level of", + "type": "text" + }, + { + "bbox": [ + 288, + 558, + 343, + 570 + ], + "score": 0.93, + "content": "l \\in \\{ 1 , . . . , L \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 556, + 347, + 571 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 52, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 568, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 187, + 583 + ], + "score": 1.0, + "content": "• Focal region size", + "type": "text" + }, + { + "bbox": [ + 188, + 569, + 198, + 581 + ], + "score": 0.89, + "content": "s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 568, + 506, + 583 + ], + "score": 1.0, + "content": "denotes the number of sub-windows that are filled up horizontally (or verti-", + "type": "text" + } + ], + "index": 53, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 581, + 266, + 593 + ], + "spans": [ + { + "bbox": [ + 114, + 581, + 257, + 593 + ], + "score": 1.0, + "content": "cally) in an attended region at level", + "type": "text" + }, + { + "bbox": [ + 258, + 581, + 262, + 591 + ], + "score": 0.73, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 581, + 266, + 593 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 54, + "is_list_end_line": true + } + ], + "index": 52, + "bbox_fs": [ + 105, + 534, + 506, + 593 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 504, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "Now, we detail how window-wise focal attention works in the following two steps, sub-window", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 608, + 241, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 241, + 622 + ], + "score": 1.0, + "content": "pooling and attention computing.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 55.5, + "bbox_fs": [ + 105, + 596, + 505, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 623, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 318, + 637 + ], + "score": 1.0, + "content": "Sub-window pooling. Consider input feature map", + "type": "text" + }, + { + "bbox": [ + 318, + 623, + 381, + 635 + ], + "score": 0.92, + "content": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 622, + 414, + 637 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 414, + 624, + 448, + 635 + ], + "score": 0.9, + "content": "M \\times N", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 622, + 506, + 637 + ], + "score": 1.0, + "content": "is the spatial", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 635, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 168, + 648 + ], + "score": 1.0, + "content": "dimension and", + "type": "text" + }, + { + "bbox": [ + 168, + 636, + 175, + 645 + ], + "score": 0.8, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 635, + 432, + 648 + ], + "score": 1.0, + "content": "the feature dimension. We perform sub-window pooling for all", + "type": "text" + }, + { + "bbox": [ + 432, + 637, + 440, + 645 + ], + "score": 0.85, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 635, + 506, + 648 + ], + "score": 1.0, + "content": "levels. At focal", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 104, + 645, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 128, + 661 + ], + "score": 1.0, + "content": "level", + "type": "text" + }, + { + "bbox": [ + 128, + 647, + 132, + 656 + ], + "score": 0.63, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 645, + 273, + 661 + ], + "score": 1.0, + "content": ", we first split the input feature map", + "type": "text" + }, + { + "bbox": [ + 274, + 649, + 281, + 656 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 645, + 428, + 661 + ], + "score": 1.0, + "content": "into a grid of sub-windows with size", + "type": "text" + }, + { + "bbox": [ + 429, + 646, + 464, + 658 + ], + "score": 0.92, + "content": "s _ { w } ^ { l } \\times s _ { w } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 645, + 506, + 661 + ], + "score": 1.0, + "content": ". Then we", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 658, + 380, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 220, + 671 + ], + "score": 1.0, + "content": "use a linear projection layer", + "type": "text" + }, + { + "bbox": [ + 220, + 659, + 230, + 672 + ], + "score": 0.9, + "content": "f _ { p } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 658, + 380, + 671 + ], + "score": 1.0, + "content": "to pool the sub-windows spatially by", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 58.5, + "bbox_fs": [ + 104, + 622, + 506, + 672 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 167, + 672, + 443, + 690 + ], + "lines": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "spans": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "score": 0.9, + "content": "\\begin{array} { r } { x ^ { l } = f _ { p } ^ { l } ( \\hat { x } ) \\in \\mathcal { R } ^ { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d } , \\quad \\hat { x } = \\mathrm { R e s h a p e } ( x ) \\in \\mathcal { R } ^ { ( \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d ) \\times ( s _ { w } ^ { l } \\times s _ { w } ^ { l } ) } . } \\end{array}", + "type": "interline_equation", + "image_path": "2a14e2c3c947dc9cd476c4ca7ba03d6c99cd76f5d3158e8561986ada02fbfb84.jpg" + } + ] + } + ], + "index": 61, + "virtual_lines": [ + { + "bbox": [ + 167, + 672, + 443, + 690 + ], + "spans": [], + "index": 61 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 503, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 209, + 713 + ], + "score": 1.0, + "content": "The pooled feature maps", + "type": "text" + }, + { + "bbox": [ + 209, + 698, + 234, + 711 + ], + "score": 0.89, + "content": "\\{ x ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 696, + 309, + 713 + ], + "score": 1.0, + "content": "at different levels", + "type": "text" + }, + { + "bbox": [ + 309, + 700, + 313, + 709 + ], + "score": 0.77, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 696, + 506, + 713 + ], + "score": 1.0, + "content": "provide rich information at both fine-grain and", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 709, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 212, + 724 + ], + "score": 1.0, + "content": "coarse-grain. Since we set", + "type": "text" + }, + { + "bbox": [ + 212, + 711, + 242, + 723 + ], + "score": 0.91, + "content": "s _ { w } ^ { l } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 709, + 506, + 724 + ], + "score": 1.0, + "content": "for the first focal level which has the same granularity as the input", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62.5, + "bbox_fs": [ + 105, + 696, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 73, + 495, + 232 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 73, + 495, + 232 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 73, + 495, + 232 + ], + "spans": [ + { + "bbox": [ + 119, + 73, + 495, + 232 + ], + "score": 0.973, + "type": "image", + "image_path": "8d7c5bc6d5803f11463c3ab8656b31fec8dbb82ddb891bec59961a9ad98dfe04.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 73, + 495, + 126.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 126.0, + 495, + 179.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 179.0, + 495, + 232.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 240, + 506, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "Figure 4: An illustration of focal attention at window level. Each of the square cells represents a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "visual token that is either from the original feature map or a summarized token formed by sub-window", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 263, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 338, + 275 + ], + "score": 1.0, + "content": "pooling. Suppose we have an input feature map of size", + "type": "text" + }, + { + "bbox": [ + 339, + 263, + 373, + 273 + ], + "score": 0.91, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 263, + 479, + 275 + ], + "score": 1.0, + "content": ". We first partition it into", + "type": "text" + }, + { + "bbox": [ + 480, + 263, + 504, + 273 + ], + "score": 0.89, + "content": "5 \\times 5", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 274, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 174, + 285 + ], + "score": 1.0, + "content": "windows of size", + "type": "text" + }, + { + "bbox": [ + 175, + 274, + 199, + 284 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 274, + 241, + 285 + ], + "score": 1.0, + "content": ". Take the", + "type": "text" + }, + { + "bbox": [ + 241, + 274, + 266, + 284 + ], + "score": 0.9, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 274, + 505, + 285 + ], + "score": 1.0, + "content": "blue window in the middle as the query set, we extract its", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "surrounding tokens at three granularity levels as its keys and values. For the first level, we extract the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 130, + 306 + ], + "score": 0.89, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "tokens which are closest to the blue window at the finest grain. At the second level, we expand", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 306, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 286, + 318 + ], + "score": 1.0, + "content": "the attention region and pool the surrounding", + "type": "text" + }, + { + "bbox": [ + 286, + 307, + 310, + 317 + ], + "score": 0.9, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 306, + 506, + 318 + ], + "score": 1.0, + "content": "sub-windows to form summarized tokens, which", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 145, + 329 + ], + "score": 1.0, + "content": "results in", + "type": "text" + }, + { + "bbox": [ + 145, + 318, + 169, + 328 + ], + "score": 0.9, + "content": "6 \\times 6", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "summarized tokens. At the third level, we attend a larger region covering the whole", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 195, + 340 + ], + "score": 1.0, + "content": "feature map and pool", + "type": "text" + }, + { + "bbox": [ + 195, + 329, + 219, + 339 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 327, + 339, + 340 + ], + "score": 1.0, + "content": "sub-windows, which leads to", + "type": "text" + }, + { + "bbox": [ + 340, + 328, + 363, + 339 + ], + "score": 0.9, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "summarized tokens. Finally, these", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 426, + 352 + ], + "score": 1.0, + "content": "three levels of tokens are concatenated to compute the keys and values for the", + "type": "text" + }, + { + "bbox": [ + 427, + 339, + 475, + 350 + ], + "score": 0.91, + "content": "4 \\times 4 = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 225, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 225, + 361 + ], + "score": 1.0, + "content": "(queries) in the blue window.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "feature map, there is no need to perform any sub-window pooling. Considering that the focal window", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "size is usually very small (7 maximally in our settings), the number of extra parameters introduced by", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 396, + 244, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 244, + 408 + ], + "score": 1.0, + "content": "sub-window pooling is negligible.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 366, + 426 + ], + "score": 1.0, + "content": "Attention computing. Once we obtain the pooled feature maps", + "type": "text" + }, + { + "bbox": [ + 366, + 411, + 392, + 424 + ], + "score": 0.93, + "content": "\\{ x ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 410, + 415, + 426 + ], + "score": 1.0, + "content": "at all", + "type": "text" + }, + { + "bbox": [ + 416, + 413, + 424, + 423 + ], + "score": 0.8, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 410, + 506, + 426 + ], + "score": 1.0, + "content": "levels, we compute", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 422, + 504, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 478, + 437 + ], + "score": 1.0, + "content": "the query at the first level, and key and value for all levels using three linear projection layers", + "type": "text" + }, + { + "bbox": [ + 479, + 423, + 504, + 435 + ], + "score": 0.57, + "content": "f _ { q } , f _ { k }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 433, + 201, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 123, + 448 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 435, + 134, + 446 + ], + "score": 0.89, + "content": "f _ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 433, + 201, + 448 + ], + "score": 1.0, + "content": ", respectively, as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 451, + 464, + 466 + ], + "lines": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "spans": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "score": 0.9, + "content": "Q = f _ { q } ( x ^ { 1 } ) , \\quad K = \\{ K ^ { l } \\} _ { 1 } ^ { L } = f _ { k } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) , \\quad V = \\{ V ^ { l } \\} _ { 1 } ^ { L } = f _ { v } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) .", + "type": "interline_equation", + "image_path": "adc5271b876d4969c2420a5cdd389cd1720ece17ee5092a72df5a2920aba267f.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 506, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "To perform focal attention, we need to first extract the surrounding tokens for each query token in the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 385, + 501 + ], + "score": 1.0, + "content": "feature map. As mentioned earlier, tokens inside a window partition", + "type": "text" + }, + { + "bbox": [ + 385, + 489, + 416, + 501 + ], + "score": 0.91, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "share the same set of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 497, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 104, + 497, + 270, + 515 + ], + "score": 1.0, + "content": "surroundings. For the queries inside the", + "type": "text" + }, + { + "bbox": [ + 270, + 502, + 275, + 511 + ], + "score": 0.73, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 497, + 322, + 515 + ], + "score": 1.0, + "content": "-th window", + "type": "text" + }, + { + "bbox": [ + 322, + 500, + 388, + 512 + ], + "score": 0.91, + "content": "Q _ { i } \\in \\mathcal { R } ^ { s _ { p } \\times s _ { p } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 497, + 451, + 515 + ], + "score": 1.0, + "content": ", we extract the", + "type": "text" + }, + { + "bbox": [ + 451, + 500, + 482, + 513 + ], + "score": 0.92, + "content": "s _ { r } ^ { l } \\times s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 497, + 506, + 515 + ], + "score": 1.0, + "content": "keys", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 171, + 525 + ], + "score": 1.0, + "content": "and values from", + "type": "text" + }, + { + "bbox": [ + 172, + 512, + 185, + 523 + ], + "score": 0.89, + "content": "K ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 511, + 203, + 525 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 512, + 215, + 523 + ], + "score": 0.88, + "content": "V ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "surrounding the window which the query lies in, and then gather the keys", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 103, + 521, + 507, + 538 + ], + "spans": [ + { + "bbox": [ + 103, + 521, + 183, + 538 + ], + "score": 1.0, + "content": "and values from all", + "type": "text" + }, + { + "bbox": [ + 184, + 524, + 192, + 533 + ], + "score": 0.8, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 521, + 254, + 538 + ], + "score": 1.0, + "content": "levels to obtain", + "type": "text" + }, + { + "bbox": [ + 255, + 523, + 372, + 536 + ], + "score": 0.91, + "content": "\\breve { K } _ { i } = \\{ K _ { i } ^ { 1 } , . . . , K _ { i } ^ { L } \\} \\in \\mathscr { R } ^ { \\bar { s } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 521, + 390, + 538 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 390, + 523, + 502, + 536 + ], + "score": 0.92, + "content": "V _ { i } = \\{ V _ { i } ^ { 1 } , . . . , \\mathbf { \\bar { V } } _ { i } ^ { L } \\} \\in \\mathcal { R } ^ { s \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 521, + 507, + 538 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 101, + 529, + 510, + 560 + ], + "spans": [ + { + "bbox": [ + 101, + 529, + 134, + 560 + ], + "score": 1.0, + "content": "where imple", + "type": "text" + }, + { + "bbox": [ + 134, + 539, + 140, + 547 + ], + "score": 0.73, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 529, + 340, + 560 + ], + "score": 1.0, + "content": "is the sum of focal regions from all levels, i.e., ntation of focal attention following Fig. 1 requires", + "type": "text" + }, + { + "bbox": [ + 340, + 535, + 406, + 550 + ], + "score": 0.93, + "content": "\\begin{array} { r } { s = \\sum _ { l = 1 } ^ { L } ( s _ { r } ^ { l } ) ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 529, + 510, + 560 + ], + "score": 1.0, + "content": ". Note that a canonicalverlapped regions across", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "score": 1.0, + "content": "different levels. In our implementation, we intentionally keep them in order to capture the pyramid", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "information for the overlapped regions. Finally, we follow [43] to include a relative position bias and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 580, + 258, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 232, + 594 + ], + "score": 1.0, + "content": "compute the focal attention for", + "type": "text" + }, + { + "bbox": [ + 232, + 581, + 244, + 592 + ], + "score": 0.89, + "content": "Q _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 580, + 258, + 594 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 598, + 399, + 623 + ], + "lines": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "spans": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "score": 0.92, + "content": "\\mathrm { A t t e n t i o n } ( Q _ { i } , K _ { i } , V _ { i } ) = \\mathrm { S o f t m a x } ( \\frac { Q _ { i } K _ { i } ^ { T } } { \\sqrt { d } } + B ) V _ { i } ,", + "type": "interline_equation", + "image_path": "a272e4bce5871346781b2ad83e37e4087f5507b82017abb37846d96c857a7475.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 629, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 507, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 133, + 643 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 630, + 183, + 642 + ], + "score": 0.94, + "content": "B = \\{ B ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 629, + 391, + 643 + ], + "score": 1.0, + "content": "is the learnable relative position bias. It consists of", + "type": "text" + }, + { + "bbox": [ + 392, + 631, + 399, + 640 + ], + "score": 0.82, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 629, + 446, + 643 + ], + "score": 1.0, + "content": "subsets for", + "type": "text" + }, + { + "bbox": [ + 447, + 631, + 455, + 640 + ], + "score": 0.82, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 629, + 507, + 643 + ], + "score": 1.0, + "content": "focal levels.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 640, + 507, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 334, + 657 + ], + "score": 1.0, + "content": "Similar to [43], for the first level, we parameterize it as", + "type": "text" + }, + { + "bbox": [ + 335, + 641, + 433, + 653 + ], + "score": 0.91, + "content": "B ^ { 1 } \\in \\mathcal { R } ^ { ( 2 s _ { p } - 1 ) \\times ( 2 s _ { p } - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 640, + 507, + 657 + ], + "score": 1.0, + "content": ", considering that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 325, + 667 + ], + "score": 1.0, + "content": "the horizontal and vertical position ranges are both in", + "type": "text" + }, + { + "bbox": [ + 325, + 654, + 396, + 666 + ], + "score": 0.91, + "content": "[ - s _ { p } + 1 , s _ { p } - 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 654, + 506, + 667 + ], + "score": 1.0, + "content": ". For the other focal levels,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 665, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 506, + 677 + ], + "score": 1.0, + "content": "considering that they have different granularity with respect to the queries, we treat all the queries", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 243, + 691 + ], + "score": 1.0, + "content": "inside a window equally and use", + "type": "text" + }, + { + "bbox": [ + 243, + 676, + 298, + 688 + ], + "score": 0.92, + "content": "B ^ { l } \\in \\mathcal { R } ^ { s _ { r } ^ { l } \\times s _ { r } ^ { l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 676, + 506, + 691 + ], + "score": 1.0, + "content": "to represent the relative position bias between the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 217, + 701 + ], + "score": 1.0, + "content": "query window and each of", + "type": "text" + }, + { + "bbox": [ + 217, + 688, + 248, + 701 + ], + "score": 0.92, + "content": "s _ { r } ^ { l } \\times s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "summarized tokens. Since the focal attention for each window", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "can be performed independent of the others, we can compute Eq. (3) in parallel. Once we obtain", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 711, + 501, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 501, + 723 + ], + "score": 1.0, + "content": "attention scores for the whole input feature map, we send them to LayerNorm and the MLP block.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 73, + 495, + 232 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 73, + 495, + 232 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 73, + 495, + 232 + ], + "spans": [ + { + "bbox": [ + 119, + 73, + 495, + 232 + ], + "score": 0.973, + "type": "image", + "image_path": "8d7c5bc6d5803f11463c3ab8656b31fec8dbb82ddb891bec59961a9ad98dfe04.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 73, + 495, + 126.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 126.0, + 495, + 179.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 179.0, + 495, + 232.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 240, + 506, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "Figure 4: An illustration of focal attention at window level. Each of the square cells represents a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "visual token that is either from the original feature map or a summarized token formed by sub-window", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 263, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 338, + 275 + ], + "score": 1.0, + "content": "pooling. Suppose we have an input feature map of size", + "type": "text" + }, + { + "bbox": [ + 339, + 263, + 373, + 273 + ], + "score": 0.91, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 263, + 479, + 275 + ], + "score": 1.0, + "content": ". We first partition it into", + "type": "text" + }, + { + "bbox": [ + 480, + 263, + 504, + 273 + ], + "score": 0.89, + "content": "5 \\times 5", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 274, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 174, + 285 + ], + "score": 1.0, + "content": "windows of size", + "type": "text" + }, + { + "bbox": [ + 175, + 274, + 199, + 284 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 274, + 241, + 285 + ], + "score": 1.0, + "content": ". Take the", + "type": "text" + }, + { + "bbox": [ + 241, + 274, + 266, + 284 + ], + "score": 0.9, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 274, + 505, + 285 + ], + "score": 1.0, + "content": "blue window in the middle as the query set, we extract its", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "surrounding tokens at three granularity levels as its keys and values. For the first level, we extract the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 130, + 306 + ], + "score": 0.89, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "tokens which are closest to the blue window at the finest grain. At the second level, we expand", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 306, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 286, + 318 + ], + "score": 1.0, + "content": "the attention region and pool the surrounding", + "type": "text" + }, + { + "bbox": [ + 286, + 307, + 310, + 317 + ], + "score": 0.9, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 306, + 506, + 318 + ], + "score": 1.0, + "content": "sub-windows to form summarized tokens, which", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 145, + 329 + ], + "score": 1.0, + "content": "results in", + "type": "text" + }, + { + "bbox": [ + 145, + 318, + 169, + 328 + ], + "score": 0.9, + "content": "6 \\times 6", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "summarized tokens. At the third level, we attend a larger region covering the whole", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 195, + 340 + ], + "score": 1.0, + "content": "feature map and pool", + "type": "text" + }, + { + "bbox": [ + 195, + 329, + 219, + 339 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 327, + 339, + 340 + ], + "score": 1.0, + "content": "sub-windows, which leads to", + "type": "text" + }, + { + "bbox": [ + 340, + 328, + 363, + 339 + ], + "score": 0.9, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "summarized tokens. Finally, these", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 426, + 352 + ], + "score": 1.0, + "content": "three levels of tokens are concatenated to compute the keys and values for the", + "type": "text" + }, + { + "bbox": [ + 427, + 339, + 475, + 350 + ], + "score": 0.91, + "content": "4 \\times 4 = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 225, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 225, + 361 + ], + "score": 1.0, + "content": "(queries) in the blue window.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "feature map, there is no need to perform any sub-window pooling. Considering that the focal window", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "size is usually very small (7 maximally in our settings), the number of extra parameters introduced by", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 396, + 244, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 244, + 408 + ], + "score": 1.0, + "content": "sub-window pooling is negligible.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 373, + 505, + 408 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 366, + 426 + ], + "score": 1.0, + "content": "Attention computing. Once we obtain the pooled feature maps", + "type": "text" + }, + { + "bbox": [ + 366, + 411, + 392, + 424 + ], + "score": 0.93, + "content": "\\{ x ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 410, + 415, + 426 + ], + "score": 1.0, + "content": "at all", + "type": "text" + }, + { + "bbox": [ + 416, + 413, + 424, + 423 + ], + "score": 0.8, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 410, + 506, + 426 + ], + "score": 1.0, + "content": "levels, we compute", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 422, + 504, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 478, + 437 + ], + "score": 1.0, + "content": "the query at the first level, and key and value for all levels using three linear projection layers", + "type": "text" + }, + { + "bbox": [ + 479, + 423, + 504, + 435 + ], + "score": 0.57, + "content": "f _ { q } , f _ { k }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 433, + 201, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 123, + 448 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 435, + 134, + 446 + ], + "score": 0.89, + "content": "f _ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 433, + 201, + 448 + ], + "score": 1.0, + "content": ", respectively, as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 410, + 506, + 448 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 451, + 464, + 466 + ], + "lines": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "spans": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "score": 0.9, + "content": "Q = f _ { q } ( x ^ { 1 } ) , \\quad K = \\{ K ^ { l } \\} _ { 1 } ^ { L } = f _ { k } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) , \\quad V = \\{ V ^ { l } \\} _ { 1 } ^ { L } = f _ { v } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) .", + "type": "interline_equation", + "image_path": "adc5271b876d4969c2420a5cdd389cd1720ece17ee5092a72df5a2920aba267f.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 147, + 451, + 464, + 466 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 506, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "To perform focal attention, we need to first extract the surrounding tokens for each query token in the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 385, + 501 + ], + "score": 1.0, + "content": "feature map. As mentioned earlier, tokens inside a window partition", + "type": "text" + }, + { + "bbox": [ + 385, + 489, + 416, + 501 + ], + "score": 0.91, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "share the same set of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 497, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 104, + 497, + 270, + 515 + ], + "score": 1.0, + "content": "surroundings. For the queries inside the", + "type": "text" + }, + { + "bbox": [ + 270, + 502, + 275, + 511 + ], + "score": 0.73, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 497, + 322, + 515 + ], + "score": 1.0, + "content": "-th window", + "type": "text" + }, + { + "bbox": [ + 322, + 500, + 388, + 512 + ], + "score": 0.91, + "content": "Q _ { i } \\in \\mathcal { R } ^ { s _ { p } \\times s _ { p } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 497, + 451, + 515 + ], + "score": 1.0, + "content": ", we extract the", + "type": "text" + }, + { + "bbox": [ + 451, + 500, + 482, + 513 + ], + "score": 0.92, + "content": "s _ { r } ^ { l } \\times s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 497, + 506, + 515 + ], + "score": 1.0, + "content": "keys", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 171, + 525 + ], + "score": 1.0, + "content": "and values from", + "type": "text" + }, + { + "bbox": [ + 172, + 512, + 185, + 523 + ], + "score": 0.89, + "content": "K ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 511, + 203, + 525 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 512, + 215, + 523 + ], + "score": 0.88, + "content": "V ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "surrounding the window which the query lies in, and then gather the keys", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 103, + 521, + 507, + 538 + ], + "spans": [ + { + "bbox": [ + 103, + 521, + 183, + 538 + ], + "score": 1.0, + "content": "and values from all", + "type": "text" + }, + { + "bbox": [ + 184, + 524, + 192, + 533 + ], + "score": 0.8, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 521, + 254, + 538 + ], + "score": 1.0, + "content": "levels to obtain", + "type": "text" + }, + { + "bbox": [ + 255, + 523, + 372, + 536 + ], + "score": 0.91, + "content": "\\breve { K } _ { i } = \\{ K _ { i } ^ { 1 } , . . . , K _ { i } ^ { L } \\} \\in \\mathscr { R } ^ { \\bar { s } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 521, + 390, + 538 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 390, + 523, + 502, + 536 + ], + "score": 0.92, + "content": "V _ { i } = \\{ V _ { i } ^ { 1 } , . . . , \\mathbf { \\bar { V } } _ { i } ^ { L } \\} \\in \\mathcal { R } ^ { s \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 521, + 507, + 538 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 101, + 529, + 510, + 560 + ], + "spans": [ + { + "bbox": [ + 101, + 529, + 134, + 560 + ], + "score": 1.0, + "content": "where imple", + "type": "text" + }, + { + "bbox": [ + 134, + 539, + 140, + 547 + ], + "score": 0.73, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 529, + 340, + 560 + ], + "score": 1.0, + "content": "is the sum of focal regions from all levels, i.e., ntation of focal attention following Fig. 1 requires", + "type": "text" + }, + { + "bbox": [ + 340, + 535, + 406, + 550 + ], + "score": 0.93, + "content": "\\begin{array} { r } { s = \\sum _ { l = 1 } ^ { L } ( s _ { r } ^ { l } ) ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 529, + 510, + 560 + ], + "score": 1.0, + "content": ". Note that a canonicalverlapped regions across", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "score": 1.0, + "content": "different levels. In our implementation, we intentionally keep them in order to capture the pyramid", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "information for the overlapped regions. Finally, we follow [43] to include a relative position bias and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 580, + 258, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 232, + 594 + ], + "score": 1.0, + "content": "compute the focal attention for", + "type": "text" + }, + { + "bbox": [ + 232, + 581, + 244, + 592 + ], + "score": 0.89, + "content": "Q _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 580, + 258, + 594 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25, + "bbox_fs": [ + 101, + 478, + 510, + 594 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 598, + 399, + 623 + ], + "lines": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "spans": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "score": 0.92, + "content": "\\mathrm { A t t e n t i o n } ( Q _ { i } , K _ { i } , V _ { i } ) = \\mathrm { S o f t m a x } ( \\frac { Q _ { i } K _ { i } ^ { T } } { \\sqrt { d } } + B ) V _ { i } ,", + "type": "interline_equation", + "image_path": "a272e4bce5871346781b2ad83e37e4087f5507b82017abb37846d96c857a7475.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 210, + 598, + 399, + 623 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 629, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 507, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 133, + 643 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 630, + 183, + 642 + ], + "score": 0.94, + "content": "B = \\{ B ^ { l } \\} _ { 1 } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 629, + 391, + 643 + ], + "score": 1.0, + "content": "is the learnable relative position bias. It consists of", + "type": "text" + }, + { + "bbox": [ + 392, + 631, + 399, + 640 + ], + "score": 0.82, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 629, + 446, + 643 + ], + "score": 1.0, + "content": "subsets for", + "type": "text" + }, + { + "bbox": [ + 447, + 631, + 455, + 640 + ], + "score": 0.82, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 629, + 507, + 643 + ], + "score": 1.0, + "content": "focal levels.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 640, + 507, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 334, + 657 + ], + "score": 1.0, + "content": "Similar to [43], for the first level, we parameterize it as", + "type": "text" + }, + { + "bbox": [ + 335, + 641, + 433, + 653 + ], + "score": 0.91, + "content": "B ^ { 1 } \\in \\mathcal { R } ^ { ( 2 s _ { p } - 1 ) \\times ( 2 s _ { p } - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 640, + 507, + 657 + ], + "score": 1.0, + "content": ", considering that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 325, + 667 + ], + "score": 1.0, + "content": "the horizontal and vertical position ranges are both in", + "type": "text" + }, + { + "bbox": [ + 325, + 654, + 396, + 666 + ], + "score": 0.91, + "content": "[ - s _ { p } + 1 , s _ { p } - 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 654, + 506, + 667 + ], + "score": 1.0, + "content": ". For the other focal levels,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 665, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 506, + 677 + ], + "score": 1.0, + "content": "considering that they have different granularity with respect to the queries, we treat all the queries", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 243, + 691 + ], + "score": 1.0, + "content": "inside a window equally and use", + "type": "text" + }, + { + "bbox": [ + 243, + 676, + 298, + 688 + ], + "score": 0.92, + "content": "B ^ { l } \\in \\mathcal { R } ^ { s _ { r } ^ { l } \\times s _ { r } ^ { l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 676, + 506, + 691 + ], + "score": 1.0, + "content": "to represent the relative position bias between the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 217, + 701 + ], + "score": 1.0, + "content": "query window and each of", + "type": "text" + }, + { + "bbox": [ + 217, + 688, + 248, + 701 + ], + "score": 0.92, + "content": "s _ { r } ^ { l } \\times s _ { r } ^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "summarized tokens. Since the focal attention for each window", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "can be performed independent of the others, we can compute Eq. (3) in parallel. Once we obtain", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 711, + 501, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 501, + 723 + ], + "score": 1.0, + "content": "attention scores for the whole input feature map, we send them to LayerNorm and the MLP block.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 629, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 70, + 496, + 247 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 70, + 496, + 247 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 70, + 493, + 247 + ], + "spans": [ + { + "bbox": [ + 119, + 70, + 493, + 247 + ], + "score": 0.986, + "html": "
Output SizeLayer NameFocal-TinyFocal-SmallFocal-Base
stage 156×56Patch Embeddingp1= 4;c1 = 96p1=4;c1= 96p1 = 4;c1 = 128
56×56Transformer Block{1,13} 三 s={7,7×2二 {1,13} ={7,7}×2{1,13} ={7,7}×2
stage 228×28Patch EmbeddingP2=2;c=192P2=2;C=192P2=2;C= 256
28×28Transformer Block{1,13} 三 swr={7,5} 1×2={1,13} sw,r={7,5} 1×2{1,13} 二 su,r={7,5}×2
stage 314 × 14Patch Embeddingp3=2;c3=384p3=2; c3= 384p3=2; c3= 512
14 × 14Transformer Block={1,13} s={7,3}×6={1,13} ={7,3}×18={1,13} ={7,3}×18
stage 47×7Patch EmbeddingP4=2;C4=768P4=2;C4=768P4= 2;C4=1024
7×7Transformer Block二 {1,7}×2{1,7} s 三 ={7,1}×2二 {1,7} su,r {7,1} 二×2
", + "type": "table", + "image_path": "2e084ddc28027667351980bf367187c19e42a436814641051a239edbb9b8f4a0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 70, + 496, + 129.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 129.0, + 496, + 188.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 188.0, + 496, + 247.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 107, + 248, + 503, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Table 1: Model configurations for Focal Transformers. We use three configurations with different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 259, + 344, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 344, + 272 + ], + "score": 1.0, + "content": "model capacities: Focal-Tiny, Focal-Small and Focal-Base.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 284, + 224, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 225, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 225, + 298 + ], + "score": 1.0, + "content": "3.2.2 Complexity analysis", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "We analyze the computational complexity for the two steps of focal attention described above. For the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 305, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 305, + 176, + 321 + ], + "score": 1.0, + "content": "input feature map", + "type": "text" + }, + { + "bbox": [ + 176, + 307, + 236, + 319 + ], + "score": 0.93, + "content": "\\dot { \\boldsymbol { x } } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 305, + 273, + 321 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 273, + 307, + 305, + 323 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 305, + 506, + 321 + ], + "score": 1.0, + "content": "sub-windows at focal level l. For each sub-window,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 311, + 336 + ], + "score": 1.0, + "content": "the pooling operation in Eq.1 has the complexity of", + "type": "text" + }, + { + "bbox": [ + 311, + 322, + 356, + 335 + ], + "score": 0.92, + "content": "\\mathcal { O } ( ( s _ { w } ^ { l } ) ^ { 2 } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 322, + 506, + 336 + ], + "score": 1.0, + "content": ". Aggregating all sub-windows brings", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 334, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 118, + 347 + ], + "score": 1.0, + "content": "us", + "type": "text" + }, + { + "bbox": [ + 118, + 334, + 167, + 346 + ], + "score": 0.91, + "content": "O ( ( M N ) d )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 334, + 379, + 347 + ], + "score": 1.0, + "content": ". Then for all focal levels, we have the complexity of", + "type": "text" + }, + { + "bbox": [ + 379, + 334, + 435, + 346 + ], + "score": 0.93, + "content": "\\mathcal { O } ( L ( M N ) d )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 334, + 506, + 347 + ], + "score": 1.0, + "content": "in total, which is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 345, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 357 + ], + "score": 1.0, + "content": "independent of the sub-window size at each focal level. Regarding the attention computation in Eq. 3,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 355, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 277, + 370 + ], + "score": 1.0, + "content": "the computational cost for a query window", + "type": "text" + }, + { + "bbox": [ + 277, + 357, + 309, + 368 + ], + "score": 0.93, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 355, + 319, + 370 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 320, + 355, + 399, + 368 + ], + "score": 0.93, + "content": "\\mathcal { O } ( ( s _ { p } ) ^ { 2 } \\textstyle \\sum _ { l } ( s _ { r } ^ { l } ) ^ { 2 } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 355, + 420, + 370 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 421, + 355, + 505, + 368 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\dot { \\sum } _ { l } ( s _ { r } ^ { l } ) ^ { 2 } ( M \\dot { N } ) d )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 365, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "for the whole input feature map. To sum up, the overall computational cost for focal attention is", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 376, + 507, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 377, + 218, + 390 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathcal { O } ( ( L + \\sum _ { l } ( s _ { r } ^ { l } ) ^ { \\bar { 2 } } ) ( M N ) d ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 376, + 349, + 391 + ], + "score": 1.0, + "content": ". In an extreme case, one can set", + "type": "text" + }, + { + "bbox": [ + 350, + 376, + 457, + 390 + ], + "score": 0.91, + "content": "s _ { r } ^ { \\hat { L } } = 2 \\times \\operatorname* { m a x } ( M , N ) / s _ { w } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 376, + 507, + 391 + ], + "score": 1.0, + "content": "to ensure a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 388, + 475, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 475, + 402 + ], + "score": 1.0, + "content": "global receptive field for all queries (including both corner and middle queries) in this layer.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 414, + 221, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 222, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 222, + 428 + ], + "score": 1.0, + "content": "3.3 Model configurations", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 429, + 505, + 571 + ], + "lines": [ + { + "bbox": [ + 105, + 429, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 506, + 442 + ], + "score": 1.0, + "content": "For fair comparison, we consider three network configurations for Focal Transformers, follow-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 440, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 452 + ], + "score": 1.0, + "content": "ing [60, 64, 43]. Specifically, we follow the design of the Tiny, Small and Base models in Swin", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 329, + 463 + ], + "score": 1.0, + "content": "Transformer [43], as shown in Table 1. Our models take", + "type": "text" + }, + { + "bbox": [ + 329, + 451, + 372, + 462 + ], + "score": 0.91, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "images as inputs and the window", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "partition size is set to 7 to make our models comparable to Swin Transformers. For the focal attention", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 473, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 506, + 486 + ], + "score": 1.0, + "content": "layer, we introduce two levels, one for fine-grained local attention and the other for coarse-grained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "global attention. Except for the last stage, the focal region size is set to 13 for the window partition", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "size of 7, which means that we expand 3 tokens for each window partition. For the last stage, since", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 206, + 518 + ], + "score": 1.0, + "content": "the whole feature map is", + "type": "text" + }, + { + "bbox": [ + 206, + 506, + 230, + 516 + ], + "score": 0.89, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 505, + 506, + 518 + ], + "score": 1.0, + "content": ", the focal region size at level 0 is set to 7, which is sufficient to cover", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "the entire feature map. For the coarse-grained global attention, we set its focal window size the same", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 527, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 445, + 540 + ], + "score": 1.0, + "content": "as the window partition size 7, but gradually decrease the focal region size to get", + "type": "text" + }, + { + "bbox": [ + 445, + 527, + 489, + 540 + ], + "score": 0.92, + "content": "\\{ 7 , 5 , 3 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 527, + 506, + 540 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 464, + 551 + ], + "score": 1.0, + "content": "the four stages, respectively. For the patch embedding layer, the spatial reduction ratio", + "type": "text" + }, + { + "bbox": [ + 464, + 541, + 474, + 550 + ], + "score": 0.85, + "content": "p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "for the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 180, + 562 + ], + "score": 1.0, + "content": "four stages are all", + "type": "text" + }, + { + "bbox": [ + 181, + 549, + 224, + 561 + ], + "score": 0.9, + "content": "\\{ 4 , 2 , 2 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 549, + 440, + 562 + ], + "score": 1.0, + "content": ". Note that Focal-Base has a higher hidden dimension", + "type": "text" + }, + { + "bbox": [ + 440, + 551, + 449, + 560 + ], + "score": 0.83, + "content": "c _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 549, + 505, + 562 + ], + "score": 1.0, + "content": ", compared to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 560, + 223, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 223, + 572 + ], + "score": 1.0, + "content": "Focal-Tiny and Focal-Small.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 588, + 191, + 602 + ], + "lines": [ + { + "bbox": [ + 104, + 586, + 193, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 586, + 193, + 605 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "title", + "bbox": [ + 107, + 608, + 286, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 288, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 288, + 622 + ], + "score": 1.0, + "content": "4.1 Image classification on ImageNet-1K", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 623, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "We compare different methods on ImageNet-1K [19]. For fair comparison, we follow the training", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "score": 1.0, + "content": "recipes in [55, 60]. All models are trained for 300 epochs with batch size 1024. The initial", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 643, + 507, + 660 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 192, + 660 + ], + "score": 1.0, + "content": "learning rate is set to", + "type": "text" + }, + { + "bbox": [ + 192, + 645, + 214, + 655 + ], + "score": 0.9, + "content": "1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 643, + 408, + 660 + ], + "score": 1.0, + "content": "with 20 epochs of linear warm-up starting from", + "type": "text" + }, + { + "bbox": [ + 408, + 645, + 430, + 655 + ], + "score": 0.89, + "content": "1 0 ^ { - 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 643, + 507, + 660 + ], + "score": 1.0, + "content": ". For optimization,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 654, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 506, + 670 + ], + "score": 1.0, + "content": "we use AdamW [44] as the optimizer with a cosine learning rate scheduler. The weight decay", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "is set to 0.05 and the maximal gradient norm is clipped to 5.0. We use the same set of data", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "augmentation and regularization strategies used in [55] after excluding random erasing [82], repeated", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 689, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 505, + 702 + ], + "score": 1.0, + "content": "augmentation [4, 34] and exponential moving average (EMA) [48]. The stochastic depth drop rates", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 698, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 104, + 698, + 506, + 715 + ], + "score": 1.0, + "content": "are set to 0.2, 0.2 and 0.3 for our tiny, small and base models, respectively. During training, we crop", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 711, + 501, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 188, + 724 + ], + "score": 1.0, + "content": "images randomly to", + "type": "text" + }, + { + "bbox": [ + 189, + 711, + 231, + 721 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 711, + 501, + 724 + ], + "score": 1.0, + "content": ", while a center crop is used during evaluation on the validation set.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 70, + 496, + 247 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 70, + 496, + 247 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 70, + 493, + 247 + ], + "spans": [ + { + "bbox": [ + 119, + 70, + 493, + 247 + ], + "score": 0.986, + "html": "
Output SizeLayer NameFocal-TinyFocal-SmallFocal-Base
stage 156×56Patch Embeddingp1= 4;c1 = 96p1=4;c1= 96p1 = 4;c1 = 128
56×56Transformer Block{1,13} 三 s={7,7×2二 {1,13} ={7,7}×2{1,13} ={7,7}×2
stage 228×28Patch EmbeddingP2=2;c=192P2=2;C=192P2=2;C= 256
28×28Transformer Block{1,13} 三 swr={7,5} 1×2={1,13} sw,r={7,5} 1×2{1,13} 二 su,r={7,5}×2
stage 314 × 14Patch Embeddingp3=2;c3=384p3=2; c3= 384p3=2; c3= 512
14 × 14Transformer Block={1,13} s={7,3}×6={1,13} ={7,3}×18={1,13} ={7,3}×18
stage 47×7Patch EmbeddingP4=2;C4=768P4=2;C4=768P4= 2;C4=1024
7×7Transformer Block二 {1,7}×2{1,7} s 三 ={7,1}×2二 {1,7} su,r {7,1} 二×2
", + "type": "table", + "image_path": "2e084ddc28027667351980bf367187c19e42a436814641051a239edbb9b8f4a0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 70, + 496, + 129.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 129.0, + 496, + 188.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 188.0, + 496, + 247.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 107, + 248, + 503, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Table 1: Model configurations for Focal Transformers. We use three configurations with different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 259, + 344, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 344, + 272 + ], + "score": 1.0, + "content": "model capacities: Focal-Tiny, Focal-Small and Focal-Base.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 284, + 224, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 225, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 225, + 298 + ], + "score": 1.0, + "content": "3.2.2 Complexity analysis", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "We analyze the computational complexity for the two steps of focal attention described above. For the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 305, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 305, + 176, + 321 + ], + "score": 1.0, + "content": "input feature map", + "type": "text" + }, + { + "bbox": [ + 176, + 307, + 236, + 319 + ], + "score": 0.93, + "content": "\\dot { \\boldsymbol { x } } \\in \\mathcal { R } ^ { M \\times N \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 305, + 273, + 321 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 273, + 307, + 305, + 323 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 305, + 506, + 321 + ], + "score": 1.0, + "content": "sub-windows at focal level l. For each sub-window,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 311, + 336 + ], + "score": 1.0, + "content": "the pooling operation in Eq.1 has the complexity of", + "type": "text" + }, + { + "bbox": [ + 311, + 322, + 356, + 335 + ], + "score": 0.92, + "content": "\\mathcal { O } ( ( s _ { w } ^ { l } ) ^ { 2 } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 322, + 506, + 336 + ], + "score": 1.0, + "content": ". Aggregating all sub-windows brings", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 334, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 118, + 347 + ], + "score": 1.0, + "content": "us", + "type": "text" + }, + { + "bbox": [ + 118, + 334, + 167, + 346 + ], + "score": 0.91, + "content": "O ( ( M N ) d )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 334, + 379, + 347 + ], + "score": 1.0, + "content": ". Then for all focal levels, we have the complexity of", + "type": "text" + }, + { + "bbox": [ + 379, + 334, + 435, + 346 + ], + "score": 0.93, + "content": "\\mathcal { O } ( L ( M N ) d )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 334, + 506, + 347 + ], + "score": 1.0, + "content": "in total, which is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 345, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 357 + ], + "score": 1.0, + "content": "independent of the sub-window size at each focal level. Regarding the attention computation in Eq. 3,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 355, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 277, + 370 + ], + "score": 1.0, + "content": "the computational cost for a query window", + "type": "text" + }, + { + "bbox": [ + 277, + 357, + 309, + 368 + ], + "score": 0.93, + "content": "s _ { p } \\times s _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 355, + 319, + 370 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 320, + 355, + 399, + 368 + ], + "score": 0.93, + "content": "\\mathcal { O } ( ( s _ { p } ) ^ { 2 } \\textstyle \\sum _ { l } ( s _ { r } ^ { l } ) ^ { 2 } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 355, + 420, + 370 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 421, + 355, + 505, + 368 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\dot { \\sum } _ { l } ( s _ { r } ^ { l } ) ^ { 2 } ( M \\dot { N } ) d )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 365, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "for the whole input feature map. To sum up, the overall computational cost for focal attention is", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 376, + 507, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 377, + 218, + 390 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathcal { O } ( ( L + \\sum _ { l } ( s _ { r } ^ { l } ) ^ { \\bar { 2 } } ) ( M N ) d ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 376, + 349, + 391 + ], + "score": 1.0, + "content": ". In an extreme case, one can set", + "type": "text" + }, + { + "bbox": [ + 350, + 376, + 457, + 390 + ], + "score": 0.91, + "content": "s _ { r } ^ { \\hat { L } } = 2 \\times \\operatorname* { m a x } ( M , N ) / s _ { w } ^ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 376, + 507, + 391 + ], + "score": 1.0, + "content": "to ensure a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 388, + 475, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 475, + 402 + ], + "score": 1.0, + "content": "global receptive field for all queries (including both corner and middle queries) in this layer.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10, + "bbox_fs": [ + 104, + 297, + 507, + 402 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 414, + 221, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 222, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 222, + 428 + ], + "score": 1.0, + "content": "3.3 Model configurations", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 429, + 505, + 571 + ], + "lines": [ + { + "bbox": [ + 105, + 429, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 506, + 442 + ], + "score": 1.0, + "content": "For fair comparison, we consider three network configurations for Focal Transformers, follow-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 440, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 452 + ], + "score": 1.0, + "content": "ing [60, 64, 43]. Specifically, we follow the design of the Tiny, Small and Base models in Swin", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 329, + 463 + ], + "score": 1.0, + "content": "Transformer [43], as shown in Table 1. Our models take", + "type": "text" + }, + { + "bbox": [ + 329, + 451, + 372, + 462 + ], + "score": 0.91, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "images as inputs and the window", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "partition size is set to 7 to make our models comparable to Swin Transformers. For the focal attention", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 473, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 506, + 486 + ], + "score": 1.0, + "content": "layer, we introduce two levels, one for fine-grained local attention and the other for coarse-grained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "global attention. Except for the last stage, the focal region size is set to 13 for the window partition", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "size of 7, which means that we expand 3 tokens for each window partition. For the last stage, since", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 206, + 518 + ], + "score": 1.0, + "content": "the whole feature map is", + "type": "text" + }, + { + "bbox": [ + 206, + 506, + 230, + 516 + ], + "score": 0.89, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 505, + 506, + 518 + ], + "score": 1.0, + "content": ", the focal region size at level 0 is set to 7, which is sufficient to cover", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "the entire feature map. For the coarse-grained global attention, we set its focal window size the same", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 527, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 445, + 540 + ], + "score": 1.0, + "content": "as the window partition size 7, but gradually decrease the focal region size to get", + "type": "text" + }, + { + "bbox": [ + 445, + 527, + 489, + 540 + ], + "score": 0.92, + "content": "\\{ 7 , 5 , 3 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 527, + 506, + 540 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 464, + 551 + ], + "score": 1.0, + "content": "the four stages, respectively. For the patch embedding layer, the spatial reduction ratio", + "type": "text" + }, + { + "bbox": [ + 464, + 541, + 474, + 550 + ], + "score": 0.85, + "content": "p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "for the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 180, + 562 + ], + "score": 1.0, + "content": "four stages are all", + "type": "text" + }, + { + "bbox": [ + 181, + 549, + 224, + 561 + ], + "score": 0.9, + "content": "\\{ 4 , 2 , 2 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 549, + 440, + 562 + ], + "score": 1.0, + "content": ". Note that Focal-Base has a higher hidden dimension", + "type": "text" + }, + { + "bbox": [ + 440, + 551, + 449, + 560 + ], + "score": 0.83, + "content": "c _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 549, + 505, + 562 + ], + "score": 1.0, + "content": ", compared to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 560, + 223, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 223, + 572 + ], + "score": 1.0, + "content": "Focal-Tiny and Focal-Small.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 429, + 506, + 572 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 588, + 191, + 602 + ], + "lines": [ + { + "bbox": [ + 104, + 586, + 193, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 586, + 193, + 605 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "title", + "bbox": [ + 107, + 608, + 286, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 288, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 288, + 622 + ], + "score": 1.0, + "content": "4.1 Image classification on ImageNet-1K", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 623, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "We compare different methods on ImageNet-1K [19]. For fair comparison, we follow the training", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "score": 1.0, + "content": "recipes in [55, 60]. All models are trained for 300 epochs with batch size 1024. The initial", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 643, + 507, + 660 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 192, + 660 + ], + "score": 1.0, + "content": "learning rate is set to", + "type": "text" + }, + { + "bbox": [ + 192, + 645, + 214, + 655 + ], + "score": 0.9, + "content": "1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 643, + 408, + 660 + ], + "score": 1.0, + "content": "with 20 epochs of linear warm-up starting from", + "type": "text" + }, + { + "bbox": [ + 408, + 645, + 430, + 655 + ], + "score": 0.89, + "content": "1 0 ^ { - 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 643, + 507, + 660 + ], + "score": 1.0, + "content": ". For optimization,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 654, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 506, + 670 + ], + "score": 1.0, + "content": "we use AdamW [44] as the optimizer with a cosine learning rate scheduler. The weight decay", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "is set to 0.05 and the maximal gradient norm is clipped to 5.0. We use the same set of data", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "augmentation and regularization strategies used in [55] after excluding random erasing [82], repeated", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 689, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 505, + 702 + ], + "score": 1.0, + "content": "augmentation [4, 34] and exponential moving average (EMA) [48]. The stochastic depth drop rates", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 698, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 104, + 698, + 506, + 715 + ], + "score": 1.0, + "content": "are set to 0.2, 0.2 and 0.3 for our tiny, small and base models, respectively. During training, we crop", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 711, + 501, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 188, + 724 + ], + "score": 1.0, + "content": "images randomly to", + "type": "text" + }, + { + "bbox": [ + 189, + 711, + 231, + 721 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 711, + 501, + 724 + ], + "score": 1.0, + "content": ", while a center crop is used during evaluation on the validation set.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35, + "bbox_fs": [ + 104, + 623, + 507, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 70, + 280, + 275 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 70, + 280, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 70, + 280, + 275 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 280, + 275 + ], + "score": 0.978, + "html": "
Model#Params. FLOPsTop-1 (%)
ResNet-50 [33]25.0 4.176.2
DeiT-Small/16 [55]22.1 4.679.9
PVT-Small [60]24.5 3.879.8
ViL-Small [76]24.6 5.182.0
CvT-13 [64]20.0 4.581.6
Swin-Tiny [43]28.3 4.581.2
Focal-Tiny (Ours)28.9 4.982.2
ResNet-101[33]45.0 7.977.4
PVT-Medium [60]44.2 6.781.2
CvT-21 [64]32.0 7.182.5
ViL-Medium [76]39.7 9.183.3
Swin-Small [43]49.6 8.783.1
Focal-Small (Ours)51.1 9.483.6
ResNet-152[33]60.0 11.078.3
ViT-Base/16 [21]86.6 17.677.9
DeiT-Base/16 [55]17.581.8
86.6
PVT-Large [60]61.4 9.881.7
ViL-Base[76]55.7 13.483.2
Swin-Base [43]87.8 15.483.4
Focal-Base (Ours)89.816.4 84.0
", + "type": "table", + "image_path": "9f9cf0ebeecc16dfc400d5f622af2df25d867b1a4030e821ba17190b276a552a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 280, + 83.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 83.66666666666667, + 280, + 97.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 97.33333333333334, + 280, + 111.00000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 111.00000000000001, + 280, + 124.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 110, + 124.66666666666669, + 280, + 138.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 138.33333333333334, + 280, + 152.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 152.0, + 280, + 165.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 165.66666666666666, + 280, + 179.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 110, + 179.33333333333331, + 280, + 192.99999999999997 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 192.99999999999997, + 280, + 206.66666666666663 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 110, + 206.66666666666663, + 280, + 220.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 110, + 220.3333333333333, + 280, + 233.99999999999994 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 110, + 233.99999999999994, + 280, + 247.6666666666666 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 110, + 247.6666666666666, + 280, + 261.33333333333326 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 261.33333333333326, + 280, + 274.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 277, + 287, + 320 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 276, + 287, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 287, + 288 + ], + "score": 1.0, + "content": "Table 2: Comparison of image classification", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 286, + 287, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 287, + 299 + ], + "score": 1.0, + "content": "on ImageNet-1K for different models. Except", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 298, + 287, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 287, + 309 + ], + "score": 1.0, + "content": "for ViT-Base/16, all other models are trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 309, + 265, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 176, + 321 + ], + "score": 1.0, + "content": "and evaluated on", + "type": "text" + }, + { + "bbox": [ + 176, + 309, + 219, + 320 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 309, + 265, + 321 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + } + ], + "index": 11.75 + }, + { + "type": "table", + "bbox": [ + 299, + 70, + 501, + 254 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 299, + 70, + 501, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 299, + 70, + 501, + 254 + ], + "spans": [ + { + "bbox": [ + 299, + 70, + 501, + 254 + ], + "score": 0.98, + "html": "
BackboneRetinaNetMask R-CNN
APbApbAPm
ResNet-50 [33]36.338.034.4
PVT-Small40.440.437.8
ViL-Small [76]41.641.838.5
Swin-Tiny [43]42.043.739.8
Focal-Tiny (Ours)43.7 (+1.7)44.8 (+1.1) 41.0 (+1.3)
ResNet-101[33]38.540.436.4
ResNeXt101-32x4d [67]39.941.937.5
PVT-Medium [60]41.942.039.0
ViL-Medium [76]42.943.439.7
Swin-Small [43]45.046.542.1
Focal-Small (Ours)45.6 (+0.6)47.4 (+0.9) 42.8 (+0.7)
ResNeXt101-64x4d[67] 41.042.838.4
PVT-Large [60]42.642.939.5
ViL-Base[76]44.345.141.0
Swin-Base [43]45.046.942.3
Focal-Base (Ours)46.3 (+1.3)47.8 (+0.9)43.2 (+0.9)
", + "type": "table", + "image_path": "2eccde7cb9cd36279d6f0e1059ce2743dcec73d9f7583e972b12720e375f7d35.jpg" + } + ] + } + ], + "index": 25.5, + "virtual_lines": [ + { + "bbox": [ + 299, + 70, + 501, + 83.14285714285714 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 299, + 83.14285714285714, + 501, + 96.28571428571428 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 299, + 96.28571428571428, + 501, + 109.42857142857142 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 299, + 109.42857142857142, + 501, + 122.57142857142856 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 299, + 122.57142857142856, + 501, + 135.7142857142857 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 299, + 135.7142857142857, + 501, + 148.85714285714283 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 299, + 148.85714285714283, + 501, + 161.99999999999997 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 299, + 161.99999999999997, + 501, + 175.1428571428571 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 299, + 175.1428571428571, + 501, + 188.28571428571425 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 299, + 188.28571428571425, + 501, + 201.4285714285714 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 299, + 201.4285714285714, + 501, + 214.57142857142853 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 299, + 214.57142857142853, + 501, + 227.71428571428567 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 299, + 227.71428571428567, + 501, + 240.8571428571428 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 299, + 240.8571428571428, + 501, + 253.99999999999994 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 297, + 255, + 505, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 295, + 255, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 295, + 255, + 506, + 267 + ], + "score": 1.0, + "content": "Table 3: Comparisons with CNN and Transformer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 295, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 295, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "baselines and SoTA methods on COCO object detec-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 296, + 277, + 504, + 289 + ], + "spans": [ + { + "bbox": [ + 296, + 277, + 379, + 289 + ], + "score": 1.0, + "content": "tion. The box mAP", + "type": "text" + }, + { + "bbox": [ + 380, + 277, + 407, + 289 + ], + "score": 0.86, + "content": "( A P ^ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 277, + 474, + 289 + ], + "score": 1.0, + "content": "and mask mAP", + "type": "text" + }, + { + "bbox": [ + 474, + 278, + 504, + 289 + ], + "score": 0.84, + "content": "( A P ^ { m } )", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 295, + 288, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 295, + 288, + 505, + 299 + ], + "score": 1.0, + "content": "are reported for RetinaNet and Mask R-CNN trained", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 295, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 295, + 298, + 316, + 311 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 317, + 300, + 331, + 310 + ], + "score": 0.87, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "schedule. More detailed comparisons with", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 296, + 310, + 408, + 321 + ], + "spans": [ + { + "bbox": [ + 296, + 311, + 311, + 321 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 310, + 408, + 321 + ], + "score": 1.0, + "content": "schedule are in Table 4.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "score": 1.0, + "content": "In Table 2, we summarize the results for baseline models and the state-of-the-art models on image", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "classification task. We can see that Focal Transformers consistently outperform other methods with", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 354, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 370 + ], + "score": 1.0, + "content": "similar model sizes (#Params.) and computational complexities (GFLOPs). Specifically, Focal-Tiny", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 366, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 340, + 379 + ], + "score": 1.0, + "content": "improves over the Transformer baseline DeiT-Small/16 by", + "type": "text" + }, + { + "bbox": [ + 340, + 367, + 362, + 377 + ], + "score": 0.86, + "content": "2 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 366, + 506, + 379 + ], + "score": 1.0, + "content": ". Meanwhile, using the same model", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "configuration (2-2-6-2) and a few extra parameters and computations, Focal-Tiny improves over", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "Swin-Tiny by 1.0 point. For small and base models, Focal-Small with 51.1M parameters can reach", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 133, + 410 + ], + "score": 0.86, + "content": "8 3 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "which is better than all the counterpart small and base models using much less parameters. By", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 326, + 423 + ], + "score": 1.0, + "content": "increasing the model size, Focal-Base model achieves", + "type": "text" + }, + { + "bbox": [ + 326, + 410, + 353, + 421 + ], + "score": 0.87, + "content": "8 4 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 409, + 505, + 423 + ], + "score": 1.0, + "content": ", surpassing all the other models with", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 421, + 252, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 252, + 434 + ], + "score": 1.0, + "content": "comparable parameters and FLOPs.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "To compare with the large-scale models, we further build Focal-Large Transformer by increasing the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "hidden dimension in Focal-Base from 128 to 196 while keeping all the other hyperparameters the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 459, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 473 + ], + "score": 1.0, + "content": "same. We follow the common practice to pretrain our Focal-Large Transformer on ImageNet-22K", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 470, + 344, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 344, + 482 + ], + "score": 1.0, + "content": "and transfer it to detection and segmentation tasks [64, 43].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49.5 + }, + { + "type": "title", + "bbox": [ + 107, + 495, + 316, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 317, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 317, + 509 + ], + "score": 1.0, + "content": "4.2 Object detection and instance segmentation", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 52 + }, + { + "type": "text", + "bbox": [ + 107, + 509, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "We benchmark our models on object detection with COCO 2017 [42]. The pretrained models are used", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "score": 1.0, + "content": "as visual backbones and then plugged into two representative pipelines, RetinaNet [41] and Mask", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 281, + 543 + ], + "score": 1.0, + "content": "R-CNN [32]. All models are trained on the", + "type": "text" + }, + { + "bbox": [ + 281, + 531, + 303, + 541 + ], + "score": 0.25, + "content": "1 1 8 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "training images and the results are reported on 5K", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 335, + 554 + ], + "score": 1.0, + "content": "validation set. We use the two standard training schedules,", + "type": "text" + }, + { + "bbox": [ + 336, + 542, + 350, + 552 + ], + "score": 0.89, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 541, + 428, + 554 + ], + "score": 1.0, + "content": "with 12 epochs and", + "type": "text" + }, + { + "bbox": [ + 428, + 542, + 442, + 552 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "with 36 epochs.", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 136, + 565 + ], + "score": 1.0, + "content": "For the", + "type": "text" + }, + { + "bbox": [ + 136, + 553, + 151, + 563 + ], + "score": 0.86, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "schedule, we resize image’s shorter side to 800 while keeping its longer side no more than", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 164, + 576 + ], + "score": 1.0, + "content": "1,333. For the", + "type": "text" + }, + { + "bbox": [ + 164, + 564, + 178, + 574 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "schedule, we use the multi-scale training strategy by randomly resizing its shorter", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 574, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 505, + 587 + ], + "score": 1.0, + "content": "side to the range of [480, 800]. Considering this higher input resolution, we adaptively increase the", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 585, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 506, + 597 + ], + "score": 1.0, + "content": "focal sizes at four stages to (15, 13, 9, 7), to ensures that the focal attention covers more than half of", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "the image region at the first two stages, and the whole image at the last two stages. With the focal size", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "increased, the relative position biases are accordingly up-sampled to the corresponding sizes using", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 617, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 631 + ], + "score": 1.0, + "content": "bilinear interpolation. During training, we use AdamW [44] for optimization with initial learning", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 104, + 627, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 104, + 627, + 124, + 642 + ], + "score": 1.0, + "content": "rate", + "type": "text" + }, + { + "bbox": [ + 124, + 628, + 146, + 639 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 627, + 506, + 642 + ], + "score": 1.0, + "content": "and weight decay 0.05. Similarly, we use 0.2, 0.3 and 0.5 stochastic depth drop rates to", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "regularize the training for our Tiny, Small and Base models, respectively. Since Swin Transformer", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "does not report the results on RetinaNet, we obtain the results by ourselves using their official code", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 106, + 662, + 358, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 358, + 674 + ], + "score": 1.0, + "content": "with the same hyper-parameters as that of Focal Transformers.", + "type": "text" + } + ], + "index": 67 + } + ], + "index": 60 + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "score": 1.0, + "content": "In Table 3, we show the performance for both CNN-based models and the current Transformer-", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 295, + 702 + ], + "score": 1.0, + "content": "based state-of-the-art models. The bbox mAP", + "type": "text" + }, + { + "bbox": [ + 295, + 688, + 322, + 700 + ], + "score": 0.87, + "content": "( A P ^ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 687, + 388, + 702 + ], + "score": 1.0, + "content": "and mask mAP", + "type": "text" + }, + { + "bbox": [ + 388, + 689, + 418, + 700 + ], + "score": 0.85, + "content": "( A P ^ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "are reported. We see", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "that Focal Transformers outperform the CNN-based models consistently with the gap of 4.8-7.1", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "score": 1.0, + "content": "points. Compared with the other methods which also use multi-scale Transformer architectures,", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 69.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 753 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 70, + 280, + 275 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 70, + 280, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 70, + 280, + 275 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 280, + 275 + ], + "score": 0.978, + "html": "
Model#Params. FLOPsTop-1 (%)
ResNet-50 [33]25.0 4.176.2
DeiT-Small/16 [55]22.1 4.679.9
PVT-Small [60]24.5 3.879.8
ViL-Small [76]24.6 5.182.0
CvT-13 [64]20.0 4.581.6
Swin-Tiny [43]28.3 4.581.2
Focal-Tiny (Ours)28.9 4.982.2
ResNet-101[33]45.0 7.977.4
PVT-Medium [60]44.2 6.781.2
CvT-21 [64]32.0 7.182.5
ViL-Medium [76]39.7 9.183.3
Swin-Small [43]49.6 8.783.1
Focal-Small (Ours)51.1 9.483.6
ResNet-152[33]60.0 11.078.3
ViT-Base/16 [21]86.6 17.677.9
DeiT-Base/16 [55]17.581.8
86.6
PVT-Large [60]61.4 9.881.7
ViL-Base[76]55.7 13.483.2
Swin-Base [43]87.8 15.483.4
Focal-Base (Ours)89.816.4 84.0
", + "type": "table", + "image_path": "9f9cf0ebeecc16dfc400d5f622af2df25d867b1a4030e821ba17190b276a552a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 280, + 83.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 83.66666666666667, + 280, + 97.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 97.33333333333334, + 280, + 111.00000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 111.00000000000001, + 280, + 124.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 110, + 124.66666666666669, + 280, + 138.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 138.33333333333334, + 280, + 152.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 152.0, + 280, + 165.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 165.66666666666666, + 280, + 179.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 110, + 179.33333333333331, + 280, + 192.99999999999997 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 192.99999999999997, + 280, + 206.66666666666663 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 110, + 206.66666666666663, + 280, + 220.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 110, + 220.3333333333333, + 280, + 233.99999999999994 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 110, + 233.99999999999994, + 280, + 247.6666666666666 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 110, + 247.6666666666666, + 280, + 261.33333333333326 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 261.33333333333326, + 280, + 274.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 277, + 287, + 320 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 276, + 287, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 287, + 288 + ], + "score": 1.0, + "content": "Table 2: Comparison of image classification", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 286, + 287, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 287, + 299 + ], + "score": 1.0, + "content": "on ImageNet-1K for different models. Except", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 298, + 287, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 287, + 309 + ], + "score": 1.0, + "content": "for ViT-Base/16, all other models are trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 309, + 265, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 176, + 321 + ], + "score": 1.0, + "content": "and evaluated on", + "type": "text" + }, + { + "bbox": [ + 176, + 309, + 219, + 320 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 309, + 265, + 321 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + } + ], + "index": 11.75 + }, + { + "type": "table", + "bbox": [ + 299, + 70, + 501, + 254 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 299, + 70, + 501, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 299, + 70, + 501, + 254 + ], + "spans": [ + { + "bbox": [ + 299, + 70, + 501, + 254 + ], + "score": 0.98, + "html": "
BackboneRetinaNetMask R-CNN
APbApbAPm
ResNet-50 [33]36.338.034.4
PVT-Small40.440.437.8
ViL-Small [76]41.641.838.5
Swin-Tiny [43]42.043.739.8
Focal-Tiny (Ours)43.7 (+1.7)44.8 (+1.1) 41.0 (+1.3)
ResNet-101[33]38.540.436.4
ResNeXt101-32x4d [67]39.941.937.5
PVT-Medium [60]41.942.039.0
ViL-Medium [76]42.943.439.7
Swin-Small [43]45.046.542.1
Focal-Small (Ours)45.6 (+0.6)47.4 (+0.9) 42.8 (+0.7)
ResNeXt101-64x4d[67] 41.042.838.4
PVT-Large [60]42.642.939.5
ViL-Base[76]44.345.141.0
Swin-Base [43]45.046.942.3
Focal-Base (Ours)46.3 (+1.3)47.8 (+0.9)43.2 (+0.9)
", + "type": "table", + "image_path": "2eccde7cb9cd36279d6f0e1059ce2743dcec73d9f7583e972b12720e375f7d35.jpg" + } + ] + } + ], + "index": 25.5, + "virtual_lines": [ + { + "bbox": [ + 299, + 70, + 501, + 83.14285714285714 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 299, + 83.14285714285714, + 501, + 96.28571428571428 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 299, + 96.28571428571428, + 501, + 109.42857142857142 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 299, + 109.42857142857142, + 501, + 122.57142857142856 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 299, + 122.57142857142856, + 501, + 135.7142857142857 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 299, + 135.7142857142857, + 501, + 148.85714285714283 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 299, + 148.85714285714283, + 501, + 161.99999999999997 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 299, + 161.99999999999997, + 501, + 175.1428571428571 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 299, + 175.1428571428571, + 501, + 188.28571428571425 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 299, + 188.28571428571425, + 501, + 201.4285714285714 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 299, + 201.4285714285714, + 501, + 214.57142857142853 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 299, + 214.57142857142853, + 501, + 227.71428571428567 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 299, + 227.71428571428567, + 501, + 240.8571428571428 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 299, + 240.8571428571428, + 501, + 253.99999999999994 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 297, + 255, + 505, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 295, + 255, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 295, + 255, + 506, + 267 + ], + "score": 1.0, + "content": "Table 3: Comparisons with CNN and Transformer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 295, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 295, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "baselines and SoTA methods on COCO object detec-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 296, + 277, + 504, + 289 + ], + "spans": [ + { + "bbox": [ + 296, + 277, + 379, + 289 + ], + "score": 1.0, + "content": "tion. The box mAP", + "type": "text" + }, + { + "bbox": [ + 380, + 277, + 407, + 289 + ], + "score": 0.86, + "content": "( A P ^ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 277, + 474, + 289 + ], + "score": 1.0, + "content": "and mask mAP", + "type": "text" + }, + { + "bbox": [ + 474, + 278, + 504, + 289 + ], + "score": 0.84, + "content": "( A P ^ { m } )", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 295, + 288, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 295, + 288, + 505, + 299 + ], + "score": 1.0, + "content": "are reported for RetinaNet and Mask R-CNN trained", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 295, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 295, + 298, + 316, + 311 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 317, + 300, + 331, + 310 + ], + "score": 0.87, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "schedule. More detailed comparisons with", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 296, + 310, + 408, + 321 + ], + "spans": [ + { + "bbox": [ + 296, + 311, + 311, + 321 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 310, + 408, + 321 + ], + "score": 1.0, + "content": "schedule are in Table 4.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "score": 1.0, + "content": "In Table 2, we summarize the results for baseline models and the state-of-the-art models on image", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "classification task. We can see that Focal Transformers consistently outperform other methods with", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 354, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 370 + ], + "score": 1.0, + "content": "similar model sizes (#Params.) and computational complexities (GFLOPs). Specifically, Focal-Tiny", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 366, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 340, + 379 + ], + "score": 1.0, + "content": "improves over the Transformer baseline DeiT-Small/16 by", + "type": "text" + }, + { + "bbox": [ + 340, + 367, + 362, + 377 + ], + "score": 0.86, + "content": "2 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 366, + 506, + 379 + ], + "score": 1.0, + "content": ". Meanwhile, using the same model", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "configuration (2-2-6-2) and a few extra parameters and computations, Focal-Tiny improves over", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "Swin-Tiny by 1.0 point. For small and base models, Focal-Small with 51.1M parameters can reach", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 133, + 410 + ], + "score": 0.86, + "content": "8 3 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "which is better than all the counterpart small and base models using much less parameters. By", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 326, + 423 + ], + "score": 1.0, + "content": "increasing the model size, Focal-Base model achieves", + "type": "text" + }, + { + "bbox": [ + 326, + 410, + 353, + 421 + ], + "score": 0.87, + "content": "8 4 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 409, + 505, + 423 + ], + "score": 1.0, + "content": ", surpassing all the other models with", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 421, + 252, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 252, + 434 + ], + "score": 1.0, + "content": "comparable parameters and FLOPs.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 333, + 506, + 434 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "To compare with the large-scale models, we further build Focal-Large Transformer by increasing the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "hidden dimension in Focal-Base from 128 to 196 while keeping all the other hyperparameters the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 459, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 473 + ], + "score": 1.0, + "content": "same. We follow the common practice to pretrain our Focal-Large Transformer on ImageNet-22K", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 470, + 344, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 344, + 482 + ], + "score": 1.0, + "content": "and transfer it to detection and segmentation tasks [64, 43].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49.5, + "bbox_fs": [ + 105, + 436, + 506, + 482 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 495, + 316, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 317, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 317, + 509 + ], + "score": 1.0, + "content": "4.2 Object detection and instance segmentation", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 52 + }, + { + "type": "text", + "bbox": [ + 107, + 509, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "We benchmark our models on object detection with COCO 2017 [42]. The pretrained models are used", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "score": 1.0, + "content": "as visual backbones and then plugged into two representative pipelines, RetinaNet [41] and Mask", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 281, + 543 + ], + "score": 1.0, + "content": "R-CNN [32]. All models are trained on the", + "type": "text" + }, + { + "bbox": [ + 281, + 531, + 303, + 541 + ], + "score": 0.25, + "content": "1 1 8 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "training images and the results are reported on 5K", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 335, + 554 + ], + "score": 1.0, + "content": "validation set. We use the two standard training schedules,", + "type": "text" + }, + { + "bbox": [ + 336, + 542, + 350, + 552 + ], + "score": 0.89, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 541, + 428, + 554 + ], + "score": 1.0, + "content": "with 12 epochs and", + "type": "text" + }, + { + "bbox": [ + 428, + 542, + 442, + 552 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "with 36 epochs.", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 136, + 565 + ], + "score": 1.0, + "content": "For the", + "type": "text" + }, + { + "bbox": [ + 136, + 553, + 151, + 563 + ], + "score": 0.86, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "schedule, we resize image’s shorter side to 800 while keeping its longer side no more than", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 164, + 576 + ], + "score": 1.0, + "content": "1,333. For the", + "type": "text" + }, + { + "bbox": [ + 164, + 564, + 178, + 574 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "schedule, we use the multi-scale training strategy by randomly resizing its shorter", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 574, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 505, + 587 + ], + "score": 1.0, + "content": "side to the range of [480, 800]. Considering this higher input resolution, we adaptively increase the", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 585, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 506, + 597 + ], + "score": 1.0, + "content": "focal sizes at four stages to (15, 13, 9, 7), to ensures that the focal attention covers more than half of", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "the image region at the first two stages, and the whole image at the last two stages. With the focal size", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "increased, the relative position biases are accordingly up-sampled to the corresponding sizes using", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 617, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 631 + ], + "score": 1.0, + "content": "bilinear interpolation. During training, we use AdamW [44] for optimization with initial learning", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 104, + 627, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 104, + 627, + 124, + 642 + ], + "score": 1.0, + "content": "rate", + "type": "text" + }, + { + "bbox": [ + 124, + 628, + 146, + 639 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 627, + 506, + 642 + ], + "score": 1.0, + "content": "and weight decay 0.05. Similarly, we use 0.2, 0.3 and 0.5 stochastic depth drop rates to", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "regularize the training for our Tiny, Small and Base models, respectively. Since Swin Transformer", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "does not report the results on RetinaNet, we obtain the results by ourselves using their official code", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 106, + 662, + 358, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 358, + 674 + ], + "score": 1.0, + "content": "with the same hyper-parameters as that of Focal Transformers.", + "type": "text" + } + ], + "index": 67 + } + ], + "index": 60, + "bbox_fs": [ + 104, + 508, + 506, + 674 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "score": 1.0, + "content": "In Table 3, we show the performance for both CNN-based models and the current Transformer-", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 295, + 702 + ], + "score": 1.0, + "content": "based state-of-the-art models. The bbox mAP", + "type": "text" + }, + { + "bbox": [ + 295, + 688, + 322, + 700 + ], + "score": 0.87, + "content": "( A P ^ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 687, + 388, + 702 + ], + "score": 1.0, + "content": "and mask mAP", + "type": "text" + }, + { + "bbox": [ + 388, + 689, + 418, + 700 + ], + "score": 0.85, + "content": "( A P ^ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "are reported. We see", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "that Focal Transformers outperform the CNN-based models consistently with the gap of 4.8-7.1", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "score": 1.0, + "content": "points. Compared with the other methods which also use multi-scale Transformer architectures,", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 69.5, + "bbox_fs": [ + 105, + 676, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 70, + 505, + 248 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 70, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 70, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 505, + 248 + ], + "score": 0.985, + "html": "
Backbone#Params (M)FLOPs (G)RetinaNet 3x schedule + MSMask R-CNN 3x schedule + MS
APAP5APAPsAPMAPLAP6APAPApmAPAP
ResNet50 [33]37.7/44.2239/26039.058.441.822.442.851.641.061.744.937.158.440.1
PVT-Small[60]34.2/44.1226/24542.262.745.026.245.257.243.065.346.939.962.542.8
ViL-Small [76]35.7/45.0252/174 42.963.845.627.846.456.343.464.947.039.662.142.4
Swin-Tiny [43]38.5/47.8245/264 45.065.948.429.748.958.146.068.150.341.665.144.9
Focal-Tiny (Ours)39.4/48.8265/291 45.566.348.831.249.258.747.269.451.942.766.5 45.9
ResNet101 [33]56.7/63.2315/33640.960.144.023.745.053.842.863.247.138.560.141.3
ResNeXt101-32x4d [67]56.4/62.8319/34041.461.044.323.945.553.744.064.448.039.261.441.9
PVT-Medium [60]53.9/63.9283/30243.263.846.127.346.358.944.266.048.240.563.143.5
ViL-Medium [76]50.8/60.1339/26143.764.646.427.947.156.944.666.348.540.763.843.7
Swin-Small [43]59.8/69.1335/354 46.467.050.131.050.160.348.570.253.543.367.346.6
Focal-Small (Ours)61.7/71.2367/40147.367.851.031.650.961.148.870.553.643.867.747.2
ResNeXt101-64x4d [67]95.5/102473/49341.861.544.425.245.454.644.464.948.839.761.942.6
PVT-Large[60]71.1/81.0345/364 43.463.646.126.146.059.544.566.048.340.763.443.7
ViL-Base [76]66.7/76.1443/365 44.765.547.629.948.058.145.767.249.941.364.444.5
Swin-Base 43]98.4/107477/496 45.866.449.129.949.460.348.569.853.243.466.846.9
Focal-Base (Ours)100.8/110.0 514/533 46.967.850.331.950.361.549.070.153.643.767.647.0
", + "type": "table", + "image_path": "ddccdbf44fb137bfc4faa3d93f6ef7c7ed8810948684d7d2e48c12d371518931.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 505, + 129.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 129.33333333333334, + 505, + 188.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 188.66666666666669, + 505, + 248.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 251, + 505, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 250, + 507, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 507, + 264 + ], + "score": 1.0, + "content": "Table 4: COCO object detection and segmentation results with RetinaNet [41] and Mask R-CNN [33].", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 217, + 273 + ], + "score": 1.0, + "content": "All models are trained with", + "type": "text" + }, + { + "bbox": [ + 217, + 262, + 231, + 272 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "schedule and multi-scale inputs (MS). The numbers before and after", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 271, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 271, + 505, + 286 + ], + "score": 1.0, + "content": "“/” at column 2 and 3 are the model size and complexity for RetinaNet and Mask R-CNN, respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "Focal Transformers show substantial gains across all settings and metrics. Particularly, Focal", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "score": 1.0, + "content": "Transformers brings 0.7-1.7 points of mAP against the current best approach Swin Transformer [43]", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "at comparable settings. Different from the other multi-scale Transformer models, Focal Transformers", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "can simultaneously enable short-range fine-grain and long-range coarse-grain interactions for each", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "visual token, and thus capture richer visual contexts at each layer for better dense predictions. To have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 363, + 361 + ], + "score": 1.0, + "content": "more comprehensive comparisons, we train all models using the", + "type": "text" + }, + { + "bbox": [ + 364, + 350, + 378, + 360 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "score": 1.0, + "content": "schedule and show the detailed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 427, + 373 + ], + "score": 1.0, + "content": "numbers for RetinaNet and Mask R-CNN in Table 4. As we can see, even with the", + "type": "text" + }, + { + "bbox": [ + 427, + 361, + 441, + 371 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "schedule, Focal", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 370, + 500, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 500, + 385 + ], + "score": 1.0, + "content": "Transformers can still achieve 0.3-1.1 gain over Swin Transformer models in comparable settings.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 388, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "score": 1.0, + "content": "Comparison with large SoTA detection models. We follow Swin Transformers to use HTC [11]", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 399, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 411 + ], + "score": 1.0, + "content": "as the detection method in that it reported SoTA performance on COCO detection when using Swin", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 410, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 422 + ], + "score": 1.0, + "content": "Transformer as the backbone. For fair comparison, we also use soft-NMS [5], instaboost [24] and a", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "multi-scale training strategy with the shorter side in range [400, 1400] and the longer side no more", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 430, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 445 + ], + "score": 1.0, + "content": "than 1600. We train the model using AdamW [44] with base learning rate 1e-4 and weight decay", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 441, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 294, + 455 + ], + "score": 1.0, + "content": "0.1. The model is trained using the standard", + "type": "text" + }, + { + "bbox": [ + 294, + 443, + 308, + 453 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 441, + 505, + 455 + ], + "score": 1.0, + "content": "schedule. The box and mask mAPs on COCO", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "validation set and test-dev are reported in Table 5, where both single-scale evaluation and multi-scale", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 465, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 476 + ], + "score": 1.0, + "content": "evaluation results are presented. Our Focal-Large model with multi-scale test achieves 58.1 box", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "score": 1.0, + "content": "mAP and 50.9 mask mAP on mini-val set, which is better than the reported numbers for Swin-Large", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 486, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 498 + ], + "score": 1.0, + "content": "in [43]. When evaluating our model on the test-dev set, it achieves 58.4 box mAP and 51.3 mask", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "score": 1.0, + "content": "mAP, which is slightly better than Swin Transformer. Note that because our model does not include", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 508, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 506, + 520 + ], + "score": 1.0, + "content": "the global self-attention layer used in Swin Transformer at the last stage, it has a smaller model size", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "and fewer FLOPs. More recently, DyHead [17] achieves new SoTA on COCO, when combined with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 530, + 504, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 489, + 542 + ], + "score": 1.0, + "content": "Swin-Large. We replace the Swin-Large model with the Focal-Large model, and use the same", + "type": "text" + }, + { + "bbox": [ + 489, + 530, + 504, + 541 + ], + "score": 0.85, + "content": "2 \\times", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 540, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 554 + ], + "score": 1.0, + "content": "training schedule as in [17]. We report the box mAPs for both mini-val and test-dev. Focal-Large", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 551, + 352, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 352, + 564 + ], + "score": 1.0, + "content": "achieves 58.7 and 59.0 on mini-val and test-dev, respectively.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 107, + 576, + 231, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 232, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 232, + 590 + ], + "score": 1.0, + "content": "4.3 Semantic Segmentation", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 603 + ], + "score": 1.0, + "content": "In addition to the instance segmentation results, we also evaluate our models on the semantic", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 600, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 104, + 600, + 505, + 616 + ], + "score": 1.0, + "content": "segmentation task which usually takes high-resolution input images and requires capturing long-range", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 611, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 625 + ], + "score": 1.0, + "content": "interactions. We benchmark our methods on ADE20K [83]. We use UperNet [65] as the segmentation", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "score": 1.0, + "content": "method and Focal Transformers as the backbones. We train three models as Focal-Tiny, Focal-Small,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 632, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 647 + ], + "score": 1.0, + "content": "Focal-Base, respectively. For all the models, we use a standard recipe that sets the input size to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 644, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 150, + 655 + ], + "score": 0.89, + "content": "5 1 2 \\times 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 644, + 506, + 658 + ], + "score": 1.0, + "content": "and trains the model for 160k iterations with batch size 16. Table 6 shows the comparison", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 654, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 669 + ], + "score": 1.0, + "content": "results. We see that Focal-Tiny, Focal-Small and Focal-Base models consistently outperform Swin", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 666, + 391, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 391, + 678 + ], + "score": 1.0, + "content": "Transformers of the similar size in single-scale and multi-scale mIoUs.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "Comparison with large SoTA semantic segmentation models. We use the pretrained Focal-Large", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "model as the backbone for semantic segmentation. Follow the setting in [43], we use input image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 700, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 125, + 711 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 125, + 700, + 169, + 711 + ], + "score": 0.9, + "content": "6 4 0 \\times 6 4 0", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 701, + 505, + 711 + ], + "score": 1.0, + "content": "and train the model for 160k iterations with a batch size of 16. We set the initial", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 712, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 712, + 506, + 723 + ], + "score": 1.0, + "content": "learning to 6e-5 and use a polynomial learning rate decay. The weight decay is set to 0.01. For", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 70, + 505, + 248 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 70, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 70, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 505, + 248 + ], + "score": 0.985, + "html": "
Backbone#Params (M)FLOPs (G)RetinaNet 3x schedule + MSMask R-CNN 3x schedule + MS
APAP5APAPsAPMAPLAP6APAPApmAPAP
ResNet50 [33]37.7/44.2239/26039.058.441.822.442.851.641.061.744.937.158.440.1
PVT-Small[60]34.2/44.1226/24542.262.745.026.245.257.243.065.346.939.962.542.8
ViL-Small [76]35.7/45.0252/174 42.963.845.627.846.456.343.464.947.039.662.142.4
Swin-Tiny [43]38.5/47.8245/264 45.065.948.429.748.958.146.068.150.341.665.144.9
Focal-Tiny (Ours)39.4/48.8265/291 45.566.348.831.249.258.747.269.451.942.766.5 45.9
ResNet101 [33]56.7/63.2315/33640.960.144.023.745.053.842.863.247.138.560.141.3
ResNeXt101-32x4d [67]56.4/62.8319/34041.461.044.323.945.553.744.064.448.039.261.441.9
PVT-Medium [60]53.9/63.9283/30243.263.846.127.346.358.944.266.048.240.563.143.5
ViL-Medium [76]50.8/60.1339/26143.764.646.427.947.156.944.666.348.540.763.843.7
Swin-Small [43]59.8/69.1335/354 46.467.050.131.050.160.348.570.253.543.367.346.6
Focal-Small (Ours)61.7/71.2367/40147.367.851.031.650.961.148.870.553.643.867.747.2
ResNeXt101-64x4d [67]95.5/102473/49341.861.544.425.245.454.644.464.948.839.761.942.6
PVT-Large[60]71.1/81.0345/364 43.463.646.126.146.059.544.566.048.340.763.443.7
ViL-Base [76]66.7/76.1443/365 44.765.547.629.948.058.145.767.249.941.364.444.5
Swin-Base 43]98.4/107477/496 45.866.449.129.949.460.348.569.853.243.466.846.9
Focal-Base (Ours)100.8/110.0 514/533 46.967.850.331.950.361.549.070.153.643.767.647.0
", + "type": "table", + "image_path": "ddccdbf44fb137bfc4faa3d93f6ef7c7ed8810948684d7d2e48c12d371518931.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 505, + 129.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 129.33333333333334, + 505, + 188.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 188.66666666666669, + 505, + 248.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 251, + 505, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 250, + 507, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 507, + 264 + ], + "score": 1.0, + "content": "Table 4: COCO object detection and segmentation results with RetinaNet [41] and Mask R-CNN [33].", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 217, + 273 + ], + "score": 1.0, + "content": "All models are trained with", + "type": "text" + }, + { + "bbox": [ + 217, + 262, + 231, + 272 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "schedule and multi-scale inputs (MS). The numbers before and after", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 271, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 271, + 505, + 286 + ], + "score": 1.0, + "content": "“/” at column 2 and 3 are the model size and complexity for RetinaNet and Mask R-CNN, respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "Focal Transformers show substantial gains across all settings and metrics. Particularly, Focal", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "score": 1.0, + "content": "Transformers brings 0.7-1.7 points of mAP against the current best approach Swin Transformer [43]", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "at comparable settings. Different from the other multi-scale Transformer models, Focal Transformers", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "can simultaneously enable short-range fine-grain and long-range coarse-grain interactions for each", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "visual token, and thus capture richer visual contexts at each layer for better dense predictions. To have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 363, + 361 + ], + "score": 1.0, + "content": "more comprehensive comparisons, we train all models using the", + "type": "text" + }, + { + "bbox": [ + 364, + 350, + 378, + 360 + ], + "score": 0.87, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "score": 1.0, + "content": "schedule and show the detailed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 427, + 373 + ], + "score": 1.0, + "content": "numbers for RetinaNet and Mask R-CNN in Table 4. As we can see, even with the", + "type": "text" + }, + { + "bbox": [ + 427, + 361, + 441, + 371 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "schedule, Focal", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 370, + 500, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 500, + 385 + ], + "score": 1.0, + "content": "Transformers can still achieve 0.3-1.1 gain over Swin Transformer models in comparable settings.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 295, + 505, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 388, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "score": 1.0, + "content": "Comparison with large SoTA detection models. We follow Swin Transformers to use HTC [11]", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 399, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 411 + ], + "score": 1.0, + "content": "as the detection method in that it reported SoTA performance on COCO detection when using Swin", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 410, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 422 + ], + "score": 1.0, + "content": "Transformer as the backbone. For fair comparison, we also use soft-NMS [5], instaboost [24] and a", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "multi-scale training strategy with the shorter side in range [400, 1400] and the longer side no more", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 430, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 445 + ], + "score": 1.0, + "content": "than 1600. We train the model using AdamW [44] with base learning rate 1e-4 and weight decay", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 441, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 294, + 455 + ], + "score": 1.0, + "content": "0.1. The model is trained using the standard", + "type": "text" + }, + { + "bbox": [ + 294, + 443, + 308, + 453 + ], + "score": 0.86, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 441, + 505, + 455 + ], + "score": 1.0, + "content": "schedule. The box and mask mAPs on COCO", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "validation set and test-dev are reported in Table 5, where both single-scale evaluation and multi-scale", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 465, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 476 + ], + "score": 1.0, + "content": "evaluation results are presented. Our Focal-Large model with multi-scale test achieves 58.1 box", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "score": 1.0, + "content": "mAP and 50.9 mask mAP on mini-val set, which is better than the reported numbers for Swin-Large", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 486, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 498 + ], + "score": 1.0, + "content": "in [43]. When evaluating our model on the test-dev set, it achieves 58.4 box mAP and 51.3 mask", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "score": 1.0, + "content": "mAP, which is slightly better than Swin Transformer. Note that because our model does not include", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 508, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 506, + 520 + ], + "score": 1.0, + "content": "the global self-attention layer used in Swin Transformer at the last stage, it has a smaller model size", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "and fewer FLOPs. More recently, DyHead [17] achieves new SoTA on COCO, when combined with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 530, + 504, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 489, + 542 + ], + "score": 1.0, + "content": "Swin-Large. We replace the Swin-Large model with the Focal-Large model, and use the same", + "type": "text" + }, + { + "bbox": [ + 489, + 530, + 504, + 541 + ], + "score": 0.85, + "content": "2 \\times", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 540, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 554 + ], + "score": 1.0, + "content": "training schedule as in [17]. We report the box mAPs for both mini-val and test-dev. Focal-Large", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 551, + 352, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 352, + 564 + ], + "score": 1.0, + "content": "achieves 58.7 and 59.0 on mini-val and test-dev, respectively.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 386, + 506, + 564 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 576, + 231, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 232, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 232, + 590 + ], + "score": 1.0, + "content": "4.3 Semantic Segmentation", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 603 + ], + "score": 1.0, + "content": "In addition to the instance segmentation results, we also evaluate our models on the semantic", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 600, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 104, + 600, + 505, + 616 + ], + "score": 1.0, + "content": "segmentation task which usually takes high-resolution input images and requires capturing long-range", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 611, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 625 + ], + "score": 1.0, + "content": "interactions. We benchmark our methods on ADE20K [83]. We use UperNet [65] as the segmentation", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "score": 1.0, + "content": "method and Focal Transformers as the backbones. We train three models as Focal-Tiny, Focal-Small,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 632, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 647 + ], + "score": 1.0, + "content": "Focal-Base, respectively. For all the models, we use a standard recipe that sets the input size to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 644, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 150, + 655 + ], + "score": 0.89, + "content": "5 1 2 \\times 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 644, + 506, + 658 + ], + "score": 1.0, + "content": "and trains the model for 160k iterations with batch size 16. Table 6 shows the comparison", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 654, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 669 + ], + "score": 1.0, + "content": "results. We see that Focal-Tiny, Focal-Small and Focal-Base models consistently outperform Swin", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 666, + 391, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 391, + 678 + ], + "score": 1.0, + "content": "Transformers of the similar size in single-scale and multi-scale mIoUs.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 590, + 506, + 678 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "Comparison with large SoTA semantic segmentation models. We use the pretrained Focal-Large", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "model as the backbone for semantic segmentation. Follow the setting in [43], we use input image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 700, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 125, + 711 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 125, + 700, + 169, + 711 + ], + "score": 0.9, + "content": "6 4 0 \\times 6 4 0", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 701, + 505, + 711 + ], + "score": 1.0, + "content": "and train the model for 160k iterations with a batch size of 16. We set the initial", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 712, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 712, + 506, + 723 + ], + "score": 1.0, + "content": "learning to 6e-5 and use a polynomial learning rate decay. The weight decay is set to 0.01. For", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 677, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 76, + 305, + 249 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 76, + 305, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 76, + 305, + 249 + ], + "spans": [ + { + "bbox": [ + 109, + 76, + 305, + 249 + ], + "score": 0.977, + "html": "
Method#Param FLOPsmini-valtest-dev
ApbAPmApbApm
X101-64x4d [67]155M1033G 52.346.0
EfficientNet-D7 [54]77M410G54.4-55.1=
GCNet*[7]-1041G51.844.752.345.4
ResNeSt-200 [75]=52.5=53.347.1
Copy-paste [28]185M1440G 55.947.256.047.4
BoTNet-200 [51]-49.7-
SpineNet-190 [22]164M1885G 52.652.8
CenterNet2 [84]-=--56.4=
Swin-L (HTC++) [43]284M1470G 57.149.557.750.2
Swin-L (DyHead)[17]213M965G56.2---
Swin-L† (HTC++) [43]284M58.050.458.751.1
Swin-L† (DyHead) [17]213M58.4-58.7
Swin-L† (QueryInst) [25]-56.1156.1
Focal-L (HTC++) (Ours)265M1165G57.049.9-=
Focal-L (DyHead) (Ours)229M1081G56.4--=
Focal-L† (HTC++) (Ours)265M-58.150.958.451.3
Focal-L† (DyHead) (Ours)229M-58.7-59.0-
", + "type": "table", + "image_path": "fbecb97ab8c83eccb4496fc4e3c85a4b9127f6ec402c3956112b387672b7d462.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 109, + 76, + 305, + 88.35714285714286 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 88.35714285714286, + 305, + 100.71428571428572 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 100.71428571428572, + 305, + 113.07142857142858 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 109, + 113.07142857142858, + 305, + 125.42857142857144 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 125.42857142857144, + 305, + 137.7857142857143 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 137.7857142857143, + 305, + 150.14285714285717 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 109, + 150.14285714285717, + 305, + 162.50000000000003 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 162.50000000000003, + 305, + 174.8571428571429 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 109, + 174.8571428571429, + 305, + 187.21428571428575 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 109, + 187.21428571428575, + 305, + 199.5714285714286 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 199.5714285714286, + 305, + 211.92857142857147 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 211.92857142857147, + 305, + 224.28571428571433 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 109, + 224.28571428571433, + 305, + 236.6428571428572 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 236.6428571428572, + 305, + 249.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 307, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 250, + 307, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 307, + 262 + ], + "score": 1.0, + "content": "Table 5: Comparison with state-of-the-art methods", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 261, + 308, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 308, + 274 + ], + "score": 1.0, + "content": "on COCO object detection and instance segmen-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 272, + 308, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 308, + 284 + ], + "score": 1.0, + "content": "tation. The numbers are reported on 5K val set", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 282, + 308, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 308, + 297 + ], + "score": 1.0, + "content": "and test-dev. Augmented HTC [11] (denoted by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 294, + 308, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 141, + 305 + ], + "score": 0.8, + "content": "\\mathrm { H T C + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 294, + 308, + 306 + ], + "score": 1.0, + "content": ") and DyHead [17] are used as the detec-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 305, + 291, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 291, + 317 + ], + "score": 1.0, + "content": "tion methods. † means multi-scale evaluation.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "table", + "bbox": [ + 319, + 70, + 498, + 254 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 319, + 70, + 498, + 254 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 319, + 70, + 498, + 254 + ], + "spans": [ + { + "bbox": [ + 319, + 70, + 498, + 254 + ], + "score": 0.978, + "html": "
BackboneMethod#Param FLOPs mIoU +MS
ResNet-101DANet [45]69M1119G45.3
ResNet-101ACNet [26]==45.9
ResNet-101DNL [69]69M1249G46.0
ResNet-101UperNet [65]86M1029G44.9
HRNet-w48 [53]OCRNet [73]71M664G45.7=
ResNeSt-200 [75]DLab.v3+ [12]88M1381G48.4=
Swin-T[43]UperNet [65]60M945G44.545.8
Swin-S [43]UperNet [65]81M1038G47.649.5
Swin-B [43]UperNet [65]121M1188G48.149.7
Twins-SVT-L[15]UperNet [65]133M48.850.2
MiT-B5 [66]SegFormer [66]85M51.051.8
ViT-L/16+ [21]SETR[80]308M50.3=
Swin-L* [43]UperNet [65]234M3230G52.153.5
ViT-L/16 [21]Segmenter [52]334M=51.853.6
Swin-L‡ [43]K-Net [77]==54.3
Swin-L‡ [43]PatchDiverse [29]234M53.154.4
VOLO-D5 [72]UperNet [65]=-54.3
Focal-T (Ours)UperNet [65]62M998G45.847.0
Focal-S (Ours)UperNet [65]85M1130G48.050.0
Focal-B (Ours)UperNet [65]126M1354G49.050.5
Focal-L‡ (Ours)UperNet [65]240M3376G54.055.4
", + "type": "table", + "image_path": "87bb36ea5cbafa44649b82024642f9740f5a663fda0b6f453aa106965b3c0739.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 319, + 70, + 498, + 82.26666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 319, + 82.26666666666667, + 498, + 94.53333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 319, + 94.53333333333333, + 498, + 106.8 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 319, + 106.8, + 498, + 119.06666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 319, + 119.06666666666666, + 498, + 131.33333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 319, + 131.33333333333334, + 498, + 143.60000000000002 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 319, + 143.60000000000002, + 498, + 155.8666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 319, + 155.8666666666667, + 498, + 168.13333333333338 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 319, + 168.13333333333338, + 498, + 180.40000000000006 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 319, + 180.40000000000006, + 498, + 192.66666666666674 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 319, + 192.66666666666674, + 498, + 204.93333333333342 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 319, + 204.93333333333342, + 498, + 217.2000000000001 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 319, + 217.2000000000001, + 498, + 229.46666666666678 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 319, + 229.46666666666678, + 498, + 241.73333333333346 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 319, + 241.73333333333346, + 498, + 254.00000000000014 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 317, + 256, + 501, + 322 + ], + "lines": [ + { + "bbox": [ + 316, + 256, + 502, + 268 + ], + "spans": [ + { + "bbox": [ + 316, + 256, + 502, + 268 + ], + "score": 1.0, + "content": "Table 6: Comparison with SoTA methods", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 316, + 267, + 501, + 279 + ], + "spans": [ + { + "bbox": [ + 316, + 267, + 501, + 279 + ], + "score": 1.0, + "content": "for semantic segmentation on ADE20K [83]", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 316, + 278, + 502, + 290 + ], + "spans": [ + { + "bbox": [ + 316, + 278, + 502, + 290 + ], + "score": 1.0, + "content": "val set. Single- and multi-scale evaluations", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 316, + 288, + 501, + 300 + ], + "spans": [ + { + "bbox": [ + 316, + 289, + 486, + 300 + ], + "score": 1.0, + "content": "are reported in the last two columns.", + "type": "text" + }, + { + "bbox": [ + 494, + 288, + 501, + 297 + ], + "score": 0.27, + "content": "^ \\ddag", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 316, + 299, + 502, + 311 + ], + "spans": [ + { + "bbox": [ + 316, + 299, + 502, + 311 + ], + "score": 1.0, + "content": "means ImageNet-22K is used as the pretrain-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 316, + 311, + 365, + 324 + ], + "spans": [ + { + "bbox": [ + 316, + 311, + 365, + 324 + ], + "score": 1.0, + "content": "ing dataset.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37.5 + }, + { + "type": "table", + "bbox": [ + 109, + 331, + 297, + 399 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 331, + 297, + 399 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 109, + 331, + 297, + 399 + ], + "spans": [ + { + "bbox": [ + 109, + 331, + 297, + 399 + ], + "score": 0.978, + "html": "
Model W-Size FLOPs Top-1(%) APb APm
Swin-Tiny74.581.2 43.739.8
144.982.1 44.0 40.5
Focal-Tiny74.982.2 44.941.1
145.282.3 45.5 41.5
", + "type": "table", + "image_path": "71e84f7ae2932bbde0991a681df51b358ad5e6359c4edd0e3856bdf12c691fef.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 109, + 331, + 297, + 344.6 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 109, + 344.6, + 297, + 358.20000000000005 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 109, + 358.20000000000005, + 297, + 371.80000000000007 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 109, + 371.80000000000007, + 297, + 385.4000000000001 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 109, + 385.4000000000001, + 297, + 399.0000000000001 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 399, + 299, + 433 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 399, + 300, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 300, + 411 + ], + "score": 1.0, + "content": "Table 7: Impact of different window sizes (W-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 410, + 300, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 300, + 421 + ], + "score": 1.0, + "content": "Size). We alter the default size 7 to 14 and ob-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 422, + 298, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 298, + 432 + ], + "score": 1.0, + "content": "serve consistent improvements for both methods.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "index": 45.0 + }, + { + "type": "table", + "bbox": [ + 325, + 331, + 490, + 399 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 325, + 331, + 490, + 399 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 325, + 331, + 490, + 399 + ], + "spans": [ + { + "bbox": [ + 325, + 331, + 490, + 399 + ], + "score": 0.976, + "html": "
Model W-Shift Top-1(%) APb APm
Swin-Tiny80.2 81.238.8 43.736.4 39.8
Focal-Tiny82.244.841.0
81.944.941.1
", + "type": "table", + "image_path": "f9b110f7022a0af916099cc762e9ee21cf07047e230774dd117201440a6f74e6.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 325, + 331, + 490, + 344.6 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 325, + 344.6, + 490, + 358.20000000000005 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 325, + 358.20000000000005, + 490, + 371.80000000000007 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 325, + 371.80000000000007, + 490, + 385.4000000000001 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 325, + 385.4000000000001, + 490, + 399.0000000000001 + ], + "spans": [], + "index": 53 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 312, + 400, + 504, + 432 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 311, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 311, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "Table 8: Impact of window shift (W-Shift) on", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 311, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 311, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Swin Transformer and Focal Transformer. Tiny", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 311, + 422, + 381, + 432 + ], + "spans": [ + { + "bbox": [ + 311, + 422, + 381, + 432 + ], + "score": 1.0, + "content": "models are used.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 55 + } + ], + "index": 53.0 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "multi-scale evaluation, we use the same scaling ratios [0.5, 0.75, 1.0, 1.25, 1.5, 1.75] as in previous", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "works. The results in Table 6 show that Focal-Large achieves significantly better performance than", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "Swin-Large. In both single-scale and multi-scale evaluations, Focal-Large leads to more than 1 point", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 477, + 431, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 431, + 488 + ], + "score": 1.0, + "content": "mIoU improvement, creating new SoTA for semantic segmentation on ADE20K.", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 58.5 + }, + { + "type": "title", + "bbox": [ + 107, + 499, + 200, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 200, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 200, + 512 + ], + "score": 1.0, + "content": "4.4 Ablation studies", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 61 + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 503, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 528 + ], + "score": 1.0, + "content": "We conduct a series of ablation studies to inspect the model’s capacity from different aspects. We use", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 106, + 526, + 372, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 372, + 537 + ], + "score": 1.0, + "content": "Focal-Tiny and the image classification and object detection tasks.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62.5 + }, + { + "type": "text", + "bbox": [ + 107, + 541, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "Effect of varying the window size. We have demonstrated that it is crucial to model both short- and", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "long-range interactions. Thus, a related question is whether increasing the window size helps as it", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "leads to a larger receptive field. Table 7 shows the performance of Swin-Tiny and Focal-Tiny with", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "score": 1.0, + "content": "window sizes 7 and 14. Clearly, a larger window size is beneficial for both methods measured in all", + "type": "text" + } + ], + "index": 67 + }, + { + "bbox": [ + 105, + 583, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 599 + ], + "score": 1.0, + "content": "three metrics, and Focal-Tiny consistently outperforms Swin-Tiny in both window sizes. Comparing", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "the second and third row, we find that Focal-Tiny outperforms Swin-Tiny even with a smaller window", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 127, + 620 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 127, + 607, + 165, + 618 + ], + "score": 0.28, + "content": "( 7 \\nu . s . \\ 1 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 606, + 506, + 620 + ], + "score": 1.0, + "content": ". We suspect that the gain is attributed to our focal attention’s superior capability of", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 105, + 618, + 336, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 336, + 631 + ], + "score": 1.0, + "content": "capturing long-range dependencies among visual tokens.", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 67.5 + }, + { + "type": "text", + "bbox": [ + 107, + 634, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 635, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 506, + 646 + ], + "score": 1.0, + "content": "The necessity of window shift. In Swin Transformer [43], window shift is proposed to capture cross-", + "type": "text" + } + ], + "index": 72 + }, + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "window interactions between two successive layers. In contrast, visual tokens in Focal Transformers", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 657, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 668 + ], + "score": 1.0, + "content": "can always communicate with each other across windows at both fine- and coarse-grain. Thus, it", + "type": "text" + } + ], + "index": 74 + }, + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "is interesting to investigate whether adding window shift to Focal Transformers can lead to any", + "type": "text" + } + ], + "index": 75 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "improvement. To answer the question, we remove window shift from Swin Transformer while adding", + "type": "text" + } + ], + "index": 76 + }, + { + "bbox": [ + 106, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "it to Focal Transformers. As shown in Table 8, Swin Transformer shows a severe degradation after", + "type": "text" + } + ], + "index": 77 + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "removing the window shift. However, adding window shift to Focal Transformer hurts classification", + "type": "text" + } + ], + "index": 78 + }, + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "performance. The result indicates that window shift is unnecessary for Focal Transformers. While in", + "type": "text" + } + ], + "index": 79 + } + ], + "index": 75.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 76, + 305, + 249 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 76, + 305, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 76, + 305, + 249 + ], + "spans": [ + { + "bbox": [ + 109, + 76, + 305, + 249 + ], + "score": 0.977, + "html": "
Method#Param FLOPsmini-valtest-dev
ApbAPmApbApm
X101-64x4d [67]155M1033G 52.346.0
EfficientNet-D7 [54]77M410G54.4-55.1=
GCNet*[7]-1041G51.844.752.345.4
ResNeSt-200 [75]=52.5=53.347.1
Copy-paste [28]185M1440G 55.947.256.047.4
BoTNet-200 [51]-49.7-
SpineNet-190 [22]164M1885G 52.652.8
CenterNet2 [84]-=--56.4=
Swin-L (HTC++) [43]284M1470G 57.149.557.750.2
Swin-L (DyHead)[17]213M965G56.2---
Swin-L† (HTC++) [43]284M58.050.458.751.1
Swin-L† (DyHead) [17]213M58.4-58.7
Swin-L† (QueryInst) [25]-56.1156.1
Focal-L (HTC++) (Ours)265M1165G57.049.9-=
Focal-L (DyHead) (Ours)229M1081G56.4--=
Focal-L† (HTC++) (Ours)265M-58.150.958.451.3
Focal-L† (DyHead) (Ours)229M-58.7-59.0-
", + "type": "table", + "image_path": "fbecb97ab8c83eccb4496fc4e3c85a4b9127f6ec402c3956112b387672b7d462.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 109, + 76, + 305, + 88.35714285714286 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 88.35714285714286, + 305, + 100.71428571428572 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 100.71428571428572, + 305, + 113.07142857142858 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 109, + 113.07142857142858, + 305, + 125.42857142857144 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 125.42857142857144, + 305, + 137.7857142857143 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 137.7857142857143, + 305, + 150.14285714285717 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 109, + 150.14285714285717, + 305, + 162.50000000000003 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 162.50000000000003, + 305, + 174.8571428571429 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 109, + 174.8571428571429, + 305, + 187.21428571428575 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 109, + 187.21428571428575, + 305, + 199.5714285714286 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 199.5714285714286, + 305, + 211.92857142857147 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 211.92857142857147, + 305, + 224.28571428571433 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 109, + 224.28571428571433, + 305, + 236.6428571428572 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 236.6428571428572, + 305, + 249.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 307, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 250, + 307, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 307, + 262 + ], + "score": 1.0, + "content": "Table 5: Comparison with state-of-the-art methods", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 261, + 308, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 308, + 274 + ], + "score": 1.0, + "content": "on COCO object detection and instance segmen-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 272, + 308, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 308, + 284 + ], + "score": 1.0, + "content": "tation. The numbers are reported on 5K val set", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 282, + 308, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 308, + 297 + ], + "score": 1.0, + "content": "and test-dev. Augmented HTC [11] (denoted by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 294, + 308, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 141, + 305 + ], + "score": 0.8, + "content": "\\mathrm { H T C + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 294, + 308, + 306 + ], + "score": 1.0, + "content": ") and DyHead [17] are used as the detec-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 305, + 291, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 291, + 317 + ], + "score": 1.0, + "content": "tion methods. † means multi-scale evaluation.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 250, + 308, + 317 + ] + }, + { + "type": "table", + "bbox": [ + 319, + 70, + 498, + 254 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 319, + 70, + 498, + 254 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 319, + 70, + 498, + 254 + ], + "spans": [ + { + "bbox": [ + 319, + 70, + 498, + 254 + ], + "score": 0.978, + "html": "
BackboneMethod#Param FLOPs mIoU +MS
ResNet-101DANet [45]69M1119G45.3
ResNet-101ACNet [26]==45.9
ResNet-101DNL [69]69M1249G46.0
ResNet-101UperNet [65]86M1029G44.9
HRNet-w48 [53]OCRNet [73]71M664G45.7=
ResNeSt-200 [75]DLab.v3+ [12]88M1381G48.4=
Swin-T[43]UperNet [65]60M945G44.545.8
Swin-S [43]UperNet [65]81M1038G47.649.5
Swin-B [43]UperNet [65]121M1188G48.149.7
Twins-SVT-L[15]UperNet [65]133M48.850.2
MiT-B5 [66]SegFormer [66]85M51.051.8
ViT-L/16+ [21]SETR[80]308M50.3=
Swin-L* [43]UperNet [65]234M3230G52.153.5
ViT-L/16 [21]Segmenter [52]334M=51.853.6
Swin-L‡ [43]K-Net [77]==54.3
Swin-L‡ [43]PatchDiverse [29]234M53.154.4
VOLO-D5 [72]UperNet [65]=-54.3
Focal-T (Ours)UperNet [65]62M998G45.847.0
Focal-S (Ours)UperNet [65]85M1130G48.050.0
Focal-B (Ours)UperNet [65]126M1354G49.050.5
Focal-L‡ (Ours)UperNet [65]240M3376G54.055.4
", + "type": "table", + "image_path": "87bb36ea5cbafa44649b82024642f9740f5a663fda0b6f453aa106965b3c0739.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 319, + 70, + 498, + 82.26666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 319, + 82.26666666666667, + 498, + 94.53333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 319, + 94.53333333333333, + 498, + 106.8 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 319, + 106.8, + 498, + 119.06666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 319, + 119.06666666666666, + 498, + 131.33333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 319, + 131.33333333333334, + 498, + 143.60000000000002 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 319, + 143.60000000000002, + 498, + 155.8666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 319, + 155.8666666666667, + 498, + 168.13333333333338 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 319, + 168.13333333333338, + 498, + 180.40000000000006 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 319, + 180.40000000000006, + 498, + 192.66666666666674 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 319, + 192.66666666666674, + 498, + 204.93333333333342 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 319, + 204.93333333333342, + 498, + 217.2000000000001 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 319, + 217.2000000000001, + 498, + 229.46666666666678 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 319, + 229.46666666666678, + 498, + 241.73333333333346 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 319, + 241.73333333333346, + 498, + 254.00000000000014 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 317, + 256, + 501, + 322 + ], + "lines": [ + { + "bbox": [ + 316, + 256, + 502, + 268 + ], + "spans": [ + { + "bbox": [ + 316, + 256, + 502, + 268 + ], + "score": 1.0, + "content": "Table 6: Comparison with SoTA methods", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 316, + 267, + 501, + 279 + ], + "spans": [ + { + "bbox": [ + 316, + 267, + 501, + 279 + ], + "score": 1.0, + "content": "for semantic segmentation on ADE20K [83]", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 316, + 278, + 502, + 290 + ], + "spans": [ + { + "bbox": [ + 316, + 278, + 502, + 290 + ], + "score": 1.0, + "content": "val set. Single- and multi-scale evaluations", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 316, + 288, + 501, + 300 + ], + "spans": [ + { + "bbox": [ + 316, + 289, + 486, + 300 + ], + "score": 1.0, + "content": "are reported in the last two columns.", + "type": "text" + }, + { + "bbox": [ + 494, + 288, + 501, + 297 + ], + "score": 0.27, + "content": "^ \\ddag", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 316, + 299, + 502, + 311 + ], + "spans": [ + { + "bbox": [ + 316, + 299, + 502, + 311 + ], + "score": 1.0, + "content": "means ImageNet-22K is used as the pretrain-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 316, + 311, + 365, + 324 + ], + "spans": [ + { + "bbox": [ + 316, + 311, + 365, + 324 + ], + "score": 1.0, + "content": "ing dataset.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37.5, + "bbox_fs": [ + 316, + 256, + 502, + 324 + ] + }, + { + "type": "table", + "bbox": [ + 109, + 331, + 297, + 399 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 109, + 331, + 297, + 399 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 109, + 331, + 297, + 399 + ], + "spans": [ + { + "bbox": [ + 109, + 331, + 297, + 399 + ], + "score": 0.978, + "html": "
Model W-Size FLOPs Top-1(%) APb APm
Swin-Tiny74.581.2 43.739.8
144.982.1 44.0 40.5
Focal-Tiny74.982.2 44.941.1
145.282.3 45.5 41.5
", + "type": "table", + "image_path": "71e84f7ae2932bbde0991a681df51b358ad5e6359c4edd0e3856bdf12c691fef.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 109, + 331, + 297, + 344.6 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 109, + 344.6, + 297, + 358.20000000000005 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 109, + 358.20000000000005, + 297, + 371.80000000000007 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 109, + 371.80000000000007, + 297, + 385.4000000000001 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 109, + 385.4000000000001, + 297, + 399.0000000000001 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 399, + 299, + 433 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 399, + 300, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 300, + 411 + ], + "score": 1.0, + "content": "Table 7: Impact of different window sizes (W-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 410, + 300, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 300, + 421 + ], + "score": 1.0, + "content": "Size). We alter the default size 7 to 14 and ob-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 422, + 298, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 298, + 432 + ], + "score": 1.0, + "content": "serve consistent improvements for both methods.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "index": 45.0 + }, + { + "type": "table", + "bbox": [ + 325, + 331, + 490, + 399 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 325, + 331, + 490, + 399 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 325, + 331, + 490, + 399 + ], + "spans": [ + { + "bbox": [ + 325, + 331, + 490, + 399 + ], + "score": 0.976, + "html": "
Model W-Shift Top-1(%) APb APm
Swin-Tiny80.2 81.238.8 43.736.4 39.8
Focal-Tiny82.244.841.0
81.944.941.1
", + "type": "table", + "image_path": "f9b110f7022a0af916099cc762e9ee21cf07047e230774dd117201440a6f74e6.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 325, + 331, + 490, + 344.6 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 325, + 344.6, + 490, + 358.20000000000005 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 325, + 358.20000000000005, + 490, + 371.80000000000007 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 325, + 371.80000000000007, + 490, + 385.4000000000001 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 325, + 385.4000000000001, + 490, + 399.0000000000001 + ], + "spans": [], + "index": 53 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 312, + 400, + 504, + 432 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 311, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 311, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "Table 8: Impact of window shift (W-Shift) on", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 311, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 311, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Swin Transformer and Focal Transformer. Tiny", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 311, + 422, + 381, + 432 + ], + "spans": [ + { + "bbox": [ + 311, + 422, + 381, + 432 + ], + "score": 1.0, + "content": "models are used.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 55 + } + ], + "index": 53.0 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "multi-scale evaluation, we use the same scaling ratios [0.5, 0.75, 1.0, 1.25, 1.5, 1.75] as in previous", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "works. The results in Table 6 show that Focal-Large achieves significantly better performance than", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "Swin-Large. In both single-scale and multi-scale evaluations, Focal-Large leads to more than 1 point", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 477, + 431, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 431, + 488 + ], + "score": 1.0, + "content": "mIoU improvement, creating new SoTA for semantic segmentation on ADE20K.", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 58.5, + "bbox_fs": [ + 106, + 443, + 505, + 488 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 499, + 200, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 200, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 200, + 512 + ], + "score": 1.0, + "content": "4.4 Ablation studies", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 61 + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 503, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 528 + ], + "score": 1.0, + "content": "We conduct a series of ablation studies to inspect the model’s capacity from different aspects. We use", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 106, + 526, + 372, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 372, + 537 + ], + "score": 1.0, + "content": "Focal-Tiny and the image classification and object detection tasks.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62.5, + "bbox_fs": [ + 105, + 512, + 505, + 537 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 541, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "Effect of varying the window size. We have demonstrated that it is crucial to model both short- and", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "long-range interactions. Thus, a related question is whether increasing the window size helps as it", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "leads to a larger receptive field. Table 7 shows the performance of Swin-Tiny and Focal-Tiny with", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "score": 1.0, + "content": "window sizes 7 and 14. Clearly, a larger window size is beneficial for both methods measured in all", + "type": "text" + } + ], + "index": 67 + }, + { + "bbox": [ + 105, + 583, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 599 + ], + "score": 1.0, + "content": "three metrics, and Focal-Tiny consistently outperforms Swin-Tiny in both window sizes. Comparing", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "the second and third row, we find that Focal-Tiny outperforms Swin-Tiny even with a smaller window", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 127, + 620 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 127, + 607, + 165, + 618 + ], + "score": 0.28, + "content": "( 7 \\nu . s . \\ 1 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 606, + 506, + 620 + ], + "score": 1.0, + "content": ". We suspect that the gain is attributed to our focal attention’s superior capability of", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 105, + 618, + 336, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 336, + 631 + ], + "score": 1.0, + "content": "capturing long-range dependencies among visual tokens.", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 67.5, + "bbox_fs": [ + 105, + 542, + 506, + 631 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 634, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 635, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 506, + 646 + ], + "score": 1.0, + "content": "The necessity of window shift. In Swin Transformer [43], window shift is proposed to capture cross-", + "type": "text" + } + ], + "index": 72 + }, + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "window interactions between two successive layers. In contrast, visual tokens in Focal Transformers", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 657, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 668 + ], + "score": 1.0, + "content": "can always communicate with each other across windows at both fine- and coarse-grain. Thus, it", + "type": "text" + } + ], + "index": 74 + }, + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "is interesting to investigate whether adding window shift to Focal Transformers can lead to any", + "type": "text" + } + ], + "index": 75 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "improvement. To answer the question, we remove window shift from Swin Transformer while adding", + "type": "text" + } + ], + "index": 76 + }, + { + "bbox": [ + 106, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "it to Focal Transformers. As shown in Table 8, Swin Transformer shows a severe degradation after", + "type": "text" + } + ], + "index": 77 + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "removing the window shift. However, adding window shift to Focal Transformer hurts classification", + "type": "text" + } + ], + "index": 78 + }, + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "performance. The result indicates that window shift is unnecessary for Focal Transformers. While in", + "type": "text" + } + ], + "index": 79 + } + ], + "index": 75.5, + "bbox_fs": [ + 104, + 635, + 506, + 722 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 73, + 296, + 163 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 73, + 296, + 163 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 73, + 296, + 163 + ], + "spans": [ + { + "bbox": [ + 108, + 73, + 296, + 163 + ], + "score": 0.97, + "type": "image", + "image_path": "bba4db737a9b7677713c20afacad8a3124328743b54298c0a2ae2792bcf81178.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 108, + 73, + 296, + 88.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 88.0, + 296, + 103.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 103.0, + 296, + 118.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 118.0, + 296, + 133.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 133.0, + 296, + 148.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 148.0, + 296, + 163.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 164, + 299, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 162, + 299, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 299, + 177 + ], + "score": 1.0, + "content": "Figure 5: Ablating Focal-Tiny model by adding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 174, + 300, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 300, + 186 + ], + "score": 1.0, + "content": "local, global and both interactions, respectively.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 185, + 300, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 300, + 197 + ], + "score": 1.0, + "content": "Blue bars are image classification results and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 196, + 299, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 299, + 209 + ], + "score": 1.0, + "content": "orange bars object detection results. This figure", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 207, + 207, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 207, + 219 + ], + "score": 1.0, + "content": "is better viewed in color.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + } + ], + "index": 5.25 + }, + { + "type": "table", + "bbox": [ + 312, + 73, + 500, + 157 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 312, + 73, + 500, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 73, + 500, + 157 + ], + "spans": [ + { + "bbox": [ + 312, + 73, + 500, + 157 + ], + "score": 0.977, + "html": "
Depths Model #Params. FLOPs Top-1(%) APb Apm
2-2-2-2Swin21.23.178.738.2 35.7
Focal21.73.479.940.5 37.6
2-2-4-2Swin24.73.880.241.2 38.1
Focal25.44.181.443.3 39.8
2-2-6-2Swin28.34.581.243.7 39.8
Focal29.14.982.244.8 41.0
", + "type": "table", + "image_path": "f441a274a5f6aa85d84879890ce822e6335cb56e3b0d5e7463cbc04311f50dfa.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 312, + 73, + 500, + 87.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 312, + 87.0, + 500, + 101.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 312, + 101.0, + 500, + 115.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 312, + 115.0, + 500, + 129.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 312, + 129.0, + 500, + 143.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 312, + 143.0, + 500, + 157.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 309, + 160, + 502, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 308, + 160, + 502, + 173 + ], + "spans": [ + { + "bbox": [ + 308, + 160, + 502, + 173 + ], + "score": 1.0, + "content": "Table 9: Impact of the change of model depth.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 309, + 172, + 502, + 183 + ], + "spans": [ + { + "bbox": [ + 309, + 172, + 502, + 183 + ], + "score": 1.0, + "content": "We gradually reduce the number of transformer", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 308, + 183, + 502, + 194 + ], + "spans": [ + { + "bbox": [ + 308, + 183, + 502, + 194 + ], + "score": 1.0, + "content": "layers at the third stage from original 6 to 4 and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 308, + 194, + 501, + 204 + ], + "spans": [ + { + "bbox": [ + 308, + 194, + 501, + 204 + ], + "score": 1.0, + "content": "further 2. Our Focal Transformers has much", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 308, + 204, + 473, + 216 + ], + "spans": [ + { + "bbox": [ + 308, + 204, + 473, + 216 + ], + "score": 1.0, + "content": "slower drop rate than Swin Transformer.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + } + ], + "index": 16.25 + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 503, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Swin Transformers, there should always be an even number of layers in each stage for the alternative", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 242, + 404, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 404, + 254 + ], + "score": 1.0, + "content": "window shift operation, Focal Transformers do not have such a constraint.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 505, + 272 + ], + "score": 1.0, + "content": "Contributions of local and global interactions. To investigate the relative contributions of capturing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 270, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 281 + ], + "score": 1.0, + "content": "local fine-grain and global coarse-grain interactions in Focal Transformers, we have developed several", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "variants of Focal-Tiny: a) Focal-Tiny-Window merely performs attention inside each window; b)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "Focal-Tiny-Local attends the additional fine-grain surrounding tokens and c) Focal-Tiny-Global", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "attends the extra coarse-grain summarized tokens. We train these models using the same setting as", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "Focal-Tiny and report their performance on image classification and object detection using Mask", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 323, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 139, + 337 + ], + "score": 1.0, + "content": "R-CNN", + "type": "text" + }, + { + "bbox": [ + 139, + 324, + 153, + 334 + ], + "score": 0.84, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 323, + 506, + 337 + ], + "score": 1.0, + "content": "schedule. As shown in Fig. 5, Focal-Tiny-Window suffers from a significant performance", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 334, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 245, + 347 + ], + "score": 1.0, + "content": "drop on both image classification", + "type": "text" + }, + { + "bbox": [ + 245, + 335, + 292, + 345 + ], + "score": 0.86, + "content": "8 2 . 2 \\substack { 8 0 . 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 334, + 382, + 347 + ], + "score": 1.0, + "content": ") and object detection", + "type": "text" + }, + { + "bbox": [ + 383, + 335, + 430, + 345 + ], + "score": 0.84, + "content": "\\cdot 4 4 . 8 \\mathrm { \\ - } \\to 3 8 . 3", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 334, + 506, + 347 + ], + "score": 1.0, + "content": "). This is expected", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "since the communication across windows is completely cut off at each Transformer layer. After", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "we enable either the local fine-grain or global coarse-grain interactions (middle two columns), we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 366, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 381 + ], + "score": 1.0, + "content": "observe significant performance boost. When we combine short- and long-range interactions, we", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 377, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 392 + ], + "score": 1.0, + "content": "observe additional improvements on both tasks. This implies that these two type of interactions are", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 389, + 357, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 357, + 402 + ], + "score": 1.0, + "content": "complementary and both are beneficial to model performance.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 506, + 417 + ], + "score": 1.0, + "content": "Model capacity against model depth. Focal attention allows a Transformer model to capture short-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "and long-range interactions at each Transformer layer. An interesting question is whether Focal", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 427, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 440 + ], + "score": 1.0, + "content": "Transformers need fewer layers to obtain a similar modeling capacity as the Transformer models", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "that does not use focal attention, such as Swin Transformer. To answer this question, we conduct", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 449, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 104, + 449, + 506, + 462 + ], + "score": 1.0, + "content": "an experiment by training a series of Swin-Tiny and Focal-Tiny models by varying the number of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 459, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 474 + ], + "score": 1.0, + "content": "Transformer layers at stage 3. As shown in Table 9, Focal-Tiny outperforms Swin-Tiny consistently", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "with the same depth. More importantly, using fewer layers, Focal-Tiny can sometimes achieve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "comparable or even better performance than Swin Transformer. For example, Focal-Tiny with", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 492, + 482, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 482, + 505 + ], + "score": 1.0, + "content": "(2-2-4-2) achieves 81.4 on image classification which is better than Swin-Tiny with (2-2-6-2).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41 + }, + { + "type": "title", + "bbox": [ + 107, + 519, + 183, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 518, + 185, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 518, + 185, + 535 + ], + "score": 1.0, + "content": "5 Conclusion", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 107, + 539, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "score": 1.0, + "content": "In this paper, we have presented a new focal attention mechanism that enables efficient long-range", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "interactions in Vision Transformers. Different from previous works, it performs the local attention at", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 561, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 506, + 573 + ], + "score": 1.0, + "content": "fine-grain and global attention at coarse-grain, providing an effective way of capturing both short-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "and long-range context with a manageable computational cost. By applying focal attention into a", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "multi-scale Transformer architecture, we propose Focal Transformers as general-purpose backbones", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "for a wide range of dense vision tasks. A comprehensive empirical study shows that our Focal", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Transformers outperform the SoTA Vision Transformers on a range of vision tasks including image", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 615, + 303, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 303, + 628 + ], + "score": 1.0, + "content": "classification, object detection and segmentation.", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 50.5 + }, + { + "type": "text", + "bbox": [ + 107, + 631, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 630, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 505, + 644 + ], + "score": 1.0, + "content": "Limitations and future work. Although our experiments show that focal attention can significantly", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "boost the performance on image classification and dense prediction tasks, focal attention does", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 652, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 505, + 663 + ], + "score": 1.0, + "content": "introduce extra computational and memory cost, since each query token needs to attend more", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 662, + 504, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 504, + 672 + ], + "score": 1.0, + "content": "(summarized) tokens in addition to tokens inside a window. A cost-effective implementation of Focal", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "score": 1.0, + "content": "Transformer is necessary to make it more applicable to many real-world scenarios. This study focuses", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "on incorporating focal attention into multi-scale Vision Transformers for CV tasks. However, we", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "notice that focal attention is an effective sparse attention mechanism that is widely applicable to all", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "attention-based neural network models that are developed for processing natural language, images,", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 710, + 295, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 295, + 724 + ], + "score": 1.0, + "content": "videos etc. This is an exciting future direction.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 59 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 301, + 742, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 73, + 296, + 163 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 73, + 296, + 163 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 73, + 296, + 163 + ], + "spans": [ + { + "bbox": [ + 108, + 73, + 296, + 163 + ], + "score": 0.97, + "type": "image", + "image_path": "bba4db737a9b7677713c20afacad8a3124328743b54298c0a2ae2792bcf81178.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 108, + 73, + 296, + 88.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 88.0, + 296, + 103.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 103.0, + 296, + 118.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 118.0, + 296, + 133.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 133.0, + 296, + 148.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 148.0, + 296, + 163.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 164, + 299, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 162, + 299, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 299, + 177 + ], + "score": 1.0, + "content": "Figure 5: Ablating Focal-Tiny model by adding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 174, + 300, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 300, + 186 + ], + "score": 1.0, + "content": "local, global and both interactions, respectively.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 185, + 300, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 300, + 197 + ], + "score": 1.0, + "content": "Blue bars are image classification results and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 196, + 299, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 299, + 209 + ], + "score": 1.0, + "content": "orange bars object detection results. This figure", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 207, + 207, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 207, + 219 + ], + "score": 1.0, + "content": "is better viewed in color.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + } + ], + "index": 5.25 + }, + { + "type": "table", + "bbox": [ + 312, + 73, + 500, + 157 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 312, + 73, + 500, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 73, + 500, + 157 + ], + "spans": [ + { + "bbox": [ + 312, + 73, + 500, + 157 + ], + "score": 0.977, + "html": "
Depths Model #Params. FLOPs Top-1(%) APb Apm
2-2-2-2Swin21.23.178.738.2 35.7
Focal21.73.479.940.5 37.6
2-2-4-2Swin24.73.880.241.2 38.1
Focal25.44.181.443.3 39.8
2-2-6-2Swin28.34.581.243.7 39.8
Focal29.14.982.244.8 41.0
", + "type": "table", + "image_path": "f441a274a5f6aa85d84879890ce822e6335cb56e3b0d5e7463cbc04311f50dfa.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 312, + 73, + 500, + 87.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 312, + 87.0, + 500, + 101.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 312, + 101.0, + 500, + 115.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 312, + 115.0, + 500, + 129.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 312, + 129.0, + 500, + 143.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 312, + 143.0, + 500, + 157.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 309, + 160, + 502, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 308, + 160, + 502, + 173 + ], + "spans": [ + { + "bbox": [ + 308, + 160, + 502, + 173 + ], + "score": 1.0, + "content": "Table 9: Impact of the change of model depth.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 309, + 172, + 502, + 183 + ], + "spans": [ + { + "bbox": [ + 309, + 172, + 502, + 183 + ], + "score": 1.0, + "content": "We gradually reduce the number of transformer", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 308, + 183, + 502, + 194 + ], + "spans": [ + { + "bbox": [ + 308, + 183, + 502, + 194 + ], + "score": 1.0, + "content": "layers at the third stage from original 6 to 4 and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 308, + 194, + 501, + 204 + ], + "spans": [ + { + "bbox": [ + 308, + 194, + 501, + 204 + ], + "score": 1.0, + "content": "further 2. Our Focal Transformers has much", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 308, + 204, + 473, + 216 + ], + "spans": [ + { + "bbox": [ + 308, + 204, + 473, + 216 + ], + "score": 1.0, + "content": "slower drop rate than Swin Transformer.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + } + ], + "index": 16.25 + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 503, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Swin Transformers, there should always be an even number of layers in each stage for the alternative", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 242, + 404, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 404, + 254 + ], + "score": 1.0, + "content": "window shift operation, Focal Transformers do not have such a constraint.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 230, + 505, + 254 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 505, + 272 + ], + "score": 1.0, + "content": "Contributions of local and global interactions. To investigate the relative contributions of capturing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 270, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 281 + ], + "score": 1.0, + "content": "local fine-grain and global coarse-grain interactions in Focal Transformers, we have developed several", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "variants of Focal-Tiny: a) Focal-Tiny-Window merely performs attention inside each window; b)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "Focal-Tiny-Local attends the additional fine-grain surrounding tokens and c) Focal-Tiny-Global", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "attends the extra coarse-grain summarized tokens. We train these models using the same setting as", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "Focal-Tiny and report their performance on image classification and object detection using Mask", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 323, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 139, + 337 + ], + "score": 1.0, + "content": "R-CNN", + "type": "text" + }, + { + "bbox": [ + 139, + 324, + 153, + 334 + ], + "score": 0.84, + "content": "1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 323, + 506, + 337 + ], + "score": 1.0, + "content": "schedule. As shown in Fig. 5, Focal-Tiny-Window suffers from a significant performance", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 334, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 245, + 347 + ], + "score": 1.0, + "content": "drop on both image classification", + "type": "text" + }, + { + "bbox": [ + 245, + 335, + 292, + 345 + ], + "score": 0.86, + "content": "8 2 . 2 \\substack { 8 0 . 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 334, + 382, + 347 + ], + "score": 1.0, + "content": ") and object detection", + "type": "text" + }, + { + "bbox": [ + 383, + 335, + 430, + 345 + ], + "score": 0.84, + "content": "\\cdot 4 4 . 8 \\mathrm { \\ - } \\to 3 8 . 3", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 334, + 506, + 347 + ], + "score": 1.0, + "content": "). This is expected", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "since the communication across windows is completely cut off at each Transformer layer. After", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "we enable either the local fine-grain or global coarse-grain interactions (middle two columns), we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 366, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 381 + ], + "score": 1.0, + "content": "observe significant performance boost. When we combine short- and long-range interactions, we", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 377, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 392 + ], + "score": 1.0, + "content": "observe additional improvements on both tasks. This implies that these two type of interactions are", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 389, + 357, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 357, + 402 + ], + "score": 1.0, + "content": "complementary and both are beneficial to model performance.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 257, + 506, + 402 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 506, + 417 + ], + "score": 1.0, + "content": "Model capacity against model depth. Focal attention allows a Transformer model to capture short-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "and long-range interactions at each Transformer layer. An interesting question is whether Focal", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 427, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 440 + ], + "score": 1.0, + "content": "Transformers need fewer layers to obtain a similar modeling capacity as the Transformer models", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "that does not use focal attention, such as Swin Transformer. To answer this question, we conduct", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 449, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 104, + 449, + 506, + 462 + ], + "score": 1.0, + "content": "an experiment by training a series of Swin-Tiny and Focal-Tiny models by varying the number of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 459, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 474 + ], + "score": 1.0, + "content": "Transformer layers at stage 3. As shown in Table 9, Focal-Tiny outperforms Swin-Tiny consistently", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "with the same depth. More importantly, using fewer layers, Focal-Tiny can sometimes achieve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "comparable or even better performance than Swin Transformer. For example, Focal-Tiny with", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 492, + 482, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 482, + 505 + ], + "score": 1.0, + "content": "(2-2-4-2) achieves 81.4 on image classification which is better than Swin-Tiny with (2-2-6-2).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41, + "bbox_fs": [ + 104, + 406, + 506, + 505 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 519, + 183, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 518, + 185, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 518, + 185, + 535 + ], + "score": 1.0, + "content": "5 Conclusion", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 107, + 539, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "score": 1.0, + "content": "In this paper, we have presented a new focal attention mechanism that enables efficient long-range", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "interactions in Vision Transformers. Different from previous works, it performs the local attention at", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 561, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 506, + 573 + ], + "score": 1.0, + "content": "fine-grain and global attention at coarse-grain, providing an effective way of capturing both short-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "and long-range context with a manageable computational cost. By applying focal attention into a", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "multi-scale Transformer architecture, we propose Focal Transformers as general-purpose backbones", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "for a wide range of dense vision tasks. A comprehensive empirical study shows that our Focal", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Transformers outperform the SoTA Vision Transformers on a range of vision tasks including image", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 615, + 303, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 303, + 628 + ], + "score": 1.0, + "content": "classification, object detection and segmentation.", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 50.5, + "bbox_fs": [ + 105, + 538, + 506, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 631, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 630, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 505, + 644 + ], + "score": 1.0, + "content": "Limitations and future work. Although our experiments show that focal attention can significantly", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "boost the performance on image classification and dense prediction tasks, focal attention does", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 652, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 505, + 663 + ], + "score": 1.0, + "content": "introduce extra computational and memory cost, since each query token needs to attend more", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 662, + 504, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 504, + 672 + ], + "score": 1.0, + "content": "(summarized) tokens in addition to tokens inside a window. A cost-effective implementation of Focal", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "score": 1.0, + "content": "Transformer is necessary to make it more applicable to many real-world scenarios. This study focuses", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "on incorporating focal attention into multi-scale Vision Transformers for CV tasks. However, we", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "notice that focal attention is an effective sparse attention mechanism that is widely applicable to all", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "attention-based neural network models that are developed for processing natural language, images,", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 710, + 295, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 295, + 724 + ], + "score": 1.0, + "content": "videos etc. This is an exciting future direction.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 59, + "bbox_fs": [ + 105, + 630, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 64, + 507, + 725 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 108, + 88, + 507, + 101 + ], + "spans": [ + { + "bbox": [ + 108, + 88, + 507, + 101 + ], + "score": 1.0, + "content": "[1] Joshua Ainslie, Santiago Ontanon, Chris Alberti, Philip Pham, Anirudh Ravula, and Sumit Sanghai. Etc:", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 124, + 99, + 458, + 111 + ], + "spans": [ + { + "bbox": [ + 124, + 99, + 458, + 111 + ], + "score": 1.0, + "content": "Encoding long and structured data in transformers. arXiv preprint arXiv:2004.08483, 2020.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 109, + 117, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 109, + 117, + 506, + 131 + ], + "score": 1.0, + "content": "[2] Irwan Bello, Barret Zoph, Ashish Vaswani, Jonathon Shlens, and Quoc V Le. Attention augmented", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 124, + 126, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 124, + 126, + 507, + 141 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 124, + 137, + 217, + 150 + ], + "spans": [ + { + "bbox": [ + 124, + 137, + 217, + 150 + ], + "score": 1.0, + "content": "pages 3286–3295, 2019.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 109, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 109, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "[3] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 123, + 165, + 252, + 178 + ], + "spans": [ + { + "bbox": [ + 123, + 165, + 252, + 178 + ], + "score": 1.0, + "content": "preprint arXiv:2004.05150, 2020.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 183, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 110, + 183, + 506, + 195 + ], + "score": 1.0, + "content": "[4] Maxim Berman, Hervé Jégou, Andrea Vedaldi, Iasonas Kokkinos, and Matthijs Douze. Multigrain: a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 124, + 192, + 460, + 205 + ], + "spans": [ + { + "bbox": [ + 124, + 192, + 460, + 205 + ], + "score": 1.0, + "content": "unified image embedding for classes and instances. arXiv preprint arXiv:1902.05509, 2019.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 109, + 210, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 109, + 210, + 506, + 224 + ], + "score": 1.0, + "content": "[5] Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S Davis. Soft-nms–improving object detection", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 124, + 219, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 124, + 219, + 506, + 235 + ], + "score": 1.0, + "content": "with one line of code. In Proceedings of the IEEE international conference on computer vision, pages", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 232, + 194, + 241 + ], + "spans": [ + { + "bbox": [ + 127, + 232, + 194, + 241 + ], + "score": 1.0, + "content": "5561–5569, 2017.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 109, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 109, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 124, + 258, + 507, + 272 + ], + "spans": [ + { + "bbox": [ + 124, + 258, + 507, + 272 + ], + "score": 1.0, + "content": "Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 268, + 274, + 281 + ], + "spans": [ + { + "bbox": [ + 124, + 268, + 274, + 281 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2005.14165, 2020.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 110, + 287, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 110, + 287, + 506, + 299 + ], + "score": 1.0, + "content": "[7] Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. Gcnet: Non-local networks meet squeeze-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 126, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 126, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "excitation networks and beyond. In Proceedings of the IEEE/CVF International Conference on Computer", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 124, + 306, + 259, + 319 + ], + "spans": [ + { + "bbox": [ + 124, + 306, + 259, + 319 + ], + "score": 1.0, + "content": "Vision Workshops, pages 0–0, 2019.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 109, + 324, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 109, + 324, + 506, + 338 + ], + "score": 1.0, + "content": "[8] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 335, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 125, + 335, + 506, + 347 + ], + "score": 1.0, + "content": "Zagoruyko. End-to-end object detection with transformers. In European Conference on Computer Vision,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 123, + 343, + 244, + 358 + ], + "spans": [ + { + "bbox": [ + 123, + 343, + 244, + 358 + ], + "score": 1.0, + "content": "pages 213–229. Springer, 2020.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 109, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "[9] Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 124, + 372, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 124, + 372, + 506, + 386 + ], + "score": 1.0, + "content": "Joulin. Emerging properties in self-supervised vision transformers. arXiv preprint arXiv:2104.14294,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 382, + 151, + 395 + ], + "spans": [ + { + "bbox": [ + 125, + 382, + 151, + 395 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "score": 1.0, + "content": "[10] Shuning Chang, Pichao Wang, Fan Wang, Hao Li, and Jiashi Feng. Augmented transformer with adaptive", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 124, + 411, + 443, + 423 + ], + "spans": [ + { + "bbox": [ + 124, + 411, + 443, + 423 + ], + "score": 1.0, + "content": "graph for temporal action proposal generation. arXiv preprint arXiv:2103.16024, 2021.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 429, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 506, + 440 + ], + "score": 1.0, + "content": "[11] Kai Chen, Jiangmiao Pang, Jiaqi Wang, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 124, + 437, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 124, + 437, + 506, + 452 + ], + "score": 1.0, + "content": "Jianping Shi, Wanli Ouyang, et al. Hybrid task cascade for instance segmentation. In Proceedings of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 125, + 448, + 467, + 461 + ], + "spans": [ + { + "bbox": [ + 125, + 448, + 467, + 461 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4974–4983, 2019.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 466, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 478 + ], + "score": 1.0, + "content": "[12] Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 124, + 475, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 124, + 475, + 506, + 490 + ], + "score": 1.0, + "content": "with atrous separable convolution for semantic image segmentation. In Proceedings of the European", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 485, + 353, + 498 + ], + "spans": [ + { + "bbox": [ + 124, + 485, + 353, + 498 + ], + "score": 1.0, + "content": "conference on computer vision (ECCV), pages 801–818, 2018.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "score": 1.0, + "content": "[13] Xin Chen, Bin Yan, Jiawen Zhu, Dong Wang, Xiaoyun Yang, and Huchuan Lu. Transformer tracking.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 124, + 514, + 274, + 527 + ], + "spans": [ + { + "bbox": [ + 124, + 514, + 274, + 527 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2103.15436, 2021.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "[14] Xinlei Chen, Saining Xie, and Kaiming He. An empirical study of training self-supervised visual", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 124, + 541, + 324, + 555 + ], + "spans": [ + { + "bbox": [ + 124, + 541, + 324, + 555 + ], + "score": 1.0, + "content": "transformers. arXiv preprint arXiv:2104.02057, 2021.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "[15] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 124, + 569, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 124, + 569, + 506, + 583 + ], + "score": 1.0, + "content": "Shen. Twins: Revisiting spatial attention design in vision transformers. arXiv preprint arXiv:2104.13840,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 125, + 578, + 151, + 593 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 151, + 593 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 597, + 507, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 507, + 611 + ], + "score": 1.0, + "content": "[16] Xiangxiang Chu, Zhi Tian, Bo Zhang, Xinlong Wang, Xiaolin Wei, Huaxia Xia, and Chunhua Shen.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 125, + 607, + 457, + 621 + ], + "spans": [ + { + "bbox": [ + 125, + 607, + 457, + 621 + ], + "score": 1.0, + "content": "Conditional positional encodings for vision transformers. Arxiv preprint 2102.10882, 2021.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 626, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 638 + ], + "score": 1.0, + "content": "[17] Xiyang Dai, Yinpeng Chen, Bin Xiao, Dongdong Chen, Mengchen Liu, Lu Yuan, and Lei Zhang. Dynamic", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 636, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 126, + 636, + 506, + 648 + ], + "score": 1.0, + "content": "head: Unifying object detection heads with attentions. In Proceedings of the IEEE/CVF Conference on", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 645, + 371, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 645, + 371, + 659 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pages 7373–7382, 2021.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "[18] Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 126, + 674, + 376, + 686 + ], + "spans": [ + { + "bbox": [ + 126, + 674, + 376, + 686 + ], + "score": 1.0, + "content": "detection with transformers. arXiv preprint arXiv:2011.09094, 2020.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "score": 1.0, + "content": "[19] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 124, + 701, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 124, + 701, + 507, + 714 + ], + "score": 1.0, + "content": "image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 712, + 169, + 723 + ], + "spans": [ + { + "bbox": [ + 125, + 712, + 169, + 723 + ], + "score": 1.0, + "content": "Ieee, 2009.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 106, + 64, + 507, + 725 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 88, + 507, + 101 + ], + "spans": [ + { + "bbox": [ + 108, + 88, + 507, + 101 + ], + "score": 1.0, + "content": "[1] Joshua Ainslie, Santiago Ontanon, Chris Alberti, Philip Pham, Anirudh Ravula, and Sumit Sanghai. Etc:", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 99, + 458, + 111 + ], + "spans": [ + { + "bbox": [ + 124, + 99, + 458, + 111 + ], + "score": 1.0, + "content": "Encoding long and structured data in transformers. arXiv preprint arXiv:2004.08483, 2020.", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 117, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 109, + 117, + 506, + 131 + ], + "score": 1.0, + "content": "[2] Irwan Bello, Barret Zoph, Ashish Vaswani, Jonathon Shlens, and Quoc V Le. Attention augmented", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 126, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 124, + 126, + 507, + 141 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 124, + 137, + 217, + 150 + ], + "spans": [ + { + "bbox": [ + 124, + 137, + 217, + 150 + ], + "score": 1.0, + "content": "pages 3286–3295, 2019.", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 109, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "[3] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 165, + 252, + 178 + ], + "spans": [ + { + "bbox": [ + 123, + 165, + 252, + 178 + ], + "score": 1.0, + "content": "preprint arXiv:2004.05150, 2020.", + "type": "text" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 183, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 110, + 183, + 506, + 195 + ], + "score": 1.0, + "content": "[4] Maxim Berman, Hervé Jégou, Andrea Vedaldi, Iasonas Kokkinos, and Matthijs Douze. Multigrain: a", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 192, + 460, + 205 + ], + "spans": [ + { + "bbox": [ + 124, + 192, + 460, + 205 + ], + "score": 1.0, + "content": "unified image embedding for classes and instances. arXiv preprint arXiv:1902.05509, 2019.", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 210, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 109, + 210, + 506, + 224 + ], + "score": 1.0, + "content": "[5] Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S Davis. Soft-nms–improving object detection", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 219, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 124, + 219, + 506, + 235 + ], + "score": 1.0, + "content": "with one line of code. In Proceedings of the IEEE international conference on computer vision, pages", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 232, + 194, + 241 + ], + "spans": [ + { + "bbox": [ + 127, + 232, + 194, + 241 + ], + "score": 1.0, + "content": "5561–5569, 2017.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 109, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 258, + 507, + 272 + ], + "spans": [ + { + "bbox": [ + 124, + 258, + 507, + 272 + ], + "score": 1.0, + "content": "Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 268, + 274, + 281 + ], + "spans": [ + { + "bbox": [ + 124, + 268, + 274, + 281 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2005.14165, 2020.", + "type": "text" + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 287, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 110, + 287, + 506, + 299 + ], + "score": 1.0, + "content": "[7] Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. Gcnet: Non-local networks meet squeeze-", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 126, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "excitation networks and beyond. In Proceedings of the IEEE/CVF International Conference on Computer", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 124, + 306, + 259, + 319 + ], + "spans": [ + { + "bbox": [ + 124, + 306, + 259, + 319 + ], + "score": 1.0, + "content": "Vision Workshops, pages 0–0, 2019.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 324, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 109, + 324, + 506, + 338 + ], + "score": 1.0, + "content": "[8] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 335, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 125, + 335, + 506, + 347 + ], + "score": 1.0, + "content": "Zagoruyko. End-to-end object detection with transformers. In European Conference on Computer Vision,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 123, + 343, + 244, + 358 + ], + "spans": [ + { + "bbox": [ + 123, + 343, + 244, + 358 + ], + "score": 1.0, + "content": "pages 213–229. Springer, 2020.", + "type": "text" + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 109, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "[9] Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 372, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 124, + 372, + 506, + 386 + ], + "score": 1.0, + "content": "Joulin. Emerging properties in self-supervised vision transformers. arXiv preprint arXiv:2104.14294,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 382, + 151, + 395 + ], + "spans": [ + { + "bbox": [ + 125, + 382, + 151, + 395 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "score": 1.0, + "content": "[10] Shuning Chang, Pichao Wang, Fan Wang, Hao Li, and Jiashi Feng. Augmented transformer with adaptive", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 411, + 443, + 423 + ], + "spans": [ + { + "bbox": [ + 124, + 411, + 443, + 423 + ], + "score": 1.0, + "content": "graph for temporal action proposal generation. arXiv preprint arXiv:2103.16024, 2021.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 429, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 506, + 440 + ], + "score": 1.0, + "content": "[11] Kai Chen, Jiangmiao Pang, Jiaqi Wang, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu,", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 437, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 124, + 437, + 506, + 452 + ], + "score": 1.0, + "content": "Jianping Shi, Wanli Ouyang, et al. Hybrid task cascade for instance segmentation. In Proceedings of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 125, + 448, + 467, + 461 + ], + "spans": [ + { + "bbox": [ + 125, + 448, + 467, + 461 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4974–4983, 2019.", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 466, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 478 + ], + "score": 1.0, + "content": "[12] Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 475, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 124, + 475, + 506, + 490 + ], + "score": 1.0, + "content": "with atrous separable convolution for semantic image segmentation. In Proceedings of the European", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 485, + 353, + 498 + ], + "spans": [ + { + "bbox": [ + 124, + 485, + 353, + 498 + ], + "score": 1.0, + "content": "conference on computer vision (ECCV), pages 801–818, 2018.", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "score": 1.0, + "content": "[13] Xin Chen, Bin Yan, Jiawen Zhu, Dong Wang, Xiaoyun Yang, and Huchuan Lu. Transformer tracking.", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 514, + 274, + 527 + ], + "spans": [ + { + "bbox": [ + 124, + 514, + 274, + 527 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2103.15436, 2021.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "[14] Xinlei Chen, Saining Xie, and Kaiming He. An empirical study of training self-supervised visual", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 541, + 324, + 555 + ], + "spans": [ + { + "bbox": [ + 124, + 541, + 324, + 555 + ], + "score": 1.0, + "content": "transformers. arXiv preprint arXiv:2104.02057, 2021.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "[15] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 569, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 124, + 569, + 506, + 583 + ], + "score": 1.0, + "content": "Shen. Twins: Revisiting spatial attention design in vision transformers. arXiv preprint arXiv:2104.13840,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 125, + 578, + 151, + 593 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 151, + 593 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 597, + 507, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 507, + 611 + ], + "score": 1.0, + "content": "[16] Xiangxiang Chu, Zhi Tian, Bo Zhang, Xinlong Wang, Xiaolin Wei, Huaxia Xia, and Chunhua Shen.", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 607, + 457, + 621 + ], + "spans": [ + { + "bbox": [ + 125, + 607, + 457, + 621 + ], + "score": 1.0, + "content": "Conditional positional encodings for vision transformers. Arxiv preprint 2102.10882, 2021.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 626, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 638 + ], + "score": 1.0, + "content": "[17] Xiyang Dai, Yinpeng Chen, Bin Xiao, Dongdong Chen, Mengchen Liu, Lu Yuan, and Lei Zhang. Dynamic", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 636, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 126, + 636, + 506, + 648 + ], + "score": 1.0, + "content": "head: Unifying object detection heads with attentions. In Proceedings of the IEEE/CVF Conference on", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 645, + 371, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 645, + 371, + 659 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pages 7373–7382, 2021.", + "type": "text" + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "[18] Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 674, + 376, + 686 + ], + "spans": [ + { + "bbox": [ + 126, + 674, + 376, + 686 + ], + "score": 1.0, + "content": "detection with transformers. arXiv preprint arXiv:2011.09094, 2020.", + "type": "text" + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "score": 1.0, + "content": "[19] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical", + "type": "text" + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 701, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 124, + 701, + 507, + 714 + ], + "score": 1.0, + "content": "image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 712, + 169, + 723 + ], + "spans": [ + { + "bbox": [ + 125, + 712, + 169, + 723 + ], + "score": 1.0, + "content": "Ieee, 2009.", + "type": "text" + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 70, + 507, + 88 + ], + "spans": [ + { + "bbox": [ + 103, + 70, + 507, + 88 + ], + "score": 1.0, + "content": "[20] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 84, + 379, + 96 + ], + "spans": [ + { + "bbox": [ + 125, + 84, + 379, + 96 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. NAACL, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 100, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 506, + 114 + ], + "score": 1.0, + "content": "[21] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 109, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 124, + 109, + 506, + 124 + ], + "score": 1.0, + "content": "Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 121, + 485, + 133 + ], + "spans": [ + { + "bbox": [ + 126, + 121, + 485, + 133 + ], + "score": 1.0, + "content": "16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 138, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 150 + ], + "score": 1.0, + "content": "[22] Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Golnaz Ghiasi, Mingxing Tan, Yin Cui, Quoc V Le, and Xiaodan", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 147, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 126, + 147, + 505, + 159 + ], + "score": 1.0, + "content": "Song. Spinenet: Learning scale-permuted backbone for recognition and localization. In Proceedings of the", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 125, + 157, + 477, + 171 + ], + "spans": [ + { + "bbox": [ + 125, + 157, + 477, + 171 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11592–11601, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "[23] Haoqi Fan, Bo Xiong, Karttikeya Mangalam, Yanghao Li, Zhicheng Yan, Jitendra Malik, and Christoph", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 186, + 442, + 198 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 442, + 198 + ], + "score": 1.0, + "content": "Feichtenhofer. Multiscale vision transformers. arXiv preprint arXiv:2104.11227, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 204, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 107, + 204, + 506, + 214 + ], + "score": 1.0, + "content": "[24] Hao-Shu Fang, Jianhua Sun, Runzhong Wang, Minghao Gou, Yong-Lu Li, and Cewu Lu. Instaboost:", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "Boosting instance segmentation via probability map guided copy-pasting. In Proceedings of the IEEE/CVF", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 124, + 222, + 378, + 237 + ], + "spans": [ + { + "bbox": [ + 124, + 222, + 378, + 237 + ], + "score": 1.0, + "content": "International Conference on Computer Vision, pages 682–691, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "[25] Yuxin Fang, Shusheng Yang, Xinggang Wang, Yu Li, Chen Fang, Ying Shan, Bin Feng, and Wenyu Liu.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 250, + 226, + 262 + ], + "spans": [ + { + "bbox": [ + 125, + 250, + 226, + 262 + ], + "score": 1.0, + "content": "Instances as queries, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 268, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 506, + 280 + ], + "score": 1.0, + "content": "[26] Jun Fu, Jing Liu, Yuhang Wang, Yong Li, Yongjun Bao, Jinhui Tang, and Hanqing Lu. Adaptive context", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 277, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 125, + 277, + 506, + 289 + ], + "score": 1.0, + "content": "network for scene parsing. In Proceedings of the IEEE/CVF International Conference on Computer Vision,", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 123, + 287, + 217, + 300 + ], + "spans": [ + { + "bbox": [ + 123, + 287, + 217, + 300 + ], + "score": 1.0, + "content": "pages 6748–6757, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "[27] Jianfeng Gao, Patrick Pantel, Michael Gamon, Xiaodong He, and Li Deng. Modeling interestingness with", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 125, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "deep neural networks. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 325, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 125, + 325, + 506, + 337 + ], + "score": 1.0, + "content": "Processing (EMNLP), pages 2–13, Doha, Qatar, October 2014. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "[28] Golnaz Ghiasi, Yin Cui, Aravind Srinivas, Rui Qian, Tsung-Yi Lin, Ekin D Cubuk, Quoc V Le, and Barret", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 126, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "Zoph. Simple copy-paste is a strong data augmentation method for instance segmentation. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 22 + }, + { + "bbox": [ + 125, + 363, + 219, + 375 + ], + "spans": [ + { + "bbox": [ + 125, + 363, + 219, + 375 + ], + "score": 1.0, + "content": "arXiv:2012.07177, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 380, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 506, + 392 + ], + "score": 1.0, + "content": "[29] Chengyue Gong, Dilin Wang, Meng Li, Vikas Chandra, and Qiang Liu. Vision transformers with patch", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 390, + 206, + 402 + ], + "spans": [ + { + "bbox": [ + 125, + 390, + 206, + 402 + ], + "score": 1.0, + "content": "diversification, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 408, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 107, + 408, + 506, + 418 + ], + "score": 1.0, + "content": "[30] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao,", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 417, + 502, + 429 + ], + "spans": [ + { + "bbox": [ + 125, + 417, + 502, + 429 + ], + "score": 1.0, + "content": "Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 434, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 507, + 448 + ], + "score": 1.0, + "content": "[31] Kai Han, An Xiao, Enhua Wu, Jianyuan Guo, Chunjing Xu, and Yunhe Wang. Transformer in transformer,", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "spans": [ + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "[32] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 471, + 381, + 485 + ], + "spans": [ + { + "bbox": [ + 123, + 471, + 381, + 485 + ], + "score": 1.0, + "content": "international conference on computer vision, pages 2961–2969, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 488, + 507, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 507, + 503 + ], + "score": 1.0, + "content": "[33] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 499, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 124, + 499, + 507, + 513 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "score": 1.0, + "content": "[34] Elad Hoffer, Tal Ben-Nun, Itay Hubara, Niv Giladi, Torsten Hoefler, and Daniel Soudry. Augment your", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 528, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 125, + 528, + 506, + 540 + ], + "score": 1.0, + "content": "batch: Improving generalization through instance repetition. In Proceedings of the IEEE/CVF Conference", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 123, + 536, + 382, + 549 + ], + "spans": [ + { + "bbox": [ + 123, + 536, + 382, + 549 + ], + "score": 1.0, + "content": "on Computer Vision and Pattern Recognition, pages 8129–8138, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 555, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 506, + 567 + ], + "score": 1.0, + "content": "[35] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 564, + 377, + 577 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 377, + 577 + ], + "score": 1.0, + "content": "on computer vision and pattern recognition, pages 7132–7141, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 582, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 506, + 594 + ], + "score": 1.0, + "content": "[36] Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, and Mubarak", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 592, + 420, + 604 + ], + "spans": [ + { + "bbox": [ + 125, + 592, + 420, + 604 + ], + "score": 1.0, + "content": "Shah. Transformers in vision: A survey. arXiv preprint arXiv:2101.01169, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "[37] Yann LeCun, Yoshua Bengio, et al. Convolutional networks for images, speech, and time series. The", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 619, + 379, + 631 + ], + "spans": [ + { + "bbox": [ + 126, + 619, + 379, + 631 + ], + "score": 1.0, + "content": "handbook of brain theory and neural networks, 3361(10):1995, 1995.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "[38] Bing Li, Cheng Zheng, Silvio Giancola, and Bernard Ghanem. Sctn: Sparse convolution-transformer", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 647, + 401, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 647, + 401, + 659 + ], + "score": 1.0, + "content": "network for scene flow estimation. arXiv preprint arXiv:2105.04447, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 664, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 676 + ], + "score": 1.0, + "content": "[39] Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Mei Gao, Bin Xiao, Xiyang Dai, Lu Yuan, and Jianfeng Gao.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 124, + 673, + 506, + 687 + ], + "score": 1.0, + "content": "Efficient self-supervised vision transformers for representation learning. arXiv preprint arXiv:2106.09785,", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 125, + 682, + 151, + 698 + ], + "spans": [ + { + "bbox": [ + 125, + 682, + 151, + 698 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 477, + 714 + ], + "score": 1.0, + "content": "[40] Xiangyu Li, Yonghong Hou, Pichao Wang, Zhimin Gao, Mingliang Xu, and Wanqing Li.", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 480, + 703, + 506, + 713 + ], + "score": 1.0, + "content": "Trear:", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 711, + 474, + 725 + ], + "spans": [ + { + "bbox": [ + 125, + 711, + 474, + 725 + ], + "score": 1.0, + "content": "Transformer-based rgb-d egocentric action recognition. arXiv preprint arXiv:2101.03904, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 104, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[41] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988,", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 93, + 151, + 106 + ], + "spans": [ + { + "bbox": [ + 125, + 93, + 151, + 106 + ], + "score": 1.0, + "content": "2017.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 111, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 506, + 123 + ], + "score": 1.0, + "content": "[42] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár,", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 121, + 450, + 133 + ], + "spans": [ + { + "bbox": [ + 125, + 121, + 450, + 133 + ], + "score": 1.0, + "content": "and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "[43] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 148, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 125, + 148, + 506, + 160 + ], + "score": 1.0, + "content": "transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030,", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 124, + 158, + 151, + 171 + ], + "spans": [ + { + "bbox": [ + 124, + 158, + 151, + 171 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "[44] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 186, + 219, + 198 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 219, + 198 + ], + "score": 1.0, + "content": "arXiv:1711.05101, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 202, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 217 + ], + "score": 1.0, + "content": "[45] Hyeonseob Nam, Jung-Woo Ha, and Jeonghee Kim. Dual attention networks for multimodal reasoning", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 212, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 123, + 212, + 505, + 227 + ], + "score": 1.0, + "content": "and matching. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 223, + 186, + 235 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 186, + 235 + ], + "score": 1.0, + "content": "299–307, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "score": 1.0, + "content": "[46] R. Pappagari, P. Zelasko, J. Villalba, Y. Carmiel, and N. Dehak. Hierarchical transformers for long", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 124, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "document classification. In 2019 IEEE Automatic Speech Recognition and Understanding Workshop", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 259, + 239, + 273 + ], + "spans": [ + { + "bbox": [ + 124, + 259, + 239, + 273 + ], + "score": 1.0, + "content": "(ASRU), pages 838–844, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "[47] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 287, + 507, + 301 + ], + "spans": [ + { + "bbox": [ + 124, + 287, + 507, + 301 + ], + "score": 1.0, + "content": "Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR,", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 125, + 298, + 151, + 310 + ], + "spans": [ + { + "bbox": [ + 125, + 298, + 151, + 310 + ], + "score": 1.0, + "content": "2018.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 316, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 327 + ], + "score": 1.0, + "content": "[48] Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 325, + 339, + 337 + ], + "spans": [ + { + "bbox": [ + 124, + 325, + 339, + 337 + ], + "score": 1.0, + "content": "journal on control and optimization, 30(4):838–855, 1992.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "[49] Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, and Timothy P Lillicrap. Compressive transformers", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 353, + 403, + 365 + ], + "spans": [ + { + "bbox": [ + 125, + 353, + 403, + 365 + ], + "score": 1.0, + "content": "for long-range sequence modelling. arXiv preprint arXiv:1911.05507, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 370, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 506, + 382 + ], + "score": 1.0, + "content": "[50] Prajit Ramachandran, Niki Parmar, Ashish Vaswani, Irwan Bello, Anselm Levskaya, and Jonathon Shlens.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 380, + 432, + 392 + ], + "spans": [ + { + "bbox": [ + 125, + 380, + 432, + 392 + ], + "score": 1.0, + "content": "Stand-alone self-attention in vision models. arXiv preprint arXiv:1906.05909, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 506, + 410 + ], + "score": 1.0, + "content": "[51] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 407, + 319, + 419 + ], + "spans": [ + { + "bbox": [ + 125, + 407, + 319, + 419 + ], + "score": 1.0, + "content": "Bottleneck transformers for visual recognition, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 425, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 437 + ], + "score": 1.0, + "content": "[52] Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. Segmenter: Transformer for semantic", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 434, + 326, + 447 + ], + "spans": [ + { + "bbox": [ + 124, + 434, + 326, + 447 + ], + "score": 1.0, + "content": "segmentation. arXiv preprint arXiv:2105.05633, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "[53] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 462, + 248, + 475 + ], + "spans": [ + { + "bbox": [ + 124, + 462, + 248, + 475 + ], + "score": 1.0, + "content": "pose estimation. In CVPR, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "[54] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 489, + 421, + 504 + ], + "spans": [ + { + "bbox": [ + 124, + 489, + 421, + 504 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, pages 6105–6114. PMLR, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "[55] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 518, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 125, + 518, + 505, + 529 + ], + "score": 1.0, + "content": "Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 527, + 219, + 538 + ], + "spans": [ + { + "bbox": [ + 125, + 527, + 219, + 538 + ], + "score": 1.0, + "content": "arXiv:2012.12877, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "[56] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 554, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 125, + 554, + 505, + 566 + ], + "score": 1.0, + "content": "Shlens. Scaling local self-attention for parameter efficient visual backbones. In Proceedings of the", + "type": "text", + "cross_page": true + } + ], + "index": 37 + }, + { + "bbox": [ + 124, + 564, + 477, + 578 + ], + "spans": [ + { + "bbox": [ + 124, + 564, + 477, + 578 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12894–12904, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "[57] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 592, + 393, + 604 + ], + "spans": [ + { + "bbox": [ + 125, + 592, + 393, + 604 + ], + "score": 1.0, + "content": "Kaiser, and Illia Polosukhin. Attention is all you need. In NeurIPS, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "[58] Huiyu Wang, Yukun Zhu, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen. Max-deeplab: End-to-end", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 619, + 447, + 632 + ], + "spans": [ + { + "bbox": [ + 123, + 619, + 447, + 632 + ], + "score": 1.0, + "content": "panoptic segmentation with mask transformers. arXiv preprint arXiv:2012.00759, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "[59] Ning Wang, Wengang Zhou, Jie Wang, and Houqaing Li. Transformer meets tracker: Exploiting temporal", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 646, + 397, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 646, + 397, + 659 + ], + "score": 1.0, + "content": "context for robust visual tracking. arXiv preprint arXiv:2103.11681, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "[60] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 125, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions.", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 123, + 683, + 273, + 696 + ], + "spans": [ + { + "bbox": [ + 123, + 683, + 273, + 696 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2102.12122, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[61] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 711, + 507, + 725 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 507, + 725 + ], + "score": 1.0, + "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 104, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "[62] Yuqing Wang, Zhaoliang Xu, Xinlong Wang, Chunhua Shen, Baoshan Cheng, Hao Shen, and Huaxia Xia.", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 84, + 488, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 488, + 96 + ], + "score": 1.0, + "content": "End-to-end video instance segmentation with transformers. arXiv preprint arXiv:2011.14503, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "[63] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Convolutional block attention", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 112, + 488, + 124 + ], + "spans": [ + { + "bbox": [ + 126, + 112, + 488, + 124 + ], + "score": 1.0, + "content": "module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 128, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 104, + 128, + 507, + 143 + ], + "score": 1.0, + "content": "[64] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt:", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 140, + 448, + 152 + ], + "spans": [ + { + "bbox": [ + 126, + 140, + 448, + 152 + ], + "score": 1.0, + "content": "Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 104, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "[65] Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 167, + 507, + 181 + ], + "spans": [ + { + "bbox": [ + 124, + 167, + 507, + 181 + ], + "score": 1.0, + "content": "understanding. In Proceedings of the European Conference on Computer Vision (ECCV), pages 418–434,", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 124, + 176, + 151, + 191 + ], + "spans": [ + { + "bbox": [ + 124, + 176, + 151, + 191 + ], + "score": 1.0, + "content": "2018.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 197, + 506, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 506, + 209 + ], + "score": 1.0, + "content": "[66] Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. Segformer:", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 207, + 414, + 219 + ], + "spans": [ + { + "bbox": [ + 126, + 207, + 414, + 219 + ], + "score": 1.0, + "content": "Simple and efficient design for semantic segmentation with transformers, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "score": 1.0, + "content": "[67] Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregated residual transforma-", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 234, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 124, + 234, + 506, + 249 + ], + "score": 1.0, + "content": "tions for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 124, + 244, + 264, + 258 + ], + "spans": [ + { + "bbox": [ + 124, + 244, + 264, + 258 + ], + "score": 1.0, + "content": "recognition, pages 1492–1500, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 262, + 507, + 277 + ], + "spans": [ + { + "bbox": [ + 104, + 262, + 507, + 277 + ], + "score": 1.0, + "content": "[68] Jianwei Yang, Zhile Ren, Chuang Gan, Hongyuan Zhu, and Devi Parikh. Cross-channel communication", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 274, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 124, + 274, + 507, + 286 + ], + "score": 1.0, + "content": "networks. In Proceedings of the 33rd International Conference on Neural Information Processing Systems,", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 124, + 283, + 218, + 296 + ], + "spans": [ + { + "bbox": [ + 124, + 283, + 218, + 296 + ], + "score": 1.0, + "content": "pages 1297–1306, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 301, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 104, + 301, + 506, + 315 + ], + "score": 1.0, + "content": "[69] Minghao Yin, Zhuliang Yao, Yue Cao, Xiu Li, Zheng Zhang, Stephen Lin, and Han Hu. Disentangled", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 311, + 507, + 324 + ], + "spans": [ + { + "bbox": [ + 124, + 311, + 507, + 324 + ], + "score": 1.0, + "content": "non-local neural networks. In European Conference on Computer Vision, pages 191–207. Springer, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "score": 1.0, + "content": "[70] Kun Yuan, Shaopeng Guo, Ziwei Liu, Aojun Zhou, Fengwei Yu, and Wei Wu. Incorporating convolution", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 339, + 394, + 353 + ], + "spans": [ + { + "bbox": [ + 124, + 339, + 394, + 353 + ], + "score": 1.0, + "content": "designs into visual transformers. arXiv preprint arXiv:2103.11816, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 357, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 104, + 357, + 506, + 372 + ], + "score": 1.0, + "content": "[71] Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi, Francis EH Tay, Jiashi Feng, and Shuicheng", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 369, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 126, + 369, + 506, + 381 + ], + "score": 1.0, + "content": "Yan. Tokens-to-token vit: Training vision transformers from scratch on imagenet. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 379, + 219, + 389 + ], + "spans": [ + { + "bbox": [ + 126, + 379, + 219, + 389 + ], + "score": 1.0, + "content": "arXiv:2101.11986, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 397, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 505, + 409 + ], + "score": 1.0, + "content": "[72] Li Yuan, Qibin Hou, Zihang Jiang, Jiashi Feng, and Shuicheng Yan. Volo: Vision outlooker for visual", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 406, + 320, + 419 + ], + "spans": [ + { + "bbox": [ + 124, + 406, + 320, + 419 + ], + "score": 1.0, + "content": "recognition. arXiv preprint arXiv:2106.13112, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 424, + 507, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 507, + 438 + ], + "score": 1.0, + "content": "[73] Yuhui Yuan, Xilin Chen, and Jingdong Wang. Object-contextual representations for semantic segmentation.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 434, + 274, + 448 + ], + "spans": [ + { + "bbox": [ + 124, + 434, + 274, + 448 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1909.11065, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 452, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 467 + ], + "score": 1.0, + "content": "[74] Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 463, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 124, + 463, + 506, + 477 + ], + "score": 1.0, + "content": "Pham, Anirudh Ravula, Qifan Wang, Li Yang, et al. Big bird: Transformers for longer sequences. arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 29 + }, + { + "bbox": [ + 123, + 473, + 252, + 486 + ], + "spans": [ + { + "bbox": [ + 123, + 473, + 252, + 486 + ], + "score": 1.0, + "content": "preprint arXiv:2007.14062, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "[75] Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Haibin Lin, Zhi Zhang, Yue Sun, Tong He, Jonas", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 501, + 503, + 515 + ], + "spans": [ + { + "bbox": [ + 124, + 501, + 503, + 515 + ], + "score": 1.0, + "content": "Mueller, R Manmatha, et al. Resnest: Split-attention networks. arXiv preprint arXiv:2004.08955, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 519, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 507, + 533 + ], + "score": 1.0, + "content": "[76] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multi-", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 529, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 124, + 529, + 506, + 544 + ], + "score": 1.0, + "content": "scale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 34 + }, + { + "bbox": [ + 126, + 540, + 219, + 550 + ], + "spans": [ + { + "bbox": [ + 126, + 540, + 219, + 550 + ], + "score": 1.0, + "content": "arXiv:2103.15358, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 557, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 104, + 557, + 505, + 572 + ], + "score": 1.0, + "content": "[77] Wenwei Zhang, Jiangmiao Pang, Kai Chen, and Chen Change Loy. K-net: Towards unified image", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 569, + 204, + 581 + ], + "spans": [ + { + "bbox": [ + 126, + 569, + 204, + 581 + ], + "score": 1.0, + "content": "segmentation, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "[78] Jiaojiao Zhao, Xinyu Li, Chunhui Liu, Shuai Bing, Hao Chen, Cees GM Snoek, and Joseph Tighe. Tuber:", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 596, + 412, + 609 + ], + "spans": [ + { + "bbox": [ + 124, + 596, + 412, + 609 + ], + "score": 1.0, + "content": "Tube-transformer for action detection. arXiv preprint arXiv:2104.00969, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "[79] Minghang Zheng, Peng Gao, Xiaogang Wang, Hongsheng Li, and Hao Dong. End-to-end object detection", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 624, + 408, + 638 + ], + "spans": [ + { + "bbox": [ + 124, + 624, + 408, + 638 + ], + "score": 1.0, + "content": "with adaptive clustering transformer. arXiv preprint arXiv:2011.09315, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "[80] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 651, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 124, + 651, + 506, + 667 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 124, + 663, + 385, + 676 + ], + "spans": [ + { + "bbox": [ + 124, + 663, + 385, + 676 + ], + "score": 1.0, + "content": "perspective with transformers. arXiv preprint arXiv:2012.15840, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 104, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "[81] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 124, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 124, + 700, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 124, + 700, + 507, + 715 + ], + "score": 1.0, + "content": "perspective with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 124, + 711, + 294, + 724 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 294, + 724 + ], + "score": 1.0, + "content": "Pattern Recognition, pages 6881–6890, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 73, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 507, + 87 + ], + "score": 1.0, + "content": "[82] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation.", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 84, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 504, + 96 + ], + "score": 1.0, + "content": "In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 13001–13008, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 100, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 506, + 115 + ], + "score": 1.0, + "content": "[83] Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 111, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 126, + 111, + 507, + 125 + ], + "score": 1.0, + "content": "through ade20k dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition,", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 124, + 122, + 208, + 135 + ], + "spans": [ + { + "bbox": [ + 124, + 122, + 208, + 135 + ], + "score": 1.0, + "content": "pages 633–641, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 138, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 153 + ], + "score": 1.0, + "content": "[84] Xingyi Zhou, Vladlen Koltun, and Philipp Krähenbühl. Probabilistic two-stage detection. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 150, + 219, + 162 + ], + "spans": [ + { + "bbox": [ + 125, + 150, + 219, + 162 + ], + "score": 1.0, + "content": "arXiv:2103.07461, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "[85] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 176, + 436, + 189 + ], + "spans": [ + { + "bbox": [ + 126, + 176, + 436, + 189 + ], + "score": 1.0, + "content": "transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 70, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 41, + 507, + 725 + ], + "lines": [ + { + "bbox": [ + 103, + 70, + 507, + 88 + ], + "spans": [ + { + "bbox": [ + 103, + 70, + 507, + 88 + ], + "score": 1.0, + "content": "[20] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 125, + 84, + 379, + 96 + ], + "spans": [ + { + "bbox": [ + 125, + 84, + 379, + 96 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. NAACL, 2019.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 100, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 506, + 114 + ], + "score": 1.0, + "content": "[21] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 124, + 109, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 124, + 109, + 506, + 124 + ], + "score": 1.0, + "content": "Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 121, + 485, + 133 + ], + "spans": [ + { + "bbox": [ + 126, + 121, + 485, + 133 + ], + "score": 1.0, + "content": "16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 150 + ], + "score": 1.0, + "content": "[22] Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Golnaz Ghiasi, Mingxing Tan, Yin Cui, Quoc V Le, and Xiaodan", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 147, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 126, + 147, + 505, + 159 + ], + "score": 1.0, + "content": "Song. Spinenet: Learning scale-permuted backbone for recognition and localization. In Proceedings of the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 125, + 157, + 477, + 171 + ], + "spans": [ + { + "bbox": [ + 125, + 157, + 477, + 171 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11592–11601, 2020.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "[23] Haoqi Fan, Bo Xiong, Karttikeya Mangalam, Yanghao Li, Zhicheng Yan, Jitendra Malik, and Christoph", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 186, + 442, + 198 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 442, + 198 + ], + "score": 1.0, + "content": "Feichtenhofer. Multiscale vision transformers. arXiv preprint arXiv:2104.11227, 2021.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 204, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 107, + 204, + 506, + 214 + ], + "score": 1.0, + "content": "[24] Hao-Shu Fang, Jianhua Sun, Runzhong Wang, Minghao Gou, Yong-Lu Li, and Cewu Lu. Instaboost:", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "Boosting instance segmentation via probability map guided copy-pasting. In Proceedings of the IEEE/CVF", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 124, + 222, + 378, + 237 + ], + "spans": [ + { + "bbox": [ + 124, + 222, + 378, + 237 + ], + "score": 1.0, + "content": "International Conference on Computer Vision, pages 682–691, 2019.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "[25] Yuxin Fang, Shusheng Yang, Xinggang Wang, Yu Li, Chen Fang, Ying Shan, Bin Feng, and Wenyu Liu.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 125, + 250, + 226, + 262 + ], + "spans": [ + { + "bbox": [ + 125, + 250, + 226, + 262 + ], + "score": 1.0, + "content": "Instances as queries, 2021.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 268, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 506, + 280 + ], + "score": 1.0, + "content": "[26] Jun Fu, Jing Liu, Yuhang Wang, Yong Li, Yongjun Bao, Jinhui Tang, and Hanqing Lu. Adaptive context", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 125, + 277, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 125, + 277, + 506, + 289 + ], + "score": 1.0, + "content": "network for scene parsing. In Proceedings of the IEEE/CVF International Conference on Computer Vision,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 123, + 287, + 217, + 300 + ], + "spans": [ + { + "bbox": [ + 123, + 287, + 217, + 300 + ], + "score": 1.0, + "content": "pages 6748–6757, 2019.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "[27] Jianfeng Gao, Patrick Pantel, Michael Gamon, Xiaodong He, and Li Deng. Modeling interestingness with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 125, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "deep neural networks. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 325, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 125, + 325, + 506, + 337 + ], + "score": 1.0, + "content": "Processing (EMNLP), pages 2–13, Doha, Qatar, October 2014. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "[28] Golnaz Ghiasi, Yin Cui, Aravind Srinivas, Rui Qian, Tsung-Yi Lin, Ekin D Cubuk, Quoc V Le, and Barret", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 126, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 126, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "Zoph. Simple copy-paste is a strong data augmentation method for instance segmentation. arXiv preprint", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 125, + 363, + 219, + 375 + ], + "spans": [ + { + "bbox": [ + 125, + 363, + 219, + 375 + ], + "score": 1.0, + "content": "arXiv:2012.07177, 2020.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 380, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 506, + 392 + ], + "score": 1.0, + "content": "[29] Chengyue Gong, Dilin Wang, Meng Li, Vikas Chandra, and Qiang Liu. Vision transformers with patch", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 125, + 390, + 206, + 402 + ], + "spans": [ + { + "bbox": [ + 125, + 390, + 206, + 402 + ], + "score": 1.0, + "content": "diversification, 2021.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 408, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 107, + 408, + 506, + 418 + ], + "score": 1.0, + "content": "[30] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 125, + 417, + 502, + 429 + ], + "spans": [ + { + "bbox": [ + 125, + 417, + 502, + 429 + ], + "score": 1.0, + "content": "Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 434, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 507, + 448 + ], + "score": 1.0, + "content": "[31] Kai Han, An Xiao, Enhua Wu, Jianyuan Guo, Chunjing Xu, and Yunhe Wang. Transformer in transformer,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "spans": [ + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "[32] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 123, + 471, + 381, + 485 + ], + "spans": [ + { + "bbox": [ + 123, + 471, + 381, + 485 + ], + "score": 1.0, + "content": "international conference on computer vision, pages 2961–2969, 2017.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 488, + 507, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 507, + 503 + ], + "score": 1.0, + "content": "[33] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 124, + 499, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 124, + 499, + 507, + 513 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "score": 1.0, + "content": "[34] Elad Hoffer, Tal Ben-Nun, Itay Hubara, Niv Giladi, Torsten Hoefler, and Daniel Soudry. Augment your", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 528, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 125, + 528, + 506, + 540 + ], + "score": 1.0, + "content": "batch: Improving generalization through instance repetition. In Proceedings of the IEEE/CVF Conference", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 123, + 536, + 382, + 549 + ], + "spans": [ + { + "bbox": [ + 123, + 536, + 382, + 549 + ], + "score": 1.0, + "content": "on Computer Vision and Pattern Recognition, pages 8129–8138, 2020.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 555, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 506, + 567 + ], + "score": 1.0, + "content": "[35] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 123, + 564, + 377, + 577 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 377, + 577 + ], + "score": 1.0, + "content": "on computer vision and pattern recognition, pages 7132–7141, 2018.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 582, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 506, + 594 + ], + "score": 1.0, + "content": "[36] Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, and Mubarak", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 125, + 592, + 420, + 604 + ], + "spans": [ + { + "bbox": [ + 125, + 592, + 420, + 604 + ], + "score": 1.0, + "content": "Shah. Transformers in vision: A survey. arXiv preprint arXiv:2101.01169, 2021.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "[37] Yann LeCun, Yoshua Bengio, et al. Convolutional networks for images, speech, and time series. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 126, + 619, + 379, + 631 + ], + "spans": [ + { + "bbox": [ + 126, + 619, + 379, + 631 + ], + "score": 1.0, + "content": "handbook of brain theory and neural networks, 3361(10):1995, 1995.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "[38] Bing Li, Cheng Zheng, Silvio Giancola, and Bernard Ghanem. Sctn: Sparse convolution-transformer", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 647, + 401, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 647, + 401, + 659 + ], + "score": 1.0, + "content": "network for scene flow estimation. arXiv preprint arXiv:2105.04447, 2021.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 664, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 676 + ], + "score": 1.0, + "content": "[39] Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Mei Gao, Bin Xiao, Xiyang Dai, Lu Yuan, and Jianfeng Gao.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 124, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 124, + 673, + 506, + 687 + ], + "score": 1.0, + "content": "Efficient self-supervised vision transformers for representation learning. arXiv preprint arXiv:2106.09785,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 125, + 682, + 151, + 698 + ], + "spans": [ + { + "bbox": [ + 125, + 682, + 151, + 698 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 477, + 714 + ], + "score": 1.0, + "content": "[40] Xiangyu Li, Yonghong Hou, Pichao Wang, Zhimin Gao, Mingliang Xu, and Wanqing Li.", + "type": "text" + }, + { + "bbox": [ + 480, + 703, + 506, + 713 + ], + "score": 1.0, + "content": "Trear:", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 711, + 474, + 725 + ], + "spans": [ + { + "bbox": [ + 125, + 711, + 474, + 725 + ], + "score": 1.0, + "content": "Transformer-based rgb-d egocentric action recognition. arXiv preprint arXiv:2101.03904, 2021.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 41, + 507, + 725 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 103, + 70, + 507, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 48, + 507, + 728 + ], + "lines": [ + { + "bbox": [ + 104, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 104, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[41] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 93, + 151, + 106 + ], + "spans": [ + { + "bbox": [ + 125, + 93, + 151, + 106 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 111, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 506, + 123 + ], + "score": 1.0, + "content": "[42] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 125, + 121, + 450, + 133 + ], + "spans": [ + { + "bbox": [ + 125, + 121, + 450, + 133 + ], + "score": 1.0, + "content": "and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, 2014.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "[43] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 125, + 148, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 125, + 148, + 506, + 160 + ], + "score": 1.0, + "content": "transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 124, + 158, + 151, + 171 + ], + "spans": [ + { + "bbox": [ + 124, + 158, + 151, + 171 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "[44] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 186, + 219, + 198 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 219, + 198 + ], + "score": 1.0, + "content": "arXiv:1711.05101, 2017.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 202, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 217 + ], + "score": 1.0, + "content": "[45] Hyeonseob Nam, Jung-Woo Ha, and Jeonghee Kim. Dual attention networks for multimodal reasoning", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 123, + 212, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 123, + 212, + 505, + 227 + ], + "score": 1.0, + "content": "and matching. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 223, + 186, + 235 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 186, + 235 + ], + "score": 1.0, + "content": "299–307, 2017.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "score": 1.0, + "content": "[46] R. Pappagari, P. Zelasko, J. Villalba, Y. Carmiel, and N. Dehak. Hierarchical transformers for long", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 124, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 124, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "document classification. In 2019 IEEE Automatic Speech Recognition and Understanding Workshop", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 259, + 239, + 273 + ], + "spans": [ + { + "bbox": [ + 124, + 259, + 239, + 273 + ], + "score": 1.0, + "content": "(ASRU), pages 838–844, 2019.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "[47] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 124, + 287, + 507, + 301 + ], + "spans": [ + { + "bbox": [ + 124, + 287, + 507, + 301 + ], + "score": 1.0, + "content": "Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 125, + 298, + 151, + 310 + ], + "spans": [ + { + "bbox": [ + 125, + 298, + 151, + 310 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 316, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 327 + ], + "score": 1.0, + "content": "[48] Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 124, + 325, + 339, + 337 + ], + "spans": [ + { + "bbox": [ + 124, + 325, + 339, + 337 + ], + "score": 1.0, + "content": "journal on control and optimization, 30(4):838–855, 1992.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "[49] Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, and Timothy P Lillicrap. Compressive transformers", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 353, + 403, + 365 + ], + "spans": [ + { + "bbox": [ + 125, + 353, + 403, + 365 + ], + "score": 1.0, + "content": "for long-range sequence modelling. arXiv preprint arXiv:1911.05507, 2019.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 370, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 506, + 382 + ], + "score": 1.0, + "content": "[50] Prajit Ramachandran, Niki Parmar, Ashish Vaswani, Irwan Bello, Anselm Levskaya, and Jonathon Shlens.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 380, + 432, + 392 + ], + "spans": [ + { + "bbox": [ + 125, + 380, + 432, + 392 + ], + "score": 1.0, + "content": "Stand-alone self-attention in vision models. arXiv preprint arXiv:1906.05909, 2019.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 506, + 410 + ], + "score": 1.0, + "content": "[51] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 125, + 407, + 319, + 419 + ], + "spans": [ + { + "bbox": [ + 125, + 407, + 319, + 419 + ], + "score": 1.0, + "content": "Bottleneck transformers for visual recognition, 2021.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 425, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 437 + ], + "score": 1.0, + "content": "[52] Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. Segmenter: Transformer for semantic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 124, + 434, + 326, + 447 + ], + "spans": [ + { + "bbox": [ + 124, + 434, + 326, + 447 + ], + "score": 1.0, + "content": "segmentation. arXiv preprint arXiv:2105.05633, 2021.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "[53] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 124, + 462, + 248, + 475 + ], + "spans": [ + { + "bbox": [ + 124, + 462, + 248, + 475 + ], + "score": 1.0, + "content": "pose estimation. In CVPR, 2019.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "[54] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 489, + 421, + 504 + ], + "spans": [ + { + "bbox": [ + 124, + 489, + 421, + 504 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, pages 6105–6114. PMLR, 2019.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "[55] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 125, + 518, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 125, + 518, + 505, + 529 + ], + "score": 1.0, + "content": "Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 527, + 219, + 538 + ], + "spans": [ + { + "bbox": [ + 125, + 527, + 219, + 538 + ], + "score": 1.0, + "content": "arXiv:2012.12877, 2020.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "[56] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 125, + 554, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 125, + 554, + 505, + 566 + ], + "score": 1.0, + "content": "Shlens. Scaling local self-attention for parameter efficient visual backbones. In Proceedings of the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 124, + 564, + 477, + 578 + ], + "spans": [ + { + "bbox": [ + 124, + 564, + 477, + 578 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12894–12904, 2021.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "[57] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 125, + 592, + 393, + 604 + ], + "spans": [ + { + "bbox": [ + 125, + 592, + 393, + 604 + ], + "score": 1.0, + "content": "Kaiser, and Illia Polosukhin. Attention is all you need. In NeurIPS, 2017.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "[58] Huiyu Wang, Yukun Zhu, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen. Max-deeplab: End-to-end", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 123, + 619, + 447, + 632 + ], + "spans": [ + { + "bbox": [ + 123, + 619, + 447, + 632 + ], + "score": 1.0, + "content": "panoptic segmentation with mask transformers. arXiv preprint arXiv:2012.00759, 2020.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "[59] Ning Wang, Wengang Zhou, Jie Wang, and Houqaing Li. Transformer meets tracker: Exploiting temporal", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 646, + 397, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 646, + 397, + 659 + ], + "score": 1.0, + "content": "context for robust visual tracking. arXiv preprint arXiv:2103.11681, 2021.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "[60] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 125, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 125, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 123, + 683, + 273, + 696 + ], + "spans": [ + { + "bbox": [ + 123, + 683, + 273, + 696 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2102.12122, 2021.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[61] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 124, + 711, + 507, + 725 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 507, + 725 + ], + "score": 1.0, + "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 48, + 507, + 728 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 104, + 72, + 507, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 40, + 507, + 728 + ], + "lines": [ + { + "bbox": [ + 104, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 104, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "[62] Yuqing Wang, Zhaoliang Xu, Xinlong Wang, Chunhua Shen, Baoshan Cheng, Hao Shen, and Huaxia Xia.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 84, + 488, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 488, + 96 + ], + "score": 1.0, + "content": "End-to-end video instance segmentation with transformers. arXiv preprint arXiv:2011.14503, 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "[63] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Convolutional block attention", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 112, + 488, + 124 + ], + "spans": [ + { + "bbox": [ + 126, + 112, + 488, + 124 + ], + "score": 1.0, + "content": "module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19, 2018.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 128, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 104, + 128, + 507, + 143 + ], + "score": 1.0, + "content": "[64] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 126, + 140, + 448, + 152 + ], + "spans": [ + { + "bbox": [ + 126, + 140, + 448, + 152 + ], + "score": 1.0, + "content": "Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 104, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "[65] Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 124, + 167, + 507, + 181 + ], + "spans": [ + { + "bbox": [ + 124, + 167, + 507, + 181 + ], + "score": 1.0, + "content": "understanding. In Proceedings of the European Conference on Computer Vision (ECCV), pages 418–434,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 124, + 176, + 151, + 191 + ], + "spans": [ + { + "bbox": [ + 124, + 176, + 151, + 191 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 197, + 506, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 506, + 209 + ], + "score": 1.0, + "content": "[66] Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. Segformer:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 126, + 207, + 414, + 219 + ], + "spans": [ + { + "bbox": [ + 126, + 207, + 414, + 219 + ], + "score": 1.0, + "content": "Simple and efficient design for semantic segmentation with transformers, 2021.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "score": 1.0, + "content": "[67] Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregated residual transforma-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 124, + 234, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 124, + 234, + 506, + 249 + ], + "score": 1.0, + "content": "tions for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 124, + 244, + 264, + 258 + ], + "spans": [ + { + "bbox": [ + 124, + 244, + 264, + 258 + ], + "score": 1.0, + "content": "recognition, pages 1492–1500, 2017.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 262, + 507, + 277 + ], + "spans": [ + { + "bbox": [ + 104, + 262, + 507, + 277 + ], + "score": 1.0, + "content": "[68] Jianwei Yang, Zhile Ren, Chuang Gan, Hongyuan Zhu, and Devi Parikh. Cross-channel communication", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 274, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 124, + 274, + 507, + 286 + ], + "score": 1.0, + "content": "networks. In Proceedings of the 33rd International Conference on Neural Information Processing Systems,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 124, + 283, + 218, + 296 + ], + "spans": [ + { + "bbox": [ + 124, + 283, + 218, + 296 + ], + "score": 1.0, + "content": "pages 1297–1306, 2019.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 301, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 104, + 301, + 506, + 315 + ], + "score": 1.0, + "content": "[69] Minghao Yin, Zhuliang Yao, Yue Cao, Xiu Li, Zheng Zhang, Stephen Lin, and Han Hu. Disentangled", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 124, + 311, + 507, + 324 + ], + "spans": [ + { + "bbox": [ + 124, + 311, + 507, + 324 + ], + "score": 1.0, + "content": "non-local neural networks. In European Conference on Computer Vision, pages 191–207. Springer, 2020.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "score": 1.0, + "content": "[70] Kun Yuan, Shaopeng Guo, Ziwei Liu, Aojun Zhou, Fengwei Yu, and Wei Wu. Incorporating convolution", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 124, + 339, + 394, + 353 + ], + "spans": [ + { + "bbox": [ + 124, + 339, + 394, + 353 + ], + "score": 1.0, + "content": "designs into visual transformers. arXiv preprint arXiv:2103.11816, 2021.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 357, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 104, + 357, + 506, + 372 + ], + "score": 1.0, + "content": "[71] Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi, Francis EH Tay, Jiashi Feng, and Shuicheng", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 126, + 369, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 126, + 369, + 506, + 381 + ], + "score": 1.0, + "content": "Yan. Tokens-to-token vit: Training vision transformers from scratch on imagenet. arXiv preprint", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 379, + 219, + 389 + ], + "spans": [ + { + "bbox": [ + 126, + 379, + 219, + 389 + ], + "score": 1.0, + "content": "arXiv:2101.11986, 2021.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 397, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 505, + 409 + ], + "score": 1.0, + "content": "[72] Li Yuan, Qibin Hou, Zihang Jiang, Jiashi Feng, and Shuicheng Yan. Volo: Vision outlooker for visual", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 124, + 406, + 320, + 419 + ], + "spans": [ + { + "bbox": [ + 124, + 406, + 320, + 419 + ], + "score": 1.0, + "content": "recognition. arXiv preprint arXiv:2106.13112, 2021.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 424, + 507, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 507, + 438 + ], + "score": 1.0, + "content": "[73] Yuhui Yuan, Xilin Chen, and Jingdong Wang. Object-contextual representations for semantic segmentation.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 124, + 434, + 274, + 448 + ], + "spans": [ + { + "bbox": [ + 124, + 434, + 274, + 448 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1909.11065, 2019.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 452, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 467 + ], + "score": 1.0, + "content": "[74] Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 124, + 463, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 124, + 463, + 506, + 477 + ], + "score": 1.0, + "content": "Pham, Anirudh Ravula, Qifan Wang, Li Yang, et al. Big bird: Transformers for longer sequences. arXiv", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 123, + 473, + 252, + 486 + ], + "spans": [ + { + "bbox": [ + 123, + 473, + 252, + 486 + ], + "score": 1.0, + "content": "preprint arXiv:2007.14062, 2020.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "[75] Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Haibin Lin, Zhi Zhang, Yue Sun, Tong He, Jonas", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 501, + 503, + 515 + ], + "spans": [ + { + "bbox": [ + 124, + 501, + 503, + 515 + ], + "score": 1.0, + "content": "Mueller, R Manmatha, et al. Resnest: Split-attention networks. arXiv preprint arXiv:2004.08955, 2020.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 519, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 507, + 533 + ], + "score": 1.0, + "content": "[76] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multi-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 124, + 529, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 124, + 529, + 506, + 544 + ], + "score": 1.0, + "content": "scale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 126, + 540, + 219, + 550 + ], + "spans": [ + { + "bbox": [ + 126, + 540, + 219, + 550 + ], + "score": 1.0, + "content": "arXiv:2103.15358, 2021.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 557, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 104, + 557, + 505, + 572 + ], + "score": 1.0, + "content": "[77] Wenwei Zhang, Jiangmiao Pang, Kai Chen, and Chen Change Loy. K-net: Towards unified image", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 126, + 569, + 204, + 581 + ], + "spans": [ + { + "bbox": [ + 126, + 569, + 204, + 581 + ], + "score": 1.0, + "content": "segmentation, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "[78] Jiaojiao Zhao, Xinyu Li, Chunhui Liu, Shuai Bing, Hao Chen, Cees GM Snoek, and Joseph Tighe. Tuber:", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 124, + 596, + 412, + 609 + ], + "spans": [ + { + "bbox": [ + 124, + 596, + 412, + 609 + ], + "score": 1.0, + "content": "Tube-transformer for action detection. arXiv preprint arXiv:2104.00969, 2021.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "[79] Minghang Zheng, Peng Gao, Xiaogang Wang, Hongsheng Li, and Hao Dong. End-to-end object detection", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 124, + 624, + 408, + 638 + ], + "spans": [ + { + "bbox": [ + 124, + 624, + 408, + 638 + ], + "score": 1.0, + "content": "with adaptive clustering transformer. arXiv preprint arXiv:2011.09315, 2020.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "[80] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 124, + 651, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 124, + 651, + 506, + 667 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 124, + 663, + 385, + 676 + ], + "spans": [ + { + "bbox": [ + 124, + 663, + 385, + 676 + ], + "score": 1.0, + "content": "perspective with transformers. arXiv preprint arXiv:2012.15840, 2020.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 104, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "[81] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 124, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 124, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 124, + 700, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 124, + 700, + 507, + 715 + ], + "score": 1.0, + "content": "perspective with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 124, + 711, + 294, + 724 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 294, + 724 + ], + "score": 1.0, + "content": "Pattern Recognition, pages 6881–6890, 2021.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 40, + 507, + 728 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 104, + 72, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 73, + 506, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 507, + 87 + ], + "score": 1.0, + "content": "[82] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 84, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 504, + 96 + ], + "score": 1.0, + "content": "In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 13001–13008, 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 100, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 506, + 115 + ], + "score": 1.0, + "content": "[83] Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 111, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 126, + 111, + 507, + 125 + ], + "score": 1.0, + "content": "through ade20k dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 124, + 122, + 208, + 135 + ], + "spans": [ + { + "bbox": [ + 124, + 122, + 208, + 135 + ], + "score": 1.0, + "content": "pages 633–641, 2017.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 153 + ], + "score": 1.0, + "content": "[84] Xingyi Zhou, Vladlen Koltun, and Philipp Krähenbühl. Probabilistic two-stage detection. arXiv preprint", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 125, + 150, + 219, + 162 + ], + "spans": [ + { + "bbox": [ + 125, + 150, + 219, + 162 + ], + "score": 1.0, + "content": "arXiv:2103.07461, 2021.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "[85] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 176, + 436, + 189 + ], + "spans": [ + { + "bbox": [ + 126, + 176, + 436, + 189 + ], + "score": 1.0, + "content": "transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 106, + 73, + 506, + 188 + ], + "lines": [], + "index": 4, + "bbox_fs": [ + 105, + 73, + 507, + 189 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/2zCRcTafea/2zCRcTafea_model.json b/parse/train/2zCRcTafea/2zCRcTafea_model.json new file mode 100644 index 0000000000000000000000000000000000000000..34cb7c1a25cbb6e301d02e92f7a53b22a67b7b95 --- /dev/null +++ b/parse/train/2zCRcTafea/2zCRcTafea_model.json @@ -0,0 +1,18352 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 766, + 1302, + 766, + 1302, + 1553, + 397, + 1553 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1664, + 1406, + 1664, + 1406, + 1877, + 298, + 1877 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 300, + 1891, + 1403, + 1891, + 1403, + 1984, + 300, + 1984 + ], + "score": 0.963 + }, + { + "category_id": 0, + "poly": [ + 368, + 271, + 1335, + 271, + 1335, + 381, + 368, + 381 + ], + "score": 0.956 + }, + { + "category_id": 0, + "poly": [ + 299, + 1610, + 531, + 1610, + 531, + 1647, + 299, + 1647 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 788, + 693, + 912, + 693, + 912, + 730, + 788, + 730 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 298, + 2033, + 1070, + 2033, + 1070, + 2061, + 298, + 2061 + ], + "score": 0.854 + }, + { + "category_id": 1, + "poly": [ + 387, + 486, + 1314, + 486, + 1314, + 617, + 387, + 617 + ], + "score": 0.739 + }, + { + "category_id": 13, + "poly": [ + 1085, + 1311, + 1210, + 1311, + 1210, + 1339, + 1085, + 1339 + ], + "score": 0.89, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 1228, + 1280, + 1304, + 1280, + 1304, + 1309, + 1228, + 1309 + ], + "score": 0.86, + "latex": "\\mathbf { 8 4 . 0 \\% }" + }, + { + "category_id": 13, + "poly": [ + 1097, + 1280, + 1173, + 1280, + 1173, + 1308, + 1097, + 1308 + ], + "score": 0.86, + "latex": "\\mathbf { 8 3 . 6 \\% }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 557, + 1137, + 557, + 1137, + 581, + 1113, + 581 + ], + "score": 0.56, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 710, + 487, + 755, + 487, + 755, + 519, + 710, + 519 + ], + "score": 0.32, + "latex": "\\mathbf { L i } ^ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 921, + 1280, + 1000, + 1280, + 1000, + 1309, + 921, + 1309 + ], + "score": 0.28, + "latex": "8 9 . 8 \\mathbf { M }" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 270.0, + 1337.0, + 270.0, + 1337.0, + 328.0, + 361.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 326.0, + 1065.0, + 326.0, + 1065.0, + 383.0, + 634.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1606.0, + 535.0, + 1606.0, + 535.0, + 1653.0, + 292.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 693.0, + 917.0, + 693.0, + 917.0, + 732.0, + 784.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2028.0, + 1075.0, + 2028.0, + 1075.0, + 2067.0, + 293.0, + 2067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 765.0, + 1305.0, + 765.0, + 1305.0, + 798.0, + 395.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 794.0, + 1306.0, + 794.0, + 1306.0, + 830.0, + 393.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 824.0, + 1306.0, + 824.0, + 1306.0, + 862.0, + 393.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 856.0, + 1307.0, + 856.0, + 1307.0, + 892.0, + 393.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 887.0, + 1306.0, + 887.0, + 1306.0, + 920.0, + 393.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 917.0, + 1305.0, + 917.0, + 1305.0, + 951.0, + 395.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 946.0, + 1305.0, + 946.0, + 1305.0, + 984.0, + 393.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 975.0, + 1307.0, + 975.0, + 1307.0, + 1015.0, + 392.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1008.0, + 1305.0, + 1008.0, + 1305.0, + 1041.0, + 394.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1037.0, + 1306.0, + 1037.0, + 1306.0, + 1073.0, + 393.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1065.0, + 1306.0, + 1065.0, + 1306.0, + 1104.0, + 393.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1098.0, + 1306.0, + 1098.0, + 1306.0, + 1133.0, + 394.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1130.0, + 1306.0, + 1130.0, + 1306.0, + 1163.0, + 394.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1158.0, + 1306.0, + 1158.0, + 1306.0, + 1191.0, + 394.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1189.0, + 1306.0, + 1189.0, + 1306.0, + 1222.0, + 394.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1219.0, + 1305.0, + 1219.0, + 1305.0, + 1252.0, + 394.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1250.0, + 1305.0, + 1250.0, + 1305.0, + 1283.0, + 395.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1280.0, + 920.0, + 1280.0, + 920.0, + 1312.0, + 392.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1280.0, + 1096.0, + 1280.0, + 1096.0, + 1312.0, + 1001.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1280.0, + 1227.0, + 1280.0, + 1227.0, + 1312.0, + 1174.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1310.0, + 1084.0, + 1310.0, + 1084.0, + 1344.0, + 392.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1310.0, + 1306.0, + 1310.0, + 1306.0, + 1344.0, + 1211.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1341.0, + 1305.0, + 1341.0, + 1305.0, + 1372.0, + 394.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1372.0, + 1305.0, + 1372.0, + 1305.0, + 1404.0, + 393.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1400.0, + 1306.0, + 1400.0, + 1306.0, + 1433.0, + 395.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1433.0, + 1304.0, + 1433.0, + 1304.0, + 1463.0, + 395.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1462.0, + 1306.0, + 1462.0, + 1306.0, + 1495.0, + 394.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1491.0, + 1308.0, + 1491.0, + 1308.0, + 1527.0, + 394.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1523.0, + 861.0, + 1523.0, + 861.0, + 1556.0, + 396.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1662.0, + 1409.0, + 1662.0, + 1409.0, + 1701.0, + 293.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1695.0, + 1408.0, + 1695.0, + 1408.0, + 1729.0, + 293.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1725.0, + 1409.0, + 1725.0, + 1409.0, + 1759.0, + 295.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1757.0, + 1407.0, + 1757.0, + 1407.0, + 1788.0, + 296.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1784.0, + 1408.0, + 1784.0, + 1408.0, + 1821.0, + 293.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1816.0, + 1409.0, + 1816.0, + 1409.0, + 1850.0, + 295.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1846.0, + 1316.0, + 1846.0, + 1316.0, + 1881.0, + 295.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1890.0, + 1405.0, + 1890.0, + 1405.0, + 1928.0, + 295.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1922.0, + 1405.0, + 1922.0, + 1405.0, + 1956.0, + 295.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1953.0, + 1404.0, + 1953.0, + 1404.0, + 1987.0, + 296.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 488.0, + 709.0, + 488.0, + 709.0, + 525.0, + 382.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 488.0, + 1314.0, + 488.0, + 1314.0, + 525.0, + 756.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 518.0, + 1009.0, + 518.0, + 1009.0, + 557.0, + 688.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 547.0, + 1112.0, + 547.0, + 1112.0, + 589.0, + 524.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 547.0, + 1178.0, + 547.0, + 1178.0, + 589.0, + 1138.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 585.0, + 1313.0, + 585.0, + 1313.0, + 620.0, + 388.0, + 620.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 914, + 1404, + 914, + 1404, + 1248, + 298, + 1248 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1262, + 1404, + 1262, + 1404, + 1596, + 298, + 1596 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 625, + 1404, + 625, + 1404, + 900, + 298, + 900 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1700, + 1405, + 1700, + 1405, + 1945, + 298, + 1945 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 313, + 213, + 1390, + 213, + 1390, + 494, + 313, + 494 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 299, + 500, + 1396, + 500, + 1396, + 594, + 299, + 594 + ], + "score": 0.954 + }, + { + "category_id": 0, + "poly": [ + 299, + 1644, + 541, + 1644, + 541, + 1681, + 299, + 1681 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 322, + 1978, + 1355, + 1978, + 1355, + 2008, + 322, + 2008 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 859, + 2062, + 859, + 2085, + 841, + 2085 + ], + "score": 0.737 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.096 + }, + { + "category_id": 13, + "poly": [ + 871, + 1385, + 946, + 1385, + 946, + 1414, + 871, + 1414 + ], + "score": 0.87, + "latex": "8 3 . 6 \\%" + }, + { + "category_id": 13, + "poly": [ + 776, + 1415, + 850, + 1415, + 850, + 1444, + 776, + 1444 + ], + "score": 0.86, + "latex": "8 4 . 0 \\%" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 223.0, + 730.0, + 223.0, + 730.0, + 231.0, + 704.0, + 231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 224.0, + 1339.0, + 224.0, + 1339.0, + 248.0, + 1241.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 251.0, + 924.0, + 251.0, + 924.0, + 408.0, + 828.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 263.0, + 1295.0, + 263.0, + 1295.0, + 274.0, + 1286.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 302.0, + 574.0, + 302.0, + 574.0, + 500.0, + 331.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 312.0, + 733.0, + 312.0, + 733.0, + 324.0, + 702.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 281.0, + 979.0, + 281.0, + 979.0, + 438.0, + 909.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 302.0, + 1363.0, + 302.0, + 1363.0, + 331.0, + 1226.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 344.0, + 1228.0, + 344.0, + 1228.0, + 357.0, + 1211.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 337.0, + 1385.0, + 337.0, + 1385.0, + 362.0, + 1289.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 451.0, + 1246.0, + 451.0, + 1246.0, + 461.0, + 1236.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 468.0, + 1379.0, + 468.0, + 1379.0, + 490.0, + 1286.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 498.0, + 1401.0, + 498.0, + 1401.0, + 536.0, + 294.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 531.0, + 1402.0, + 531.0, + 1402.0, + 565.0, + 295.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 560.0, + 1372.0, + 560.0, + 1372.0, + 599.0, + 293.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1641.0, + 545.0, + 1641.0, + 545.0, + 1686.0, + 292.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1973.0, + 1360.0, + 1973.0, + 1360.0, + 2011.0, + 335.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2094.0, + 838.0, + 2094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 915.0, + 1404.0, + 915.0, + 1404.0, + 949.0, + 294.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 945.0, + 1404.0, + 945.0, + 1404.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 975.0, + 1404.0, + 975.0, + 1404.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1006.0, + 1404.0, + 1006.0, + 1404.0, + 1040.0, + 293.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1033.0, + 1405.0, + 1033.0, + 1405.0, + 1074.0, + 292.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1066.0, + 1406.0, + 1066.0, + 1406.0, + 1101.0, + 294.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1097.0, + 1406.0, + 1097.0, + 1406.0, + 1132.0, + 294.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1128.0, + 1404.0, + 1128.0, + 1404.0, + 1159.0, + 294.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1155.0, + 1405.0, + 1155.0, + 1405.0, + 1191.0, + 293.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1188.0, + 1406.0, + 1188.0, + 1406.0, + 1220.0, + 293.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1219.0, + 1155.0, + 1219.0, + 1155.0, + 1250.0, + 296.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1263.0, + 1408.0, + 1263.0, + 1408.0, + 1297.0, + 296.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1295.0, + 1406.0, + 1295.0, + 1406.0, + 1329.0, + 294.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1323.0, + 1405.0, + 1323.0, + 1405.0, + 1357.0, + 294.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1351.0, + 1405.0, + 1351.0, + 1405.0, + 1388.0, + 293.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1382.0, + 870.0, + 1382.0, + 870.0, + 1419.0, + 292.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1382.0, + 1405.0, + 1382.0, + 1405.0, + 1419.0, + 947.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1413.0, + 775.0, + 1413.0, + 775.0, + 1450.0, + 293.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1413.0, + 1406.0, + 1413.0, + 1406.0, + 1450.0, + 851.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1444.0, + 1406.0, + 1444.0, + 1406.0, + 1478.0, + 294.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1473.0, + 1406.0, + 1473.0, + 1406.0, + 1509.0, + 292.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1504.0, + 1406.0, + 1504.0, + 1406.0, + 1541.0, + 293.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1534.0, + 1406.0, + 1534.0, + 1406.0, + 1571.0, + 293.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1566.0, + 1165.0, + 1566.0, + 1165.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 626.0, + 1409.0, + 626.0, + 1409.0, + 662.0, + 294.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 656.0, + 1406.0, + 656.0, + 1406.0, + 691.0, + 295.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 683.0, + 1408.0, + 683.0, + 1408.0, + 726.0, + 292.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 716.0, + 1406.0, + 716.0, + 1406.0, + 753.0, + 293.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 747.0, + 1408.0, + 747.0, + 1408.0, + 783.0, + 294.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 777.0, + 1406.0, + 777.0, + 1406.0, + 813.0, + 294.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 807.0, + 1405.0, + 807.0, + 1405.0, + 843.0, + 291.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 838.0, + 1405.0, + 838.0, + 1405.0, + 874.0, + 294.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 868.0, + 1371.0, + 868.0, + 1371.0, + 904.0, + 294.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1702.0, + 1405.0, + 1702.0, + 1405.0, + 1735.0, + 296.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1733.0, + 1405.0, + 1733.0, + 1405.0, + 1766.0, + 295.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1763.0, + 1405.0, + 1763.0, + 1405.0, + 1796.0, + 293.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1791.0, + 1406.0, + 1791.0, + 1406.0, + 1827.0, + 292.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1824.0, + 1405.0, + 1824.0, + 1405.0, + 1857.0, + 296.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1854.0, + 1407.0, + 1854.0, + 1407.0, + 1887.0, + 295.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1886.0, + 1406.0, + 1886.0, + 1406.0, + 1916.0, + 296.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1948.0, + 293.0, + 1948.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 908, + 1405, + 908, + 1405, + 1458, + 297, + 1458 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1630, + 1405, + 1630, + 1405, + 1934, + 297, + 1934 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 774, + 1403, + 774, + 1403, + 896, + 299, + 896 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 307, + 201, + 1389, + 201, + 1389, + 603, + 307, + 603 + ], + "score": 0.969 + }, + { + "category_id": 2, + "poly": [ + 332, + 1977, + 996, + 1977, + 996, + 2006, + 332, + 2006 + ], + "score": 0.916 + }, + { + "category_id": 4, + "poly": [ + 294, + 642, + 1402, + 642, + 1402, + 706, + 294, + 706 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 299, + 1570, + 589, + 1570, + 589, + 1603, + 299, + 1603 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 298, + 1511, + 461, + 1511, + 461, + 1548, + 298, + 1548 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.667 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.413 + }, + { + "category_id": 13, + "poly": [ + 502, + 1689, + 592, + 1689, + 592, + 1729, + 502, + 1729 + ], + "score": 0.94, + "latex": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } }" + }, + { + "category_id": 13, + "poly": [ + 781, + 1783, + 946, + 1783, + 946, + 1816, + 781, + 1816 + ], + "score": 0.93, + "latex": "i \\in \\{ 1 , 2 , 3 , 4 \\}" + }, + { + "category_id": 13, + "poly": [ + 822, + 1659, + 987, + 1659, + 987, + 1690, + 822, + 1690 + ], + "score": 0.93, + "latex": "I \\in \\mathcal { R } ^ { H \\times W \\times 3 }" + }, + { + "category_id": 13, + "poly": [ + 920, + 1692, + 1030, + 1692, + 1030, + 1721, + 920, + 1721 + ], + "score": 0.9, + "latex": "4 \\times 4 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 297, + 1694, + 359, + 1694, + 359, + 1720, + 297, + 1720 + ], + "score": 0.88, + "latex": "4 \\times 4" + }, + { + "category_id": 13, + "poly": [ + 298, + 1814, + 331, + 1814, + 331, + 1843, + 298, + 1843 + ], + "score": 0.86, + "latex": "N _ { i }" + }, + { + "category_id": 13, + "poly": [ + 780, + 1754, + 797, + 1754, + 797, + 1780, + 780, + 1780 + ], + "score": 0.66, + "latex": "d" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 203.0, + 540.0, + 203.0, + 540.0, + 248.0, + 455.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 202.0, + 766.0, + 202.0, + 766.0, + 249.0, + 672.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 202.0, + 1012.0, + 202.0, + 1012.0, + 249.0, + 910.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 202.0, + 1254.0, + 202.0, + 1254.0, + 255.0, + 1146.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 255.0, + 580.0, + 255.0, + 580.0, + 264.0, + 570.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 284.0, + 613.0, + 284.0, + 613.0, + 308.0, + 529.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 282.0, + 851.0, + 282.0, + 851.0, + 308.0, + 767.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 281.0, + 1085.0, + 281.0, + 1085.0, + 308.0, + 1000.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 284.0, + 1320.0, + 284.0, + 1320.0, + 308.0, + 1236.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 300.0, + 613.0, + 300.0, + 613.0, + 327.0, + 528.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 300.0, + 851.0, + 300.0, + 851.0, + 327.0, + 767.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 300.0, + 1086.0, + 300.0, + 1086.0, + 327.0, + 1000.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 302.0, + 1320.0, + 302.0, + 1320.0, + 326.0, + 1236.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 319.0, + 484.0, + 319.0, + 484.0, + 463.0, + 455.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 349.0, + 613.0, + 349.0, + 613.0, + 373.0, + 529.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 315.0, + 729.0, + 315.0, + 729.0, + 466.0, + 693.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 349.0, + 852.0, + 349.0, + 852.0, + 373.0, + 769.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 317.0, + 959.0, + 317.0, + 959.0, + 465.0, + 927.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 349.0, + 1084.0, + 349.0, + 1084.0, + 373.0, + 1001.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 319.0, + 1192.0, + 319.0, + 1192.0, + 464.0, + 1163.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 349.0, + 1320.0, + 349.0, + 1320.0, + 373.0, + 1237.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 388.0, + 1062.0, + 388.0, + 1062.0, + 416.0, + 1030.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 389.0, + 1296.0, + 389.0, + 1296.0, + 413.0, + 1265.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 433.0, + 410.0, + 433.0, + 410.0, + 458.0, + 326.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 426.0, + 624.0, + 426.0, + 624.0, + 470.0, + 516.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 426.0, + 862.0, + 426.0, + 862.0, + 470.0, + 755.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 426.0, + 1094.0, + 426.0, + 1094.0, + 468.0, + 988.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 425.0, + 1330.0, + 425.0, + 1330.0, + 470.0, + 1223.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 488.0, + 613.0, + 488.0, + 613.0, + 515.0, + 529.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 488.0, + 852.0, + 488.0, + 852.0, + 515.0, + 768.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 488.0, + 1086.0, + 488.0, + 1086.0, + 515.0, + 1001.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 488.0, + 1321.0, + 488.0, + 1321.0, + 515.0, + 1236.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 543.0, + 678.0, + 543.0, + 678.0, + 569.0, + 475.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 542.0, + 922.0, + 542.0, + 922.0, + 571.0, + 718.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 544.0, + 1151.0, + 544.0, + 1151.0, + 571.0, + 948.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 544.0, + 1384.0, + 544.0, + 1384.0, + 571.0, + 1183.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 573.0, + 618.0, + 573.0, + 618.0, + 606.0, + 541.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 573.0, + 853.0, + 573.0, + 853.0, + 606.0, + 775.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 573.0, + 1089.0, + 573.0, + 1089.0, + 606.0, + 1013.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 573.0, + 1323.0, + 573.0, + 1323.0, + 606.0, + 1244.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.75, + 392.0, + 583.75, + 392.0, + 583.75, + 406.5, + 552.75, + 406.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1972.0, + 998.0, + 1972.0, + 998.0, + 2011.0, + 331.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 638.0, + 1407.0, + 638.0, + 1407.0, + 682.0, + 293.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 1122.0, + 673.0, + 1122.0, + 709.0, + 296.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1568.0, + 592.0, + 1568.0, + 592.0, + 1607.0, + 294.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1508.0, + 467.0, + 1508.0, + 467.0, + 1553.0, + 290.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 909.0, + 1405.0, + 909.0, + 1405.0, + 948.0, + 294.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 941.0, + 1405.0, + 941.0, + 1405.0, + 979.0, + 294.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 972.0, + 1405.0, + 972.0, + 1405.0, + 1008.0, + 294.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1038.0, + 295.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1029.0, + 1405.0, + 1029.0, + 1405.0, + 1070.0, + 294.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1060.0, + 1405.0, + 1060.0, + 1405.0, + 1102.0, + 294.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1092.0, + 1406.0, + 1092.0, + 1406.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1160.0, + 294.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1151.0, + 1405.0, + 1151.0, + 1405.0, + 1191.0, + 294.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1182.0, + 1408.0, + 1182.0, + 1408.0, + 1221.0, + 294.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 1408.0, + 1215.0, + 1408.0, + 1250.0, + 294.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1243.0, + 1407.0, + 1243.0, + 1407.0, + 1281.0, + 292.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1272.0, + 1407.0, + 1272.0, + 1407.0, + 1313.0, + 292.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1306.0, + 1406.0, + 1306.0, + 1406.0, + 1342.0, + 295.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1334.0, + 1406.0, + 1334.0, + 1406.0, + 1372.0, + 291.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1368.0, + 1402.0, + 1368.0, + 1402.0, + 1400.0, + 295.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1392.0, + 1406.0, + 1392.0, + 1406.0, + 1434.0, + 291.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1424.0, + 1246.0, + 1424.0, + 1246.0, + 1464.0, + 292.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1628.0, + 1406.0, + 1628.0, + 1406.0, + 1666.0, + 292.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1653.0, + 821.0, + 1653.0, + 821.0, + 1700.0, + 290.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1653.0, + 1408.0, + 1653.0, + 1408.0, + 1700.0, + 988.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 1675.0, + 296.0, + 1675.0, + 296.0, + 1744.0, + 282.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1675.0, + 501.0, + 1675.0, + 501.0, + 1744.0, + 360.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1675.0, + 919.0, + 1675.0, + 919.0, + 1744.0, + 593.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1675.0, + 1405.0, + 1675.0, + 1405.0, + 1744.0, + 1031.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1721.0, + 1403.0, + 1721.0, + 1403.0, + 1757.0, + 294.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1749.0, + 779.0, + 1749.0, + 779.0, + 1788.0, + 292.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1749.0, + 1406.0, + 1749.0, + 1406.0, + 1788.0, + 798.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1782.0, + 780.0, + 1782.0, + 780.0, + 1817.0, + 294.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1782.0, + 1407.0, + 1782.0, + 1407.0, + 1817.0, + 947.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1810.0, + 297.0, + 1810.0, + 297.0, + 1848.0, + 294.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1810.0, + 1406.0, + 1810.0, + 1406.0, + 1848.0, + 332.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1845.0, + 1407.0, + 1845.0, + 1407.0, + 1877.0, + 296.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1873.0, + 1406.0, + 1873.0, + 1406.0, + 1908.0, + 295.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1903.0, + 1408.0, + 1903.0, + 1408.0, + 1939.0, + 295.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 775.0, + 1404.0, + 775.0, + 1404.0, + 811.0, + 294.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 807.0, + 1404.0, + 807.0, + 1404.0, + 840.0, + 295.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 834.0, + 1407.0, + 834.0, + 1407.0, + 872.0, + 293.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 864.0, + 1366.0, + 864.0, + 1366.0, + 902.0, + 293.0, + 902.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1058, + 1405, + 1058, + 1405, + 1271, + 297, + 1271 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 559, + 936, + 559, + 936, + 1044, + 299, + 1044 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 279, + 1404, + 279, + 1404, + 469, + 298, + 469 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1350, + 1405, + 1350, + 1405, + 1475, + 298, + 1475 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1733, + 1405, + 1733, + 1405, + 1862, + 298, + 1862 + ], + "score": 0.973 + }, + { + "category_id": 4, + "poly": [ + 957, + 771, + 1406, + 771, + 1406, + 1014, + 957, + 1014 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 955, + 559, + 1382, + 559, + 1382, + 758, + 955, + 758 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 1489, + 1406, + 1489, + 1406, + 1646, + 298, + 1646 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 294, + 202, + 1401, + 202, + 1401, + 265, + 294, + 265 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 298, + 1940, + 1399, + 1940, + 1399, + 2009, + 298, + 2009 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 463, + 1869, + 1235, + 1869, + 1235, + 1920, + 463, + 1920 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 294, + 1658, + 1402, + 1658, + 1402, + 1723, + 294, + 1723 + ], + "score": 0.934 + }, + { + "category_id": 0, + "poly": [ + 299, + 502, + 673, + 502, + 673, + 534, + 299, + 534 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 299, + 1302, + 720, + 1302, + 720, + 1333, + 299, + 1333 + ], + "score": 0.917 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1888, + 1400, + 1888, + 1400, + 1916, + 1369, + 1916 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.784 + }, + { + "category_id": 13, + "poly": [ + 1084, + 337, + 1224, + 337, + 1224, + 376, + 1084, + 376 + ], + "score": 0.93, + "latex": "{ \\frac { H } { 4 } } \\times { \\frac { W } { 4 } } \\times d" + }, + { + "category_id": 13, + "poly": [ + 801, + 1551, + 953, + 1551, + 953, + 1584, + 801, + 1584 + ], + "score": 0.93, + "latex": "l \\in \\{ 1 , . . . , L \\}" + }, + { + "category_id": 13, + "poly": [ + 1192, + 1797, + 1289, + 1797, + 1289, + 1830, + 1192, + 1830 + ], + "score": 0.92, + "latex": "s _ { w } ^ { l } \\times s _ { w } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 506, + 372, + 690, + 372, + 690, + 411, + 506, + 411 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\mathcal { O } ( ( \\frac { H } { 4 } \\times \\frac { W } { 4 } ) ^ { 2 } d ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 573, + 1350, + 746, + 1350, + 746, + 1382, + 573, + 1382 + ], + "score": 0.92, + "latex": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }" + }, + { + "category_id": 13, + "poly": [ + 886, + 1732, + 1060, + 1732, + 1060, + 1764, + 886, + 1764 + ], + "score": 0.92, + "latex": "\\boldsymbol { x } \\in \\mathcal { R } ^ { M \\times N \\times d }" + }, + { + "category_id": 13, + "poly": [ + 591, + 1977, + 674, + 1977, + 674, + 2009, + 591, + 2009 + ], + "score": 0.91, + "latex": "s _ { w } ^ { l } = 1" + }, + { + "category_id": 13, + "poly": [ + 347, + 407, + 483, + 407, + 483, + 439, + 347, + 439 + ], + "score": 0.91, + "latex": "\\operatorname* { m i n } ( H , W )" + }, + { + "category_id": 13, + "poly": [ + 520, + 1387, + 609, + 1387, + 609, + 1419, + 520, + 1419 + ], + "score": 0.9, + "latex": "s _ { p } \\times s _ { p }" + }, + { + "category_id": 13, + "poly": [ + 613, + 1831, + 640, + 1831, + 640, + 1867, + 613, + 1867 + ], + "score": 0.9, + "latex": "f _ { p } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 946, + 1354, + 1041, + 1354, + 1041, + 1383, + 946, + 1383 + ], + "score": 0.9, + "latex": "M \\times N" + }, + { + "category_id": 14, + "poly": [ + 466, + 1869, + 1232, + 1869, + 1232, + 1919, + 466, + 1919 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { x ^ { l } = f _ { p } ^ { l } ( \\hat { x } ) \\in \\mathcal { R } ^ { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d } , \\quad \\hat { x } = \\mathrm { R e s h a p e } ( x ) \\in \\mathcal { R } ^ { ( \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } \\times d ) \\times ( s _ { w } ^ { l } \\times s _ { w } ^ { l } ) } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 541, + 1519, + 574, + 1519, + 574, + 1554, + 541, + 1554 + ], + "score": 0.9, + "latex": "s _ { w } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 1736, + 1246, + 1736, + 1246, + 1764, + 1152, + 1764 + ], + "score": 0.9, + "latex": "M \\times N" + }, + { + "category_id": 13, + "poly": [ + 523, + 1581, + 551, + 1581, + 551, + 1616, + 523, + 1616 + ], + "score": 0.89, + "latex": "s _ { r } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 583, + 1941, + 652, + 1941, + 652, + 1976, + 583, + 1976 + ], + "score": 0.89, + "latex": "\\{ x ^ { l } \\} _ { 1 } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 1202, + 1770, + 1223, + 1770, + 1223, + 1792, + 1202, + 1792 + ], + "score": 0.85, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 734, + 235, + 752, + 235, + 752, + 261, + 734, + 261 + ], + "score": 0.82, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 469, + 1768, + 487, + 1768, + 487, + 1793, + 469, + 1793 + ], + "score": 0.8, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 860, + 1947, + 870, + 1947, + 870, + 1970, + 860, + 1970 + ], + "score": 0.77, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 762, + 1803, + 781, + 1803, + 781, + 1824, + 762, + 1824 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 452, + 1492, + 474, + 1492, + 474, + 1518, + 452, + 1518 + ], + "score": 0.73, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 717, + 1616, + 729, + 1616, + 729, + 1642, + 717, + 1642 + ], + "score": 0.73, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 357, + 1798, + 368, + 1798, + 368, + 1824, + 357, + 1824 + ], + "score": 0.63, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 1128, + 838, + 1144, + 838, + 1144, + 860, + 1128, + 860 + ], + "score": 0.47, + "latex": "\\mathbf { \\bar { x } }" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 769.0, + 1408.0, + 769.0, + 1408.0, + 806.0, + 956.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 803.0, + 1405.0, + 803.0, + 1405.0, + 833.0, + 954.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 832.0, + 1127.0, + 832.0, + 1127.0, + 865.0, + 955.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 832.0, + 1408.0, + 832.0, + 1408.0, + 865.0, + 1145.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 861.0, + 1408.0, + 861.0, + 1408.0, + 894.0, + 956.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 892.0, + 1409.0, + 892.0, + 1409.0, + 926.0, + 954.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 923.0, + 1405.0, + 923.0, + 1405.0, + 955.0, + 954.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 953.0, + 1409.0, + 953.0, + 1409.0, + 986.0, + 954.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 984.0, + 1209.0, + 984.0, + 1209.0, + 1013.0, + 954.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 565.0, + 1367.0, + 565.0, + 1367.0, + 579.0, + 1354.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 572.0, + 992.0, + 572.0, + 992.0, + 588.0, + 971.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 574.0, + 1297.0, + 574.0, + 1297.0, + 589.0, + 1284.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 578.0, + 1001.0, + 578.0, + 1001.0, + 611.0, + 987.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 580.0, + 974.0, + 580.0, + 974.0, + 711.0, + 951.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 585.0, + 1228.0, + 585.0, + 1228.0, + 602.0, + 1213.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 603.0, + 1158.0, + 603.0, + 1158.0, + 617.0, + 1145.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 619.0, + 1000.0, + 619.0, + 1000.0, + 656.0, + 971.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 635.0, + 1107.0, + 635.0, + 1107.0, + 650.0, + 1089.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 639.0, + 1174.0, + 639.0, + 1174.0, + 656.0, + 1154.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 633.0, + 1229.0, + 633.0, + 1229.0, + 650.0, + 1212.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 628.0, + 1295.0, + 628.0, + 1295.0, + 644.0, + 1278.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 625.0, + 1367.0, + 625.0, + 1367.0, + 639.0, + 1354.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 656.0, + 997.0, + 656.0, + 997.0, + 662.0, + 991.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 653.0, + 1088.0, + 653.0, + 1088.0, + 672.0, + 1060.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 650.0, + 1125.0, + 650.0, + 1125.0, + 662.0, + 1110.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 667.0, + 1002.0, + 667.0, + 1002.0, + 703.0, + 969.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 670.0, + 1055.0, + 670.0, + 1055.0, + 683.0, + 1041.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 677.0, + 1151.0, + 677.0, + 1151.0, + 689.0, + 1129.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 673.0, + 1295.0, + 673.0, + 1295.0, + 693.0, + 1155.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 701.0, + 999.0, + 701.0, + 999.0, + 710.0, + 989.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 692.0, + 1151.0, + 692.0, + 1151.0, + 707.0, + 1130.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 691.0, + 1280.0, + 691.0, + 1280.0, + 707.0, + 1156.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 723.0, + 1149.0, + 723.0, + 1149.0, + 746.0, + 1129.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 724.0, + 1187.0, + 724.0, + 1187.0, + 732.0, + 1177.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 724.0, + 1273.0, + 724.0, + 1273.0, + 732.0, + 1265.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 730.0, + 1015.0, + 730.0, + 1015.0, + 744.0, + 1003.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 729.0, + 1061.0, + 729.0, + 1061.0, + 745.0, + 1043.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 729.0, + 1105.0, + 729.0, + 1105.0, + 745.0, + 1086.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 729.0, + 1192.0, + 729.0, + 1192.0, + 745.0, + 1173.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 729.0, + 1235.0, + 729.0, + 1235.0, + 745.0, + 1216.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 729.0, + 1279.0, + 729.0, + 1279.0, + 745.0, + 1259.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 729.0, + 1322.0, + 729.0, + 1322.0, + 745.0, + 1303.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 729.0, + 1365.0, + 729.0, + 1365.0, + 745.0, + 1346.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 741.0, + 1241.0, + 741.0, + 1241.0, + 760.0, + 1131.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 710.0, + 1001.0, + 710.0, + 1001.0, + 732.0, + 970.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 714.5, + 1025.0, + 714.5, + 1025.0, + 725.0, + 1001.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 500.0, + 674.0, + 500.0, + 674.0, + 537.0, + 294.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1299.0, + 722.0, + 1299.0, + 722.0, + 1338.0, + 294.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1058.0, + 1405.0, + 1058.0, + 1405.0, + 1092.0, + 295.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1090.0, + 1405.0, + 1090.0, + 1405.0, + 1123.0, + 294.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1153.0, + 294.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1149.0, + 1407.0, + 1149.0, + 1407.0, + 1184.0, + 292.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1178.0, + 1405.0, + 1178.0, + 1405.0, + 1215.0, + 292.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1210.0, + 1405.0, + 1210.0, + 1405.0, + 1244.0, + 295.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1237.0, + 1336.0, + 1237.0, + 1336.0, + 1276.0, + 294.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 554.0, + 941.0, + 554.0, + 941.0, + 595.0, + 293.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 586.0, + 940.0, + 586.0, + 940.0, + 625.0, + 294.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 620.0, + 937.0, + 620.0, + 937.0, + 651.0, + 296.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 647.0, + 938.0, + 647.0, + 938.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 680.0, + 938.0, + 680.0, + 938.0, + 714.0, + 294.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 709.0, + 938.0, + 709.0, + 938.0, + 743.0, + 294.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 740.0, + 939.0, + 740.0, + 939.0, + 774.0, + 294.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 772.0, + 938.0, + 772.0, + 938.0, + 802.0, + 295.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 801.0, + 937.0, + 801.0, + 937.0, + 833.0, + 296.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 832.0, + 937.0, + 832.0, + 937.0, + 861.0, + 295.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 862.0, + 940.0, + 862.0, + 940.0, + 894.0, + 296.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 892.0, + 939.0, + 892.0, + 939.0, + 926.0, + 294.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 921.0, + 937.0, + 921.0, + 937.0, + 955.0, + 294.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 953.0, + 938.0, + 953.0, + 938.0, + 984.0, + 296.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 982.0, + 941.0, + 982.0, + 941.0, + 1017.0, + 295.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1012.0, + 876.0, + 1012.0, + 876.0, + 1046.0, + 294.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 279.0, + 1404.0, + 279.0, + 1404.0, + 312.0, + 297.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 308.0, + 1402.0, + 308.0, + 1402.0, + 340.0, + 296.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 321.0, + 1083.0, + 321.0, + 1083.0, + 384.0, + 286.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 321.0, + 1411.0, + 321.0, + 1411.0, + 384.0, + 1225.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 367.0, + 505.0, + 367.0, + 505.0, + 415.0, + 291.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 367.0, + 1406.0, + 367.0, + 1406.0, + 415.0, + 691.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 404.0, + 346.0, + 404.0, + 346.0, + 440.0, + 294.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 404.0, + 1405.0, + 404.0, + 1405.0, + 440.0, + 484.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 437.0, + 1100.0, + 437.0, + 1100.0, + 470.0, + 296.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1346.0, + 572.0, + 1346.0, + 572.0, + 1390.0, + 292.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1346.0, + 945.0, + 1346.0, + 945.0, + 1390.0, + 747.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1346.0, + 1408.0, + 1346.0, + 1408.0, + 1390.0, + 1042.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1383.0, + 519.0, + 1383.0, + 519.0, + 1420.0, + 292.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1383.0, + 1406.0, + 1383.0, + 1406.0, + 1420.0, + 610.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1414.0, + 1406.0, + 1414.0, + 1406.0, + 1450.0, + 293.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1443.0, + 559.0, + 1443.0, + 559.0, + 1477.0, + 295.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1728.0, + 885.0, + 1728.0, + 885.0, + 1770.0, + 291.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1728.0, + 1151.0, + 1728.0, + 1151.0, + 1770.0, + 1061.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 1728.0, + 1408.0, + 1728.0, + 1408.0, + 1770.0, + 1247.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1764.0, + 468.0, + 1764.0, + 468.0, + 1801.0, + 295.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1764.0, + 1201.0, + 1764.0, + 1201.0, + 1801.0, + 488.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1764.0, + 1406.0, + 1764.0, + 1406.0, + 1801.0, + 1224.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1793.0, + 356.0, + 1793.0, + 356.0, + 1837.0, + 291.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1793.0, + 761.0, + 1793.0, + 761.0, + 1837.0, + 369.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1793.0, + 1191.0, + 1793.0, + 1191.0, + 1837.0, + 782.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 1793.0, + 1407.0, + 1793.0, + 1407.0, + 1837.0, + 1290.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1829.0, + 612.0, + 1829.0, + 612.0, + 1865.0, + 295.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1829.0, + 1058.0, + 1829.0, + 1058.0, + 1865.0, + 641.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1486.0, + 451.0, + 1486.0, + 451.0, + 1525.0, + 292.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1486.0, + 1358.0, + 1486.0, + 1358.0, + 1525.0, + 475.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1517.0, + 540.0, + 1517.0, + 540.0, + 1555.0, + 295.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1517.0, + 1406.0, + 1517.0, + 1406.0, + 1555.0, + 575.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1547.0, + 800.0, + 1547.0, + 800.0, + 1588.0, + 313.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1547.0, + 964.0, + 1547.0, + 964.0, + 1588.0, + 954.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1580.0, + 522.0, + 1580.0, + 522.0, + 1621.0, + 296.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1580.0, + 1408.0, + 1580.0, + 1408.0, + 1621.0, + 552.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1614.0, + 716.0, + 1614.0, + 716.0, + 1648.0, + 317.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1614.0, + 739.0, + 1614.0, + 739.0, + 1648.0, + 730.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 203.0, + 1404.0, + 203.0, + 1404.0, + 239.0, + 296.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 231.0, + 733.0, + 231.0, + 733.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 231.0, + 1256.0, + 231.0, + 1256.0, + 270.0, + 753.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1936.0, + 582.0, + 1936.0, + 582.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1936.0, + 859.0, + 1936.0, + 859.0, + 1981.0, + 653.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1936.0, + 1406.0, + 1936.0, + 1406.0, + 1981.0, + 871.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 590.0, + 1972.0, + 590.0, + 2013.0, + 294.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2013.0, + 675.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1656.0, + 1404.0, + 1656.0, + 1404.0, + 1696.0, + 293.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1689.0, + 671.0, + 1689.0, + 671.0, + 1729.0, + 293.0, + 1729.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1325, + 1406, + 1325, + 1406, + 1646, + 296, + 1646 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1749, + 1406, + 1749, + 1406, + 2009, + 297, + 2009 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1145, + 1403, + 1145, + 1403, + 1239, + 298, + 1239 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 296, + 668, + 1406, + 668, + 1406, + 1005, + 296, + 1005 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 332, + 205, + 1377, + 205, + 1377, + 645, + 332, + 645 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1039, + 1404, + 1039, + 1404, + 1133, + 298, + 1133 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 588, + 1664, + 1111, + 1664, + 1111, + 1733, + 588, + 1733 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 409, + 1254, + 1285, + 1254, + 1285, + 1298, + 409, + 1298 + ], + "score": 0.923 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1684, + 1400, + 1684, + 1400, + 1712, + 1369, + 1712 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1264, + 1400, + 1264, + 1400, + 1291, + 1369, + 1291 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.726 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.09 + }, + { + "category_id": 13, + "poly": [ + 373, + 1750, + 509, + 1750, + 509, + 1786, + 373, + 1786 + ], + "score": 0.94, + "latex": "B = \\{ B ^ { l } \\} _ { 1 } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 1144, + 1089, + 1144, + 1089, + 1180, + 1019, + 1180 + ], + "score": 0.93, + "latex": "\\{ x ^ { l } \\} _ { 1 } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 947, + 1487, + 1130, + 1487, + 1130, + 1528, + 947, + 1528 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { s = \\sum _ { l = 1 } ^ { L } ( s _ { r } ^ { l } ) ^ { 2 } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 586, + 1662, + 1110, + 1662, + 1110, + 1733, + 586, + 1733 + ], + "score": 0.92, + "latex": "\\mathrm { A t t e n t i o n } ( Q _ { i } , K _ { i } , V _ { i } ) = \\mathrm { S o f t m a x } ( \\frac { Q _ { i } K _ { i } ^ { T } } { \\sqrt { d } } + B ) V _ { i } ," + }, + { + "category_id": 13, + "poly": [ + 676, + 1879, + 830, + 1879, + 830, + 1912, + 676, + 1912 + ], + "score": 0.92, + "latex": "B ^ { l } \\in \\mathcal { R } ^ { s _ { r } ^ { l } \\times s _ { r } ^ { l } }" + }, + { + "category_id": 13, + "poly": [ + 1085, + 1454, + 1397, + 1454, + 1397, + 1489, + 1085, + 1489 + ], + "score": 0.92, + "latex": "V _ { i } = \\{ V _ { i } ^ { 1 } , . . . , \\mathbf { \\bar { V } } _ { i } ^ { L } \\} \\in \\mathcal { R } ^ { s \\times d }" + }, + { + "category_id": 13, + "poly": [ + 604, + 1913, + 689, + 1913, + 689, + 1948, + 604, + 1948 + ], + "score": 0.92, + "latex": "s _ { r } ^ { l } \\times s _ { r } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1389, + 1341, + 1389, + 1341, + 1425, + 1255, + 1425 + ], + "score": 0.92, + "latex": "s _ { r } ^ { l } \\times s _ { r } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 931, + 1783, + 1203, + 1783, + 1203, + 1816, + 931, + 1816 + ], + "score": 0.91, + "latex": "B ^ { 1 } \\in \\mathcal { R } ^ { ( 2 s _ { p } - 1 ) \\times ( 2 s _ { p } - 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 709, + 1453, + 1035, + 1453, + 1035, + 1489, + 709, + 1489 + ], + "score": 0.91, + "latex": "\\breve { K } _ { i } = \\{ K _ { i } ^ { 1 } , . . . , K _ { i } ^ { L } \\} \\in \\mathscr { R } ^ { \\bar { s } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 897, + 1390, + 1079, + 1390, + 1079, + 1424, + 897, + 1424 + ], + "score": 0.91, + "latex": "Q _ { i } \\in \\mathcal { R } ^ { s _ { p } \\times s _ { p } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 942, + 732, + 1037, + 732, + 1037, + 761, + 942, + 761 + ], + "score": 0.91, + "latex": "2 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 1187, + 944, + 1320, + 944, + 1320, + 973, + 1187, + 973 + ], + "score": 0.91, + "latex": "4 \\times 4 = 1 6" + }, + { + "category_id": 13, + "poly": [ + 1071, + 1360, + 1158, + 1360, + 1158, + 1393, + 1071, + 1393 + ], + "score": 0.91, + "latex": "s _ { p } \\times s _ { p }" + }, + { + "category_id": 13, + "poly": [ + 905, + 1819, + 1102, + 1819, + 1102, + 1852, + 905, + 1852 + ], + "score": 0.91, + "latex": "[ - s _ { p } + 1 , s _ { p } - 1 ]" + }, + { + "category_id": 13, + "poly": [ + 405, + 884, + 471, + 884, + 471, + 912, + 405, + 912 + ], + "score": 0.9, + "latex": "6 \\times 6" + }, + { + "category_id": 13, + "poly": [ + 797, + 853, + 863, + 853, + 863, + 882, + 797, + 882 + ], + "score": 0.9, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 672, + 763, + 739, + 763, + 739, + 790, + 672, + 790 + ], + "score": 0.9, + "latex": "4 \\times 4" + }, + { + "category_id": 14, + "poly": [ + 411, + 1254, + 1289, + 1254, + 1289, + 1296, + 411, + 1296 + ], + "score": 0.9, + "latex": "Q = f _ { q } ( x ^ { 1 } ) , \\quad K = \\{ K ^ { l } \\} _ { 1 } ^ { L } = f _ { k } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) , \\quad V = \\{ V ^ { l } \\} _ { 1 } ^ { L } = f _ { v } ( \\{ x ^ { 1 } , . . . , x ^ { L } \\} ) ." + }, + { + "category_id": 13, + "poly": [ + 945, + 913, + 1010, + 913, + 1010, + 942, + 945, + 942 + ], + "score": 0.9, + "latex": "5 \\times 5" + }, + { + "category_id": 13, + "poly": [ + 487, + 763, + 554, + 763, + 554, + 791, + 487, + 791 + ], + "score": 0.89, + "latex": "4 \\times 4" + }, + { + "category_id": 13, + "poly": [ + 543, + 914, + 609, + 914, + 609, + 942, + 543, + 942 + ], + "score": 0.89, + "latex": "4 \\times 4" + }, + { + "category_id": 13, + "poly": [ + 297, + 822, + 363, + 822, + 363, + 851, + 297, + 851 + ], + "score": 0.89, + "latex": "8 \\times 8" + }, + { + "category_id": 13, + "poly": [ + 646, + 1615, + 679, + 1615, + 679, + 1646, + 646, + 1646 + ], + "score": 0.89, + "latex": "Q _ { i }" + }, + { + "category_id": 13, + "poly": [ + 345, + 1209, + 373, + 1209, + 373, + 1240, + 345, + 1240 + ], + "score": 0.89, + "latex": "f _ { v }" + }, + { + "category_id": 13, + "poly": [ + 478, + 1423, + 515, + 1423, + 515, + 1453, + 478, + 1453 + ], + "score": 0.89, + "latex": "K ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1334, + 732, + 1402, + 732, + 1402, + 761, + 1334, + 761 + ], + "score": 0.89, + "latex": "5 \\times 5" + }, + { + "category_id": 13, + "poly": [ + 565, + 1423, + 599, + 1423, + 599, + 1453, + 565, + 1453 + ], + "score": 0.88, + "latex": "V ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1242, + 1754, + 1264, + 1754, + 1264, + 1780, + 1242, + 1780 + ], + "score": 0.82, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1754, + 1111, + 1754, + 1111, + 1780, + 1089, + 1780 + ], + "score": 0.82, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1148, + 1178, + 1148, + 1178, + 1175, + 1156, + 1175 + ], + "score": 0.8, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 512, + 1457, + 534, + 1457, + 534, + 1483, + 512, + 1483 + ], + "score": 0.8, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 374, + 1499, + 391, + 1499, + 391, + 1520, + 374, + 1520 + ], + "score": 0.73, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 752, + 1395, + 765, + 1395, + 765, + 1420, + 752, + 1420 + ], + "score": 0.73, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1331, + 1177, + 1400, + 1177, + 1400, + 1211, + 1331, + 1211 + ], + "score": 0.57, + "latex": "f _ { q } , f _ { k }" + }, + { + "category_id": 13, + "poly": [ + 459, + 481, + 495, + 481, + 495, + 498, + 459, + 498 + ], + "score": 0.41, + "latex": "s _ { p } = 4" + }, + { + "category_id": 13, + "poly": [ + 1331, + 1178, + 1359, + 1178, + 1359, + 1211, + 1331, + 1211 + ], + "score": 0.39, + "latex": "f _ { q }" + }, + { + "category_id": 13, + "poly": [ + 433, + 278, + 513, + 278, + 513, + 290, + 433, + 290 + ], + "score": 0.39, + "latex": "M = 2 0 ; N = 2 0" + }, + { + "category_id": 13, + "poly": [ + 1371, + 1178, + 1400, + 1178, + 1400, + 1209, + 1371, + 1209 + ], + "score": 0.37, + "latex": "f _ { k }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 669.0, + 1407.0, + 669.0, + 1407.0, + 705.0, + 294.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 701.0, + 1406.0, + 701.0, + 1406.0, + 736.0, + 294.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 731.0, + 941.0, + 731.0, + 941.0, + 766.0, + 293.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 731.0, + 1333.0, + 731.0, + 1333.0, + 766.0, + 1038.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 731.0, + 1407.0, + 731.0, + 1407.0, + 766.0, + 1403.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 763.0, + 486.0, + 763.0, + 486.0, + 794.0, + 295.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 763.0, + 671.0, + 763.0, + 671.0, + 794.0, + 555.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 763.0, + 1404.0, + 763.0, + 1404.0, + 794.0, + 740.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 792.0, + 1403.0, + 792.0, + 1403.0, + 823.0, + 295.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 822.0, + 1406.0, + 822.0, + 1406.0, + 856.0, + 364.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 851.0, + 796.0, + 851.0, + 796.0, + 885.0, + 293.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 851.0, + 1406.0, + 851.0, + 1406.0, + 885.0, + 864.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 882.0, + 404.0, + 882.0, + 404.0, + 916.0, + 293.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 882.0, + 1404.0, + 882.0, + 1404.0, + 916.0, + 472.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 910.0, + 542.0, + 910.0, + 542.0, + 947.0, + 293.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 910.0, + 944.0, + 910.0, + 944.0, + 947.0, + 610.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 910.0, + 1404.0, + 910.0, + 1404.0, + 947.0, + 1011.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 943.0, + 1186.0, + 943.0, + 1186.0, + 978.0, + 295.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 943.0, + 1406.0, + 943.0, + 1406.0, + 978.0, + 1321.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 976.0, + 625.0, + 976.0, + 625.0, + 1005.0, + 295.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 204.0, + 771.0, + 204.0, + 771.0, + 227.0, + 688.0, + 227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 204.0, + 1006.0, + 204.0, + 1006.0, + 225.0, + 969.0, + 225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 226.0, + 747.0, + 226.0, + 747.0, + 244.0, + 690.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 215.0, + 1110.0, + 215.0, + 1110.0, + 309.0, + 1091.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 249.0, + 703.0, + 249.0, + 703.0, + 263.0, + 691.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 236.0, + 917.0, + 236.0, + 917.0, + 261.0, + 808.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 241.0, + 1082.0, + 241.0, + 1082.0, + 265.0, + 1038.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 274.0, + 432.0, + 274.0, + 432.0, + 294.0, + 340.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 274.0, + 517.0, + 274.0, + 517.0, + 294.0, + 514.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 284.0, + 708.0, + 284.0, + 708.0, + 298.0, + 691.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 284.0, + 826.0, + 284.0, + 826.0, + 304.0, + 790.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 286.0, + 933.0, + 286.0, + 933.0, + 303.0, + 900.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 285.0, + 1273.0, + 285.0, + 1273.0, + 306.0, + 1240.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 311.0, + 770.0, + 311.0, + 770.0, + 330.0, + 693.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 308.0, + 1125.0, + 308.0, + 1125.0, + 327.0, + 1072.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 302.0, + 1264.0, + 302.0, + 1264.0, + 402.0, + 1248.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 325.0, + 1005.0, + 325.0, + 1005.0, + 345.0, + 968.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 365.0, + 448.0, + 365.0, + 448.0, + 391.0, + 406.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 368.0, + 597.0, + 368.0, + 597.0, + 385.0, + 558.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 365.0, + 925.0, + 365.0, + 925.0, + 384.0, + 820.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 367.0, + 1080.0, + 367.0, + 1080.0, + 388.0, + 1037.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 349.0, + 1107.0, + 349.0, + 1107.0, + 385.0, + 1092.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 360.0, + 1182.0, + 360.0, + 1182.0, + 391.0, + 1169.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 338.0, + 1245.0, + 338.0, + 1245.0, + 371.0, + 1188.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 338.0, + 1339.0, + 338.0, + 1339.0, + 465.0, + 1320.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 382.0, + 1107.0, + 382.0, + 1107.0, + 431.0, + 1092.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 400.0, + 1185.0, + 400.0, + 1185.0, + 461.0, + 1166.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 401.0, + 1295.0, + 401.0, + 1295.0, + 505.0, + 1276.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 413.0, + 1237.0, + 413.0, + 1237.0, + 430.0, + 1200.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 419.0, + 845.0, + 419.0, + 845.0, + 436.0, + 812.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 419.0, + 933.0, + 419.0, + 933.0, + 436.0, + 901.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 425.0, + 1247.0, + 425.0, + 1247.0, + 444.0, + 1190.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 427.0, + 1126.0, + 427.0, + 1126.0, + 447.0, + 1072.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 444.0, + 770.0, + 444.0, + 770.0, + 461.0, + 693.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 456.0, + 1208.0, + 456.0, + 1208.0, + 478.0, + 1147.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 480.0, + 458.0, + 480.0, + 458.0, + 500.0, + 349.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 480.0, + 499.0, + 480.0, + 499.0, + 500.0, + 496.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 507.0, + 1005.0, + 507.0, + 1005.0, + 527.0, + 968.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 525.0, + 941.0, + 525.0, + 941.0, + 552.0, + 836.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 528.0, + 1079.0, + 528.0, + 1079.0, + 552.0, + 1036.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 538.0, + 748.0, + 538.0, + 748.0, + 565.0, + 709.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 558.0, + 1104.0, + 558.0, + 1104.0, + 573.0, + 1094.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 543.0, + 1233.0, + 543.0, + 1233.0, + 594.0, + 1132.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 542.0, + 1307.0, + 542.0, + 1307.0, + 576.0, + 1234.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 585.0, + 871.0, + 585.0, + 871.0, + 602.0, + 838.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 588.0, + 938.0, + 588.0, + 938.0, + 603.0, + 907.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 571.0, + 1125.0, + 571.0, + 1125.0, + 606.0, + 1070.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 579.0, + 1181.0, + 579.0, + 1181.0, + 615.0, + 1140.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 571.0, + 1234.0, + 571.0, + 1234.0, + 603.0, + 1218.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 571.0, + 1308.0, + 571.0, + 1308.0, + 605.0, + 1292.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 603.0, + 1253.0, + 603.0, + 1253.0, + 623.0, + 1197.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 602.0, + 1320.0, + 602.0, + 1320.0, + 624.0, + 1279.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 271.5, + 603.0, + 271.5, + 603.0, + 293.0, + 555.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 386.0, + 1295.0, + 386.0, + 1295.0, + 405.0, + 1273.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 476.0, + 599.0, + 476.0, + 599.0, + 495.5, + 560.0, + 495.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1328.0, + 1404.0, + 1328.0, + 1404.0, + 1361.0, + 296.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1357.0, + 1070.0, + 1357.0, + 1070.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 1357.0, + 1406.0, + 1357.0, + 1406.0, + 1393.0, + 1159.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1383.0, + 751.0, + 1383.0, + 751.0, + 1432.0, + 290.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1383.0, + 896.0, + 1383.0, + 896.0, + 1432.0, + 766.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1383.0, + 1254.0, + 1383.0, + 1254.0, + 1432.0, + 1080.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1383.0, + 1408.0, + 1383.0, + 1408.0, + 1432.0, + 1342.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1421.0, + 477.0, + 1421.0, + 477.0, + 1461.0, + 294.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1421.0, + 564.0, + 1421.0, + 564.0, + 1461.0, + 516.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1421.0, + 1406.0, + 1421.0, + 1406.0, + 1461.0, + 600.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1449.0, + 511.0, + 1449.0, + 511.0, + 1495.0, + 287.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1449.0, + 708.0, + 1449.0, + 708.0, + 1495.0, + 535.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1449.0, + 1084.0, + 1449.0, + 1084.0, + 1495.0, + 1036.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1449.0, + 1410.0, + 1449.0, + 1410.0, + 1495.0, + 1398.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1471.0, + 373.0, + 1471.0, + 373.0, + 1557.0, + 283.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1471.0, + 946.0, + 1471.0, + 946.0, + 1557.0, + 392.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1471.0, + 1417.0, + 1471.0, + 1417.0, + 1557.0, + 1131.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1550.0, + 1406.0, + 1550.0, + 1406.0, + 1589.0, + 294.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1584.0, + 1404.0, + 1584.0, + 1404.0, + 1617.0, + 295.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1613.0, + 645.0, + 1613.0, + 645.0, + 1650.0, + 295.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1613.0, + 717.0, + 1613.0, + 717.0, + 1650.0, + 680.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1749.0, + 372.0, + 1749.0, + 372.0, + 1788.0, + 294.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1749.0, + 1088.0, + 1749.0, + 1088.0, + 1788.0, + 510.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1749.0, + 1241.0, + 1749.0, + 1241.0, + 1788.0, + 1112.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1749.0, + 1409.0, + 1749.0, + 1409.0, + 1788.0, + 1265.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1778.0, + 930.0, + 1778.0, + 930.0, + 1825.0, + 292.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1778.0, + 1411.0, + 1778.0, + 1411.0, + 1825.0, + 1204.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1818.0, + 904.0, + 1818.0, + 904.0, + 1853.0, + 295.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1818.0, + 1407.0, + 1818.0, + 1407.0, + 1853.0, + 1103.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1848.0, + 1406.0, + 1848.0, + 1406.0, + 1883.0, + 295.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1878.0, + 675.0, + 1878.0, + 675.0, + 1920.0, + 292.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1878.0, + 1407.0, + 1878.0, + 1407.0, + 1920.0, + 831.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1913.0, + 603.0, + 1913.0, + 603.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1913.0, + 1406.0, + 1913.0, + 1406.0, + 1949.0, + 690.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1945.0, + 1404.0, + 1945.0, + 1404.0, + 1980.0, + 295.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1975.0, + 1392.0, + 1975.0, + 1392.0, + 2010.0, + 295.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1140.0, + 1018.0, + 1140.0, + 1018.0, + 1185.0, + 292.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1140.0, + 1155.0, + 1140.0, + 1155.0, + 1185.0, + 1090.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1140.0, + 1407.0, + 1140.0, + 1407.0, + 1185.0, + 1179.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1173.0, + 1330.0, + 1173.0, + 1330.0, + 1214.0, + 293.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1204.0, + 344.0, + 1204.0, + 344.0, + 1245.0, + 293.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1204.0, + 561.0, + 1204.0, + 561.0, + 1245.0, + 374.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1037.0, + 1405.0, + 1037.0, + 1405.0, + 1076.0, + 292.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1068.0, + 1405.0, + 1068.0, + 1405.0, + 1108.0, + 292.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1101.0, + 679.0, + 1101.0, + 679.0, + 1135.0, + 296.0, + 1135.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 333, + 196, + 1370, + 196, + 1370, + 688, + 333, + 688 + ], + "score": 0.986, + "html": "
Output SizeLayer NameFocal-TinyFocal-SmallFocal-Base
stage 156×56Patch Embeddingp1= 4;c1 = 96p1=4;c1= 96p1 = 4;c1 = 128
56×56Transformer Block{1,13} 三 s={7,7×2二 {1,13} ={7,7}×2{1,13} ={7,7}×2
stage 228×28Patch EmbeddingP2=2;c=192P2=2;C=192P2=2;C= 256
28×28Transformer Block{1,13} 三 swr={7,5} 1×2={1,13} sw,r={7,5} 1×2{1,13} 二 su,r={7,5}×2
stage 314 × 14Patch Embeddingp3=2;c3=384p3=2; c3= 384p3=2; c3= 512
14 × 14Transformer Block={1,13} s={7,3}×6={1,13} ={7,3}×18={1,13} ={7,3}×18
stage 47×7Patch EmbeddingP4=2;C4=768P4=2;C4=768P4= 2;C4=1024
7×7Transformer Block二 {1,7}×2{1,7} s 三 ={7,1}×2二 {1,7} su,r {7,1} 二×2
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1194, + 1405, + 1194, + 1405, + 1588, + 297, + 1588 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 826, + 1405, + 826, + 1405, + 1112, + 297, + 1112 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1733, + 1405, + 1733, + 1405, + 2007, + 298, + 2007 + ], + "score": 0.981 + }, + { + "category_id": 7, + "poly": [ + 299, + 691, + 1399, + 691, + 1399, + 753, + 299, + 753 + ], + "score": 0.936 + }, + { + "category_id": 0, + "poly": [ + 299, + 1152, + 616, + 1152, + 616, + 1184, + 299, + 1184 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 299, + 789, + 623, + 789, + 623, + 821, + 299, + 821 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 299, + 1691, + 796, + 1691, + 796, + 1723, + 299, + 1723 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 298, + 1635, + 533, + 1635, + 533, + 1673, + 298, + 1673 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 859, + 2062, + 859, + 2085, + 840, + 2085 + ], + "score": 0.781 + }, + { + "category_id": 13, + "poly": [ + 889, + 987, + 1111, + 987, + 1111, + 1024, + 889, + 1024 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( ( s _ { p } ) ^ { 2 } \\textstyle \\sum _ { l } ( s _ { r } ^ { l } ) ^ { 2 } d )" + }, + { + "category_id": 13, + "poly": [ + 761, + 855, + 849, + 855, + 849, + 899, + 761, + 899 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { M } { s _ { w } ^ { l } } \\times \\frac { N } { s _ { w } ^ { l } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1055, + 929, + 1209, + 929, + 1209, + 963, + 1055, + 963 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( L ( M N ) d )" + }, + { + "category_id": 13, + "poly": [ + 491, + 855, + 657, + 855, + 657, + 887, + 491, + 887 + ], + "score": 0.93, + "latex": "\\dot { \\boldsymbol { x } } \\in \\mathcal { R } ^ { M \\times N \\times d }" + }, + { + "category_id": 13, + "poly": [ + 771, + 992, + 859, + 992, + 859, + 1024, + 771, + 1024 + ], + "score": 0.93, + "latex": "s _ { p } \\times s _ { p }" + }, + { + "category_id": 13, + "poly": [ + 865, + 895, + 989, + 895, + 989, + 932, + 865, + 932 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( ( s _ { w } ^ { l } ) ^ { 2 } d )" + }, + { + "category_id": 13, + "poly": [ + 1238, + 1466, + 1359, + 1466, + 1359, + 1500, + 1238, + 1500 + ], + "score": 0.92, + "latex": "\\{ 7 , 5 , 3 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 1170, + 988, + 1404, + 988, + 1404, + 1024, + 1170, + 1024 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\dot { \\sum } _ { l } ( s _ { r } ^ { l } ) ^ { 2 } ( M \\dot { N } ) d )" + }, + { + "category_id": 13, + "poly": [ + 330, + 929, + 466, + 929, + 466, + 963, + 330, + 963 + ], + "score": 0.91, + "latex": "O ( ( M N ) d )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1049, + 607, + 1049, + 607, + 1085, + 298, + 1085 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\mathcal { O } ( ( L + \\sum _ { l } ( s _ { r } ^ { l } ) ^ { \\bar { 2 } } ) ( M N ) d ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 973, + 1047, + 1272, + 1047, + 1272, + 1084, + 973, + 1084 + ], + "score": 0.91, + "latex": "s _ { r } ^ { \\hat { L } } = 2 \\times \\operatorname* { m a x } ( M , N ) / s _ { w } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 915, + 1255, + 1036, + 1255, + 1036, + 1284, + 915, + 1284 + ], + "score": 0.91, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 525, + 1976, + 644, + 1976, + 644, + 2005, + 525, + 2005 + ], + "score": 0.9, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 503, + 1527, + 624, + 1527, + 624, + 1561, + 503, + 1561 + ], + "score": 0.9, + "latex": "\\{ 4 , 2 , 2 , 2 \\}" + }, + { + "category_id": 13, + "poly": [ + 536, + 1792, + 596, + 1792, + 596, + 1822, + 536, + 1822 + ], + "score": 0.9, + "latex": "1 0 ^ { - 3 }" + }, + { + "category_id": 13, + "poly": [ + 574, + 1407, + 639, + 1407, + 639, + 1436, + 574, + 1436 + ], + "score": 0.89, + "latex": "7 \\times 7" + }, + { + "category_id": 13, + "poly": [ + 1135, + 1792, + 1195, + 1792, + 1195, + 1822, + 1135, + 1822 + ], + "score": 0.89, + "latex": "1 0 ^ { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 1290, + 1504, + 1318, + 1504, + 1318, + 1530, + 1290, + 1530 + ], + "score": 0.85, + "latex": "p _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1170, + 360, + 1322, + 360, + 1322, + 379, + 1170, + 379 + ], + "score": 0.84, + "latex": "p _ { 2 } = 2 ; c _ { 2 } = 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 1170, + 248, + 1322, + 248, + 1322, + 267, + 1170, + 267 + ], + "score": 0.84, + "latex": "p _ { 1 } = 4 ; c _ { 1 } = 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1533, + 1248, + 1533, + 1248, + 1558, + 1224, + 1558 + ], + "score": 0.83, + "latex": "c _ { i }" + }, + { + "category_id": 13, + "poly": [ + 936, + 360, + 1088, + 360, + 1088, + 379, + 936, + 379 + ], + "score": 0.79, + "latex": "p _ { 2 } = 2 ; c _ { 2 } = 1 9 2" + }, + { + "category_id": 14, + "poly": [ + 340, + 617, + 1360, + 617, + 1360, + 680, + 340, + 680 + ], + "score": 0.72, + "latex": "\\begin{array} { r l r l r l r l } { \\mathrm { s t a g e ~ 4 } } & { } & { { } { { 7 } \\times 7 } } & { } & { { \\mathrm { T r a n s f o r m e r } } } & { } & { { } \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 7 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 1 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 7 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 1 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 7 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 1 \\} } \\end{array} \\right] \\times 2 } & { } & { { } \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 7 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 1 \\} } \\end{array} \\right] \\times 2 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 941, + 248, + 1082, + 248, + 1082, + 267, + 941, + 267 + ], + "score": 0.67, + "latex": "p _ { 1 } = 4 ; c _ { 1 } = 9 6" + }, + { + "category_id": 14, + "poly": [ + 446, + 506, + 1367, + 506, + 1367, + 566, + 446, + 566 + ], + "score": 0.59, + "latex": "\\begin{array} { r l } { \\times 1 4 \\quad } & { \\mathrm { T r a n s f o r m e r } \\quad \\left[ \\begin{array} { c } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 3 \\} } \\end{array} \\right] \\times 6 \\left[ \\begin{array} { c } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 3 \\} } \\end{array} \\right] \\times 1 8 \\left[ \\begin{array} { c } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 3 \\} } \\end{array} \\right] \\times 1 8 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1170, + 473, + 1322, + 473, + 1322, + 493, + 1170, + 493 + ], + "score": 0.56, + "latex": "p _ { 3 } = 2 ; c _ { 3 } = 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 936, + 473, + 1088, + 473, + 1088, + 493, + 936, + 493 + ], + "score": 0.52, + "latex": "p _ { 3 } = 2 ; c _ { 3 } = 3 8 4" + }, + { + "category_id": 13, + "poly": [ + 706, + 359, + 860, + 359, + 860, + 379, + 706, + 379 + ], + "score": 0.49, + "latex": "p _ { 2 } = 2 ; c _ { 2 } = 1 9 2" + }, + { + "category_id": 14, + "poly": [ + 360, + 576, + 1366, + 576, + 1366, + 604, + 360, + 604 + ], + "score": 0.42, + "latex": "{ \\begin{array} { r l r l r l r l r l } { { 7 } \\times 7 } & { } & { { \\mathrm { P a t c h ~ E m b e d d i n g } } } & & { } & { p _ { 4 } = 2 ; c _ { 4 } = 7 6 8 } & & { } & { p _ { 4 } = 2 ; c _ { 4 } = 7 6 8 } & & { } & { p _ { 4 } = 2 ; c _ { 4 } = 1 0 2 4 } \\end{array} }" + }, + { + "category_id": 14, + "poly": [ + 305, + 393, + 1373, + 393, + 1373, + 453, + 305, + 453 + ], + "score": 0.37, + "latex": "\\begin{array} { r l r l r l r l } { \\mathrm { s t a g e ~ 2 } } & { } & & { \\mathrm { T r a n s f o r m e r } } & { \\quad \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 5 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 5 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 5 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 5 \\} } \\end{array} \\right] \\times 2 } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 330, + 280, + 1379, + 280, + 1379, + 341, + 330, + 341 + ], + "score": 0.34, + "latex": "\\begin{array} { r l r l r l r l } { \\mathrm { s t a g e ~ l ~ } } & { } & & { \\mathrm { T r a n s f o r m e r } } & { } & & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 7 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 7 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 7 \\} } \\end{array} \\right] \\times 2 } & { \\left[ \\begin{array} { l } { s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} } \\\\ { s _ { w , r } ^ { 1 } = \\{ 7 , 7 \\} } \\end{array} \\right] \\times 2 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 707, + 472, + 860, + 472, + 860, + 493, + 707, + 493 + ], + "score": 0.33, + "latex": "p _ { 3 } = 2 ; c _ { 3 } = 3 8 4" + }, + { + "category_id": 13, + "poly": [ + 691, + 282, + 892, + 282, + 892, + 341, + 691, + 341 + ], + "score": 0.29, + "latex": "s _ { w , r } ^ { 0 } = \\{ 1 , 1 3 \\} ~ ] \\times 2" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 689.0, + 1404.0, + 689.0, + 1404.0, + 725.0, + 295.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 721.0, + 958.0, + 721.0, + 958.0, + 756.0, + 295.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1147.0, + 618.0, + 1147.0, + 618.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 785.0, + 626.0, + 785.0, + 626.0, + 828.0, + 294.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1689.0, + 800.0, + 1689.0, + 800.0, + 1729.0, + 294.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1630.0, + 537.0, + 1630.0, + 537.0, + 1681.0, + 289.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 862.0, + 2060.0, + 862.0, + 2091.0, + 840.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1192.0, + 1408.0, + 1192.0, + 1408.0, + 1229.0, + 294.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1224.0, + 1406.0, + 1224.0, + 1406.0, + 1258.0, + 292.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1253.0, + 914.0, + 1253.0, + 914.0, + 1288.0, + 295.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1253.0, + 1406.0, + 1253.0, + 1406.0, + 1288.0, + 1037.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1283.0, + 1403.0, + 1283.0, + 1403.0, + 1319.0, + 294.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1315.0, + 1406.0, + 1315.0, + 1406.0, + 1351.0, + 295.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1347.0, + 1403.0, + 1347.0, + 1403.0, + 1379.0, + 296.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1412.0, + 292.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1404.0, + 573.0, + 1404.0, + 573.0, + 1441.0, + 292.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1404.0, + 1407.0, + 1404.0, + 1407.0, + 1441.0, + 640.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 1407.0, + 1436.0, + 1407.0, + 1472.0, + 294.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1466.0, + 1237.0, + 1466.0, + 1237.0, + 1502.0, + 294.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 1466.0, + 1406.0, + 1466.0, + 1406.0, + 1502.0, + 1360.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1499.0, + 1289.0, + 1499.0, + 1289.0, + 1531.0, + 295.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1499.0, + 1403.0, + 1499.0, + 1403.0, + 1531.0, + 1319.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1527.0, + 502.0, + 1527.0, + 502.0, + 1563.0, + 295.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1527.0, + 1223.0, + 1527.0, + 1223.0, + 1563.0, + 625.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 1527.0, + 1405.0, + 1527.0, + 1405.0, + 1563.0, + 1249.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1558.0, + 621.0, + 1558.0, + 621.0, + 1590.0, + 295.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 825.0, + 1406.0, + 825.0, + 1406.0, + 859.0, + 292.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 848.0, + 490.0, + 848.0, + 490.0, + 893.0, + 290.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 848.0, + 760.0, + 848.0, + 760.0, + 893.0, + 658.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 848.0, + 1407.0, + 848.0, + 1407.0, + 893.0, + 850.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 895.0, + 864.0, + 895.0, + 864.0, + 936.0, + 294.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 895.0, + 1406.0, + 895.0, + 1406.0, + 936.0, + 990.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 928.0, + 329.0, + 928.0, + 329.0, + 965.0, + 295.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 928.0, + 1054.0, + 928.0, + 1054.0, + 965.0, + 467.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 928.0, + 1406.0, + 928.0, + 1406.0, + 965.0, + 1210.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 959.0, + 1408.0, + 959.0, + 1408.0, + 993.0, + 296.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 987.0, + 770.0, + 987.0, + 770.0, + 1028.0, + 292.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 987.0, + 888.0, + 987.0, + 888.0, + 1028.0, + 860.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 987.0, + 1169.0, + 987.0, + 1169.0, + 1028.0, + 1112.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 987.0, + 1410.0, + 987.0, + 1410.0, + 1028.0, + 1405.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1016.0, + 1407.0, + 1016.0, + 1407.0, + 1056.0, + 292.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1047.0, + 297.0, + 1047.0, + 297.0, + 1088.0, + 294.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 1047.0, + 972.0, + 1047.0, + 972.0, + 1088.0, + 608.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1047.0, + 1410.0, + 1047.0, + 1410.0, + 1088.0, + 1273.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1080.0, + 1322.0, + 1080.0, + 1322.0, + 1117.0, + 295.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.75, + 871.5, + 787.75, + 871.5, + 787.75, + 899.5, + 766.75, + 899.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 871.0, + 852.0, + 871.0, + 852.0, + 905.0, + 809.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1731.0, + 1405.0, + 1731.0, + 1405.0, + 1770.0, + 295.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1762.0, + 1406.0, + 1762.0, + 1406.0, + 1799.0, + 293.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1787.0, + 535.0, + 1787.0, + 535.0, + 1834.0, + 291.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1787.0, + 1134.0, + 1787.0, + 1134.0, + 1834.0, + 597.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1787.0, + 1411.0, + 1787.0, + 1411.0, + 1834.0, + 1196.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1819.0, + 1406.0, + 1819.0, + 1406.0, + 1862.0, + 291.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1884.0, + 1406.0, + 1884.0, + 1406.0, + 1921.0, + 293.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 1405.0, + 1914.0, + 1405.0, + 1951.0, + 295.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1940.0, + 1406.0, + 1940.0, + 1406.0, + 1987.0, + 291.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1975.0, + 524.0, + 1975.0, + 524.0, + 2012.0, + 296.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1975.0, + 1393.0, + 1975.0, + 1393.0, + 2012.0, + 645.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1415, + 1405, + 1415, + 1405, + 1870, + 298, + 1870 + ], + "score": 0.984 + }, + { + "category_id": 5, + "poly": [ + 833, + 196, + 1394, + 196, + 1394, + 706, + 833, + 706 + ], + "score": 0.98, + "html": "
BackboneRetinaNetMask R-CNN
APbApbAPm
ResNet-50 [33]36.338.034.4
PVT-Small40.440.437.8
ViL-Small [76]41.641.838.5
Swin-Tiny [43]42.043.739.8
Focal-Tiny (Ours)43.7 (+1.7)44.8 (+1.1) 41.0 (+1.3)
ResNet-101[33]38.540.436.4
ResNeXt101-32x4d [67]39.941.937.5
PVT-Medium [60]41.942.039.0
ViL-Medium [76]42.943.439.7
Swin-Small [43]45.046.542.1
Focal-Small (Ours)45.6 (+0.6)47.4 (+0.9) 42.8 (+0.7)
ResNeXt101-64x4d[67] 41.042.838.4
PVT-Large [60]42.642.939.5
ViL-Base[76]44.345.141.0
Swin-Base [43]45.046.942.3
Focal-Base (Ours)46.3 (+1.3)47.8 (+0.9)43.2 (+0.9)
" + }, + { + "category_id": 1, + "poly": [ + 297, + 927, + 1404, + 927, + 1404, + 1202, + 297, + 1202 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 307, + 197, + 780, + 197, + 780, + 766, + 307, + 766 + ], + "score": 0.978, + "html": "
Model#Params. FLOPsTop-1 (%)
ResNet-50 [33]25.0 4.176.2
DeiT-Small/16 [55]22.1 4.679.9
PVT-Small [60]24.5 3.879.8
ViL-Small [76]24.6 5.182.0
CvT-13 [64]20.0 4.581.6
Swin-Tiny [43]28.3 4.581.2
Focal-Tiny (Ours)28.9 4.982.2
ResNet-101[33]45.0 7.977.4
PVT-Medium [60]44.2 6.781.2
CvT-21 [64]32.0 7.182.5
ViL-Medium [76]39.7 9.183.3
Swin-Small [43]49.6 8.783.1
Focal-Small (Ours)51.1 9.483.6
ResNet-152[33]60.0 11.078.3
ViT-Base/16 [21]86.6 17.677.9
DeiT-Base/16 [55]17.581.8
86.6
PVT-Large [60]61.4 9.881.7
ViL-Base[76]55.7 13.483.2
Swin-Base [43]87.8 15.483.4
Focal-Base (Ours)89.816.4 84.0
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1216, + 1403, + 1216, + 1403, + 1339, + 298, + 1339 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 300, + 1884, + 1403, + 1884, + 1403, + 2006, + 300, + 2006 + ], + "score": 0.975 + }, + { + "category_id": 0, + "poly": [ + 298, + 1375, + 878, + 1375, + 878, + 1407, + 298, + 1407 + ], + "score": 0.932 + }, + { + "category_id": 7, + "poly": [ + 296, + 770, + 798, + 770, + 798, + 891, + 296, + 891 + ], + "score": 0.923 + }, + { + "category_id": 6, + "poly": [ + 825, + 711, + 1405, + 711, + 1405, + 893, + 825, + 893 + ], + "score": 0.86 + }, + { + "category_id": 2, + "poly": [ + 842, + 2061, + 858, + 2061, + 858, + 2084, + 842, + 2084 + ], + "score": 0.717 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2084, + 841, + 2084 + ], + "score": 0.099 + }, + { + "category_id": 13, + "poly": [ + 347, + 1747, + 407, + 1747, + 407, + 1777, + 347, + 1777 + ], + "score": 0.9, + "latex": "1 0 ^ { - 4 }" + }, + { + "category_id": 13, + "poly": [ + 491, + 861, + 611, + 861, + 611, + 890, + 491, + 890 + ], + "score": 0.89, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 934, + 1507, + 973, + 1507, + 973, + 1535, + 934, + 1535 + ], + "score": 0.89, + "latex": "1 \\times" + }, + { + "category_id": 13, + "poly": [ + 821, + 1913, + 895, + 1913, + 895, + 1945, + 821, + 1945 + ], + "score": 0.87, + "latex": "( A P ^ { b } )" + }, + { + "category_id": 13, + "poly": [ + 881, + 834, + 920, + 834, + 920, + 862, + 881, + 862 + ], + "score": 0.87, + "latex": "1 \\times" + }, + { + "category_id": 13, + "poly": [ + 1191, + 1507, + 1229, + 1507, + 1229, + 1535, + 1191, + 1535 + ], + "score": 0.87, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 907, + 1141, + 982, + 1141, + 982, + 1170, + 907, + 1170 + ], + "score": 0.87, + "latex": "8 4 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 824, + 864, + 864, + 864, + 864, + 892, + 824, + 892 + ], + "score": 0.87, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 1056, + 770, + 1131, + 770, + 1131, + 803, + 1056, + 803 + ], + "score": 0.86, + "latex": "( A P ^ { b } )" + }, + { + "category_id": 13, + "poly": [ + 946, + 1020, + 1007, + 1020, + 1007, + 1049, + 946, + 1049 + ], + "score": 0.86, + "latex": "2 . 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 380, + 1537, + 420, + 1537, + 420, + 1566, + 380, + 1566 + ], + "score": 0.86, + "latex": "1 \\times" + }, + { + "category_id": 13, + "poly": [ + 297, + 1111, + 371, + 1111, + 371, + 1140, + 297, + 1140 + ], + "score": 0.86, + "latex": "8 3 . 6 \\%" + }, + { + "category_id": 13, + "poly": [ + 457, + 1568, + 497, + 1568, + 497, + 1596, + 457, + 1596 + ], + "score": 0.86, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1915, + 1163, + 1915, + 1163, + 1945, + 1080, + 1945 + ], + "score": 0.85, + "latex": "( A P ^ { m } )" + }, + { + "category_id": 13, + "poly": [ + 1319, + 773, + 1402, + 773, + 1402, + 803, + 1319, + 803 + ], + "score": 0.84, + "latex": "( A P ^ { m } )" + }, + { + "category_id": 13, + "poly": [ + 782, + 1476, + 842, + 1476, + 842, + 1505, + 782, + 1505 + ], + "score": 0.25, + "latex": "1 1 8 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1371.0, + 881.0, + 1371.0, + 881.0, + 1415.0, + 293.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 768.0, + 799.0, + 768.0, + 799.0, + 802.0, + 295.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 797.0, + 799.0, + 797.0, + 799.0, + 831.0, + 296.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 830.0, + 799.0, + 830.0, + 799.0, + 860.0, + 296.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 860.0, + 490.0, + 860.0, + 490.0, + 894.0, + 296.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 860.0, + 738.0, + 860.0, + 738.0, + 894.0, + 612.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 709.0, + 1406.0, + 709.0, + 1406.0, + 744.0, + 822.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 738.0, + 1407.0, + 738.0, + 1407.0, + 774.0, + 822.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 770.0, + 1055.0, + 770.0, + 1055.0, + 804.0, + 823.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 770.0, + 1318.0, + 770.0, + 1318.0, + 804.0, + 1132.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 770.0, + 1406.0, + 770.0, + 1406.0, + 804.0, + 1403.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 802.0, + 1405.0, + 802.0, + 1405.0, + 832.0, + 821.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 830.0, + 880.0, + 830.0, + 880.0, + 866.0, + 822.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 830.0, + 1404.0, + 830.0, + 1404.0, + 866.0, + 921.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 863.0, + 1136.0, + 863.0, + 1136.0, + 893.0, + 865.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2092.0, + 840.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2092.0, + 840.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1412.0, + 1406.0, + 1412.0, + 1406.0, + 1449.0, + 293.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1446.0, + 1406.0, + 1446.0, + 1406.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1475.0, + 781.0, + 1475.0, + 781.0, + 1510.0, + 293.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1510.0, + 843.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1504.0, + 933.0, + 1504.0, + 933.0, + 1541.0, + 293.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1504.0, + 1190.0, + 1504.0, + 1190.0, + 1541.0, + 974.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1504.0, + 1407.0, + 1504.0, + 1407.0, + 1541.0, + 1230.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1535.0, + 379.0, + 1535.0, + 379.0, + 1572.0, + 292.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1535.0, + 1405.0, + 1535.0, + 1405.0, + 1572.0, + 421.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1567.0, + 456.0, + 1567.0, + 456.0, + 1602.0, + 295.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1567.0, + 1406.0, + 1567.0, + 1406.0, + 1602.0, + 498.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1597.0, + 1403.0, + 1597.0, + 1403.0, + 1631.0, + 295.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1626.0, + 1406.0, + 1626.0, + 1406.0, + 1661.0, + 295.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1657.0, + 1403.0, + 1657.0, + 1403.0, + 1692.0, + 295.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1686.0, + 1405.0, + 1686.0, + 1405.0, + 1724.0, + 292.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1714.0, + 1405.0, + 1714.0, + 1405.0, + 1755.0, + 292.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1743.0, + 346.0, + 1743.0, + 346.0, + 1785.0, + 291.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1743.0, + 1406.0, + 1743.0, + 1406.0, + 1785.0, + 408.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1778.0, + 1403.0, + 1778.0, + 1403.0, + 1812.0, + 295.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1810.0, + 1405.0, + 1810.0, + 1405.0, + 1844.0, + 295.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1839.0, + 995.0, + 1839.0, + 995.0, + 1874.0, + 295.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 925.0, + 1404.0, + 925.0, + 1404.0, + 964.0, + 292.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 958.0, + 1404.0, + 958.0, + 1404.0, + 992.0, + 295.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 984.0, + 1404.0, + 984.0, + 1404.0, + 1028.0, + 292.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1019.0, + 945.0, + 1019.0, + 945.0, + 1055.0, + 295.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1019.0, + 1406.0, + 1019.0, + 1406.0, + 1055.0, + 1008.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1051.0, + 1406.0, + 1051.0, + 1406.0, + 1084.0, + 296.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1081.0, + 1404.0, + 1081.0, + 1404.0, + 1114.0, + 296.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1107.0, + 1404.0, + 1107.0, + 1404.0, + 1147.0, + 372.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1137.0, + 906.0, + 1137.0, + 906.0, + 1176.0, + 294.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1137.0, + 1404.0, + 1137.0, + 1404.0, + 1176.0, + 983.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1171.0, + 701.0, + 1171.0, + 701.0, + 1206.0, + 295.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1212.0, + 1404.0, + 1212.0, + 1404.0, + 1252.0, + 292.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1247.0, + 1404.0, + 1247.0, + 1404.0, + 1280.0, + 294.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1277.0, + 1407.0, + 1277.0, + 1407.0, + 1314.0, + 293.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1308.0, + 957.0, + 1308.0, + 957.0, + 1340.0, + 294.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 1408.0, + 1880.0, + 1408.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 820.0, + 1910.0, + 820.0, + 1950.0, + 293.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1910.0, + 1079.0, + 1910.0, + 1079.0, + 1950.0, + 896.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1910.0, + 1407.0, + 1910.0, + 1407.0, + 1950.0, + 1164.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1945.0, + 1407.0, + 1945.0, + 1407.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1975.0, + 1410.0, + 1975.0, + 1410.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 307, + 196, + 1405, + 196, + 1405, + 689, + 307, + 689 + ], + "score": 0.985, + "html": "
Backbone#Params (M)FLOPs (G)RetinaNet 3x schedule + MSMask R-CNN 3x schedule + MS
APAP5APAPsAPMAPLAP6APAPApmAPAP
ResNet50 [33]37.7/44.2239/26039.058.441.822.442.851.641.061.744.937.158.440.1
PVT-Small[60]34.2/44.1226/24542.262.745.026.245.257.243.065.346.939.962.542.8
ViL-Small [76]35.7/45.0252/174 42.963.845.627.846.456.343.464.947.039.662.142.4
Swin-Tiny [43]38.5/47.8245/264 45.065.948.429.748.958.146.068.150.341.665.144.9
Focal-Tiny (Ours)39.4/48.8265/291 45.566.348.831.249.258.747.269.451.942.766.5 45.9
ResNet101 [33]56.7/63.2315/33640.960.144.023.745.053.842.863.247.138.560.141.3
ResNeXt101-32x4d [67]56.4/62.8319/34041.461.044.323.945.553.744.064.448.039.261.441.9
PVT-Medium [60]53.9/63.9283/30243.263.846.127.346.358.944.266.048.240.563.143.5
ViL-Medium [76]50.8/60.1339/26143.764.646.427.947.156.944.666.348.540.763.843.7
Swin-Small [43]59.8/69.1335/354 46.467.050.131.050.160.348.570.253.543.367.346.6
Focal-Small (Ours)61.7/71.2367/40147.367.851.031.650.961.148.870.553.643.867.747.2
ResNeXt101-64x4d [67]95.5/102473/49341.861.544.425.245.454.644.464.948.839.761.942.6
PVT-Large[60]71.1/81.0345/364 43.463.646.126.146.059.544.566.048.340.763.443.7
ViL-Base [76]66.7/76.1443/365 44.765.547.629.948.058.145.767.249.941.364.444.5
Swin-Base 43]98.4/107477/496 45.866.449.129.949.460.348.569.853.243.466.846.9
Focal-Base (Ours)100.8/110.0 514/533 46.967.850.331.950.361.549.070.153.643.767.647.0
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1078, + 1404, + 1078, + 1404, + 1565, + 297, + 1565 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1641, + 1404, + 1641, + 1404, + 1884, + 297, + 1884 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 820, + 1404, + 820, + 1404, + 1065, + 297, + 1065 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 300, + 1886, + 1402, + 1886, + 1402, + 2007, + 300, + 2007 + ], + "score": 0.969 + }, + { + "category_id": 0, + "poly": [ + 298, + 1601, + 642, + 1601, + 642, + 1633, + 298, + 1633 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.799 + }, + { + "category_id": 6, + "poly": [ + 298, + 698, + 1403, + 698, + 1403, + 791, + 298, + 791 + ], + "score": 0.455 + }, + { + "category_id": 13, + "poly": [ + 349, + 1946, + 472, + 1946, + 472, + 1975, + 349, + 1975 + ], + "score": 0.9, + "latex": "6 4 0 \\times 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 297, + 1793, + 418, + 1793, + 418, + 1822, + 297, + 1822 + ], + "score": 0.89, + "latex": "5 1 2 \\times 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 1012, + 974, + 1052, + 974, + 1052, + 1002, + 1012, + 1002 + ], + "score": 0.87, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 604, + 730, + 644, + 730, + 644, + 758, + 604, + 758 + ], + "score": 0.86, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1005, + 1227, + 1005, + 1227, + 1033, + 1188, + 1033 + ], + "score": 0.86, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 818, + 1232, + 858, + 1232, + 858, + 1260, + 818, + 1260 + ], + "score": 0.86, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 1361, + 1473, + 1402, + 1473, + 1402, + 1503, + 1361, + 1503 + ], + "score": 0.85, + "latex": "2 \\times" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1595.0, + 645.0, + 1595.0, + 645.0, + 1641.0, + 293.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 695.0, + 1409.0, + 695.0, + 1409.0, + 734.0, + 292.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 726.0, + 603.0, + 726.0, + 603.0, + 761.0, + 294.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 726.0, + 1405.0, + 726.0, + 1405.0, + 761.0, + 645.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 753.0, + 1405.0, + 753.0, + 1405.0, + 795.0, + 290.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 1405.0, + 1074.0, + 1405.0, + 1115.0, + 294.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1110.0, + 1405.0, + 1110.0, + 1405.0, + 1144.0, + 294.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1139.0, + 1406.0, + 1139.0, + 1406.0, + 1174.0, + 295.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1167.0, + 1407.0, + 1167.0, + 1407.0, + 1207.0, + 292.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1197.0, + 1405.0, + 1197.0, + 1405.0, + 1237.0, + 292.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1227.0, + 817.0, + 1227.0, + 817.0, + 1266.0, + 294.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1227.0, + 1405.0, + 1227.0, + 1405.0, + 1266.0, + 859.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1259.0, + 1406.0, + 1259.0, + 1406.0, + 1295.0, + 294.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1324.0, + 296.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1318.0, + 1405.0, + 1318.0, + 1405.0, + 1358.0, + 292.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1351.0, + 1406.0, + 1351.0, + 1406.0, + 1384.0, + 294.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1381.0, + 1402.0, + 1381.0, + 1402.0, + 1412.0, + 296.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1412.0, + 1406.0, + 1412.0, + 1406.0, + 1446.0, + 295.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1443.0, + 1404.0, + 1443.0, + 1404.0, + 1477.0, + 295.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1473.0, + 1360.0, + 1473.0, + 1360.0, + 1507.0, + 295.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1501.0, + 1405.0, + 1501.0, + 1405.0, + 1540.0, + 292.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1532.0, + 979.0, + 1532.0, + 979.0, + 1569.0, + 294.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1639.0, + 1404.0, + 1639.0, + 1404.0, + 1675.0, + 294.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1669.0, + 1405.0, + 1669.0, + 1405.0, + 1712.0, + 291.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1698.0, + 1406.0, + 1698.0, + 1406.0, + 1738.0, + 294.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1732.0, + 1406.0, + 1732.0, + 1406.0, + 1765.0, + 295.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1758.0, + 1406.0, + 1758.0, + 1406.0, + 1799.0, + 292.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1789.0, + 296.0, + 1789.0, + 296.0, + 1830.0, + 292.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 1789.0, + 1406.0, + 1789.0, + 1406.0, + 1830.0, + 419.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1819.0, + 1406.0, + 1819.0, + 1406.0, + 1859.0, + 292.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1087.0, + 1851.0, + 1087.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 822.0, + 1405.0, + 822.0, + 1405.0, + 855.0, + 295.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 853.0, + 1405.0, + 853.0, + 1405.0, + 886.0, + 295.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 883.0, + 1404.0, + 883.0, + 1404.0, + 916.0, + 295.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 914.0, + 1405.0, + 914.0, + 1405.0, + 947.0, + 295.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 944.0, + 1405.0, + 944.0, + 1405.0, + 977.0, + 295.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 975.0, + 1011.0, + 975.0, + 1011.0, + 1005.0, + 293.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 975.0, + 1405.0, + 975.0, + 1405.0, + 1005.0, + 1053.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1004.0, + 1187.0, + 1004.0, + 1187.0, + 1037.0, + 294.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 1004.0, + 1405.0, + 1004.0, + 1405.0, + 1037.0, + 1228.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1030.0, + 1389.0, + 1030.0, + 1389.0, + 1071.0, + 294.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1406.0, + 1882.0, + 1406.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1914.0, + 1406.0, + 1914.0, + 1406.0, + 1949.0, + 294.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1948.0, + 348.0, + 1948.0, + 348.0, + 1976.0, + 296.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1948.0, + 1404.0, + 1948.0, + 1404.0, + 1976.0, + 473.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1978.0, + 1406.0, + 1978.0, + 1406.0, + 2010.0, + 296.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1504, + 1403, + 1504, + 1403, + 1749, + 298, + 1749 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 1762, + 1405, + 1762, + 1405, + 2007, + 298, + 2007 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1231, + 1404, + 1231, + 1404, + 1354, + 299, + 1354 + ], + "score": 0.978 + }, + { + "category_id": 5, + "poly": [ + 303, + 920, + 826, + 920, + 826, + 1109, + 303, + 1109 + ], + "score": 0.978, + "html": "
Model W-Size FLOPs Top-1(%) APb APm
Swin-Tiny74.581.2 43.739.8
144.982.1 44.0 40.5
Focal-Tiny74.982.2 44.941.1
145.282.3 45.5 41.5
" + }, + { + "category_id": 5, + "poly": [ + 888, + 197, + 1385, + 197, + 1385, + 708, + 888, + 708 + ], + "score": 0.978, + "html": "
BackboneMethod#Param FLOPs mIoU +MS
ResNet-101DANet [45]69M1119G45.3
ResNet-101ACNet [26]==45.9
ResNet-101DNL [69]69M1249G46.0
ResNet-101UperNet [65]86M1029G44.9
HRNet-w48 [53]OCRNet [73]71M664G45.7=
ResNeSt-200 [75]DLab.v3+ [12]88M1381G48.4=
Swin-T[43]UperNet [65]60M945G44.545.8
Swin-S [43]UperNet [65]81M1038G47.649.5
Swin-B [43]UperNet [65]121M1188G48.149.7
Twins-SVT-L[15]UperNet [65]133M48.850.2
MiT-B5 [66]SegFormer [66]85M51.051.8
ViT-L/16+ [21]SETR[80]308M50.3=
Swin-L* [43]UperNet [65]234M3230G52.153.5
ViT-L/16 [21]Segmenter [52]334M=51.853.6
Swin-L‡ [43]K-Net [77]==54.3
Swin-L‡ [43]PatchDiverse [29]234M53.154.4
VOLO-D5 [72]UperNet [65]=-54.3
Focal-T (Ours)UperNet [65]62M998G45.847.0
Focal-S (Ours)UperNet [65]85M1130G48.050.0
Focal-B (Ours)UperNet [65]126M1354G49.050.5
Focal-L‡ (Ours)UperNet [65]240M3376G54.055.4
" + }, + { + "category_id": 5, + "poly": [ + 303, + 212, + 848, + 212, + 848, + 694, + 303, + 694 + ], + "score": 0.977, + "html": "
Method#Param FLOPsmini-valtest-dev
ApbAPmApbApm
X101-64x4d [67]155M1033G 52.346.0
EfficientNet-D7 [54]77M410G54.4-55.1=
GCNet*[7]-1041G51.844.752.345.4
ResNeSt-200 [75]=52.5=53.347.1
Copy-paste [28]185M1440G 55.947.256.047.4
BoTNet-200 [51]-49.7-
SpineNet-190 [22]164M1885G 52.652.8
CenterNet2 [84]-=--56.4=
Swin-L (HTC++) [43]284M1470G 57.149.557.750.2
Swin-L (DyHead)[17]213M965G56.2---
Swin-L† (HTC++) [43]284M58.050.458.751.1
Swin-L† (DyHead) [17]213M58.4-58.7
Swin-L† (QueryInst) [25]-56.1156.1
Focal-L (HTC++) (Ours)265M1165G57.049.9-=
Focal-L (DyHead) (Ours)229M1081G56.4--=
Focal-L† (HTC++) (Ours)265M-58.150.958.451.3
Focal-L† (DyHead) (Ours)229M-58.7-59.0-
" + }, + { + "category_id": 5, + "poly": [ + 905, + 920, + 1362, + 920, + 1362, + 1110, + 905, + 1110 + ], + "score": 0.976, + "html": "
Model W-Shift Top-1(%) APb APm
Swin-Tiny80.2 81.238.8 43.736.4 39.8
Focal-Tiny82.244.841.0
81.944.941.1
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1430, + 1399, + 1430, + 1399, + 1491, + 298, + 1491 + ], + "score": 0.954 + }, + { + "category_id": 0, + "poly": [ + 298, + 1388, + 556, + 1388, + 556, + 1419, + 298, + 1419 + ], + "score": 0.913 + }, + { + "category_id": 7, + "poly": [ + 868, + 1112, + 1402, + 1112, + 1402, + 1202, + 868, + 1202 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 883, + 713, + 1393, + 713, + 1393, + 896, + 883, + 896 + ], + "score": 0.909 + }, + { + "category_id": 7, + "poly": [ + 297, + 1111, + 833, + 1111, + 833, + 1203, + 297, + 1203 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 297, + 697, + 853, + 697, + 853, + 881, + 297, + 881 + ], + "score": 0.806 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 858, + 2061, + 858, + 2084, + 840, + 2084 + ], + "score": 0.798 + }, + { + "category_id": 13, + "poly": [ + 297, + 819, + 393, + 819, + 393, + 849, + 297, + 849 + ], + "score": 0.8, + "latex": "\\mathrm { H T C + + }" + }, + { + "category_id": 13, + "poly": [ + 372, + 471, + 442, + 471, + 442, + 491, + 372, + 491 + ], + "score": 0.44, + "latex": "\\scriptstyle ( \\mathrm { H T C + + } )" + }, + { + "category_id": 13, + "poly": [ + 375, + 595, + 445, + 595, + 445, + 615, + 375, + 615 + ], + "score": 0.42, + "latex": "\\mathrm { ( H T C + + ) }" + }, + { + "category_id": 13, + "poly": [ + 383, + 640, + 454, + 640, + 454, + 661, + 383, + 661 + ], + "score": 0.41, + "latex": "\\mathrm { ( H T C + + ) }" + }, + { + "category_id": 13, + "poly": [ + 355, + 1688, + 460, + 1688, + 460, + 1718, + 355, + 1718 + ], + "score": 0.28, + "latex": "( 7 \\nu . s . \\ 1 4 )" + }, + { + "category_id": 13, + "poly": [ + 1374, + 801, + 1393, + 801, + 1393, + 827, + 1374, + 827 + ], + "score": 0.27, + "latex": "^ \\ddag" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1383.0, + 558.0, + 1383.0, + 558.0, + 1424.0, + 293.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1108.0, + 1404.0, + 1108.0, + 1404.0, + 1145.0, + 866.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1139.0, + 1404.0, + 1139.0, + 1404.0, + 1176.0, + 866.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1173.0, + 1060.0, + 1173.0, + 1060.0, + 1202.0, + 866.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1109.0, + 836.0, + 1109.0, + 836.0, + 1143.0, + 294.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1140.0, + 836.0, + 1140.0, + 836.0, + 1172.0, + 295.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1174.0, + 830.0, + 1174.0, + 830.0, + 1202.0, + 295.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 860.0, + 2060.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1506.0, + 1405.0, + 1506.0, + 1405.0, + 1539.0, + 294.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1537.0, + 1407.0, + 1537.0, + 1407.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1567.0, + 1403.0, + 1567.0, + 1403.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1597.0, + 1405.0, + 1597.0, + 1405.0, + 1630.0, + 294.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1622.0, + 1407.0, + 1622.0, + 1407.0, + 1665.0, + 292.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1658.0, + 1405.0, + 1658.0, + 1405.0, + 1691.0, + 294.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1686.0, + 354.0, + 1686.0, + 354.0, + 1724.0, + 293.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1686.0, + 1407.0, + 1686.0, + 1407.0, + 1724.0, + 461.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1719.0, + 935.0, + 1719.0, + 935.0, + 1753.0, + 293.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1764.0, + 1408.0, + 1764.0, + 1408.0, + 1797.0, + 295.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1795.0, + 1406.0, + 1795.0, + 1406.0, + 1828.0, + 295.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1826.0, + 1405.0, + 1826.0, + 1405.0, + 1856.0, + 296.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1851.0, + 1407.0, + 1851.0, + 1407.0, + 1892.0, + 291.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 1406.0, + 1883.0, + 1406.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 1407.0, + 1914.0, + 1407.0, + 1948.0, + 295.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1946.0, + 1403.0, + 1946.0, + 1403.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1976.0, + 1405.0, + 1976.0, + 1405.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1233.0, + 1404.0, + 1233.0, + 1404.0, + 1266.0, + 295.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1263.0, + 1404.0, + 1263.0, + 1404.0, + 1296.0, + 295.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1295.0, + 1404.0, + 1295.0, + 1404.0, + 1327.0, + 295.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1325.0, + 1199.0, + 1325.0, + 1199.0, + 1358.0, + 295.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1424.0, + 1404.0, + 1424.0, + 1404.0, + 1467.0, + 293.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1462.0, + 1036.0, + 1462.0, + 1036.0, + 1493.0, + 297.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 712.0, + 1396.0, + 712.0, + 1396.0, + 746.0, + 879.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 742.0, + 1394.0, + 742.0, + 1394.0, + 775.0, + 880.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 773.0, + 1397.0, + 773.0, + 1397.0, + 806.0, + 879.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 805.0, + 1351.0, + 805.0, + 1351.0, + 835.0, + 878.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1364.0, + 801.0, + 1373.0, + 801.0, + 1373.0, + 828.0, + 1364.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 833.0, + 1396.0, + 833.0, + 1396.0, + 866.0, + 878.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 864.0, + 1016.0, + 864.0, + 1016.0, + 900.0, + 879.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 697.0, + 854.0, + 697.0, + 854.0, + 728.0, + 295.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 727.0, + 858.0, + 727.0, + 858.0, + 762.0, + 294.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 757.0, + 857.0, + 757.0, + 857.0, + 791.0, + 295.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 785.0, + 856.0, + 785.0, + 856.0, + 825.0, + 293.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 818.0, + 858.0, + 818.0, + 858.0, + 850.0, + 394.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 848.0, + 809.0, + 848.0, + 809.0, + 882.0, + 295.0, + 882.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1499, + 1404, + 1499, + 1404, + 1743, + 298, + 1743 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 718, + 1404, + 718, + 1404, + 1112, + 298, + 1112 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1754, + 1404, + 1754, + 1404, + 2007, + 298, + 2007 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1126, + 1404, + 1126, + 1404, + 1401, + 298, + 1401 + ], + "score": 0.981 + }, + { + "category_id": 5, + "poly": [ + 867, + 205, + 1391, + 205, + 1391, + 438, + 867, + 438 + ], + "score": 0.977, + "html": "
Depths Model #Params. FLOPs Top-1(%) APb Apm
2-2-2-2Swin21.23.178.738.2 35.7
Focal21.73.479.940.5 37.6
2-2-4-2Swin24.73.880.241.2 38.1
Focal25.44.181.443.3 39.8
2-2-6-2Swin28.34.581.243.7 39.8
Focal29.14.982.244.8 41.0
" + }, + { + "category_id": 3, + "poly": [ + 301, + 203, + 823, + 203, + 823, + 453, + 301, + 453 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 296, + 456, + 832, + 456, + 832, + 607, + 296, + 607 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 297, + 642, + 1398, + 642, + 1398, + 703, + 297, + 703 + ], + "score": 0.952 + }, + { + "category_id": 0, + "poly": [ + 298, + 1444, + 509, + 1444, + 509, + 1481, + 298, + 1481 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 837, + 2062, + 865, + 2062, + 865, + 2085, + 837, + 2085 + ], + "score": 0.853 + }, + { + "category_id": 7, + "poly": [ + 859, + 447, + 1395, + 447, + 1395, + 599, + 859, + 599 + ], + "score": 0.442 + }, + { + "category_id": 1, + "poly": [ + 859, + 447, + 1395, + 447, + 1395, + 599, + 859, + 599 + ], + "score": 0.425 + }, + { + "category_id": 13, + "poly": [ + 682, + 931, + 813, + 931, + 813, + 961, + 682, + 961 + ], + "score": 0.86, + "latex": "8 2 . 2 \\substack { 8 0 . 1 }" + }, + { + "category_id": 13, + "poly": [ + 1064, + 931, + 1197, + 931, + 1197, + 961, + 1064, + 961 + ], + "score": 0.84, + "latex": "\\cdot 4 4 . 8 \\mathrm { \\ - } \\to 3 8 . 3" + }, + { + "category_id": 13, + "poly": [ + 388, + 901, + 427, + 901, + 427, + 930, + 388, + 930 + ], + "score": 0.84, + "latex": "1 \\times" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 202.0, + 339.0, + 202.0, + 339.0, + 218.0, + 320.0, + 218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 201.0, + 811.0, + 201.0, + 811.0, + 220.0, + 788.0, + 220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 219.0, + 738.0, + 219.0, + 738.0, + 246.0, + 693.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 229.0, + 811.0, + 229.0, + 811.0, + 248.0, + 788.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 241.0, + 341.0, + 241.0, + 341.0, + 280.0, + 301.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 253.0, + 512.0, + 253.0, + 512.0, + 278.0, + 470.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 243.0, + 625.0, + 243.0, + 625.0, + 271.0, + 580.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 259.0, + 772.0, + 259.0, + 772.0, + 284.0, + 729.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 255.0, + 811.0, + 255.0, + 811.0, + 276.0, + 787.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 269.0, + 340.0, + 269.0, + 340.0, + 345.0, + 301.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 272.0, + 547.0, + 272.0, + 547.0, + 297.0, + 505.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 274.0, + 826.0, + 274.0, + 826.0, + 354.0, + 789.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 298.0, + 656.0, + 298.0, + 656.0, + 319.0, + 617.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 308.0, + 399.0, + 308.0, + 399.0, + 333.0, + 356.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 313.0, + 810.0, + 313.0, + 810.0, + 332.0, + 790.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 326.0, + 339.0, + 326.0, + 339.0, + 343.0, + 320.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 335.0, + 324.0, + 335.0, + 324.0, + 381.0, + 302.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 345.0, + 433.0, + 345.0, + 433.0, + 368.0, + 393.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 340.0, + 809.0, + 340.0, + 809.0, + 357.0, + 790.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 369.0, + 342.0, + 369.0, + 342.0, + 385.0, + 321.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 368.0, + 811.0, + 368.0, + 811.0, + 387.0, + 787.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 395.0, + 811.0, + 395.0, + 811.0, + 415.0, + 788.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 411.0, + 340.0, + 411.0, + 340.0, + 426.0, + 321.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 418.0, + 427.0, + 418.0, + 427.0, + 442.0, + 365.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 418.0, + 536.0, + 418.0, + 536.0, + 442.0, + 480.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 418.0, + 652.0, + 418.0, + 652.0, + 442.0, + 589.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 418.0, + 786.0, + 418.0, + 786.0, + 441.0, + 681.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 436.0, + 642.0, + 436.0, + 642.0, + 455.0, + 486.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 450.0, + 833.0, + 450.0, + 833.0, + 492.0, + 294.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 486.0, + 834.0, + 486.0, + 834.0, + 519.0, + 295.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 516.0, + 834.0, + 516.0, + 834.0, + 549.0, + 295.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 546.0, + 832.0, + 546.0, + 832.0, + 581.0, + 296.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 577.0, + 576.0, + 577.0, + 576.0, + 610.0, + 295.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1440.0, + 514.0, + 1440.0, + 514.0, + 1488.0, + 291.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 445.0, + 1397.0, + 445.0, + 1397.0, + 481.0, + 857.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 479.0, + 1395.0, + 479.0, + 1395.0, + 509.0, + 859.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 509.0, + 1395.0, + 509.0, + 1395.0, + 540.0, + 858.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 539.0, + 1394.0, + 539.0, + 1394.0, + 569.0, + 858.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 568.0, + 1315.0, + 568.0, + 1315.0, + 602.0, + 858.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1496.0, + 1405.0, + 1496.0, + 1405.0, + 1536.0, + 292.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1529.0, + 1405.0, + 1529.0, + 1405.0, + 1562.0, + 294.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1560.0, + 1408.0, + 1560.0, + 1408.0, + 1594.0, + 296.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1591.0, + 1406.0, + 1591.0, + 1406.0, + 1625.0, + 294.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1620.0, + 1405.0, + 1620.0, + 1405.0, + 1654.0, + 293.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1684.0, + 294.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1678.0, + 1404.0, + 1678.0, + 1404.0, + 1717.0, + 293.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1711.0, + 843.0, + 1711.0, + 843.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 714.0, + 1405.0, + 714.0, + 1405.0, + 756.0, + 295.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 750.0, + 1405.0, + 750.0, + 1405.0, + 782.0, + 296.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 777.0, + 1406.0, + 777.0, + 1406.0, + 815.0, + 292.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 811.0, + 1404.0, + 811.0, + 1404.0, + 843.0, + 296.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 839.0, + 1406.0, + 839.0, + 1406.0, + 876.0, + 292.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 870.0, + 1406.0, + 870.0, + 1406.0, + 906.0, + 293.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 898.0, + 387.0, + 898.0, + 387.0, + 937.0, + 292.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 898.0, + 1406.0, + 898.0, + 1406.0, + 937.0, + 428.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 928.0, + 681.0, + 928.0, + 681.0, + 966.0, + 293.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 928.0, + 1063.0, + 928.0, + 1063.0, + 966.0, + 814.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 928.0, + 1406.0, + 928.0, + 1406.0, + 966.0, + 1198.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 963.0, + 1405.0, + 963.0, + 1405.0, + 995.0, + 296.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 990.0, + 1405.0, + 990.0, + 1405.0, + 1028.0, + 293.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1019.0, + 1405.0, + 1019.0, + 1405.0, + 1059.0, + 293.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1049.0, + 1405.0, + 1049.0, + 1405.0, + 1089.0, + 295.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1082.0, + 993.0, + 1082.0, + 993.0, + 1117.0, + 294.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1752.0, + 1404.0, + 1752.0, + 1404.0, + 1789.0, + 293.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1780.0, + 1404.0, + 1780.0, + 1404.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1812.0, + 1405.0, + 1812.0, + 1405.0, + 1843.0, + 296.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1839.0, + 1402.0, + 1839.0, + 1402.0, + 1869.0, + 296.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1865.0, + 1406.0, + 1865.0, + 1406.0, + 1899.0, + 294.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1926.0, + 294.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1921.0, + 1405.0, + 1921.0, + 1405.0, + 1955.0, + 294.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1945.0, + 1408.0, + 1945.0, + 1408.0, + 1985.0, + 293.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1974.0, + 820.0, + 1974.0, + 820.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1128.0, + 1406.0, + 1128.0, + 1406.0, + 1161.0, + 296.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1157.0, + 1406.0, + 1157.0, + 1406.0, + 1193.0, + 294.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1188.0, + 1406.0, + 1188.0, + 1406.0, + 1224.0, + 294.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1219.0, + 1405.0, + 1219.0, + 1405.0, + 1252.0, + 296.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1248.0, + 1407.0, + 1248.0, + 1407.0, + 1284.0, + 291.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1275.0, + 1405.0, + 1275.0, + 1405.0, + 1317.0, + 292.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1309.0, + 1406.0, + 1309.0, + 1406.0, + 1346.0, + 294.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1339.0, + 1404.0, + 1339.0, + 1404.0, + 1373.0, + 293.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1369.0, + 1341.0, + 1369.0, + 1341.0, + 1405.0, + 294.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 640.0, + 1403.0, + 640.0, + 1403.0, + 678.0, + 295.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 673.0, + 1123.0, + 673.0, + 1123.0, + 706.0, + 293.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 445.0, + 1397.0, + 445.0, + 1397.0, + 481.0, + 857.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 479.0, + 1395.0, + 479.0, + 1395.0, + 509.0, + 859.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 509.0, + 1395.0, + 509.0, + 1395.0, + 540.0, + 858.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 539.0, + 1394.0, + 539.0, + 1394.0, + 569.0, + 858.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 568.0, + 1315.0, + 568.0, + 1315.0, + 602.0, + 858.0, + 602.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 299, + 200, + 455, + 200, + 455, + 234, + 299, + 234 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 295, + 180, + 1410, + 180, + 1410, + 2015, + 295, + 2015 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.558 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.537 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 196.0, + 460.0, + 196.0, + 460.0, + 240.0, + 295.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 196.0, + 461.0, + 196.0, + 461.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 247.0, + 1410.0, + 247.0, + 1410.0, + 283.0, + 301.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 275.0, + 1274.0, + 275.0, + 1274.0, + 311.0, + 347.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 326.0, + 1408.0, + 326.0, + 1408.0, + 364.0, + 305.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 352.0, + 1410.0, + 352.0, + 1410.0, + 392.0, + 345.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 382.0, + 605.0, + 382.0, + 605.0, + 418.0, + 345.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 430.0, + 1406.0, + 430.0, + 1406.0, + 469.0, + 305.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 461.0, + 700.0, + 461.0, + 700.0, + 495.0, + 343.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 511.0, + 1406.0, + 511.0, + 1406.0, + 543.0, + 307.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 535.0, + 1278.0, + 535.0, + 1278.0, + 571.0, + 347.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 586.0, + 1408.0, + 586.0, + 1408.0, + 624.0, + 305.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 610.0, + 1406.0, + 610.0, + 1406.0, + 654.0, + 347.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 646.0, + 541.0, + 646.0, + 541.0, + 672.0, + 353.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 690.0, + 1408.0, + 690.0, + 1408.0, + 727.0, + 303.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 717.0, + 1410.0, + 717.0, + 1410.0, + 757.0, + 347.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 747.0, + 762.0, + 747.0, + 762.0, + 783.0, + 345.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 799.0, + 1408.0, + 799.0, + 1408.0, + 831.0, + 307.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 827.0, + 1406.0, + 827.0, + 1406.0, + 860.0, + 351.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 852.0, + 720.0, + 852.0, + 720.0, + 888.0, + 347.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 900.0, + 1406.0, + 900.0, + 1406.0, + 940.0, + 303.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 932.0, + 1408.0, + 932.0, + 1408.0, + 964.0, + 349.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 954.0, + 680.0, + 954.0, + 680.0, + 997.0, + 343.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1007.0, + 1408.0, + 1007.0, + 1408.0, + 1045.0, + 305.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1035.0, + 1408.0, + 1035.0, + 1408.0, + 1073.0, + 347.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1063.0, + 421.0, + 1063.0, + 421.0, + 1098.0, + 349.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1112.0, + 1406.0, + 1112.0, + 1406.0, + 1150.0, + 293.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1142.0, + 1232.0, + 1142.0, + 1232.0, + 1176.0, + 347.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1192.0, + 1406.0, + 1192.0, + 1406.0, + 1224.0, + 295.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1216.0, + 1406.0, + 1216.0, + 1406.0, + 1257.0, + 345.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1245.0, + 1298.0, + 1245.0, + 1298.0, + 1283.0, + 349.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1297.0, + 1404.0, + 1297.0, + 1404.0, + 1329.0, + 295.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1321.0, + 1406.0, + 1321.0, + 1406.0, + 1362.0, + 347.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1349.0, + 982.0, + 1349.0, + 982.0, + 1386.0, + 345.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1398.0, + 1408.0, + 1398.0, + 1408.0, + 1436.0, + 293.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1428.0, + 762.0, + 1428.0, + 762.0, + 1464.0, + 347.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1480.0, + 1404.0, + 1480.0, + 1404.0, + 1513.0, + 295.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1505.0, + 902.0, + 1505.0, + 902.0, + 1543.0, + 347.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1555.0, + 1408.0, + 1555.0, + 1408.0, + 1593.0, + 293.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1583.0, + 1408.0, + 1583.0, + 1408.0, + 1622.0, + 347.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1607.0, + 420.0, + 1607.0, + 420.0, + 1649.0, + 349.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1660.0, + 1410.0, + 1660.0, + 1410.0, + 1698.0, + 293.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1688.0, + 1272.0, + 1688.0, + 1272.0, + 1726.0, + 349.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1773.0, + 297.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1769.0, + 1406.0, + 1769.0, + 1406.0, + 1801.0, + 351.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1793.0, + 1032.0, + 1793.0, + 1032.0, + 1831.0, + 349.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1845.0, + 1404.0, + 1845.0, + 1404.0, + 1878.0, + 295.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1874.0, + 1046.0, + 1874.0, + 1046.0, + 1906.0, + 351.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1924.0, + 1404.0, + 1924.0, + 1404.0, + 1956.0, + 295.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1948.0, + 1410.0, + 1948.0, + 1410.0, + 1986.0, + 347.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1978.0, + 471.0, + 1978.0, + 471.0, + 2011.0, + 349.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 293, + 115, + 1411, + 115, + 1411, + 2015, + 293, + 2015 + ], + "score": 0.746 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 196.0, + 1410.0, + 196.0, + 1410.0, + 246.0, + 288.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 234.0, + 1055.0, + 234.0, + 1055.0, + 267.0, + 349.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 280.0, + 1408.0, + 280.0, + 1408.0, + 317.0, + 292.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 305.0, + 1408.0, + 305.0, + 1408.0, + 346.0, + 347.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 338.0, + 1349.0, + 338.0, + 1349.0, + 371.0, + 351.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 386.0, + 1406.0, + 386.0, + 1406.0, + 419.0, + 294.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 411.0, + 1404.0, + 411.0, + 1404.0, + 444.0, + 351.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 438.0, + 1326.0, + 438.0, + 1326.0, + 477.0, + 349.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 490.0, + 1404.0, + 490.0, + 1404.0, + 523.0, + 296.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 517.0, + 1228.0, + 517.0, + 1228.0, + 550.0, + 351.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 569.0, + 1406.0, + 569.0, + 1406.0, + 596.0, + 298.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 594.0, + 1404.0, + 594.0, + 1404.0, + 627.0, + 351.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 619.0, + 1051.0, + 619.0, + 1051.0, + 659.0, + 347.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 665.0, + 1406.0, + 665.0, + 1406.0, + 705.0, + 294.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 696.0, + 628.0, + 696.0, + 628.0, + 730.0, + 349.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 746.0, + 1406.0, + 746.0, + 1406.0, + 780.0, + 296.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 771.0, + 1406.0, + 771.0, + 1406.0, + 805.0, + 349.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 798.0, + 604.0, + 798.0, + 604.0, + 836.0, + 344.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 846.0, + 1408.0, + 846.0, + 1408.0, + 886.0, + 294.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 877.0, + 1404.0, + 877.0, + 1404.0, + 911.0, + 349.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 905.0, + 1406.0, + 905.0, + 1406.0, + 938.0, + 349.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 952.0, + 1406.0, + 952.0, + 1406.0, + 986.0, + 296.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 982.0, + 1406.0, + 982.0, + 1406.0, + 1015.0, + 351.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1009.0, + 611.0, + 1009.0, + 611.0, + 1042.0, + 349.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1057.0, + 1406.0, + 1057.0, + 1406.0, + 1090.0, + 296.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1084.0, + 573.0, + 1084.0, + 573.0, + 1117.0, + 349.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1136.0, + 1406.0, + 1136.0, + 1406.0, + 1163.0, + 298.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1159.0, + 1395.0, + 1159.0, + 1395.0, + 1192.0, + 349.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1207.0, + 1410.0, + 1207.0, + 1410.0, + 1246.0, + 292.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1236.0, + 421.0, + 1236.0, + 421.0, + 1271.0, + 349.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1286.0, + 1404.0, + 1286.0, + 1404.0, + 1319.0, + 296.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1311.0, + 1061.0, + 1311.0, + 1061.0, + 1348.0, + 344.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1357.0, + 1410.0, + 1357.0, + 1410.0, + 1398.0, + 292.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1388.0, + 1410.0, + 1388.0, + 1410.0, + 1427.0, + 347.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1438.0, + 1406.0, + 1438.0, + 1406.0, + 1471.0, + 296.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1467.0, + 1406.0, + 1467.0, + 1406.0, + 1500.0, + 349.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1490.0, + 1063.0, + 1490.0, + 1063.0, + 1527.0, + 344.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1542.0, + 1406.0, + 1542.0, + 1406.0, + 1575.0, + 296.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1569.0, + 1048.0, + 1569.0, + 1048.0, + 1605.0, + 344.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1617.0, + 1406.0, + 1617.0, + 1406.0, + 1650.0, + 294.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1646.0, + 1169.0, + 1646.0, + 1169.0, + 1680.0, + 349.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1694.0, + 1404.0, + 1694.0, + 1404.0, + 1728.0, + 296.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1721.0, + 1053.0, + 1721.0, + 1053.0, + 1755.0, + 351.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1769.0, + 1404.0, + 1769.0, + 1404.0, + 1802.0, + 296.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1798.0, + 1114.0, + 1798.0, + 1114.0, + 1832.0, + 349.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1846.0, + 1406.0, + 1846.0, + 1406.0, + 1880.0, + 296.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1871.0, + 1408.0, + 1871.0, + 1408.0, + 1911.0, + 347.0, + 1911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1895.0, + 422.0, + 1895.0, + 422.0, + 1939.0, + 349.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1326.0, + 1946.0, + 1326.0, + 1986.0, + 294.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1953.0, + 1408.0, + 1953.0, + 1408.0, + 1982.0, + 1334.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1975.0, + 1319.0, + 1975.0, + 1319.0, + 2015.0, + 349.0, + 2015.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 293, + 134, + 1410, + 134, + 1410, + 2023, + 293, + 2023 + ], + "score": 0.756 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.725 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.329 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 202.0, + 1407.0, + 202.0, + 1407.0, + 239.0, + 289.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 233.0, + 1407.0, + 233.0, + 1407.0, + 266.0, + 351.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 260.0, + 420.0, + 260.0, + 420.0, + 295.0, + 349.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 310.0, + 1407.0, + 310.0, + 1407.0, + 343.0, + 296.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 337.0, + 1251.0, + 337.0, + 1251.0, + 370.0, + 349.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 386.0, + 1405.0, + 386.0, + 1405.0, + 420.0, + 294.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 413.0, + 1407.0, + 413.0, + 1407.0, + 447.0, + 349.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 440.0, + 420.0, + 440.0, + 420.0, + 476.0, + 347.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 486.0, + 1407.0, + 486.0, + 1407.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 517.0, + 611.0, + 517.0, + 611.0, + 550.0, + 351.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 563.0, + 1407.0, + 563.0, + 1407.0, + 604.0, + 292.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 590.0, + 1405.0, + 590.0, + 1405.0, + 631.0, + 344.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 621.0, + 518.0, + 621.0, + 518.0, + 654.0, + 351.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 666.0, + 1407.0, + 666.0, + 1407.0, + 708.0, + 292.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 693.0, + 1407.0, + 693.0, + 1407.0, + 735.0, + 347.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 722.0, + 665.0, + 722.0, + 665.0, + 759.0, + 346.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 772.0, + 1405.0, + 772.0, + 1405.0, + 805.0, + 296.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 799.0, + 1409.0, + 799.0, + 1409.0, + 838.0, + 347.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 828.0, + 420.0, + 828.0, + 420.0, + 863.0, + 349.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 878.0, + 1405.0, + 878.0, + 1405.0, + 911.0, + 296.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 904.0, + 944.0, + 904.0, + 944.0, + 938.0, + 346.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 952.0, + 1405.0, + 952.0, + 1405.0, + 985.0, + 296.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 981.0, + 1120.0, + 981.0, + 1120.0, + 1014.0, + 349.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1029.0, + 1407.0, + 1029.0, + 1407.0, + 1062.0, + 296.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1058.0, + 1202.0, + 1058.0, + 1202.0, + 1091.0, + 349.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1105.0, + 1407.0, + 1105.0, + 1407.0, + 1139.0, + 296.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1132.0, + 887.0, + 1132.0, + 887.0, + 1166.0, + 349.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1182.0, + 1405.0, + 1182.0, + 1405.0, + 1215.0, + 296.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1207.0, + 908.0, + 1207.0, + 908.0, + 1244.0, + 346.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1253.0, + 1407.0, + 1253.0, + 1407.0, + 1294.0, + 292.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1284.0, + 689.0, + 1284.0, + 689.0, + 1321.0, + 346.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1333.0, + 1407.0, + 1333.0, + 1407.0, + 1366.0, + 296.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1360.0, + 1171.0, + 1360.0, + 1171.0, + 1400.0, + 347.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1410.0, + 1405.0, + 1410.0, + 1405.0, + 1443.0, + 296.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1439.0, + 1405.0, + 1439.0, + 1405.0, + 1472.0, + 349.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1464.0, + 611.0, + 1464.0, + 611.0, + 1497.0, + 349.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1514.0, + 1405.0, + 1514.0, + 1405.0, + 1547.0, + 296.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1541.0, + 1403.0, + 1541.0, + 1403.0, + 1574.0, + 349.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1567.0, + 1325.0, + 1567.0, + 1325.0, + 1607.0, + 347.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1617.0, + 1405.0, + 1617.0, + 1405.0, + 1650.0, + 296.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1646.0, + 1092.0, + 1646.0, + 1092.0, + 1679.0, + 349.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1694.0, + 1405.0, + 1694.0, + 1405.0, + 1727.0, + 296.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1721.0, + 1242.0, + 1721.0, + 1242.0, + 1756.0, + 344.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1771.0, + 1405.0, + 1771.0, + 1405.0, + 1804.0, + 296.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1797.0, + 1105.0, + 1797.0, + 1105.0, + 1831.0, + 349.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1843.0, + 1407.0, + 1843.0, + 1407.0, + 1882.0, + 292.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1874.0, + 1407.0, + 1874.0, + 1407.0, + 1907.0, + 349.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1899.0, + 761.0, + 1899.0, + 761.0, + 1936.0, + 344.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1947.0, + 1407.0, + 1947.0, + 1407.0, + 1986.0, + 294.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1976.0, + 1409.0, + 1976.0, + 1409.0, + 2015.0, + 347.0, + 2015.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.822 + }, + { + "category_id": 1, + "poly": [ + 294, + 112, + 1410, + 112, + 1410, + 2023, + 294, + 2023 + ], + "score": 0.653 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 871.0, + 2058.0, + 871.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 202.0, + 1411.0, + 202.0, + 1411.0, + 240.0, + 290.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 234.0, + 1357.0, + 234.0, + 1357.0, + 267.0, + 350.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 284.0, + 1405.0, + 284.0, + 1405.0, + 318.0, + 295.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 313.0, + 1357.0, + 313.0, + 1357.0, + 347.0, + 350.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 357.0, + 1411.0, + 357.0, + 1411.0, + 399.0, + 290.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 391.0, + 1247.0, + 391.0, + 1247.0, + 424.0, + 350.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 439.0, + 1407.0, + 439.0, + 1407.0, + 477.0, + 290.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 466.0, + 1409.0, + 466.0, + 1409.0, + 504.0, + 345.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 491.0, + 420.0, + 491.0, + 420.0, + 532.0, + 347.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 548.0, + 1407.0, + 548.0, + 1407.0, + 582.0, + 295.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 577.0, + 1152.0, + 577.0, + 1152.0, + 611.0, + 350.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 628.0, + 1407.0, + 628.0, + 1407.0, + 661.0, + 295.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 651.0, + 1407.0, + 651.0, + 1407.0, + 693.0, + 345.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 680.0, + 734.0, + 680.0, + 734.0, + 718.0, + 347.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 728.0, + 1409.0, + 728.0, + 1409.0, + 770.0, + 290.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 762.0, + 1409.0, + 762.0, + 1409.0, + 795.0, + 347.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 787.0, + 607.0, + 787.0, + 607.0, + 824.0, + 345.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 837.0, + 1407.0, + 837.0, + 1407.0, + 875.0, + 290.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 864.0, + 1409.0, + 864.0, + 1409.0, + 902.0, + 345.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 915.0, + 1407.0, + 915.0, + 1407.0, + 954.0, + 293.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 944.0, + 1095.0, + 944.0, + 1095.0, + 982.0, + 345.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 992.0, + 1407.0, + 992.0, + 1407.0, + 1034.0, + 290.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1026.0, + 1407.0, + 1026.0, + 1407.0, + 1059.0, + 350.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1055.0, + 611.0, + 1055.0, + 611.0, + 1082.0, + 352.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1103.0, + 1405.0, + 1103.0, + 1405.0, + 1137.0, + 295.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1128.0, + 890.0, + 1128.0, + 890.0, + 1166.0, + 345.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1179.0, + 1409.0, + 1179.0, + 1409.0, + 1218.0, + 293.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1208.0, + 763.0, + 1208.0, + 763.0, + 1245.0, + 345.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1256.0, + 1405.0, + 1256.0, + 1405.0, + 1298.0, + 293.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1287.0, + 1407.0, + 1287.0, + 1407.0, + 1327.0, + 347.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1315.0, + 702.0, + 1315.0, + 702.0, + 1350.0, + 343.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1367.0, + 1405.0, + 1367.0, + 1405.0, + 1401.0, + 295.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1392.0, + 1399.0, + 1392.0, + 1399.0, + 1432.0, + 347.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1442.0, + 1409.0, + 1442.0, + 1409.0, + 1482.0, + 293.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1470.0, + 1407.0, + 1470.0, + 1407.0, + 1512.0, + 345.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1501.0, + 609.0, + 1501.0, + 609.0, + 1528.0, + 352.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1589.0, + 290.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1581.0, + 567.0, + 1581.0, + 567.0, + 1614.0, + 350.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1631.0, + 1407.0, + 1631.0, + 1407.0, + 1665.0, + 295.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1656.0, + 1147.0, + 1656.0, + 1147.0, + 1694.0, + 347.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1744.0, + 295.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1736.0, + 1135.0, + 1736.0, + 1135.0, + 1773.0, + 345.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1786.0, + 1407.0, + 1786.0, + 1407.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1811.0, + 1407.0, + 1811.0, + 1407.0, + 1853.0, + 345.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1843.0, + 1071.0, + 1843.0, + 1071.0, + 1878.0, + 345.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1891.0, + 1407.0, + 1891.0, + 1407.0, + 1933.0, + 290.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1920.0, + 1405.0, + 1920.0, + 1405.0, + 1960.0, + 347.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1947.0, + 1409.0, + 1947.0, + 1409.0, + 1987.0, + 347.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1975.0, + 818.0, + 1975.0, + 818.0, + 2012.0, + 345.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.823 + }, + { + "category_id": 1, + "poly": [ + 295, + 204, + 1408, + 204, + 1408, + 523, + 295, + 523 + ], + "score": 0.741 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 205.0, + 1409.0, + 205.0, + 1409.0, + 242.0, + 296.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 234.0, + 1400.0, + 234.0, + 1400.0, + 267.0, + 350.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 280.0, + 1406.0, + 280.0, + 1406.0, + 322.0, + 292.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 311.0, + 1409.0, + 311.0, + 1409.0, + 348.0, + 350.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 340.0, + 578.0, + 340.0, + 578.0, + 375.0, + 347.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 386.0, + 1406.0, + 386.0, + 1406.0, + 426.0, + 293.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 417.0, + 609.0, + 417.0, + 609.0, + 450.0, + 349.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 467.0, + 1404.0, + 467.0, + 1404.0, + 500.0, + 296.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 491.0, + 1212.0, + 491.0, + 1212.0, + 527.0, + 350.0, + 527.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_layout.pdf b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9ef4d920ba2e41c2f3fc0cabc88d269ce7c996d1 --- /dev/null +++ b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:038646ed1a87064699fc6d2cec60c87652915aedb6f60adb10b966ed88842d78 +size 1115032 diff --git a/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_origin.pdf b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4de1edd4bd0cef19b337f13d114586b448be707c --- /dev/null +++ b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:045b808816d7280364f4fd91d38530085e1023404ccdc9a401331e50233a0176 +size 886331 diff --git a/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_span.pdf b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..63bdfca5d2277dc21e1f6b5cc734577c9cbf6af8 --- /dev/null +++ b/parse/train/5CGPY2VeEGb/5CGPY2VeEGb_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fda6eb7f587de665617da8500ec4c254177a8cb252946eaa3d8bba75923bfe3b +size 1121252 diff --git a/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_layout.pdf b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..af66ae2552d99d452e1dcd32561867cc8b5e1cbe --- /dev/null +++ b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:483f25878e5b2ff4e54365f4ec678262050564f5a298bd6ea00b160960f690d1 +size 493544 diff --git a/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_origin.pdf b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..65aef4242dc97719db6c441dc1ec0649ce94687d --- /dev/null +++ b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aba48fd8e403e275f7a31d46a8d7768bded6388baff0701da82eb4a46e206e23 +size 327211 diff --git a/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_span.pdf b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..31018bec4c2d2eeb2049d9de5d4a8438308131ec --- /dev/null +++ b/parse/train/6MaBrlQ5JM/6MaBrlQ5JM_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39fe06bdfed54361643133b269642946f6dc1b84d65fcdba79c4763e8185f00b +size 504006 diff --git a/parse/train/7J-fKoXiReA/7J-fKoXiReA_layout.pdf b/parse/train/7J-fKoXiReA/7J-fKoXiReA_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..dddb7033791248fa260bcfbcec4098b9980f06f6 --- /dev/null +++ b/parse/train/7J-fKoXiReA/7J-fKoXiReA_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d88a85f5ecce0cf80e83f699cf3d1d8dc918ad02ffc9b8a3075e9e5d19211c6b +size 2235836 diff --git a/parse/train/7J-fKoXiReA/7J-fKoXiReA_origin.pdf b/parse/train/7J-fKoXiReA/7J-fKoXiReA_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..823e92d88284c6da777dceba19500d2004d467f3 --- /dev/null +++ b/parse/train/7J-fKoXiReA/7J-fKoXiReA_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7e840d68159eace3fa3704c865ae68f7a836a09bdc4354ff2c5170a21aa3227 +size 1995067 diff --git a/parse/train/7J-fKoXiReA/7J-fKoXiReA_span.pdf b/parse/train/7J-fKoXiReA/7J-fKoXiReA_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4d27b3643049988e020dac0cd7a8fe613fd2f529 --- /dev/null +++ b/parse/train/7J-fKoXiReA/7J-fKoXiReA_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:281b19961d42892a617f775ce0d69ece788be508d204eab8e584b70f2c0a3287 +size 2250827 diff --git a/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_layout.pdf b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fd13c295b2b7a30b32242ef9cf85cd49b45623a5 --- /dev/null +++ b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b43a91dfa71fc68f1f75720c4927d265607d7880eee1d43c05a54d0b1db1256f +size 670893 diff --git a/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_origin.pdf b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d820aa073830affef51240f4404d27b8801f919d --- /dev/null +++ b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0f26524e4f11642d71be3efad7bdb4827cd923f52913b7914cb6724c2d35d08 +size 469981 diff --git a/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_span.pdf b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fcc63e5390db8edf494985fd2ffd0ccbb8cbf0c6 --- /dev/null +++ b/parse/train/Ad7Gv5NkBPz/Ad7Gv5NkBPz_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ab5c25d0101ed3cc934b5f0d8e01075ed5cabf86c8e3e55b86a3b35eff7b836 +size 705628 diff --git a/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_layout.pdf b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c8a4eaa4ad369428a9c831f204088c211e89f00f --- /dev/null +++ b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:446e1066b468a4810f11d43c48d379dad06d3d2d94cde4cd90605d050d0d4f51 +size 838376 diff --git a/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_origin.pdf b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..21b7331e30eb145da58d3fec6287872e9ea97490 --- /dev/null +++ b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4304d7fb4b975c6c23c05dc8223d1e45ee4375fa56f6e638744848dc2437eb6d +size 605057 diff --git a/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_span.pdf b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..481d6057a3dd46b369772b953f44de0a6bb0ecf0 --- /dev/null +++ b/parse/train/Arn2E4IRjEB/Arn2E4IRjEB_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8e3d9579527afd745da6ba934b486858cd0dfbe4602711b95815e8e811503df +size 851609 diff --git a/parse/train/B1eXygBFPH/B1eXygBFPH_layout.pdf b/parse/train/B1eXygBFPH/B1eXygBFPH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d0535d319cd4bc7032b6aef9e68019fe1bb3a4ed --- /dev/null +++ b/parse/train/B1eXygBFPH/B1eXygBFPH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:298e19a875a0281fb1191dfa6adc8507fe033e77ef85df039fb1d303e16d50b1 +size 580450 diff --git a/parse/train/B1eXygBFPH/B1eXygBFPH_origin.pdf b/parse/train/B1eXygBFPH/B1eXygBFPH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e36f558b346abb493e66d4e2881f8ffdb2d64473 --- /dev/null +++ b/parse/train/B1eXygBFPH/B1eXygBFPH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:daeeed83cdcd8e26e8c64c06c444af9830feb727b532fcdb9334a75cca29a206 +size 428701 diff --git a/parse/train/B1eXygBFPH/B1eXygBFPH_span.pdf b/parse/train/B1eXygBFPH/B1eXygBFPH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8b884515c9ca854acc7a1e731bc5f875a8afff89 --- /dev/null +++ b/parse/train/B1eXygBFPH/B1eXygBFPH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6482ad6578f443979e505c2d9d15b80e17203c1aee24880b611301bd2807ee24 +size 590310 diff --git a/parse/train/BJg7x1HFvB/BJg7x1HFvB_origin.pdf b/parse/train/BJg7x1HFvB/BJg7x1HFvB_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..14254def8b07dad7c9db56c83307816b0ba72a60 --- /dev/null +++ b/parse/train/BJg7x1HFvB/BJg7x1HFvB_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a45e1fd097db5489bbcdcfca254e087d9d7cc74d56d059fcbab91462b77a7791 +size 379561 diff --git a/parse/train/BJg7x1HFvB/BJg7x1HFvB_span.pdf b/parse/train/BJg7x1HFvB/BJg7x1HFvB_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f94f226b0c16204ea15d940dba48d430a35fc7e7 --- /dev/null +++ b/parse/train/BJg7x1HFvB/BJg7x1HFvB_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5917369f61f5a0670e875b2e9cb14934816dd17cf0d3e710dc7868edeb757c2a +size 584100 diff --git a/parse/train/BJlxmAKlg/BJlxmAKlg_layout.pdf b/parse/train/BJlxmAKlg/BJlxmAKlg_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1226eee7a092568c2cfd52ba531340a884530ecc --- /dev/null +++ b/parse/train/BJlxmAKlg/BJlxmAKlg_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adfe12daa702a113d6e6cf671292f54a4aff75ad8dfaee017116e017f21665f8 +size 1184896 diff --git a/parse/train/BJlxmAKlg/BJlxmAKlg_origin.pdf b/parse/train/BJlxmAKlg/BJlxmAKlg_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..085c0f58e38433cb63bd61e1a14ead9098dc074b --- /dev/null +++ b/parse/train/BJlxmAKlg/BJlxmAKlg_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5be958bb57ee2a868ac6289d9e6aa3b26a37c8f21a7a47f46a8fac843d335c9b +size 1054776 diff --git a/parse/train/BJlxmAKlg/BJlxmAKlg_span.pdf b/parse/train/BJlxmAKlg/BJlxmAKlg_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..17581f5cb19752410537dd2b5805557413cd443d --- /dev/null +++ b/parse/train/BJlxmAKlg/BJlxmAKlg_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e134a990d830ba6d4d42abc86dd7bfb5364f8f1e497741536d1bfa0d09ed76ec +size 1189439 diff --git a/parse/train/BJlzm64tDH/BJlzm64tDH_layout.pdf b/parse/train/BJlzm64tDH/BJlzm64tDH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d9254ab5b6855ae6de1e28b6ab9008d73a9a42dc --- /dev/null +++ b/parse/train/BJlzm64tDH/BJlzm64tDH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9aea0964ed97c05d87299fc9084e4be4e6c87a5296440d847a9ffd5b98c04c02 +size 1736423 diff --git a/parse/train/BJlzm64tDH/BJlzm64tDH_origin.pdf b/parse/train/BJlzm64tDH/BJlzm64tDH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..67f597f316c74589e12e08286960dd4d2d7fc554 --- /dev/null +++ b/parse/train/BJlzm64tDH/BJlzm64tDH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d659f1aaf0dd4a3703adf9f8cd13ae6a70580f5d391b0abd45ab4fd4c6aec77 +size 1611506 diff --git a/parse/train/BJlzm64tDH/BJlzm64tDH_span.pdf b/parse/train/BJlzm64tDH/BJlzm64tDH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..836c7db9a0aeb72da856da82710886d271f4a62d --- /dev/null +++ b/parse/train/BJlzm64tDH/BJlzm64tDH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60730f1c7c4303f167f922b0560b82dd56568b099acf5998d003aac0c50733d9 +size 1736827 diff --git a/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_layout.pdf b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f805f2a7e047ede1eb1fed3dee13626bf400bb7a --- /dev/null +++ b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b443860622b127e09f3058bbf5d45383ce842c51c81aa439214af39615b5867 +size 7522928 diff --git a/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_origin.pdf b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d14671966f121440543009ed0fdcfbc28f3c581c --- /dev/null +++ b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64dd90a04a7baca3ff880777e8acb1fa3d4bf9e043a438125ac3d7690e46c925 +size 7318943 diff --git a/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_span.pdf b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..14467e0931d68f95d098bac337faf3ce6451ca0d --- /dev/null +++ b/parse/train/Bk8ZcAxR-/Bk8ZcAxR-_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b9436c50d802a56cebd8759e4cf0bf7539311ce2f384fba9e30d784c8a378a6 +size 7526082 diff --git a/parse/train/BkfbpsAcF7/BkfbpsAcF7_layout.pdf b/parse/train/BkfbpsAcF7/BkfbpsAcF7_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e10840e4887e4c7e5b3e4f00187a694fc16e23e0 --- /dev/null +++ b/parse/train/BkfbpsAcF7/BkfbpsAcF7_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a4e3660e5ce1b0697e728aca2abd1fcb712f50f318ba5908a0312d9d9a546d4 +size 6286763 diff --git a/parse/train/BkfbpsAcF7/BkfbpsAcF7_origin.pdf b/parse/train/BkfbpsAcF7/BkfbpsAcF7_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d9a8efc5b90054c053e8ad2a5690371cf8c7d0b4 --- /dev/null +++ b/parse/train/BkfbpsAcF7/BkfbpsAcF7_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36b69fdc80dbfe4d7b21d2e590d206fdd3804cc15e27630bec906b31580b55c1 +size 6088008 diff --git a/parse/train/BkfbpsAcF7/BkfbpsAcF7_span.pdf b/parse/train/BkfbpsAcF7/BkfbpsAcF7_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..057754f2a120d9680bd589c987dad68b9cf89cc6 --- /dev/null +++ b/parse/train/BkfbpsAcF7/BkfbpsAcF7_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97eca7c6f362eebd852eb5cd5e67d4695ed2e733d9d317b1d714d17f759a5001 +size 6295612 diff --git a/parse/train/ByZvfijeg/ByZvfijeg_layout.pdf b/parse/train/ByZvfijeg/ByZvfijeg_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1a6b5f07fd7ad4bae0e977ab39dade1be4465bae --- /dev/null +++ b/parse/train/ByZvfijeg/ByZvfijeg_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1f313f12d739684d9620d4b3d7a16277e9a2acafbdd8506787c4b6e1a2657ec +size 363213 diff --git a/parse/train/ByZvfijeg/ByZvfijeg_origin.pdf b/parse/train/ByZvfijeg/ByZvfijeg_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c3101d3f1d71e4adc03341e27e9297d5a3abbdb9 --- /dev/null +++ b/parse/train/ByZvfijeg/ByZvfijeg_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aba67280166c3ad53d8aac3c94dd7c99f5a1dd4c363b421f38b98c2d5826e333 +size 277756 diff --git a/parse/train/ByZvfijeg/ByZvfijeg_span.pdf b/parse/train/ByZvfijeg/ByZvfijeg_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bc42ef60030b7e864d74ccf728bbfdfce9debd4b --- /dev/null +++ b/parse/train/ByZvfijeg/ByZvfijeg_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdcd1b8cebff773baf4f99f731cad2100afdbb8e30f78fe0af7ab195dc9e57bc +size 367058 diff --git a/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_layout.pdf b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a6b3b68a66031bd038fc711d3fd731a6e3cda9b7 --- /dev/null +++ b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df85998b64c552bf8d403ff584c2105fcb8c68bac5890677ad984d0e97801d32 +size 1112568 diff --git a/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_origin.pdf b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6e0aa96dee2621644c7c59e39dacfc458c0c3a44 --- /dev/null +++ b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f0d8a27198d22783d72026c9ebba892b3fb5faa73c1d1f55b389b2ddbb4e326 +size 878744 diff --git a/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_span.pdf b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6b7182c343187985eb4e11f3c0f3f3122c43ddfb --- /dev/null +++ b/parse/train/ByeZ5jC5YQ/ByeZ5jC5YQ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c4563e5d345b90c37363fcdede519f3e36e63b88a77ebfef0caa0c1eeaa9418 +size 1122033 diff --git a/parse/train/BygWRaVYwH/BygWRaVYwH_layout.pdf b/parse/train/BygWRaVYwH/BygWRaVYwH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2f45d7fb96ed52bba3d317b87ece6142eb2bfea6 --- /dev/null +++ b/parse/train/BygWRaVYwH/BygWRaVYwH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abb062b37d281c7ccd29c0bdcad16d14f9743d78af5d784cc5311ed41bba2e09 +size 2637909 diff --git a/parse/train/BygWRaVYwH/BygWRaVYwH_origin.pdf b/parse/train/BygWRaVYwH/BygWRaVYwH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e3ae29751dce366f95ef78360c4e6b16a7bb8c70 --- /dev/null +++ b/parse/train/BygWRaVYwH/BygWRaVYwH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f219c1a8dcaa36f7f8152e16a306ed05af252245cb43645e4e47cc0f941fffd8 +size 2454055 diff --git a/parse/train/BygWRaVYwH/BygWRaVYwH_span.pdf b/parse/train/BygWRaVYwH/BygWRaVYwH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ed93365e94568d9a8fd46c5fd8c6c8a2412b0a14 --- /dev/null +++ b/parse/train/BygWRaVYwH/BygWRaVYwH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9e42b4f175aa576613e2c3a0d9a5482cfc2f0c0795958748b1e39125f0ff410 +size 2645993 diff --git a/parse/train/Bygh9j09KX/Bygh9j09KX_layout.pdf b/parse/train/Bygh9j09KX/Bygh9j09KX_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2fd09263b7c20e893889176a2201eebe5dd4df12 --- /dev/null +++ b/parse/train/Bygh9j09KX/Bygh9j09KX_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76552d089d9c469b93826c01e45e9220d8fec068b848bd2cfb008e9ef50f5c4c +size 6462578 diff --git a/parse/train/Bygh9j09KX/Bygh9j09KX_origin.pdf b/parse/train/Bygh9j09KX/Bygh9j09KX_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..abe03308ce8a841c7c92c25110ba6238d24539e9 --- /dev/null +++ b/parse/train/Bygh9j09KX/Bygh9j09KX_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23cdff8ef2034d12a1e0b424fe6d4021bf6f876e69a3d15971ab40e205a58aab +size 6297869 diff --git a/parse/train/Bygh9j09KX/Bygh9j09KX_span.pdf b/parse/train/Bygh9j09KX/Bygh9j09KX_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..03dfc03dc80a24bee663d2e548ac78166bbe04c8 --- /dev/null +++ b/parse/train/Bygh9j09KX/Bygh9j09KX_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6eb227265556bf5ae856a8b20646aa396131650abba21d68a4e4337da2d271cc +size 6465880 diff --git a/parse/train/Bygq-H9eg/Bygq-H9eg_layout.pdf b/parse/train/Bygq-H9eg/Bygq-H9eg_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..434d4340960ad34ceeb9cb4f63bf48767420d544 --- /dev/null +++ b/parse/train/Bygq-H9eg/Bygq-H9eg_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:600647a02cca64167d65d56bdae2f20c1ee4fae1dd6b5ecd51381e31bbcaa5c2 +size 365577 diff --git a/parse/train/Bygq-H9eg/Bygq-H9eg_origin.pdf b/parse/train/Bygq-H9eg/Bygq-H9eg_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ac4ed830b69647c2fb138c99dc1f86d9cb5f30cd --- /dev/null +++ b/parse/train/Bygq-H9eg/Bygq-H9eg_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0519d3c7fe27db47ff3a7be7994bf603e8502f626e25a47dca472fa7dce6eccb +size 312482 diff --git a/parse/train/Bygq-H9eg/Bygq-H9eg_span.pdf b/parse/train/Bygq-H9eg/Bygq-H9eg_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0398e72d9dfa71728f931b07ec93a9654c34673e --- /dev/null +++ b/parse/train/Bygq-H9eg/Bygq-H9eg_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9164d69bf48c66fca91a5372ffa6e310224feaa133be768bb45c8728a95f96bf +size 365847 diff --git a/parse/train/Bys_NzbC-/Bys_NzbC-_layout.pdf b/parse/train/Bys_NzbC-/Bys_NzbC-_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b5690e59516491d61b6e97193ec9e798e482a72c --- /dev/null +++ b/parse/train/Bys_NzbC-/Bys_NzbC-_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2ff4964feecfbd802fb30b49d0ba49e1b8b77659f6a8f3b26319214af8c5bfb +size 1409145 diff --git a/parse/train/Bys_NzbC-/Bys_NzbC-_origin.pdf b/parse/train/Bys_NzbC-/Bys_NzbC-_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f3377329ceed3fd928f0b4bc152eafb3e6e6c151 --- /dev/null +++ b/parse/train/Bys_NzbC-/Bys_NzbC-_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:005246f5c3a151b1baa94b49a29d33c340437b92de8f6f9f268246ca896d859b +size 393712 diff --git a/parse/train/Bys_NzbC-/Bys_NzbC-_span.pdf b/parse/train/Bys_NzbC-/Bys_NzbC-_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b63544889f6ab11986d045b59b658f258e3abed2 --- /dev/null +++ b/parse/train/Bys_NzbC-/Bys_NzbC-_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e8f4a0c079c16184bd1f38995c1d1d9f45d6b1e722ee5e800e5d723a58d6976 +size 1412344 diff --git a/parse/train/ByxBFsRqYm/ByxBFsRqYm_layout.pdf b/parse/train/ByxBFsRqYm/ByxBFsRqYm_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b2639529eb1c48ee1c20912261bda20263fff6d6 --- /dev/null +++ b/parse/train/ByxBFsRqYm/ByxBFsRqYm_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1558927159c62cbdc8faba1e1379505602276de49ccf8e1dfd25ce87784ef8d7 +size 1749150 diff --git a/parse/train/ByxBFsRqYm/ByxBFsRqYm_origin.pdf b/parse/train/ByxBFsRqYm/ByxBFsRqYm_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..eaa09769511335f12ac5db8eed671e3f92bf3e06 --- /dev/null +++ b/parse/train/ByxBFsRqYm/ByxBFsRqYm_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c1cc45a2a3b0421fef7158d4c9c183822bce9c6a326fbf071cbc44975bdd2ed +size 1420200 diff --git a/parse/train/ByxBFsRqYm/ByxBFsRqYm_span.pdf b/parse/train/ByxBFsRqYm/ByxBFsRqYm_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d994465e37681d8049784339fef0e833b605dacd --- /dev/null +++ b/parse/train/ByxBFsRqYm/ByxBFsRqYm_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5437a0ccf15c345ffaede5266a69e0ec9d6e305fc765a63b55df457c48513230 +size 1768050 diff --git a/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_layout.pdf b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2ac55dd84077914cc6cc7e3467000df586000abf --- /dev/null +++ b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7abb22d4a23f82de23747695e89c4f0453ae80c546820f54acbb02eccede30da +size 2152895 diff --git a/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_origin.pdf b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..09c278a8497c8ef4552f09c86090b4474bcaf86e --- /dev/null +++ b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:412912059af7e05ee4a9a3f970c6aeebaa8b1de4a8f5707ac5cddbdd56cce377 +size 1966787 diff --git a/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_span.pdf b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0617ea4c71f9f87854e0634e41190e51c6c7f77b --- /dev/null +++ b/parse/train/Cnon5ezMHtu/Cnon5ezMHtu_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4093139588d4c42bed33b19d75610e526fdddfaded95a2cfb37fa06b9d811337 +size 2158118 diff --git a/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_layout.pdf b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f12d648e0f668dc6549cea55afe7a20325309360 --- /dev/null +++ b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30984a088a70196d73b2c5de340731abf211100f876a718c352a252420189a8c +size 1624820 diff --git a/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_origin.pdf b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..de430165076a723f5e47f1ed984832454111db76 --- /dev/null +++ b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5dfc6c6724c4261b980493608a918becf24eb942c07cbbc607adce935433f91 +size 1376447 diff --git a/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_span.pdf b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4d85a7cbcd1b4821ba4300db72e89396c831018c --- /dev/null +++ b/parse/train/DAaaaqPv9-q/DAaaaqPv9-q_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04b025979c3288932c93e6b074c74408b8981907ef78031165f53285562add6d +size 1624113 diff --git a/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_layout.pdf b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..af8565511ad5c7241e7cac830375061a9bbc05ae --- /dev/null +++ b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e44b5a176c76ce6121cde8a0b4673af8c79e7967272d1932769a7c68bf18f03a +size 11933566 diff --git a/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_origin.pdf b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4df4ff6ad1733c0f6c6e738c5f0622e9e2e1796e --- /dev/null +++ b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0f3e83df4151ead6e286930c2b6dee0926cf7f4ca5f213e80922d0ca984f49f +size 11458407 diff --git a/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_span.pdf b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..93ada06a94bde63f4e6dcfa8d0e51c60f0797035 --- /dev/null +++ b/parse/train/EbIDjBynYJ8/EbIDjBynYJ8_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ccbc2d15974fcbbaebe3b9a6ed44a2fb5e87382da8c5ecb61b306ac63a7bb907 +size 11948675 diff --git a/parse/train/H1gBsgBYwH/H1gBsgBYwH.md b/parse/train/H1gBsgBYwH/H1gBsgBYwH.md new file mode 100644 index 0000000000000000000000000000000000000000..40739c80e580ed99ae2c50c1a53b49035b333eea --- /dev/null +++ b/parse/train/H1gBsgBYwH/H1gBsgBYwH.md @@ -0,0 +1,1569 @@ +# GENERALIZATION OF TWO-LAYER NEURAL NET-WORKS: AN ASYMPTOTIC VIEWPOINT + +Jimmy $\mathbf { B a } ^ { 1 , 2 }$ , Murat A. Erdogdu1,2, Taiji Suzuki3,4, Denny $\mathbf { W _ { u } } 1 , 2 , 4$ , Tianzong Zhang2,5 University of Toronto1, Vector Institute2, University of Tokyo3, RIKEN AIP4, Tsinghua University5 {jba,erdogdu,dennywu}@cs.toronto.edu, taiji@mist.i.u-tokyo.ac.jp, ztz16@mails.tsinghua.edu.cn + +# ABSTRACT + +This paper investigates the generalization properties of two-layer neural networks in high-dimensions, i.e. when the number of samples $n$ , features $d$ , and neurons $h$ tend to infinity at the same rate. Specifically, we derive the exact population risk of the unregularized least squares regression problem with two-layer neural networks when either the first or the second layer is trained using a gradient flow under different initialization setups. When only the second layer coefficients are optimized, we recover the double descent phenomenon: a cusp in the population risk appears at $h \approx n$ and further overparameterization decreases the risk. In contrast, when the first layer weights are optimized, we highlight how different scales of initialization lead to different inductive bias, and show that the resulting risk is independent of overparameterization. Our theoretical and experimental results suggest that previously studied model setups that provably give rise to double descent might not translate to optimizing two-layer neural networks. + +# 1 INTRODUCTION + +In modern neural networks, the number of parameters can easily exceed the number of training samples, yet in many circumstances, there is little sign of overfitting even in the absence of explicit regularization (Zhang et al., 2016). This phenomenon is usually explained by the interplay between the model architecture and the optimization method. Existing works have analyzed the implicit regularization of gradient descent on simple models (Gunasekar et al., 2018; Ji and Telgarsky, 2018), and provided generalization guarantees (Arora et al., 2018; Bartlett et al., 2017; Dziugaite and Roy, 2017) that align with the empirical observations. + +Recently, a series of works highlighted the implicit regularization of interpolators in the overparameterized regime (Belkin et al., 2018; Spigler et al., 2018; Geiger et al., 2018; Advani and Saxe, 2017). Specifically, a second decrease in the population risk is observed when the model is further overparameterized beyond the interpolation limit, i.e. when the model achieves zero training error. This phenomenon is known as double descent, and can be precisely quantified for certain linear models (Hastie et al., 2019; Mei and Montanari, 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Among the recent works, Hastie et al. (2019) and Mei and Montanari (2019) explicitly derived the population risk of linear regression and random features regression models in high dimensions using tools from random matrix theory. + +However, there is still a gap between the practical benefit of overparameterization and the recently proved double descent phenomenon, which is typically established under models that exhibits the following structure: the trained model solves a linear inverse problem, and the “cusp” in the risk arises from the instability of the inverse at the interpolation threshold. Moreover, given a dataset or fixed $n , d .$ , the number of parameters in the linear regression model is also fixed, i.e. the level of overparameterization cannot be altered. It is therefore unclear if the trend persists in the optimization of more complex models, for instance in two-layer neural networks where overparameterization can be controlled simply by adding more neurons. + +In this work, we analyze the generalization properties of two-layer neural networks in the unregularized least squares regression setting and examine the presence/absence of the double descent phenomenon. We consider the proportional asymptotic limit where the number of samples $n$ , input features $d$ , and neurons $h$ tend to infinity at the same rate, under which overparameterization corresponds to increasing the limit of $h / n$ (network “width”). This regime is particularly interesting because even though $n \to \infty$ , the empirical risk is not equivalent to the population risk. In addition, the joint scaling of $n , d , h$ is parallel to the practical choice of model architectures, where it is common to train a larger network when the number of samples and input features are larger. Following Hastie et al. (2019), we assume unit Gaussian input and noisy linear observations, and analytically derive the population risk of the solution of gradient flow on either the first or the second layer parameters when the flow is initialized close to zero. + +Our findings can be summarized as follows (see Figure 1): + +• When only the second layer is optimized, we derive the risk in its bias-variance decomposition and demonstrate the presence of the double descent phenomenon. • When the first layer is optimized, we compare two solutions of gradient flow from different scales of initialization, which we term as vanishing and non-vanishing initialization, and show in both cases the population risk is independent to overparameterization. • For the vanishing initialization, we show that the risk of the gradient flow solution is asymptotically close to that of a rank-1 model. For non-vanishing initialization, we show that the gradient flow solution is wellapproximated by a kernel model and derive the risk. + +![](images/fdb2f7294504073f6249cfffccb6993d35f35a22379caec2e8d08b45a4316c5a.jpg) +Figure 1: Illustration of the double descent risk curve in two-layer linear networks $( \mathrm { S N R } = 1 6 $ ). Brighter color indicates larger $\gamma _ { 1 } = d / \bar { n }$ . Double descent is observed when the second layer coefficients are optimized (main figure), but not when the first layer weights are optimized (subfigure). + +# 1.1 RELATED WORKS + +Global Convergence of Two-layer Networks. A plethora of recent works have explored the global convergence of shallow neural networks. Mei et al. (2018; 2019); Chizat and Bach (2018a); Rotskoff and Vanden-Eijnden (2018); Sirignano and Spiliopoulos (2018); Nitanda and Suzuki (2017) studied the mean-field limit where the number of neurons $h \to \infty$ and the second layer scaled by $1 / h$ , and established correspondence between the main-particle limit of gradient descent and Wasserstein gradient flow to demonsrate global convergence. On the other hand, Jacot et al. (2018); Du et al. (2018); Oymak and Soltanolkotabi (2019); Allen-Zhu et al. (2018b); Song and Yang (2019) considered a different scaling and showed that gradient descent on overparameterized models converges to global minimizer at a linear rate; key to these results is an observation that optimization via gradient descent is asymptotically equivalent to kernel regression with respect to the neural tangent kernel. + +Active vs. Lazy Training. Following Chizat and Bach (2018b), we refer to the two aforementioned scalings as the active and lazy (kernel) regime. It has been observed that different regimes lead to contrasting inductive biases. Williams et al. (2019); Woodworth et al. (2019); Li et al. (2017) showed that for certain two-layer network or overparameterized linear model, the scale of initialization controls the implicit regularization of gradient descent (from sparse to smooth solution). In the student-teacher setup (Tian, 2017; Zhong et al., 2017), Ghorbani et al. (2019b;a) showed that kernel models in high dimensions perform no better than low-degree polynomials on the input or fullytrained two-layer network. Additionally, Suzuki (2018); Allen-Zhu and Li (2019); Yehudai and Shamir (2019); Wei et al. (2018) demonstrated that neural network outperforms linear estimators (including kernel method) in learning various target functions. The difference between fixed bases and adaptive bases mirrors the difference in optimizing the first or second layer in our setup. + +Generalization of Overparameterized Models. It is often observed that overparameterization does not result in overfitting (Neyshabur et al., 2014). In the lazy regime, generalization guarantees can be derived from the distance traveled by the parameters (Neyshabur et al., 2018; Nagarajan and Kolter, 2019), which becomes small if the model is sufficiently overparameterized (Arora et al., 2019b; Li and Liang, 2018; Allen-Zhu et al., 2018a; Cao and Gu, 2019). Compared to these guarantees that usually require significant overparameterization, our result relies on stronger data assumptions, but consequently we obtain the exact population risk instead of a vacuous upper-bound. Beyond the kernel regime, Advani and Saxe (2017); Goldt et al. (2019) analyzed the generalization dynamics of overparameterized models in the student-teacher setup. + +Double Descent. The term double descent refers to the phenomenon that the population risk of an empirical risk minimizer manifests a "cusp" at the interpolation threshold, and further overparameterization decreases the risk. First observed in Krogh and Hertz (1992), the phenomenon has been recently connected to the benefit of overparameterization (Belkin et al., 2018; Geiger et al., 2018; Spigler et al., 2018; Advani and Saxe, 2017), and can be precisely characterized for certain simple models (Hastie et al., 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Our work is inspired by Hastie et al. (2019) which uses random matrix theory to derive the asymptotic risk for linear and random feature models. Concurrent to our work, Mei and Montanari (2019) analyzed the random features model and derived its population risk for which double descent occurs both in bias and variance. This aligns with our results on optimizing the second layer in Section 4 although we do not derive the bias component explicitly. Compared to Hastie et al. (2019); Mei and Montanari (2019), the focus of this work is to highlight the different generalization property of models obtained from optimizing different layers of the network and from different initialization. + +Random Matrix Theory. High-dimensional models, including kernel models and neural networks, can be analyzed by studying the properties of random matrices. El Karoui et al. (2010); Cheng and Singer (2013); Fan and Montanari (2019) studied the spectral properties of kernel matrix via decomposing the nonlinearity with Taylor series or Hermite polynomials, which in turn explains the generalization of high-dimensional kernel ridgeless interpolators (Liang and Rakhlin, 2018). In addition, similar tools have been used to study two-layer neural networks (Louart et al., 2018; Pennington and Worah, 2017) and related quantities such as the Fisher information matrix (Karakida et al., 2018; Pennington and Worah, 2018). + +# 2 PRELIMINARIES: TWO-LAYER NEURAL NETWORK + +Consider the following bias-free two-layer neural network $f : { \mathbb { R } ^ { d } } \to { \mathbb { R } }$ with $h$ hidden units + +$$ +f ( \pmb { x } ) = \sum _ { i = 1 } ^ { h } a _ { i } \phi ( \langle \pmb { x } , \pmb { w } _ { i } \rangle ) , +$$ + +where $\pmb { x } \in \mathbb { R } ^ { d }$ is the input, $\boldsymbol { w } _ { i } \in \mathbb { R } ^ { d }$ is the weights corresponding to neuron $i$ , $a _ { i } \in \mathbb { R }$ is the $i$ -th coefficient of the second layer, and $\phi : \mathbb { R } \mathbb { R }$ is a Lipschitz continuous activation function with bounded Gaussian moments, i.e. $\mathbb { E } [ \phi ( G ) ^ { k } ] < \infty$ , $\forall k \in \mathbb { Z } _ { + }$ for $G \sim \mathcal { N } ( 0 , 1 )$ . For concise notation, we write $W = [ { \pmb w } _ { 1 } , . . . { \pmb w } _ { h } ] \in \mathbb { R } ^ { d \times h }$ for the weight matrix, $\pmb { a } = [ a _ { 1 } , . . . a _ { h } ] \in \mathbb { R } ^ { h }$ for the coefficient vector, $X = [ { \pmb x } _ { 1 } , . . . { \pmb x } _ { n } ] \in \mathbb { R } ^ { d \times n }$ for the data matrix, $\ b { y } \in \mathbb { R } ^ { n }$ for the corresponding vector of labels, and $\Phi = \phi ( W ^ { \top } X ) \in \mathbb { R } ^ { h \times n }$ for the feature matrix at the first layer. We omit arguments of $f$ when they are clear from the context. + +We consider a student-teacher setup, in which data is generated by a teacher model $F : { \mathbb { R } ^ { d } } \to { \mathbb { R } }$ with additive noise, and the student model aims to minimize the squared loss: + +$$ +( { \pmb x } _ { i } , \varepsilon _ { i } ) \overset { \mathrm { i . i . d . } } { \sim } P _ { { \pmb x } } \times P _ { \varepsilon } , \quad y _ { i } = F ( { \pmb x } _ { i } ) + \varepsilon _ { i } , \quad L ( { \boldsymbol X } ; f ) = \frac { 1 } { 2 n } \sum _ { i = 1 } ^ { n } \left( y _ { i } - f ( { \pmb x } _ { i } ) \right) ^ { 2 } , +$$ + +where $\mathbb { E } [ { \pmb x } _ { i } ] = 0$ , $\mathrm { C o v } ( { \pmb x } _ { i } ) = \Sigma$ , $\mathbb { E } [ \varepsilon _ { i } ] = 0$ , $\mathrm { V a r } ( \varepsilon _ { i } ) = \sigma ^ { 2 }$ . We are interested in the population risk $R ( f ) = \mathbb { E } _ { P _ { x } } [ ( F ( { \pmb x } ) - f ( { \pmb x } ) ) ^ { 2 } ]$ . Our analysis will be made under the proportional asymptotics: + +$$ +n , d , h \infty ; \quad d / n \gamma _ { 1 } , h / n \gamma _ { 2 } ; \quad \gamma _ { 1 } , \gamma _ { 2 } \in ( 0 , \infty ) , +$$ + +in which overparameterization corresponds to increasing $\gamma _ { 2 }$ . Thus the characteristics of double descent considered in this work are: 1) large population risk as $\gamma _ { 2 } 1 ; 2$ ) decrease in the risk for $\gamma _ { 2 } > 1$ . While the empirical risk can be minimized in various ways, we analyze the solution of gradient flow, in which we update either the first layer $W$ or the second layer $\textbf { \em a }$ : + +$$ +\mathrm { d } W ( t ) = - \nabla _ { W } L ( X ; f ) \mathrm { d } t \quad \mathrm { o r } \quad \mathrm { d } a ( t ) = - \nabla _ { a } L ( X ; f ) \mathrm { d } t , +$$ + +from small initialization. The rest of the paper is organized as follows. In Section 3, we start with a simple example of two-layer linear network as warm-up. In Section 4, we consider optimizing the second layer coefficients (flow over $\textbf { \em a }$ ) of a non-linear two-layer neural network under fixed Gaussian first layer, which is a random feature model. Section 5 considers optimizing the first layer weights (flow over $W$ ) of such network under fixed Rademacher second layer. We defer all proofs and details on experiments to appendix. + +# 3 WARM-UP: LINEAR NETWORK + +We begin with a simple linear model with $\phi ( { \pmb x } ) = { \pmb x }$ , i.e. $\Phi = W ^ { \top } X$ . We remark that although the model is linear, the solution obtained by gradient flow on the two-layer model can be different than that from directly solving the linear regression problem on input features. + +Training the Second Layer. Following Hastie et al. (2019), we fix the first layer parameters to be randomly drawn from a unit Gaussian and optimize the coefficients $\textbf { \em a }$ by minimizing $\left| \left| \pmb { a } ^ { \top } \Phi - \pmb { y } \right| \right| _ { 2 } ^ { 2 }$ . The following lemma characterizes the solution of the gradient flow. + +Lemma 1 (Least squares solution). Given data matrix $X _ { i }$ , response vector $\textbf { { y } }$ and model $f ( { \pmb x } ) =$ $\langle \phi ( \pmb { x } ^ { \top } W ) , \hat { \pmb { a } } \rangle$ with fixed first layer coefficients $W$ , gradient flow on the coefficients $\textbf { \em a }$ starting from zero initialization converges to $\mathbf { \bar { a } } = \Phi ^ { \dagger } \mathbf { y }$ , where $\dagger$ stands for the Moore-Penrose inverse. + +We make two assumptions on the data and the teacher model to simplify the computation. + +(A1) Gaussian Features: $\pmb { x } _ { i } \sim \mathcal { N } ( 0 , I _ { d } )$ ; (A2) Linear Teacher: $F ( { \pmb x } ) = \langle { \pmb x } , { \pmb \beta } \rangle$ , $\| { \boldsymbol { \beta } } \| = r$ + +Denote the linear student network as $f ( \pmb { x } ) = \langle \pmb { x } , \hat { \beta } \rangle$ , where $\hat { \boldsymbol { \beta } } = W \hat { \mathbf { a } }$ and $\hat { \textbf { \textit a } }$ is the least-square solution defined by Lemma 1. We write the population risk in its bias-variance decomposition. + +$$ +R = \mathbb { E } _ { { \mathbf { x } } \sim P _ { \mathbf { x } } } [ \Vert \hat { \beta } - \beta \Vert _ { \Sigma } ^ { 2 } \vert { \cal X } , { \cal W } ] = \underbrace { \Vert \mathbb { E } [ \hat { \beta } \vert { \cal X } , { \cal W } ] - \beta \Vert _ { 2 } ^ { 2 } } _ { B = \mathrm { b i a s } } + \underbrace { \mathrm { t r } \left( \mathrm { C o v } ( \hat { \beta } \vert { \cal X } , { \cal W } ) \right) } _ { { \cal V } = \mathrm { v a r i a n c e } } , +$$ + +where $\left\| \pmb { x } \right\| _ { \Sigma } ^ { 2 } = \pmb { x } ^ { \top } \Sigma \pmb { x }$ . We compute the bias and the variance separately to obtain the risk. + +Theorem 2. Given (A1)(A2) and let ${ \pmb w } _ { i }$ i.i.d. ∼ $\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )$ , at $n , d , h \infty$ we have + +$$ +R _ { ( \gamma _ { 1 } < 1 ) } \to \{ \begin{array} { l l } { \frac { \gamma _ { 1 } - \gamma _ { 2 } } { \gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \frac { \gamma _ { 2 } } { g _ { 2 } } \sigma ^ { 2 } , } & { \gamma _ { 2 } < \gamma _ { 1 } , } \\ { \frac { \gamma _ { 1 } } { g _ { 1 } } \sigma ^ { 2 } , } & R _ { ( \gamma _ { 1 } > 1 ) } \to \{ \begin{array} { l l } { \frac { \gamma _ { 1 } - \gamma _ { 2 } } { \gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \frac { \gamma _ { 2 } } { g _ { 2 } } \sigma ^ { 2 } , } & { \gamma _ { 2 } < 1 , } \\ { ~ } & { R _ { ( \gamma _ { 1 } > 1 ) } \to \{ \begin{array} { l l } { \frac { \gamma _ { 1 } - \gamma _ { 2 } } { \gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \frac { \gamma _ { 2 } } { g _ { 2 } } \sigma ^ { 2 } , } & { \gamma _ { 2 } < 1 , } \\ { ~ } & { R _ { ( \gamma _ { 1 } > 1 ) } \to \{ \begin{array} { l l } { \frac { \gamma _ { 2 } g _ { 1 } } { \gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \frac { g _ { 1 } + g _ { 2 } } { g _ { 1 } g _ { 2 } } \sigma ^ { 2 } , } & { \gamma _ { 2 } > 1 . } \end{array} } \end{array} } \end{array} \end{array} +$$ + +where $d / n \to \gamma _ { 1 } , h / n \to \gamma _ { 2 } , g _ { 1 } = | \gamma _ { 1 } - 1 | ,$ , and $g _ { 2 } = | \gamma _ { 2 } - 1 |$ . + +We observe that when $d > n$ (i.e. $\gamma _ { 1 } > 1$ ), we obtain the double descent risk curve, i.e., the population risk achieves its maximum at $\gamma _ { 2 } 1$ and further overparameterization $( \gamma _ { 2 } > 1 )$ ) reduces both the bias and the variance. Conversely when $n > d$ and $h > d$ (i.e. $\gamma _ { 1 } < \operatorname* { m i n } ( 1 , \gamma _ { 2 } ) )$ , the population risk becomes constant and equals to that of the minimum-norm solution ${ \hat { \boldsymbol { \beta } } } _ { \operatorname* { m i n } } = { \boldsymbol { X } } ^ { \dagger } { \boldsymbol { y } }$ on the input features. + +Training the First Layer. When the first layer of a linear network is optimized via gradient flow and the second layer is fixed, the following holds for zero-initialization of $W$ . + +Proposition 3. Given $W ( 0 ) = 0$ and fixed $\pm \ : 0$ , at any time $t > 0$ of the gradient flow on $W$ , $W ( t )$ is rank-1. Further, ${ \hat { \boldsymbol { \beta } } } = { \widehat { \boldsymbol { W } } } \mathbf { a }$ converges to the least squares solution of $\boldsymbol { y } = \boldsymbol { X } ^ { \intercal } \hat { \boldsymbol { \beta } }$ , the population risk of which is given in (Hastie et al., 2019, Thm. 1 & 3)) as + +$$ +R _ { ( \gamma _ { 1 } < 1 ) } \to \frac { \gamma _ { 1 } } { 1 - \gamma _ { 1 } } \sigma ^ { 2 } ; \quad R _ { ( \gamma _ { 1 } > 1 ) } \to \frac { \gamma _ { 1 } - 1 } { \gamma _ { 1 } } r ^ { 2 } + \frac { 1 } { \gamma _ { 1 } - 1 } \sigma ^ { 2 } . +$$ + +In this case, overparameterization by increasing $\gamma _ { 2 }$ does not influence the population risk. In addition, since the obtained two-layer linear model is equivalent to the minimum-norm solution on the input features $\hat { \beta } _ { \mathrm { m i n } }$ , optimizing the first layer always results in smaller or equal population risk compared to optimizing the second layer. + +In this simple scenario for two-layer linear networks, double descent is observed only when the second layer is optimized, which reduces the objective to least squares regression on the intermediate features. On the other hand, training the first layer from zero-initialization always yields the same solution that is independent to overparameterization. One natural question to ask is: does this phenomenon generalize to nonlinear two-layer neural networks? The following sections answer this question in the affirmative under certain conditions. + +![](images/277c0189331694fef2e6d8835188d44b8e63f3a5b84f2bea72662ce57820d839.jpg) +Figure 2: Population risk of two-layer neural networks with optimized second layer under (A1)(A2). Brighter color indicates larger $\gamma _ { 1 }$ . (a) risk of linear network with $r ^ { 2 } / \sigma ^ { 2 } \overset { \cdot } { = } 1 6$ and $\gamma _ { 1 } < 1$ . $( \gamma _ { 1 } > 1$ is shown in Figure 1) (b) variance of network with ReLU activation. Black line corresponds to $\gamma _ { 1 } \to \infty$ predicted by Corollary 5. (c) bias of network with ReLU activation. Black line corresponds to $\gamma _ { 1 } \to \infty$ for linear network, which is empirically observed as an upper-bound. Note that as $\gamma _ { 2 } 1$ both bias and variance becomes unbounded. + +# 4 NONLINEAR MODEL: OPTIMIZING THE SECOND LAYER + +In this section, we analyze the case when the second layer $\textbf { \em a }$ is learned under fixed $W$ and a nonlinear activation function $\phi$ (a random feature model). We first observe that by Lemma 1, the gradient flow finds the solution $\hat { \mathbf { a } } = \Phi ^ { \dagger } \mathbf { y }$ . We again consider the following bias-variance decomposition. + +$$ +\begin{array} { r l } & { R = \mathbb { E } _ { \boldsymbol { x } \sim P _ { \boldsymbol { x } } } [ \| \phi ( \boldsymbol { x } ^ { \top } \boldsymbol { W } ) \hat { \boldsymbol { a } } - \boldsymbol { F } ( \boldsymbol { x } ) \| _ { 2 } ^ { 2 } | \boldsymbol { X } , \boldsymbol { W } ] } \\ & { \quad = \underbrace { \mathbb { E } _ { \boldsymbol { x } } [ \| \mathbb { E } [ \phi ( \boldsymbol { x } ^ { \top } \boldsymbol { W } ) \hat { \boldsymbol { a } } | \boldsymbol { X } , \boldsymbol { W } ] - \boldsymbol { F } ( \boldsymbol { x } ) \| _ { 2 } ^ { 2 } ] } _ { B = \mathrm { b i a s } } + \underbrace { \mathbb { E } _ { \boldsymbol { x } } [ \| \phi ( \boldsymbol { x } ^ { \top } \boldsymbol { W } ) \hat { \boldsymbol { a } } - \mathbb { E } [ \phi ( \boldsymbol { x } ^ { \top } \boldsymbol { W } ) \hat { \boldsymbol { a } } ] \| _ { 2 } ^ { 2 } \big | \boldsymbol { X } , \boldsymbol { W } ] } _ { \boldsymbol { V } = \mathrm { v a r i a n c e } } . } \end{array} +$$ + +We highlight that the variance term does not depend on the target function. The following result characterizes the variance of the random feature model. + +Theorem 4. Given (A1) and ${ \pmb w } _ { i } \overset { \mathrm { i . i . d . } } { \sim } \mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )$ , when $n , d , h \infty$ , we have + +$$ +V = \{ \begin{array} { l l } { \sigma ^ { 2 } \frac { \gamma _ { 2 } } { 1 - \gamma _ { 2 } } , } & { \gamma _ { 2 } < 1 , } \\ { \quad } & { \gamma _ { 2 } < 1 \leq r \leq r , } \\ { \sigma ^ { 2 } \displaystyle \operatorname* { l i m } _ { \xi 0 } - [ \gamma _ { 2 } \frac { \partial } { \partial x } m _ { 1 } ( \xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \frac { \partial } { \partial x } m _ { 2 } ( \xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \frac { \gamma _ { 2 } - 1 } { \xi ^ { 2 } } ] } & { \gamma _ { 2 } > 1 . } \end{array} +$$ + +in which $m _ { 1 } ( \xi , \rho , \tau )$ , $m _ { 2 } ( \xi , \rho , \tau )$ is the unique solution in $\{ | m _ { 1 } | , | m _ { 2 } | < 1 / { \mathfrak { T } } \xi \}$ of + +$$ +\begin{array} { r l } & { m _ { 1 } ^ { - 1 } = - \xi - \rho - \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } \tau ^ { 2 } m _ { 1 } - c _ { 1 } m _ { 2 } + \frac { \tau ^ { 2 } \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } m _ { 1 } ^ { 2 } \left( c _ { 2 } m _ { 2 } - \tau \right) - 2 \tau c _ { 2 } m _ { 1 } m _ { 2 } + c _ { 2 } ^ { 2 } m _ { 1 } m _ { 2 } ^ { 2 } } { m _ { 1 } \left( c _ { 2 } m _ { 2 } - \tau \right) - \gamma _ { 1 } \gamma _ { 2 } ^ { - 1 } } , } \\ & { m _ { 2 } ^ { - 1 } = - \xi - r \gamma _ { 2 } m _ { 1 } + \frac { \gamma _ { 2 } c _ { 2 } m _ { 1 } ^ { 2 } \left( c _ { 2 } m _ { 2 } - \tau \right) } { m _ { 1 } \left( c _ { 2 } m _ { 2 } - \tau \right) - \gamma _ { 1 } \gamma _ { 2 } ^ { - 1 } } , } \end{array} +$$ + +where variables $\xi , \rho , \tau$ satisfies $\Im \xi > 0$ or $\xi < 0$ , $\rho > \tau > 0$ , and constants $c _ { 1 } , c _ { 2 }$ defined as, + +$$ +c _ { 1 } = \mathbb { E } [ \phi ( G ) ^ { 2 } ] - \mathbb { E } [ \phi ( G ) ] ^ { 2 } , \quad c _ { 2 } = \mathbb { E } [ G \phi ( G ) ] ^ { 2 } , +$$ + +for $G \sim \mathcal { N } ( 0 , 1 )$ and $\Im \xi$ denoting the imaginary part of $\xi$ . + +Remark. $c _ { 1 } \geq c _ { 2 }$ and the equality holds iff $\phi$ is linear. + +Corollary 5. If we let $\gamma _ { 1 } \to \infty$ , the variance is equal to the lowest value of the variance of the linear model $V _ { ( \gamma _ { 1 } \to \infty ) } = \sigma ^ { 2 } \mathrm { m i n } \{ \gamma _ { 2 } , 1 \} / | 1 - \gamma _ { 2 } |$ . + +The proof of Theorem 4 largely follows from Hastie et al. (2019) with techniques similar to Cheng and Singer (2013), but with modifications in otder to handle unnormalized and uncentered activation functions. The above theorem holds irrespective of the underlying teacher model, and is consistent with the double descent risk curve as it suggests that for all $\gamma _ { 1 }$ , variance of the random feature model peaks at $h = n$ then drops as $\gamma _ { 2 }$ further increases. Note that as $\gamma _ { 1 } \to \infty$ , a linear and nonlinear network would have the same asymptotic variance. + +Since double descent is observed in the variance term, we do not derive the bias for all $\gamma _ { 1 } , \gamma _ { 2 }$ . Instead, we show that for linear teacher, the bias also becomes unbounded as $\gamma _ { 2 } 1$ . + +Proposition 6. Given $( A I ) ( A 2 )$ and ${ \pmb w } _ { i }$ i.i.d. ∼ $\mathcal { N } ( 0 , I _ { d } )$ , then $B \to \infty$ as $\gamma _ { 2 } 1$ . Furthermore, $B$ is finite when $\gamma _ { 2 } > 1$ . + +Thus we have shown that a “cusp” in the population risk appears at $h = n$ , which aligns with the double descent phenomenon. Empirically as $\gamma _ { 1 } \to \infty$ the nonlinear model also shares the same asymptotic bias with the linear model, as shown in Figure 2. We note that (Mei and Montanari, 2019, Thm. 1 & 3) analytically solved the risk of random feature model for a larger class of target functions than ours and confirmed that double descent appears in both the bias and the variance. + +# 5 NONLINEAR MODEL: OPTIMIZING THE FIRST LAYER + +Having observed the double descent phenomenon in optimizing the second layer, in the sequel we consider a two-layer neural network with fixed second layer coefficients initialized from a Rademacher√ √ distribution $a _ { i } \sim \operatorname { U n i f } \{ - 1 / \sqrt { h } , 1 \sqrt { h } \}$ , and the first layer $W$ is optimized with the following update + +$$ +\frac { \partial W ( t ) } { \partial t } = - \frac { \partial L ( X ; W ) } { \partial W } = \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \left[ y _ { i } - \pmb { a } ^ { \top } \phi ( W ^ { \top } \pmb { x } _ { i } ) \right] \pmb { x } _ { i } [ \phi ^ { \prime } ( \pmb { x } _ { i } ^ { \top } W ) \circ \pmb { a } ] , +$$ + +which potentially has different stationary solutions with no explicit form, depending on the initialization. We denote the solution of this flow at time $t$ started from designated initialization by $W ^ { \mathrm { i n i t } } ( t )$ , its stationary solution by $\widehat { W }$ , and the corresponding network by $\hat { f }$ . + +Remark. Although we let $n \to \infty$ , this dynamics does not corresponds to the population gradient flow considered in Tian (2017). For instance when $\pmb { x } \sim \mathcal { N } ( 0 , I / d )$ , the spectrum of the data covariance is Marcenko–Pastur, whereas the population covariance is identity. ˇ + +As we cannot characterize the gradient flow solution from all possible initializations, we consider two specific scales of initialization: + +Note that neither of the two initializations correspond to the “mean-field” regime (e.g. analyzed in√ Mei et al. (2018)) due to the $1 / { \sqrt { h } }$ scaling of the second layer. In other words, as $h$ increases, we expect the distance traveled by each parameter to decrease under both initializations. The difference, however, is the “relative” amount the parameters traveled compared to their initialized magnitude, which leads to solutions with contrasting properties. As we will see, under (A1)(A2) and vanishing initialization we have $\lVert W ( 0 ) - \widehat { W } \rVert _ { F } / \lVert W ( 0 ) \rVert _ { F } \gg 1$ , i.e. the contribution of initialization vanishes at the end of training, whereas for non-vanishing initialization the inequality is in the opposite direction, i.e. $\widehat { W }$ “barely moves” and resembles the initialization $W ( 0 )$ . + +# 5.1 VANISHING INITIALIZATION + +As $d , h \infty$ , the vanishing initialization becomes arbitrarily close to zero-initialization. We thus expect the gradient flow under vanishing initialization to "resemble" that of starting from exactly zero if the flow converges sufficiently fast and the gradient being Lipschitz. The Lipschitz condition (Lemma 18) can be established under the following assumption on the activation. + +(A3): $\phi$ is smooth, Lipschitz and monotone with $\phi ^ { \prime } ( 0 ) \neq 0 ; | \phi ^ { \prime } ( \pm x ) - \phi ^ { \prime } ( \pm \infty ) | = O ( e ^ { - x } ) .$ + +The above assumption requires that the derivative of the nonlinearity $\phi$ saturates beyond a O(1) region, which holds true for the commonly-used smooth activations such as sigmoid and SoftPlus. In addition, the choice of scaling ensures that the gradient flow converges sufficiently fast. We thus have the following characterization of the population risk: + +Theorem 7. Given (A1-3). Let $T = O ( \log \log h )$ and $\hat { f } ( \cdot ) = f ^ { \nu a n } ( \cdot , W ( T ) )$ , then as $n , d , h \infty$ , the gradient flow reaches a $o ( 1 )$ first-order stationary point at time $T ,$ i.e. $\| \partial W ( T ) / \partial t \| _ { F } \in o ( 1 ) ,$ , at which point the population risk is given as + +$$ +R ( \hat { f } ) \operatorname* { m a x } \{ 0 , \frac { \gamma _ { 1 } - 1 } { \gamma _ { 1 } } \} r ^ { 2 } + \frac { \operatorname* { m i n } \{ \gamma _ { 1 } , 1 \} } { | 1 - \gamma _ { 1 } | } \sigma ^ { 2 } . +$$ + +The expression above is the same as the risk of the least squares solution on input ${ \hat { \boldsymbol { \beta } } } = X ^ { \dagger } \boldsymbol { y }$ ; therefore the risk is independent to overparameterization (increasing $\gamma _ { 2 }$ ). The intuition is that when the weights are initialized sufficiently small and travel infinitesimally, then the activation can be linearized around 0 and thus the model is equivalent to a two-layer linear network. Note that this result does not apply to the non-smooth ReLU activation. Instead, in Appendix $\mathrm { E }$ we heuristically show that under the additional assumption that the data is symmetric, the risk of ReLU network is also independent to $\gamma _ { 2 }$ + +# 5.2 NON-VANISHING INITIALIZATION + +When initialization is sufficiently large, the amount each parameter travels to minimize the empirical risk becomes asymptotically negligible compared to the magnitude of initialization. In this case we establish under (A1-3) that (11) is asymptotically equivalent to the kernel gradient flow on the tangent kernel: $k ( { \pmb x } , { \pmb y } ) = \langle \nabla _ { W ^ { \mathrm { i n i t } } } f ( { \pmb x } ) , \nabla _ { W ^ { \mathrm { i n i t } } } f ( { \pmb y } ) \rangle$ . The converged parameters under this linearized dynamics has the following closed-form: + +$$ +\mathrm { v e c } ( W ^ { * } ) \approx \mathrm { v e c } ( W ^ { \mathrm { i n i t } } ) + \Delta ; \quad \Delta = J ^ { \dagger } ( { \pmb y } - f ^ { \mathrm { i n i t } } ( { \pmb X } ) ) ; \quad J _ { [ i , j ] } = \nabla _ { \mathrm { v e c } ( W ^ { \mathrm { i n i t } } ) _ { j } } f ^ { \mathrm { i n i t } } ( { \pmb x } _ { i } ) , +$$ + +where $J \in \mathbb { R } ^ { n \times ( d \times h ) }$ is the Jacobian matrix w.r.t. to the model parameters. We remark that in contrast to most NTK-type global convergence results that require the width of the model to grow faster than the number of data points (e.g. Du et al. (2018)), our result is not built upon the overparameterization on width, but instead an anti-concentration that relies on the scale of initialization. Consequently the above initialization is larger than the scale that is commonly used in practice. + +One may naturally expect the double descent phenomenon to appear in this kernel solution, as $\Delta$ exhibits the form of a least squares solution which contains a pseudo-inverse. However, we show that this is not the case under the same assumptions in Section 4; in fact, the risk is also independent to $\gamma _ { 2 }$ + +An obstacle in computing the risk of the kernel model is the potentially non-zero $f ^ { \mathrm { i n i t } } ( X )$ . We thus adopt the "doubling-trick" from Chizat and Bach (2018b) to ensure $f ^ { \mathrm { i n i t } } ( \cdot ) = 0$ , i.e. we assume the following symmetric property on the initialized weights: + +(A4) Symmetric Initialization: $\forall i \in [ 1 , h ]$ , ∃! $j \in [ 1 , h ]$ s.t. $a _ { i } \mathbf { w } _ { i } ^ { \mathrm { i n i t } } = - a _ { j } \mathbf { w } _ { j } ^ { \mathrm { i n i t } }$ + +Theorem 8. Given (A1-4) and let $n , d , h \infty$ , the stationary solution $\hat { f }$ has the following risk + +$$ +\begin{array} { c } { { R ( \hat { f } ) ( \frac { \gamma _ { 1 } - 1 } { 2 \gamma _ { 1 } } + \frac { \gamma _ { 1 } ( \gamma _ { 1 } + \gamma _ { 1 } m + m - 2 ) + 1 } { 2 \gamma _ { 1 } \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) r ^ { 2 } } } \\ { { + ( \frac { \gamma _ { 1 } + \gamma _ { 1 } m + 1 } { 4 \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } - \frac { 1 } { 4 } ) \sigma ^ { 2 } , } } \end{array} +$$ + +where $m = b _ { 1 } ^ { 2 } / b _ { 0 } ^ { 2 }$ , $b _ { 0 } ^ { 2 } = \mathbb { E } [ \phi ^ { \prime } ( G ) ] ^ { 2 }$ , and $b _ { 1 } ^ { 2 } = \mathbb { E } [ \phi ^ { \prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }$ , $G \sim \mathcal { N } ( 0 , 1 )$ . + +Note that the population risk is again independent to $\gamma _ { 2 }$ , and thus double descent does not appear for this initialization. Roughly speaking, the reason that the risk does not become unbounded at some point is that in the asymptotic limit the pseudo-inverse $( J J ^ { \top } ) ^ { \dagger }$ is stable due to the nonlinearity and $d h \gg n$ . We make two additional observations: + +• the stability of the inverse at $\gamma _ { 1 } 1$ depends on the lowest eigenvalue of the tangent kernel matrix (smaller $\lambda _ { \operatorname* { m i n } } ( K )$ entails larger variance), which is determined by the nonlinearity; While our result only holds for network with zero initial output, for non-symmetric (i.i.d.) initialization we also observe that the risk is independent to $\gamma _ { 2 }$ , but the bias is higher than that in symmetric initialization as shown in Figure 8. We comment that the non-zero $f ^ { \mathrm { i n i t } } ( X )$ in (13) behaves as zero-mean Gaussian due to central limit theorem; therefore, in the kernel regime the function output at initialization is equivalent to additive noise to the labels $\textbf { { y } }$ , and we thus expect the magnitude of $f ^ { \mathrm { i n i t } } ( X )$ to negatively influence the model generalization. + +![](images/389923c17d029c0a38688561f57f18607a9ed4055a5409c2180d94b356e6ec61.jpg) +Figure 3: Bias and variance of two-layer sigmoid network with optimized first layer under (A1)(A2). Individual dotted lines correspond to different $\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. The bias and variance for both initializations is well-aligned with Theorem 7 and Theorem 8. + +# 5.3 COMPARING THE INITIALIZATIONS + +Figure 3 shows the agreement between theoretical prediction and experimental results. Although in both cases the risk is independent to overparameterization $( \gamma _ { 2 } )$ , the two initializations lead to models with contrasting properties, as demonstrated by the following comparison on the risk. + +Corollary 9. For any $\gamma _ { 1 } \in ( 0 , \infty )$ and nonlinearity $\phi$ , $B ( \hat { f } ^ { V a n } ) \le B ( \hat { f } ^ { N V } ) \le 1 .$ . On the other hand, for all $m > 0$ , $V ( \hat { f } ^ { N V } ) = { \cal { O } } ( 1 )$ , whereas $V ( \hat { f } ^ { V a n } )$ can be arbitrarily large as $\gamma _ { 1 } 1$ . + +Remark. $m \geq 0$ for all smooth activations $\phi$ , and the equality holds if $\phi$ is linear. + +Intuitively, small initialization enables the model "evolve" more during optimization and better align with the data and target function. This potentially results in a lower bias, at the expense of overfitting more to the noise (high variance). In contrast, with sufficiently large initialization the final model becomes close to the initialized model, and thus we may expect it to be less “aligned” to the target (high bias) but is more stable (lower variance). + +In illustrate the different inductive bias of the two initializations, we plot the trajectory of neurons in Appendix A Figure 4. Observe that for vanishing initialization the neurons stay close to one another throughout the trajectory, which results in a low-rank weight matrix, as predicted by Theorem 7. In contrast, for non-vanishing initialization the neurons stay close to initialization (therefore full-rank), which validates the kernel approximation. Last but not least, although the derived risk is only for learning a linear target function, we empirically observe that when the teacher is also a two-layer network, the population risk follows the same trend, i.e. double descent occurs when only the second layer is optimized, as shown in Figure 7. + +# 6 DISCUSSION AND FUTURE WORKS + +We derived the exact population risk of high-dimensional two-layer neural networks in learning a linear target function over Gaussian data with additive label noise, and showed that optimizing the first or the second layer via gradient flow results in solutions with contrasting properties. Specifically, double descent is present when the second layer coefficients are optimized, but not when the first layer weights are optimized under certain initializations. Moreover, we highlight that the scale of initialization leads to different inductive bias in optimizing the first layer. + +It should be noted that our analysis only applies to the unregularized objective: it has been shown that explicit regularization (such as $\ell _ { 2 }$ penalty) stabilizes the singularity at $\gamma _ { 2 } ~ ~ 1$ (Mei and Montanari, 2019), and algorithmic regularization (Li et al., 2019a; Dong et al., 2019) also provides robustness against noisy observations. We further remark that our findings do not directly contradict the experimental double descent phenomenon, nor the practical benefit of overparameterization. In particular, the interpolation limit could occur at $\gamma _ { 2 } 0$ which is beyond the regime we consider (such as Figure 4 in (Belkin et al., 2018)). Thus what we conclude is that under the studied proportional asymptotics, the mechanism that provably gives rise to double descent from previous works on least squares regression might not translate to neural networks trained with gradient descent. + +To simplify the computation, we rely on a set of strong assumptions similar to those in Hastie et al. (2019), some of which we believe can be relaxed in future works, such as isotropic Gaussian input and linear target. Importantly, the two specific scales of initialization studied in Section 5 are by no means exhaustive, and thus one would expect that under a different initialization of the first layer, or the mean-field $1 / h$ scaling of the second layer, the risk of the trained model can be very different. Changing the loss function may also alter the generalization behavior of the network. Another challenging problem is to extend the current analysis to beyond two layers. Last but not least, our result characterizes gradient flow which resembles gradient descent with small stepsize, and thus it would be interesting to study the effect of learning rate schedule, which has a known impact on generalization (Smith and Le, 2017; Li et al., 2019b). + +# ACKNOWLEDGEMENT + +We thank Xiuyuan Cheng, Xuechen Li, Yiping Lu, Atsushi Nitanda, Shengyang Sun and anonymous reviewers for helpful comments and feedback. JB and DW were partially funded by LG Electronics and NSERC. JB and MAE were supported by the CIFAR AI Chairs program. TS was partially supported by JSPS Kakenhi (26280009, 15H05707 and 18H03201), Japan Digital Design and JST-CREST. + +# REFERENCES + +Madhu S Advani and Andrew M Saxe. High-dimensional dynamics of generalization error in neural networks. arXiv preprint arXiv:1710.03667, 2017. +Zeyuan Allen-Zhu and Yuanzhi Li. What can resnet learn efficiently, going beyond kernels? arXiv preprint arXiv:1905.10337, 2019. +Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized neural networks, going beyond two layers. arXiv preprint arXiv:1811.04918, 2018a. +Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via overparameterization. arXiv preprint arXiv:1811.03962, 2018b. +Sanjeev Arora, Rong Ge, Behnam Neyshabur, and Yi Zhang. Stronger generalization bounds for deep nets via a compression approach. arXiv preprint arXiv:1802.05296, 2018. +Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Ruslan Salakhutdinov, and Ruosong Wang. On exact computation with an infinitely wide neural net. arXiv preprint arXiv:1904.11955, 2019a. +Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. arXiv preprint arXiv:1901.08584, 2019b. +Zhidong Bai and Jack W Silverstein. Spectral analysis of large dimensional random matrices, volume 20. Springer, 2010. +Peter L Bartlett, Dylan J Foster, and Matus J Telgarsky. Spectrally-normalized margin bounds for neural networks. In Advances in Neural Information Processing Systems, pages 6240–6249, 2017. +Peter L Bartlett, Philip M Long, Gábor Lugosi, and Alexander Tsigler. Benign overfitting in linear regression. arXiv preprint arXiv:1906.11300, 2019. +Mikhail Belkin, Daniel Hsu, Siyuan Ma, and Soumik Mandal. Reconciling modern machine learning and the bias-variance trade-off. arXiv preprint arXiv:1812.11118, 2018. +Mikhail Belkin, Daniel Hsu, and Ji Xu. Two models of double descent for weak features. arXiv preprint arXiv:1903.07571, 2019. + +Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning overparameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019. + +Xiuyuan Cheng and Amit Singer. The spectrum of random inner-product kernel matrices. Random Matrices: Theory and Applications, 2(04):1350010, 2013. + +Lenaic Chizat and Francis Bach. On the global convergence of gradient descent for over-parameterized models using optimal transport. In Advances in neural information processing systems, pages 3036–3046, 2018a. + +Lenaic Chizat and Francis Bach. A note on lazy training in supervised differentiable programming. arXiv preprint arXiv:1812.07956, 2018b. + +Bin Dong, Jikai Hou, Yiping Lu, and Zhihua Zhang. Distillation $\approx$ early stopping? harvesting dark knowledge utilizing anisotropic information retrieval for overparameterized neural network. arXiv preprint arXiv:1910.01255, 2019. + +Simon S Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes over-parameterized neural networks. arXiv preprint arXiv:1810.02054, 2018. + +Gintare Karolina Dziugaite and Daniel M Roy. Computing nonvacuous generalization bounds for deep (stochastic) neural networks with many more parameters than training data. arXiv preprint arXiv:1703.11008, 2017. + +Noureddine El Karoui et al. The spectrum of kernel random matrices. The Annals of Statistics, 38(1): 1–50, 2010. + +Zhou Fan and Andrea Montanari. The spectral norm of random inner-product kernel matrices. Probability Theory and Related Fields, 173(1-2):27–85, 2019. + +Mario Geiger, Stefano Spigler, Stéphane d’Ascoli, Levent Sagun, Marco Baity-Jesi, Giulio Biroli, and Matthieu Wyart. The jamming transition as a paradigm to understand the loss landscape of deep neural networks. arXiv preprint arXiv:1809.09349, 2018. + +Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Limitations of lazy training of two-layers neural networks. arXiv preprint arXiv:1906.08899, 2019a. + +Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Linearized two-layers neural networks in high dimension. arXiv preprint arXiv:1904.12191, 2019b. + +Sebastian Goldt, Madhu S Advani, Andrew M Saxe, Florent Krzakala, and Lenka Zdeborová. Generalisation dynamics of online learning in over-parameterised neural networks. arXiv preprint arXiv:1901.09085, 2019. + +Suriya Gunasekar, Jason D Lee, Daniel Soudry, and Nati Srebro. Implicit bias of gradient descent on linear convolutional networks. In Advances in Neural Information Processing Systems, pages 9461–9471, 2018. + +Trevor Hastie, Andrea Montanari, Saharon Rosset, and Ryan J Tibshirani. Surprises in highdimensional ridgeless least squares interpolation. arXiv preprint arXiv:1903.08560, 2019. + +Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and generalization in neural networks. In Advances in neural information processing systems, pages 8571–8580, 2018. + +Ziwei Ji and Matus Telgarsky. Gradient descent aligns the layers of deep linear networks. arXiv preprint arXiv:1810.02032, 2018. + +Ryo Karakida, Shotaro Akaho, and Shun-ichi Amari. Universal statistics of fisher information in deep neural networks: Mean field approach. arXiv preprint arXiv:1806.01316, 2018. + +Anders Krogh and John A. Hertz. A simple weight decay can improve generalization. pages 950–957, 1992. + +Mingchen Li, Mahdi Soltanolkotabi, and Samet Oymak. Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks. arXiv preprint arXiv:1903.11680, 2019a. + +Yuanzhi Li and Yingyu Liang. Learning overparameterized neural networks via stochastic gradient descent on structured data. In Advances in Neural Information Processing Systems, pages 8157– 8166, 2018. + +Yuanzhi Li, Tengyu Ma, and Hongyang Zhang. Algorithmic regularization in over-parameterized matrix sensing and neural networks with quadratic activations. arXiv preprint arXiv:1712.09203, 2017. + +Yuanzhi Li, Colin Wei, and Tengyu Ma. Towards explaining the regularization effect of initial large learning rate in training neural networks. arXiv preprint arXiv:1907.04595, 2019b. + +Tengyuan Liang and Alexander Rakhlin. Just interpolate: Kernel" ridgeless" regression can generalize. arXiv preprint arXiv:1808.00387, 2018. + +Zhenyu Liao and Romain Couillet. On the spectrum of random features maps of high dimensional data. arXiv preprint arXiv:1805.11916, 2018. + +Cosme Louart, Zhenyu Liao, Romain Couillet, et al. A random matrix approach to neural networks. The Annals of Applied Probability, 28(2):1190–1248, 2018. + +V.A. Marcenko and Leonid Pastur. Distribution of eigenvalues for some sets of random matrices. ˇ Math USSR Sb, 1:457–483, 01 1967. + +Song Mei and Andrea Montanari. The generalization error of random features regression: Precise asymptotics and double descent curve. arXiv preprint arXiv:1908.05355, 2019. + +Song Mei, Andrea Montanari, and Phan-Minh Nguyen. A mean field view of the landscape of twolayer neural networks. Proceedings of the National Academy of Sciences, 115(33):E7665–E7671, 2018. + +Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Mean-field theory of two-layers neural networks: dimension-free bounds and kernel limit. arXiv preprint arXiv:1902.06015, 2019. + +Vaishnavh Nagarajan and J Zico Kolter. Generalization in deep networks: The role of distance from initialization. arXiv preprint arXiv:1901.01672, 2019. + +Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the role of implicit regularization in deep learning. arXiv preprint arXiv:1412.6614, 2014. + +Behnam Neyshabur, Zhiyuan Li, Srinadh Bhojanapalli, Yann LeCun, and Nathan Srebro. Towards understanding the role of over-parametrization in generalization of neural networks. arXiv preprint arXiv:1805.12076, 2018. + +Atsushi Nitanda and Taiji Suzuki. Stochastic particle gradient descent for infinite ensembles. arXiv preprint arXiv:1712.05438, 2017. + +Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global convergence guarantees for training shallow neural networks. arXiv preprint arXiv:1902.04674, 2019. + +Jeffrey Pennington and Pratik Worah. Nonlinear random matrix theory for deep learning. In Advances in Neural Information Processing Systems, pages 2637–2646, 2017. + +Jeffrey Pennington and Pratik Worah. The spectrum of the fisher information matrix of a singlehidden-layer neural network. In Advances in Neural Information Processing Systems, pages 5410–5419, 2018. + +Grant M Rotskoff and Eric Vanden-Eijnden. Neural networks as interacting particle systems: Asymptotic convexity of the loss landscape and universal scaling of the approximation error. arXiv preprint arXiv:1805.00915, 2018. + +Justin Sirignano and Konstantinos Spiliopoulos. Mean field analysis of neural networks: A central limit theorem. arXiv preprint arXiv:1808.09372, 2018. + +Samuel L Smith and Quoc V Le. A bayesian perspective on generalization and stochastic gradient descent. arXiv preprint arXiv:1710.06451, 2017. + +Zhao Song and Xin Yang. Quadratic suffices for over-parametrization via matrix chernoff bound. arXiv preprint arXiv:1906.03593, 2019. + +Stefano Spigler, Mario Geiger, Stéphane d’Ascoli, Levent Sagun, Giulio Biroli, and Matthieu Wyart. A jamming transition from under-to over-parametrization affects loss landscape and generalization. arXiv preprint arXiv:1810.09665, 2018. + +Taiji Suzuki. Adaptivity of deep relu network for learning in besov and mixed smooth besov spaces: optimal rate and curse of dimensionality. arXiv preprint arXiv:1810.08033, 2018. + +Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012. + +Yuandong Tian. An analytical formula of population gradient for two-layered relu network and its applications in convergence and critical point analysis. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 3404–3413. JMLR. org, 2017. + +Colin Wei, Jason D Lee, Qiang Liu, and Tengyu Ma. On the margin theory of feedforward neural networks. arXiv preprint arXiv:1810.05369, 2018. + +Francis Williams, Matthew Trager, Claudio Silva, Daniele Panozzo, Denis Zorin, and Joan Bruna. Gradient dynamics of shallow univariate relu networks. arXiv preprint arXiv:1906.07842, 2019. + +Blake Woodworth, Suriya Gunasekar, Jason Lee, Daniel Soudry, and Nathan Srebro. Kernel and deep regimes in overparametrized models. arXiv preprint arXiv:1906.05827, 2019. + +Ji Xu and Daniel Hsu. On the number of variables to use in principal component regression. 2019. + +Gilad Yehudai and Ohad Shamir. On the power and limitations of random features for understanding neural networks. arXiv preprint arXiv:1904.00687, 2019. + +Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016. + +Kai Zhong, Zhao Song, Prateek Jain, Peter L Bartlett, and Inderjit S Dhillon. Recovery guarantees for one-hidden-layer neural networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 4140–4149. JMLR. org, 2017. + +![](images/8be263e99069d2ff2756929d4f59b8c08885627a89e302d107583385553bcd06.jpg) +Figure 4: trajectory of neurons from initialization (dark blue) to optimum (orange) on the first two dimensions (two-layer SoftPlus student and linear teacher; $\mathrm { S N R } { = } 1 / 4 \mathrm { \Omega }$ . For vanishing initialization the neurons stay close to one another throughout the trajectory, whereas for non-vanishing initialization the neurons stay close to initialization. + +![](images/67d5f63ec5bc3ba3b56af6a95e487287d5f924b0f1e30f617bcb3ec06520d044.jpg) +Figure 5: Population risk of two-layer linear network with fixed random 1st layer with $S _ { \mathrm { { N R = } } 2 5 / 1 6 }$ under Gaussian input and linear teacher. Brighter color indicates larger $\gamma _ { 1 }$ . + +SUMMARY OF THE PRESENCE / ABSENCE OF DOUBLE DESCENT + +
Singularity in2nd Layer Trained (RF)Vanishing Init.Non-vanishing Init.
Bias1: No; Y2: Yesγ1: No; γ2: Noγ1: No; 2: No
VarianceY1: No; 2: Yes1: Yes; Y2: NoY1: No; 2: No
+ +![](images/487235692f241523abf1600424158205f71f19e5f44d0aed9c5fb66d2d8f7cb9.jpg) +Figure 6: Bias and variance of two-layer SoftPlus network with optimized first layer under (A1)(A2). Individual dotted lines correspond to different $\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. The bias and variance for both initializations is well-aligned with Theorem 7 and Theorem 8, respectively. + +![](images/0a41e6cc5c4bd20c793bbe18e10c8b870f315c2459db916ab633ef96972f668a.jpg) +Figure 7: Population risk (scaled by $1 / d )$ of two-layer ReLU network trained to fit a two-layer ReLU teacher model with $h = d$ neurons. Brighter color corresponds to larger $\gamma _ { 1 }$ . Similar to the linear teacher case, double descent is observed when the second layer is optimized (a) but not when the first layer is optimized (b). + +![](images/c366b578fe991ae26f542e54255e90708f233ee5480d69cce84eed842879b63d.jpg) +Figure 8: Bias of (a) SoftPlus and (b) sigmoid two-layer network with optimized first layer under (A1)(A2). Note that bias under i.i.d. initialization is also independent to overparameterization $\left( \gamma _ { 2 } \right)$ , but is higher than the bias under symmetric initialization (“doubling trick”) and not always upper-bounded by the null risk $r ^ { 2 }$ . + +# B BACKGROUND + +# B.1 ROTATIONAL INVARIANCE + +The rotational invariance of Gaussian distribution is crucial in our analysis throughout this paper. A basic observation is that for a random Gaussian matrix $X$ and any fixed unitary matrix $U$ , the distribution of $X$ and $U X$ are the same. + +Lemma 10 (Rotational Invariance). Denote $A ( X ) \in \mathbb { R } ^ { d \times d }$ a matrix function of $X \in \mathbb { R } ^ { d \times n }$ . If $A ( X )$ satisfies that $A ( U X ) = U A ( X ) U ^ { T }$ for all unitary $U$ , then + +$$ +\operatorname { \mathbb { E } } _ { X } [ \beta ^ { T } A ( X ) \beta ] = { \frac { 1 } { d } } \beta ^ { T } \beta \operatorname { \mathbb { E } } _ { X } [ \operatorname { t r } \left( A ( X ) \right) ] . +$$ + +for any fixed nonzero $\beta \in \mathbb { R } ^ { d }$ and random matrix $X$ with each entry i.i.d. $X _ { i j } \sim \mathcal { N } ( 0 , \sigma ^ { 2 } )$ + +Proof. Choose a set of Unitary matrices $\{ U _ { i } \} _ { i = 1 } ^ { d }$ such that $U _ { i } ^ { \top } \beta = \| \beta \| e _ { i }$ , where $e _ { i }$ is the $i$ -th canonical vector in $\mathbb { R } ^ { d }$ . Since $U _ { i } X \sim X$ , we have $\mathbb { E } [ A ( X ) ] = \mathbb { E } [ A ( U _ { i } X ) ] = U _ { i } \mathbb { E } [ A ( X ) ] U _ { i } ^ { \top }$ and hence + +$$ +\mathbb { E } [ \beta ^ { T } A ( X ) \beta ] = \frac { 1 } { d } \sum _ { i = 1 } ^ { d } \mathbb { E } [ \beta ^ { T } U _ { i } A ( X ) U _ { i } ^ { \top } \beta ] = \frac { 1 } { d } \sum _ { i = 1 } ^ { d } e _ { i } ^ { T } \mathbb { E } [ A ( X ) ] e _ { i } = \frac { \beta ^ { T } \beta } { d } \mathbb { E } [ \mathrm { t r } ( A ( X ) ) ] . +$$ + +Note that the property also holds for matrix function $A ( X )$ that satisfies $A ( X U ) = U A ( X ) U ^ { T }$ , and can be extended to matrix function $A$ that takes multiple matrices as input. □ + +For brevity we will refer to rotational invariance instead of equation (15). + +# B.2 MARCENKO ˇ –PASTUR LAW + +For a real symmetric random matrix $A \in \mathbb { R } ^ { p \times p }$ , define its empirical spectral density as + +$$ +\mu _ { A } ( d \lambda ) = \frac { 1 } { p } \sum _ { i = 1 } ^ { p } \delta _ { \lambda _ { i } ( A ) } ( \lambda ) d \lambda , +$$ + +where $\delta _ { a } ( x ) = \delta ( x - a )$ is the Dirac delta function. Assume $A = W _ { p } \sim W _ { p } ( I , n )$ is a Wishart matrix, i.e. $W _ { p } = X ^ { \top } X / n$ and $\ b { X } \in \mathbb { R } ^ { n \times p }$ is random Gaussian matrix with each column i.i.d. $X _ { i } \sim \mathcal { N } ( \mathbf { 0 } , I )$ . Marcenko and Pastur ˇ (1967) showed that as $n , p \to \infty$ and $p / n = \gamma \in ( 0 , \infty )$ , the empirical spectral density $\mu _ { W _ { p } } ( d \lambda )$ converges weakly to a limiting density $\mu _ { \mathrm { M P } ( \gamma ) } ( \lambda )$ : + +$$ +\mu _ { \mathrm { M P } ( \gamma ) } ( d \lambda ) = [ 1 - \gamma ^ { - 1 } ] _ { + } \delta _ { 0 } ( \lambda ) d \lambda + \frac { 1 } { 2 \pi \gamma \lambda } \sqrt { ( ( 1 + \sqrt \gamma ) ^ { 2 } - \lambda ) ( \lambda - ( 1 - \sqrt \gamma ) ^ { 2 } ) } d \lambda . +$$ + +We say $\mu _ { \mathrm { M P } ( \gamma ) }$ is the density of the Marˇcenko–Pastur distribution with support $S = [ ( 1 - \sqrt { \gamma } ) ^ { 2 } , ( 1 +$ $\sqrt { \gamma } ) ^ { 2 } ]$ (for $0 < \gamma < 1$ ) or $S = \{ 0 \} \cup [ ( 1 - \sqrt { \gamma } ) ^ { 2 } , ( 1 + \sqrt { \gamma } ) ^ { 2 } ]$ (for $\gamma \geq 1$ ). Note that this implies that the smallest non-zero eigenvalue of $A$ is bounded away from 0 a.s. for $\gamma \neq 1$ . + +The explicit form of Marcenko–Pastur distribution allows us to investigate the asymptotic properties ˇ of random matrices. Generally speaking, by Pormanteau theorem one can translate any bounded continuous function on the empirical spectral density to the one on Marcenko–Pastur distribution, i.e. ˇ for any bounded $f ( \lambda ) \in C ( S )$ , as $n , p \to \infty$ with $p / n = \gamma$ , almost surely + +$$ +\int _ { S } f ( \lambda ) \cdot \mu _ { W } ( d \lambda ) \to \int f _ { S } ( \lambda ) \cdot \mu _ { M P } ( d \lambda ) . +$$ + +One implication is the following trace concentration on the inverse Wishart matrix for $\gamma < 1$ , + +$$ +\operatorname { t r } ( X ^ { \top } X ) = \operatorname { t r } ( { \frac { 1 } { p } } W _ { p } ^ { - 1 } ) = { \frac { 1 } { p } } \sum _ { i = 1 } ^ { p } { \frac { 1 } { \lambda _ { i } ( W _ { p } ) } } = \int _ { S } { \frac { 1 } { \lambda } } \mu _ { W _ { p } } ( d \lambda ) \int _ { S } { \frac { 1 } { \lambda } } \mu _ { \operatorname { M P } ( \gamma ) } ( d \lambda ) = { \frac { 1 } { 1 - \gamma } } . +$$ + +We remark that instability of the trace of the invert Wishart matrix as $\gamma 1$ plays an important role in the double descent phenomenon. + +# B.3 ORTHOGONAL POLYNOMIALS + +Orthogonal polynomials are useful in the analysis of nonlinear random matrices. Suppose $\phi$ is a function in $L ^ { 2 } ( \mathbb { R } , \mu _ { G } )$ , where $G \sim \mathcal { N } ( 0 , 1 )$ , $\mu _ { G } ( d x ) = ( \sqrt { 2 \pi } ) ^ { - 1 } e ^ { - x ^ { 2 } / 2 } d x$ is the Gaussian measure. For $n \geq 0$ , define the Hermite polynomials + +$$ +H _ { n } ( x ) = ( - 1 ) ^ { n } e ^ { - x ^ { 2 } / 2 } \frac { \partial ^ { n } } { \partial x ^ { n } } e ^ { - x ^ { 2 } / 2 } , +$$ + +Note that orthogonality can be easily verified: + +$$ +\mathbb { E } [ H _ { j } ( G ) H _ { k } ( G ) ] = \int _ { \mathbb { R } } H _ { j } ( x ) H _ { k } ( x ) \mu _ { G } ( d x ) = j ! \cdot \delta _ { j k } . +$$ + +Since $\{ H _ { i } ( x ) \} _ { i = 0 } ^ { \infty }$ forms a set of orthogonal basis in $L ^ { 2 } ( \mathbb { R } , \mu _ { G } )$ , the function $\phi ( x )$ can be expanded under the Hermite basis as + +$$ +\phi ( x ) = \sum _ { i = 0 } ^ { \infty } c _ { i } H _ { i } ( x ) = \sum _ { i = 0 } ^ { \infty } \left( \frac { 1 } { k ! } \int _ { \mathbb { R } } \phi ( a ) H _ { i } ( a ) \mu _ { G } ( d a ) \right) H _ { i } ( x ) . +$$ + +Following Cheng and Singer (2013), we mainly focus on the first few terms of the Hermite expansion. One can check that $H _ { 0 } ( x ) = 1$ , $H _ { 1 } ( x ) = x$ , and $\phi ( x )$ can be expanded as + +$$ +\phi ( x ) = c _ { 0 } + c _ { 1 } x + \phi _ { \perp } ( x ) , +$$ + +where $c _ { 0 } = \mathbb { E } [ \phi ( G ) ]$ , $c _ { 1 } = \mathbb { E } [ G \phi ( G ) ]$ , and terms in the RHS are orthogonal to one another in $L ^ { 2 } ( \mathbb { R } , \mu _ { G } )$ . Taking square and expectation over both sides yields + +$$ +\begin{array} { r } { \mathbb { E } [ \phi ( G ) ^ { 2 } ] = \mathbb { E } [ \phi ( G ) ] ^ { 2 } + \mathbb { E } [ G \phi ( G ) ] ^ { 2 } + \mathbb { E } [ \phi _ { \perp } ( G ) ^ { 2 } ] , } \end{array} +$$ + +which indicates $\begin{array} { r } { \mathbb { E } [ \phi ( G ) ^ { 2 } ] - \mathbb { E } [ \phi ( G ) ] ^ { 2 } - \mathbb { E } [ G \phi ( G ) ] ^ { 2 } = \mathbb { E } [ \phi _ { \perp } ( G ) ^ { 2 } ] \ge 0 } \end{array}$ . Note that the equality holds if and only if $\mathbb { E } [ \phi _ { \bot } ( G ) ^ { 2 } ] = 0$ , i.e. $\phi ( x ) = c _ { 0 } + c _ { 1 } x$ is linear. One application of this decomposition is to “linearize” a non-linear matrix in high dimensions, which will be useful for the following sections. + +# C PROOF OF MAIN RESULTS + +# C.1 PROOF OF LEMMA 1 + +Given features $X \in \mathbb { R } ^ { d \times n }$ , labels $\pmb { y } \in \mathbb { R } ^ { d }$ and model parameters $\pmb \theta$ , the gradient flow of $\pmb \theta$ on the squared loss $\left\| \pmb { y } - X ^ { \top } \pmb { \theta } \right\| _ { 2 } ^ { 2 }$ can be written as + +$$ +\frac { \partial \pmb { \theta } ( t ) } { \partial t } = \frac { 1 } { n } X ( y - X ^ { \top } \pmb { \theta } ( t ) ) . +$$ + +Thus with initialization $\pmb { \theta } _ { 0 }$ , the solution of this ODE at time $t$ can be written in explicit form + +$$ +\pmb { \theta } ( t ) = e ^ { - \frac { t } { n } X X ^ { \top } } \pmb { \theta } _ { 0 } + ( X X ^ { \top } ) ^ { \dagger } \left( I - e ^ { - \frac { t } { n } X X ^ { \top } } \right) X \pmb { y } . +$$ + +Since $\pmb { \theta } _ { 0 } = 0$ , taking $t \to \infty$ yields the desired result. + +# C.2 PROOF OF THEOREM 2 + +We compute the bias and variance for different cases of $\gamma _ { 1 } , \gamma _ { 2 }$ . We first discuss the case where the random feature $\Phi _ { X }$ is not full rank (Case I). Otherwise when $\Phi _ { X }$ is full rank, we discuss whether it is full column rank (Case II) or full row rank (Case III). + +Case I: $W ^ { \top } X$ is not full rank, i.e. $\gamma _ { 1 } < 1 , \gamma _ { 2 } > \gamma _ { 1 }$ . In this case rank $( \Phi _ { X } ) = d < \operatorname* { m i n } ( n , h )$ , and thus by taking the Moore-Penrose inverse we obtain + +$$ +{ \hat { \boldsymbol { \beta } } } = W ( X ^ { \top } W ) ^ { \dagger } { \boldsymbol { y } } = ( X X ^ { \top } ) ^ { - 1 } X { \boldsymbol { y } } . +$$ + +It is clear that the mean and variance is identical to the underparameterized regime in Hastie et al. (2019), i.e. when $n , d , h \infty$ , + +$$ +B 0 ; \quad V \frac { \gamma _ { 1 } } { 1 - \gamma _ { 1 } } \sigma ^ { 2 } . +$$ + +Case II: $W ^ { \top } X$ has full column rank, i.e. $\gamma _ { 2 } < 1 , \gamma _ { 1 } > \gamma _ { 2 }$ . By Lemma 1, the solution of the second layer coefficients is + +$$ +\begin{array} { r } { \hat { \beta } = W ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X y . } \end{array} +$$ + +Denote $W = U \Sigma V ^ { \top }$ the singular value decomposition. We perform the block decomposition: + +$$ +\Sigma = \left[ \Sigma _ { 0 } \right] , X = \left[ \Sigma _ { 1 } \right] , +$$ + +where $\Sigma _ { 0 } \in \mathbb { R } ^ { h \times h } , X _ { 0 } \in \mathbb { R } ^ { h \times n } , X _ { 1 } \in \mathbb { R } ^ { ( d - h ) \times n }$ , and notice that $X _ { 0 } , X _ { 1 }$ are independent. By a concentration of measure argument (e.g. Tao (2012); Hastie et al. (2019)) one can show that the quantity below tightly concentrates at its expectation. For the variance we have + +$$ +\begin{array} { r l } & { V = \mathrm { t r } \left( W \left( W ^ { \top } X X ^ { \top } W \right) ^ { - 1 } W ^ { \top } X ^ { \top } \sigma ^ { 2 } X W \left( \left( W ^ { \top } X X ^ { \top } W \right) ^ { - 1 } W ^ { \top } \right) \right. } \\ & { \left. = \sigma ^ { 2 } \mathrm { t r } \left( W ^ { \top } W \left( W ^ { \top } X X ^ { \top } W \right) ^ { - 1 } \right) = \sigma ^ { 2 } \mathrm { t r } \left( \Sigma ^ { \top } \Sigma \left( \Sigma ^ { \top } U ^ { \top } X X ^ { \top } U \Sigma \right) ^ { - 1 } \right) \right. } \\ & { \left. \sim \sigma ^ { 2 } \mathrm { t r } \left( \Sigma ^ { \top } \Sigma \left( \Sigma ^ { \top } X X ^ { \top } \Sigma \right) ^ { - 1 } \right) = \sigma ^ { 2 } \mathrm { t r } \left( \left( X _ { 0 } X _ { 0 } ^ { \top } \right) ^ { - 1 } \right) \right. \sigma ^ { 2 } \frac { \gamma _ { 2 } } { 1 - \gamma _ { 2 } } , } \end{array} +$$ + +where the last equality follows from Appendix B.2. Similarly for the bias term we have + +$$ +\begin{array} { r l } & { B = \| { \cal { W } } ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X X ^ { \top } \beta - \beta \| _ { 2 } ^ { 2 } } \\ & { \quad = \beta ^ { \top } \Big ( { \cal { W } } ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X X ^ { \top } - I _ { d } ) ^ { \top } \Big ( { \cal { W } } ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X X ^ { \top } - I _ { d } \Big ) \beta } \\ & { \stackrel { ( \psi ) } { = } \frac { { r ^ { 2 } } } { d } \mathrm { t r } ( \Big ( { \cal { W } } ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X X ^ { \top } - I _ { d } \Big ) ^ { \top } \Big ( { \cal { W } } ( W ^ { \top } X X ^ { \top } W ) ^ { - 1 } W ^ { \top } X X ^ { \top } - I _ { d } \Big ) ) } \\ & { \quad = \frac { { r ^ { 2 } } } { d } \mathrm { t r } ( \Big ( { \cal { W } } \Sigma X ^ { \top } ( V \Sigma X ^ { \top } X X ^ { \top } U \Sigma \Sigma { \cal { W } } ^ { \top } ) ^ { - 1 } V \Sigma \Sigma ^ { \top } { \cal { U } } ^ { \top } X X ^ { \top } - I _ { d } \Big ) ^ { \top } \Big ( \cdots ) ) } \\ & { \quad = \frac { { r ^ { 2 } } } { d } \mathrm { t r } ( \Big ( { \cal { W } } \Sigma ( \Sigma ^ { \top } X X ^ { \top } \Sigma ) ^ { - 1 } \Sigma ^ { \top } X X ^ { \top } { \cal { U } } ^ { \top } - { \cal { W } } \Sigma ^ { \top } \Sigma \sigma \Sigma ^ { \top } \Big ) ^ { \top } \Big ( \cdots \Big ) ) } \\ & { \quad = \frac { { r ^ { 2 } } } { d } \mathrm { t r } ( \Big ( \Sigma \Sigma ^ { \top } X X ^ { \top } \Sigma \Sigma \Big ) ^ { - 1 } \Sigma ^ { \top } X X ^ { \top } { \cal { U } } ^ { \top } - { \cal { W } } \Sigma \Big ) ^ { \top } \Big ( \cdots \Big ) ) } \\ & \quad = \frac { { r ^ { 2 } } } { d } \mathrm { t r } ( ( \Sigma ^ { \top } X ^ { \top } \Sigma ) ^ { - 1 } \end{array} +$$ + +where symmetric arguments are omitted as $( \cdot \cdot \cdot )$ , and (i) follows from the rotational invariance argument introduced in Lemma 10 and that $\beta ^ { \top } \beta = r ^ { 2 }$ . By the block decomposition (30), + +$$ +\Sigma \left( \Sigma ^ { \top } X X ^ { \top } \Sigma \right) ^ { - 1 } \Sigma ^ { \top } X X ^ { \top } - I _ { d } = \left[ \begin{array} { l l } { 0 } & { ( X _ { 0 } X _ { 0 } ^ { \top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \top } } \\ { 0 } & { - I _ { d - h } } \end{array} \right] . +$$ + +Therefore the bias term simplifies to + +$$ +\begin{array} { l } { { \displaystyle B = \frac { r ^ { 2 } } { d } \mathrm { t r } \left( \left( \Sigma ( \Sigma ^ { \top } X X ^ { \top } \Sigma ) ^ { - 1 } \Sigma ^ { \top } X X ^ { \top } - I _ { d } \right) ^ { \top } \left( \Sigma ( \Sigma ^ { \top } X X ^ { \top } \Sigma ) ^ { - 1 } \Sigma ^ { \top } X X ^ { \top } - I _ { d } \right) \right) } } \\ { { \mathrm { } = \frac { r ^ { 2 } } { d } \left( \mathrm { t r } \left( ( X _ { 0 } X _ { 0 } ^ { \top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \top } X _ { 1 } X _ { 0 } ^ { \top } ( X _ { 0 } X _ { 0 } ^ { \top } ) ^ { - 1 } \right) + ( d - h ) \right) } } \\ { { \mathrm { } \to \frac { r ^ { 2 } } { d } \left( \frac { ( d - h ) h } { n - h - 1 } + d - h \right) \to \frac { \gamma _ { 1 } - \gamma _ { 2 } } { \gamma _ { 1 } \left( 1 - \gamma _ { 2 } \right) } r ^ { 2 } . } } \end{array} +$$ + +Thus we have obtained that as $n , d , h \infty$ + +$$ +B \frac { \gamma _ { 1 } - \gamma _ { 2 } } { \gamma _ { 1 } ( 1 - \gamma _ { 2 } ) } r ^ { 2 } . +$$ + +Case III: $W ^ { \top } X$ has full row rank, i.e. $\gamma _ { 1 } > 1 , \gamma _ { 2 } > 1$ . Similarly, the least squares solution is + +$$ +\hat { \beta } = W W ^ { \top } X ( X ^ { \top } W W ^ { \top } X ) ^ { - 1 } \pmb { y } , +$$ + +Simplifying the variance: + +$$ +\begin{array} { r l } & { V = \mathrm { t r } \left( W W ^ { \top } X \left( X ^ { \top } W W ^ { \top } X \right) ^ { - 1 } \sigma ^ { 2 } \left( X ^ { \top } W W ^ { \top } X \right) ^ { - 1 } X ^ { \top } W W ^ { \top } \right) } \\ & { \quad = \sigma ^ { 2 } \mathrm { t r } \left( W W ^ { \top } U \Sigma V ^ { \top } \left( V \Sigma ^ { \top } U ^ { \top } W W ^ { \top } U \Sigma V ^ { \top } \right) ^ { - 2 } V \Sigma ^ { \top } U ^ { \top } W W ^ { \top } \right) } \\ & { \quad \sim \sigma ^ { 2 } \mathrm { t r } \left( W W ^ { \top } \Sigma \left( \Sigma ^ { \top } W W ^ { \top } \Sigma \right) ^ { - 2 } \Sigma ^ { \top } W W ^ { \top } \right) . } \end{array} +$$ + +where we applied the SVD of $X = U \Sigma V ^ { \top }$ and the rotational invariance argument. Using a similar block decomposition on $W$ : + +$$ +\Sigma = \left[ \stackrel { \Sigma _ { 0 } } { 0 } \right] , W = \left[ \stackrel { W _ { 0 } } { W _ { 1 } } \right] , +$$ + +where $\Sigma _ { 0 } \in \mathbb { R } ^ { n \times n } , W _ { 0 } \in \mathbb { R } ^ { n \times h } , W _ { 1 } \in \mathbb { R } ^ { ( d - n ) \times h }$ , and $W _ { 0 } , W _ { 1 }$ independent. We thus simplify $V$ as + +$$ +\begin{array} { r l } & { V = \sigma ^ { 2 } \mathrm { t r } ( W W ^ { \top } \Sigma ( \Sigma ^ { \top } W W ^ { \top } \Sigma ) ^ { - 2 } \Sigma ^ { \top } W W ^ { \top } ) } \\ & \quad = \sigma ^ { 2 } \mathrm { t r } ( [ \begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ( \Sigma _ { 0 } ^ { \top } W _ { 0 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ) ^ { - 2 } \Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \top } } & { \cdots } & { \cdots } \\ { ( \begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ( \Sigma _ { 0 } ^ { \top } W _ { 0 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ) ^ { - 2 } \Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \top } } & { W _ { 1 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ( \Sigma _ { 0 } ^ { \top } W _ { 0 } W _ { 0 } ^ { \top } \Sigma _ { 0 } ) ^ { - 2 } \Sigma _ { 0 } W _ { 0 } W _ { 1 } ^ { \top } } \end{array} ) } \\ & { \quad = \sigma ^ { 2 } ( \mathrm { t r } ( \Sigma _ { 0 } ^ { - T } \Sigma _ { 0 } ^ { - 1 } ) + \mathrm { t r } ( W _ { 1 } W _ { 0 } ^ { \top } ( W _ { 0 } W _ { 0 } ^ { \top } ) ^ { - 1 } \Sigma _ { 0 } ^ { - T } \Sigma _ { 0 } ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \top } ) ^ { - 1 } W _ { 0 } W _ { 1 } ^ { \top } ) ) } \\ & { \quad = \sigma ^ { 2 } \mathrm { t r } ( ( X ^ { \top } X ) ^ { - 1 } ) + \sigma ^ { 2 } \mathrm { t r } ( W _ { 1 } ^ { \top } W _ { 1 } W _ { 0 } ^ { \top } ( W _ { 0 } W _ { 0 } ^ { \top } ) ^ { - 1 } ( \Sigma _ { 0 } ^ { \top } \Sigma _ { 0 } ) ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \top } ) ^ { - 1 } W _ { 0 } ) \cdot \ ( 3 9 ) } \end{array} \end{array} +$$ + +Hence we obtain the following expression on the variance + +$$ +V \sigma ^ { 2 } \frac { 1 } { \gamma _ { 1 } - 1 } + \sigma ^ { 2 } ( d - n ) \mathbb { E } _ { W , X } V \mathrm { t r } ( ( W _ { 0 } W _ { 0 } ^ { \top } ) ^ { - 1 } ( \Sigma _ { 0 } ^ { \top } \Sigma _ { 0 } ) ^ { - 1 } ) \sigma ^ { 2 } ( \frac { 1 } { \gamma _ { 1 } - 1 } + \frac { 1 } { \gamma _ { 2 } - 1 } ) . +$$ + +We omit the derivation of bias, which follows a similar derivation: + +$$ +B \frac { \gamma _ { 2 } ( \gamma _ { 1 } - 1 ) } { \gamma _ { 1 } ( \gamma _ { 2 } - 1 ) } r ^ { 2 } . +$$ + +Combining Case I, II, III yields theorem 2. + +# C.3 PROOF OF PROPOSITION 3 + +Given the squared loss, one can derive the dynamics of $W$ with fixed second layer $^ { a }$ w.r.t the loss: + +$$ +\frac { \partial W ( t ) } { \partial t } = - \frac { 1 } { n } X ( \pmb { y } - X ^ { \top } W ( t ) \pmb { a } ) \pmb { a } ^ { \top } . +$$ + +Note that the update of $W$ can be written as a linear combination of $^ { a }$ . Since $W ( 0 ) = 0$ , we can write $W ( t ) = \bar { \hat { w } } ( t ) \mathbf { { a } } ^ { \top }$ for some $\hat { \textbf { \textit { w } } }$ . The corresponding flow on $\hat { w }$ is + +$$ +\frac { \partial \pmb { \hat { w } } ( t ) } { \partial t } = - \frac { 1 } { n } X ( \pmb { y } - \boldsymbol { X } ^ { \top } \pmb { \hat { w } } ( t ) \left\| \pmb { a } \right\| _ { 2 } ^ { 2 } ) , +$$ + +which gives the following solution + +$$ +{ \hat { \pmb { w } } } ^ { * } = { \frac { 1 } { \left\| \pmb { a } \right\| _ { 2 } ^ { 2 } } } { \boldsymbol { X } } ^ { \dagger } \pmb { y } \Rightarrow { \hat { \beta } } = { \boldsymbol { W } } ^ { * } \pmb { a } = { \boldsymbol { X } } ^ { \dagger } \pmb { y } . +$$ + +Thus gradient flow on the first layer leads to the minimum-norm solution on the input features. + +# C.4 PROOF OF THEOREM 4 + +Following the bias-variance decomposition (7), the variance term can be written as ( $\cdot \sigma ^ { 2 }$ omitted) + +$$ +\begin{array} { r l } & { V = \mathbb { E } _ { \pmb { x } , \pmb { \varepsilon } } \Big [ \| \pmb { a } ^ { \top } \phi ( W ^ { \top } \pmb { x } ) - \mathbb { E } _ { \pmb { \varepsilon } } \pmb { a } ^ { \top } \phi ( W ^ { \top } \pmb { x } ) \| _ { 2 } ^ { 2 } \Big ] } \\ & { ~ = \mathbb { E } _ { \pmb { x } } \Big [ \Big \| \big [ \phi ( W ^ { \top } \pmb { X } ) \big ] ^ { \dagger } \phi ( W ^ { \top } \pmb { x } ) \Big \| _ { 2 } ^ { 2 } \Big ] } \\ & { ~ = \mathrm { t r } \left( \big [ \phi ( X ^ { \top } W ) \big ] ^ { \dagger } \big [ \phi ( W ^ { \top } \pmb { X } ) \big ] ^ { \dagger } \mathbb { E } _ { \pmb { x } } \Big [ \phi ( W ^ { \top } \pmb { x } ) \phi ( W ^ { \top } \pmb { x } ) ^ { \top } \Big ] \right) } \\ & { ~ = \mathrm { t r } \left( \big [ \phi ( X ^ { \top } W ) \big ] ^ { \dagger } \big [ \phi ( W ^ { \top } \pmb { X } ) \big ] ^ { \dagger } K _ { W } \right) , } \end{array} +$$ + +where we define the expected non-linear Gram matrix $K _ { W } \in \mathbb { R } ^ { h \times h }$ as + +$$ +K _ { W } = \mathbb { E } _ { \pmb { x } } \Big [ \phi ( W ^ { \top } \pmb { x } ) \phi ( W ^ { \top } \pmb { x } ) ^ { \top } \Big ] . +$$ + +and for each entry we have $( K _ { W } ) _ { [ i , j ] } = \mathbb { E } _ { \pmb { x } } \Big [ \phi ( \pmb { w } _ { i } ^ { \top } \pmb { x } ) \phi ( \pmb { w } _ { j } ^ { \top } \pmb { x } ) \Big ] .$ + +Random matrix in the form of covariance matrix of nonlinear features has been studied in many works Hastie et al. (2019); Mei and Montanari (2019); Liao and Couillet (2018); Louart et al. (2018); Pennington and Worah (2017). We note that our setup for the variance term is very similar to that for nonlinear features in Hastie et al. (2019) with modifications mentioned below. + +In contrast to the linear network in Section C.2, the Gram matrix of a nonlinear activation is almost surely full-rank, as specified in the following lemma from Pennington and Worah (2017): + +Lemma 11. Suppose √ $\phi$ is not linear. Then the smallest singular value of $\Phi = \phi ( W ^ { \top } X ) \in \mathbb { R } ^ { h \times n }$ is of order $O ( { \sqrt { n } } )$ . To be precise, consider the empirical spectral density $\mu _ { \Phi \Phi ^ { \top } / n } ( d \lambda )$ , when $n , d , h \infty$ with $d / n \gamma _ { 1 }$ and $h / n \gamma _ { 2 }$ , $\mu _ { \Phi \Phi ^ { \top } / n } ( d \lambda )$ converges weakly to + +$$ +\mu _ { \Phi \Phi ^ { \top } / n } ( d \lambda ) [ 1 - \gamma _ { 2 } ^ { - 1 } ] _ { + } \delta _ { 0 } ( \lambda ) d \lambda + \mu ^ { + } ( d \lambda ) , +$$ + +where $\mu ^ { + } ( d \lambda )$ has non-negative support $\lbrack \rho , \infty )$ with $\rho > 0$ . + +We therefore consider two scenarios: $\Phi$ is full column rank (Case I) or full row rank (Case II). + +Case 1. $h < n$ . In this case (45) simplifies into + +$$ +V = \operatorname { t r } \left( \left( \phi ( W ^ { \top } X ) \phi ( X ^ { \top } W ) \right) ^ { - 1 } K _ { W } \right) = \operatorname* { l i m } _ { \xi \to 0 ^ { - } } \operatorname { t r } \left( \left( \Phi \Phi ^ { \top } - \xi I \right) ^ { - 1 } K _ { W } \right) = \operatorname* { l i m } _ { \xi \to 0 ^ { - } } V _ { \xi } , +$$ + +where the continuity and boundness of $V _ { \xi }$ at $\xi = 0 ^ { - }$ is guaranteed by Lemma 11 when $\gamma _ { 2 } = h / n \neq 1$ . A ridge $\xi$ is added make use of (Louart et al., 2018, Theorem 1), which derived the asymptotic equivalent of the resolvent $\left( \Phi \Phi ^ { \top } - \xi I \right) ^ { - 1 }$ . It follows that as $n , d , h \infty$ , + +$$ +\begin{array} { r l } & { \mathrm { t r } \left( h ^ { - 1 } \big ( \Phi \Phi ^ { \top } - \xi I \big ) ^ { - 1 } \left( \frac { n } { h } \frac { K _ { W } } { 1 + h ^ { - 1 } \mathrm { t r } \big ( h \big ( \Phi \Phi ^ { \top } - \xi I \big ) ^ { - 1 } K _ { W } \big ) } \right) - h ^ { - 1 } I \right) } \\ & { \le \displaystyle \frac { 1 } { h } \left\| \big ( \Phi \Phi ^ { \top } - \xi I \big ) ^ { - 1 } - \left( \frac { n } { h } \frac { K _ { W } } { 1 + h ^ { - 1 } \mathrm { t r } \big ( h \big ( \Phi \Phi ^ { \top } - \xi I \big ) ^ { - 1 } K _ { W } \big ) } \right) ^ { - 1 } \right\| _ { F } } \\ & { \quad \cdot \left\| \frac { n } { h } \frac { K _ { W } } { 1 + h ^ { - 1 } \mathrm { t r } \big ( h \big ( \Phi \Phi ^ { \top } - \xi I \big ) ^ { - 1 } K _ { W } \big ) } \right\| _ { 2 } } \\ & { \le \displaystyle \frac { 1 } { h } O ( n ^ { - 1 / 2 + \varepsilon } ) O ( n ^ { 1 / 2 } ) \to 0 . } \end{array} +$$ + +where we have used the inequality $\operatorname { t r } \left( A B \right) \leq \left\| A \right\| _ { F } \left\| B \right\| _ { 2 }$ , and equivalently by taking $\xi 0$ we get + +$$ +\operatorname* { l i m } _ { \xi \to 0 } \operatorname* { l i m } _ { n , d , h \to \infty } \frac { n } { h } \frac { \mathrm { t r } \left( \left( \Phi \Phi ^ { \top } - \xi I \right) ^ { - 1 } K _ { W } \right) } { 1 + \mathrm { t r } \left( \left( \Phi \Phi ^ { \top } - \xi I \right) ^ { - 1 } K _ { W } \right) } = 1 . +$$ + +Therefore $\begin{array} { r } { \operatorname* { l i m } _ { \xi \to 0 } \operatorname* { l i m } _ { n , d , h \to \infty } n / h \cdot \mathrm { t r } \left( \left( \Phi \Phi ^ { \top } - \xi I \right) ^ { - 1 } K _ { W } \right) = \gamma _ { 2 } / ( 1 - \gamma _ { 2 } ) . } \end{array}$ . Note that $\partial V _ { \xi } / \partial \xi =$ $\mathrm { t r } \left( \xi ( \Phi \Phi ^ { \top } - \xi I ) ^ { - 2 } K _ { W } \right)$ is bounded around the neighbourhood of $\xi = 0$ , hence following the same argument as (Hastie et al., 2019, Theorem 4), we exchange the limit of $\xi 0$ and $n , d , h 0$ : + +$$ +V \frac { \gamma _ { 2 } } { 1 - \gamma _ { 2 } } . +$$ + +Case 2. $h > n$ . Techniques used in the current proof are largely borrowed from Hastie et al. (2019); Cheng and Singer (2013), and we include the full proof for completeness. It should be noted that compared to Hastie et al. (2019) we handle the non-zero expectation of the nonlinearity under Gaussian distribution, i.e. the off-diagonal entries of the kernel matrix is no longer zero-centered. For simplicity we mainly adhere to the notations in Hastie et al. (2019). + +We briefly summarizes the procedure for deriving $V$ . Instead of calculating the variance directly, we analyze a modified quantity $V _ { \xi }$ and then take $\xi 0$ , which can be connected to the trace of the resolvent of matrix $\tilde { A }$ defined in (56); this translates the calculation of $V _ { \xi }$ into the calculation of the Stieltjes transform of $\tilde { A }$ (57), (61), (62). + +C.5 DERIVING THE VARIANCE $V$ FOR $h > n$ + +Step 1. An equivalent expression. For notational simplicity we omit the magnitude $\sigma$ : + +$$ +\begin{array} { r } { V = \mathrm { t r } \bigg ( \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 2 } \phi ( X ^ { \top } W ) K _ { W } \bigg ) . } \end{array} +$$ + +and due to the same continuity argument as in Case 1 we have + +$$ +V = \operatorname * { l i m } _ { \xi 0 } \frac { 1 } { n } \Big [ \mathrm { t r } ( S ( S ^ { \top } S - \xi I _ { n } ) ^ { - 2 } S ^ { \top } K _ { W } ) \Big ] = \operatorname * { l i m } _ { \xi 0 } V _ { \xi } . +$$ + +where $S = \phi ( W ^ { \top } X ) / \sqrt { n } = \Phi / \sqrt { n } , \xi \in \mathbb { C }$ and $\Im \xi > 0$ or $\xi < 0$ . + +We decompose the normalized feature matrix $S = \phi ( W ^ { \top } X ) / \sqrt { n }$ as $S = U \Sigma V ^ { \top }$ , where $\Sigma =$ $\mathrm { d i a g } _ { h \times n } \big ( \mathring { \phi _ { 1 } } , \cdot \cdot \cdot , \phi _ { n } \big ) \in \mathbb { R } ^ { h \times n }$ is a tall diagonal matrix, and $U \overset { ^ { \prime } } { = } [ \pmb { u } _ { 1 } , \cdots , \pmb { u } _ { h } ] \in \mathbb { R } ^ { h \times h }$ is the set of orthogonal eigenvectors of $S \underline { { S } } ^ { \top } = \phi ( \underline { { W } } ^ { \top } X ) \phi ( X ^ { \top } W ) / n$ , and $V \in \mathbb { R } ^ { n \times n }$ is the set of orthogonal eigenvectors of $S ^ { \top } S = \phi ( X ^ { \top } W ) \phi ( \dot { W } ^ { \top } X ) / n$ . Now the variance can be written as + +$$ +V _ { \xi } = \frac { 1 } { n } \mathrm { t r } \left( U \Sigma ( \Sigma ^ { \top } \Sigma + \xi I _ { n } ) ^ { - 2 } \Sigma U ^ { \top } K _ { W } \right) . +$$ + +By the same argument as in Lemma 13 of Hastie et al. (2019) one can show that when $n , d , h \infty$ + +$$ +V _ { \xi } = \frac { 1 } { n } \Big [ \mathrm { t r } \left( U \Sigma ( \Sigma ^ { \top } \Sigma + \xi I _ { n } ) ^ { - 2 } \Sigma U ^ { \top } K _ { W } \right) \Big ] \to \frac { 1 } { n } \Big [ \mathrm { t r } \left( U \Sigma ( \Sigma ^ { \top } \Sigma + \xi I _ { n } ) ^ { - 2 } \Sigma U ^ { \top } \tilde { K } _ { W } \right) \Big ] , +$$ + +in which $\tilde { K } _ { W }$ is the approxmation of $K _ { W }$ defined in Lemma 16. Writing the trace explicitly (denote eigenvalues $\lambda _ { i } = \phi _ { i } ^ { 2 }$ , and $\phi _ { n + 1 } = \cdot \cdot \cdot = \phi _ { h } = 0 )$ ), we have + +$$ +V _ { \xi } \to \frac { 1 } { n } \mathrm { t r } \left( U \Sigma ( \Sigma ^ { \top } \Sigma + \xi I _ { n } ) ^ { - 2 } \Sigma U ^ { \top } \tilde { K } _ { W } \right) = \gamma _ { 2 } \frac { 1 } { h } \sum _ { i = 1 } ^ { h } \frac { \lambda _ { i } } { ( \lambda _ { i } + \xi ) ^ { 2 } } { \pmb { u } } _ { i } ^ { \top } \tilde { K } _ { W } { \pmb { u } } _ { i } . +$$ + +Since the positive support of spectrum $\lambda$ is lower bounded and the density at 0 is $1 - \gamma _ { 2 } ^ { - 1 }$ , we have + +$$ +\begin{array} { r } { \gamma _ { \xi } \to \gamma _ { 2 } \displaystyle \frac { 1 } { h } \operatorname* { l i m } _ { h , d , n \to \infty } \sum _ { i = 1 } ^ { h } \frac { \lambda _ { i } } { ( \lambda _ { i } + \xi ) ^ { 2 } } u _ { i } ^ { \top } \tilde { K } _ { W } u _ { i } = \gamma _ { 2 } \displaystyle \int \frac { \lambda } { ( \lambda + \xi ) ^ { 2 } } \mu _ { \infty } ( d \lambda ) = \gamma _ { 2 } \displaystyle \int _ { \lambda > \rho } \frac { \lambda } { ( \lambda + \xi ) ^ { 2 } } \mu _ { \infty } ^ { + } ( d \lambda ) , } \end{array} +$$ + +where we define $\begin{array} { r } { \mu _ { n } ( x ) = \frac { 1 } { h } \sum _ { i = 1 } ^ { h } \delta _ { \lambda _ { i } } ( x ) \pmb { u } _ { i } ^ { \top } \tilde { K } _ { W } \pmb { u } _ { i } } \end{array}$ and its positive part $\mu _ { n } ^ { + } ( x )$ . Hence we have + +$$ +V = \operatorname* { l i m } _ { \xi \to 0 } V _ { \xi } = \operatorname* { l i m } _ { \xi \to 0 } \gamma _ { 2 } \int _ { \lambda > \rho } \frac { \lambda } { ( \lambda + \xi ) ^ { 2 } } \mu _ { \infty } ^ { + } ( d \lambda ) = \gamma _ { 2 } \int _ { \lambda > \rho } \frac { 1 } { \lambda } \mu _ { \infty } ^ { + } ( d \lambda ) . +$$ + +We define the following matrix $\tilde { A } _ { n } ( \rho , \varsigma , \tau ) \in \mathbb { R } ^ { N \times N }$ where $N = n + h$ : + +$$ +\boldsymbol { \tilde { A } _ { n } } ( \rho , \varsigma , \tau ) = \left[ \begin{array} { c c } { \rho I _ { h } + \varsigma \mathbf { 1 } _ { h } \mathbf { 1 } _ { h } ^ { \top } + \tau Q } & { S } \\ { S ^ { \top } } & { 0 _ { n } } \end{array} \right] . +$$ + +And denote the Stieltjes transform of ${ \tilde { A } } _ { n }$ as + +$$ +\tilde { m } _ { n } ( \xi , \rho , \varsigma , \tau ) = \frac { 1 } { n } \mathrm { t r } \left( ( \tilde { A } _ { n } ( \rho , \varsigma , \tau ) - \xi I _ { N } ) ^ { - 1 } \right) . +$$ + +Then following the definition of $\tilde { K } _ { W }$ one can show that + +$$ +\tilde { m } _ { n } ( \xi , r x , s x , t x ) = \frac { 1 } { n } \mathrm { t r } \left( \left[ \begin{array} { c c } { \tilde { K } _ { W } x - \xi I _ { h } } & { S } \\ { S ^ { \top } } & { - I _ { n } } \end{array} \right] ^ { - 1 } \right) , +$$ + +and taking matrix derivative gives + +$$ +\begin{array} { r l } & { - \displaystyle \frac { \partial } { \partial x } \tilde { m } _ { n } ( \xi , r x , s x , t x ) \Big | _ { x = 0 } = \frac { 1 } { n } \mathrm { t r } \left( \left[ \begin{array} { c c } { - \xi I _ { h } } & { S } \\ { S ^ { \top } } & { - I _ { n } } \end{array} \right] ^ { - 1 } \left[ \begin{array} { c c } { \tilde { K } _ { W } } & { 0 } \\ { 0 } & { 0 } \end{array} \right] \left[ \begin{array} { c c } { - \xi I _ { h } } & { S } \\ { S ^ { \top } } & { - I _ { n } } \end{array} \right] ^ { - 1 } \right) } \\ & { \quad \quad \quad \quad \quad \quad \quad \quad \quad = \frac { 1 } { n } \mathrm { t r } \left( \left[ \begin{array} { c c } { \xi ^ { 2 } I _ { h } + S S ^ { \top } } & { 0 } \\ { 0 } & { I _ { n } + S ^ { \top } S } \end{array} \right] ^ { - 1 } \left[ \begin{array} { c c } { \tilde { K } _ { W } } & { 0 } \\ { 0 } & { 0 } \end{array} \right] \right) } \\ & { \quad \quad \quad \quad \quad \quad \quad \quad \quad = \frac { 1 } { n } \mathrm { t r } \left( U ( \Sigma \Sigma ^ { \top } + \xi ^ { 2 } I _ { h } ) ^ { - 1 } U ^ { \top } \tilde { K } _ { W } \right) = \frac { 1 } { n } \displaystyle \sum _ { i = 1 } ^ { h } \frac { 1 } { \lambda + \xi ^ { 2 } } u _ { i } ^ { \top } \tilde { K } _ { W } u _ { i } . } \end{array} +$$ + +Denote the limit $\begin{array} { r } { \tilde { m } ( \xi , \rho , \varsigma , \tau ) = \operatorname* { l i m } _ { n , h , d \infty } \tilde { m } _ { n } ( \xi , \rho , \varsigma , \tau ) } \end{array}$ , and the derivative is given as + +$$ +\begin{array} { l } { \displaystyle \left. - \frac { \partial } { \partial x } \tilde { m } ( \xi , r x , s x , t x ) \right. _ { x = 0 } = \displaystyle \operatorname* { l i m } _ { h , d , n \to \infty } \frac { 1 } { n } \sum _ { i = 1 } ^ { h } \frac { 1 } { \lambda + \xi ^ { 2 } } { \boldsymbol u } _ { i } ^ { \top } \tilde { K } _ { W } { \boldsymbol u } _ { i } } \\ { \displaystyle = \gamma _ { 2 } \int _ { \lambda \geq 0 } \frac { 1 } { \lambda + \xi ^ { 2 } } \mu _ { \infty } ( d \lambda ) = \frac { \gamma _ { 2 } - 1 } { \xi ^ { 2 } } + \gamma _ { 2 } \int _ { \lambda > \rho } \frac { 1 } { \lambda + \xi ^ { 2 } } \mu _ { \infty } ^ { + } ( d \lambda ) . } \end{array} +$$ + +For simplicity we define the following function on $\xi$ : + +$$ +q ( \xi ) = - \frac { \partial } { \partial x } \tilde { m } ( \xi , r x , s x , t x ) \Big | _ { x = 0 } , +$$ + +$\begin{array} { r } { q _ { + } ( \xi ) = q ( \xi ) - \frac { \gamma _ { 2 } - 1 } { \xi ^ { 2 } } = \gamma _ { 2 } \int _ { \lambda > \rho } \frac 1 { \lambda + \xi ^ { 2 } } \mu _ { \infty } ^ { + } ( d \lambda ) } \end{array}$ + +$$ +V = \operatorname* { l i m } _ { \xi \to 0 } q _ { + } ( \xi ) . +$$ + +Step 2. Calculating $q ( \xi )$ and $m _ { n } ( \xi , \rho , \varsigma , \tau )$ This subsection aims to calculate $\tilde { m } ( \xi , \rho , \varsigma , \tau )$ and $q ( \xi ) = - \tilde { m } _ { x } ^ { \prime } ( \xi , r x , s x , t x ) | _ { x = 0 }$ , from which the variance can be computed from (57)(61)(62). + +We define $A _ { n }$ by subtracting the off-diagonal entries of the upper-left block of ${ \tilde { A } } _ { n }$ : + +$$ +A _ { n } ( \rho , \tau ) = \left[ \begin{array} { c c } { \rho I _ { h } + \tau Q } & { S } \\ { S ^ { \top } } & { 0 _ { n } } \end{array} \right] , +$$ + +where $S = \tilde { S } - a _ { 0 } I _ { p \times n }$ , i.e. $S _ { i k } = \phi ( \pmb { w } _ { i } ^ { \top } \pmb { x } _ { k } ) - a _ { 0 } = \varphi ( \pmb { w } _ { i } ^ { \top } \pmb { x } _ { k } )$ , $a _ { 0 } = \mathbb { E } [ \phi ( x ) ]$ . The Stieltjes transform of $A _ { n }$ given by $\begin{array} { r } { m _ { n } ( \xi , \rho , \tau ) = \frac { 1 } { n } \mathrm { t r } \left( ( A _ { n } ( \rho , \tau ) - \xi I _ { N } ) ^ { - 1 } \right) } \end{array}$ . The following Lemma shows that $\tilde { m } _ { n }$ and $m _ { n }$ have the same limit: + +Lemma 12. when $n \to \infty$ and for $\Im \xi > 0 o r \xi < 0 $ , we have $m _ { n } ( \xi , \rho , \tau ) \tilde { m } _ { n } ( \xi , \rho , \varsigma , \tau ) .$ + +Proof. By definition of ${ \bar { A } } _ { n }$ and $A _ { n }$ , + +$$ +\begin{array} { r } { \tilde { A } _ { n } ( \rho , \varsigma , \tau ) - A _ { n } ( \rho , \tau ) = \left[ \begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\ { 0 _ { n } } & { 1 _ { n } } \end{array} \right] \left[ \begin{array} { c c } { \xi } & { a _ { 0 } } \\ { a _ { 0 } } & { 0 } \end{array} \right] \left[ \begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\ { 0 _ { n } } & { 1 _ { n } } \end{array} \right] ^ { \top } } \end{array} +$$ + +which is a rank-2 matrix. By theorem A.43 from Bai and Silverstein (2010), which characterizes the effect of finite-rank perturbation on the e.s.d. of random matrices: + +$$ +\operatorname* { s u p } _ { x } | F ^ { \tilde { A } _ { n } } ( x ) - F ^ { A _ { n } } ( x ) | \leq O \left( n ^ { - 1 } \right) , +$$ + +where $F ^ { M }$ is the empirical spectral distribution of $M \in \mathbb { R } ^ { n \times n }$ . The claim follows from the Stieltjes continuity theorem (e.g. Section 2.4 in Tao (2012)). □ + +To calculate the Stieltjes transform $m _ { n }$ , we take advantage of the block structure of $A _ { n }$ + +$$ +\begin{array} { l } { { m _ { 1 , n } ( \xi , \rho , \tau ) = \displaystyle \frac { 1 } { p } \mathrm { t r } \left( ( A _ { n } ( \rho , \tau ) - \xi I _ { N } ) _ { [ 1 . . p , 1 . . p ] } ^ { - 1 } \right) , } } \\ { { \displaystyle m _ { 2 , n } ( \xi , \rho , \tau ) = \displaystyle \frac { 1 } { n } \mathrm { t r } \left( ( A _ { n } ( \rho , \tau ) - \xi I _ { N } ) _ { [ p + 1 . . p + n , p + 1 . . p + n ] } ^ { - 1 } \right) . } } \end{array} +$$ + +One can observe that $\begin{array} { r } { m _ { n } ( \xi , \rho , \tau ) = \gamma _ { 2 } m _ { 1 , n } ( \xi , \rho , \tau ) + m _ { 2 , n } ( \xi , \rho , \tau ) . } \end{array}$ + +In the following equations we omit the subscript $n$ , as well as dependency on $\rho , \varsigma , \tau$ . Following Hastie et al. (2019), we rewrite the matrix $A _ { n } = A = { \left[ \begin{array} { l l } { A _ { * } } & { a } \\ { \mathbf { 1 } } & { 0 } \end{array} \right] }$ , where $A ^ { * }$ is a $\left( N - 1 \right) \times \left( N - 1 \right)$ matrix with last column and row of $A$ removed and $\pmb { s }$ the activation vector: + +$$ +\begin{array} { r } { A _ { * } = \left[ \begin{array} { c c } { \rho I _ { h } + \tau Q } & { S _ { * } } \\ { S _ { * } ^ { \top } } & { 0 _ { n - 1 } } \end{array} \right] ; \quad \pmb { a } ^ { \top } = [ \phi ( \boldsymbol { W } ^ { \top } \mathbf { x } _ { n } ) ^ { \top } \quad \mathbf { 0 } _ { n - 1 } ^ { \top } ] = [ \boldsymbol { s } ^ { \top } \quad \mathbf { 0 } _ { n - 1 } ^ { \top } ] . } \end{array} +$$ + +Hence by the block matrix inverse formula + +$$ +\begin{array} { r } { ( A - \xi I _ { N } ) ^ { - 1 } = \left[ \begin{array} { l l } { * } & { * } \\ { * } & { [ - \xi - { \pmb a } ^ { \top } ( A _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } { \pmb a } ] ^ { - 1 } } \end{array} \right] . } \end{array} +$$ + +Plugging this back in the Stieltjes transform we have + +$$ +\begin{array} { l } { \displaystyle m _ { 2 , n } ( \xi , \rho , \tau ) = \frac { 1 } { n } \mathrm { t r } \left( ( A _ { n } ( \rho , \varsigma , \tau ) - \xi I _ { N } ) _ { [ h + 1 , N ] } ^ { - 1 } \right) = \mathbb { E } _ { a } \left[ ( A _ { n } ( \rho , \tau ) - \xi I _ { N } ) ^ { - 1 } \right] _ { N N } } \\ { \displaystyle = \mathbb { E } _ { a } \bigg [ \Big ( - \xi - a ^ { \top } ( A ^ { * } - \xi I _ { N - 1 } ) ^ { - 1 } a \Big ) ^ { - 1 } \bigg ] . } \end{array} +$$ + +To obtain an asymptotic description of $m _ { 2 , n }$ , we perform the orthonormal decomposition on the nonlinearity $\varphi$ introduced in Cheng and Singer (2013): $\varphi ( x ) \ = \ a _ { 1 } x + \varphi _ { \perp } ( x )$ , where $a _ { 1 } = \mathbb { E } _ { x \sim \mathcal { N } ( 0 , 1 ) } [ x \varphi ( x ) ]$ . For each ${ \pmb w } _ { i } ( \bar { 1 } \leq i \leq \bar { h } )$ , we perform the following orthonormal decomposition (along the direction of ${ \bf { x } } _ { n }$ and the direction of $\tilde { \mathbf { \pmb { w } } } _ { i }$ perpendicular to ${ \mathbf { \mathcal { x } } } _ { n }$ ): + +$$ +\pmb { w } _ { i } = \underbrace { \pmb { w } _ { i } ^ { \top } \pmb { x } _ { n } } _ { \eta _ { i } } \frac { \pmb { x } _ { n } } { \| \pmb { x } _ { n } \| } + \tilde { \pmb { w } } _ { i } = \eta _ { i } \frac { \pmb { x } _ { n } } { \| \pmb { x } _ { n } \| } + \tilde { \pmb { w } } _ { i } . +$$ + +We can thus simplify the activation vector $\textbf { \em a }$ as + +$$ +\begin{array} { r l } { a ^ { \top } = \Bigg [ \frac { 1 } { \sqrt { n } } \varphi ( \| x _ { n } \| \eta _ { 1 } ) } & { \cdots \cdot \frac { 1 } { \sqrt { n } } \varphi ( \| x _ { n } \| \eta _ { h } ) \quad \underbrace { 0 \cdots \cdot 0 } _ { n - 1 } \Bigg ] } \\ & { = \underbrace { \Bigg [ \frac { 1 } { \sqrt { n } } a _ { 1 } \| x _ { n } \| \eta _ { 1 } } _ { \alpha _ { 1 } ^ { \top } = \| x _ { n } ^ { \top } \| ^ { \alpha } } \cdots \quad \frac { 1 } { \sqrt { n } } a _ { 1 } \| x _ { n } \| \eta _ { h } \quad \underbrace { 0 \cdots \cdot 0 } _ { n - 1 } \Bigg ] } \\ & { + \underbrace { \Bigg [ \frac { 1 } { \sqrt { n } } \varphi _ { \bot } ( \| x _ { n } \| \eta _ { 1 } ) } _ { \alpha _ { 1 } ^ { \top } = \| x _ { n } ^ { \top } \| ^ { \alpha } } \cdots \quad \frac { 1 } { \sqrt { n } } \varphi _ { \bot } ( \| x _ { n } \| \eta _ { h } ) \quad \underbrace { 0 \cdot \cdots \cdot 0 } _ { n - 1 } \Bigg ] . } \end{array} +$$ + +and for $1 \leq i \neq j \leq h , 1 \leq k \leq n - 1$ , + +$$ +Q _ { i j } = \left( \eta _ { i } \frac { \pmb { x _ { n } } } { \lVert \pmb { x _ { n } } \rVert } + \tilde { \pmb { w } } _ { i } \right) ^ { \top } \left( \eta _ { j } \frac { \pmb { x _ { n } } } { \lVert \pmb { x _ { n } } \rVert } + \tilde { \pmb { w } } _ { j } \right) = \eta _ { i } \eta _ { j } + \underbrace { \tilde { \pmb { w } } _ { i } ^ { \top } \tilde { \pmb { w } } _ { j } } _ { \tilde { Q } _ { i j } } . +$$ + +Similarly we decompose the activation function in $S$ , + +$$ +\begin{array} { l } { { \displaystyle { S _ { i k } = \frac { 1 } { \sqrt n } \varphi \left( \eta _ { i } \frac { x _ { n } ^ { \top } x _ { k } } { \| x _ { n } \| } + \tilde { w } _ { i } ^ { \top } x _ { k } \right) = \frac { 1 } { \sqrt n } a _ { 1 } \eta _ { i } \frac { x _ { n } ^ { \top } x _ { k } } { \| x _ { n } \| } + \frac { 1 } { \sqrt n } a _ { 1 } \tilde { w } _ { i } ^ { \top } x _ { k } + \frac { 1 } { \sqrt n } \varphi _ { \bot } \left( \eta _ { i } \frac { x _ { n } ^ { \top } x _ { k } } { \| x _ { n } \| } + \tilde { w } _ { i } ^ { \top } x _ { k } \right) } } \\ { { \displaystyle { \quad = \frac { 1 } { \sqrt n } \varphi ( \tilde { w } _ { i } ^ { \top } x _ { k } ) + \frac { 1 } { \sqrt n } a _ { 1 } \eta _ { i } \frac { x _ { n } ^ { \top } x _ { k } } { \| x _ { n } \| } + \frac { 1 } { \sqrt n } \left[ \varphi _ { \bot } \left( \eta _ { i } \frac { x _ { n } ^ { \top } x _ { k } } { \| x _ { n } \| } + \tilde { w } _ { i } ^ { \top } x _ { k } \right) - \varphi _ { \bot } ( \tilde { w } _ { i } ^ { \top } x _ { k } ) \right] } } \cdot { \displaystyle ( 7 4 ) } } } \end{array} +$$ + +We thus have an equivalent expression of matrix $A _ { * }$ + +$$ +\begin{array} { r l } & { A _ { * } = \left[ \begin{array} { c c } { \rho I _ { h } + \tau \tilde { Q } } & { \tilde { S } _ { * } } \\ { \tilde { S } _ { * } ^ { \top } } & { 0 _ { n - 1 } } \end{array} \right] + \left[ \begin{array} { c c } { t \eta \eta ^ { \top } } & { a _ { 1 } \eta u ^ { \top } } \\ { a _ { 1 } u \eta ^ { \top } } & { 0 _ { n - 1 } } \end{array} \right] + \left[ \begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\ { E _ { 1 } ^ { \top } } & { 0 _ { n - 1 } } \end{array} \right] } \\ & { \quad = \underbrace { \left[ \begin{array} { c c } { \rho I _ { h } + \tau \tilde { Q } } & { \tilde { S } _ { * } } \\ { \tilde { S } _ { * } ^ { \top } } & { 0 _ { n - 1 } } \end{array} \right] } _ { \tilde { A } _ { * } } + \underbrace { \left[ \begin{array} { c c } { \eta } & { \mathbf { 0 } _ { h } } \\ { \mathbf { 0 } _ { n - 1 } } & { u } \end{array} \right] } _ { U } \underbrace { \left[ \begin{array} { c c } { \tau } & { a _ { 1 } } \\ { a _ { 1 } } & { 0 } \end{array} \right] } _ { C } \underbrace { \left[ \begin{array} { c c } { \eta } & { \mathbf { 0 } _ { h } } \\ { \mathbf { 0 } _ { n - 1 } } & { u } \end{array} \right] ^ { \top } } _ { U ^ { \top } } + \underbrace { \left[ \begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\ { E _ { 1 } ^ { \top } } & { \mathbf { 0 } _ { n - 1 } } \end{array} \right] } _ { E } } \\ & { \quad = \tilde { A } _ { * } + U C U ^ { \top } + E . } \end{array} +$$ + +By argument similar to (Hastie et al., 2019, B.1.2), $E$ diminishes to 0 as $n \to \infty$ with respect to the Frobenius norm, therefore by the Woodbury’s identity and the expression of $m _ { 2 , n }$ in (70) + +$$ +\begin{array} { r l } & { n _ { 2 , n } ( \xi , \rho , \tau ) = \mathbb { E } _ { a } \bigg [ \bigg ( - \xi - a ^ { \top } ( A _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } a \bigg ) ^ { - 1 } \bigg ] } \\ & { \qquad \mathbb { E } _ { a } \bigg [ \bigg ( - \xi - a ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } + U C U ^ { \top } ) ^ { - 1 } a \bigg ) ^ { - 1 } \bigg ] } \\ & { \qquad = \mathbb { E } _ { a } \bigg [ \bigg ( - \xi - \underbrace { a ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } } _ { u } + } \\ & { \qquad \underbrace { a ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } U } _ { v ^ { \top } } \underbrace { ( C ^ { - 1 } + U ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } U ) ^ { - 1 } } _ { S } \underbrace { U ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } a } _ { v } \bigg ) ^ { - 1 } \bigg ] } \end{array} +$$ + +We bound each term $u , v , S$ to compute $m _ { 2 , n }$ . For $u$ + +$$ +\begin{array} { r } { \mathbb { E } _ { a } u = \mathbb { E } _ { a } \Big [ a ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } a \Big ] = \mathrm { t r } \left( \mathbb { E } _ { s } \big [ s s ^ { \top } \big ] ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) _ { [ 1 . { h } , 1 . { h } ] } ^ { - 1 } \right) = b \gamma _ { 2 } m _ { 1 , n } ( \xi , \rho , \tau ) , } \end{array} +$$ + +where $b = \mathbb { E } _ { \boldsymbol { x } \sim \mathcal { N } ( 0 , 1 ) } [ \varphi ( \boldsymbol { x } ) ^ { 2 } ] = \mathbb { E } _ { \boldsymbol { x } \sim \mathcal { N } ( 0 , 1 ) } [ ( \phi ( \boldsymbol { x } ) - \mathbb { E } \phi ( \boldsymbol { x } ) ) ^ { 2 } ] = r _ { ! }$ . From a standard concentration of measure argument we have that as $n , h , d \infty$ + +$$ +u \mathbb { E } _ { \pmb { a } } u = r \gamma _ { 2 } m _ { 1 , n } ( \xi , \rho , \tau ) . +$$ + +And for $\textbf { { v } }$ (note that $U$ is dependent on $\textbf { \em a }$ ) + +$$ +\begin{array} { r l } { \mathbb { E } _ { a } \boldsymbol { v } ^ { \top } = \mathbb { E } _ { a } [ a ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } \boldsymbol { U } ] = \mathbb { E } _ { a } [ [ \boldsymbol { s } ^ { \top } , \boldsymbol { 0 } _ { n - 1 } ^ { \top } ] ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) ^ { - 1 } [ \begin{array} { c c } { \boldsymbol { \eta } } & { \mathbf { 0 } _ { h } } \\ { \mathbf { 0 } _ { n - 1 } } & { \boldsymbol { u } } \end{array} ] ] } & { } \\ { = [ \mathbb { E } _ { s } [ \boldsymbol { s } ^ { \top } ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \boldsymbol { \eta } ] ] } & { 0 ] = [ \underbrace { \mathrm { t r } ( ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \mathbb { E } _ { s } [ \boldsymbol { \eta } \boldsymbol { s } ^ { \top } ] ) } _ { \boldsymbol { v } } } & { 0 ] . } \end{array} +$$ + +where $\begin{array} { r } { v = \mathrm { t r } \left( ( \tilde { A } _ { * } - \xi I _ { N - 1 } ) _ { [ 1 . . h , 1 . . h ] } ^ { - 1 } \mathbb { E } _ { s } [ \eta s ^ { \top } ] \right) = a _ { 1 } \sqrt { \gamma _ { 2 } ^ { 2 } / \gamma _ { 1 } } m _ { 1 , n } ( \xi , \rho , \tau ) . } \end{array}$ We thus have + +$$ +v \mathbb { E } _ { a } v = [ a _ { 1 } \sqrt { \gamma _ { 2 } ^ { 2 } / \gamma _ { 1 } } m _ { 1 , n } ( \xi , \rho , \tau ) \quad 0 ] . +$$ + +as $n , d , p \to \infty$ . And finally for $S$ , + +$$ +\begin{array} { r l } { \iota _ { \alpha } S ^ { - 1 } = \mathbb { E } _ { \alpha } [ C ^ { - 1 } + U ^ { \top } ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) ^ { - 1 } U ] } \\ { = } & { [ \begin{array} { c c } { \tau } & { a _ { 1 } } \\ { a _ { 1 } } & { 0 } \end{array} ] ^ { - 1 } + \mathbb { E } _ { \alpha } [ [ \begin{array} { c c } { \eta } & { 0 _ { h } } \\ { 0 _ { h - 1 } } & { u } \end{array} ] ^ { \top } ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) ^ { - 1 } [ \begin{array} { c c } { \eta } & { 0 _ { h } } \\ { 0 _ { h - 1 } } & { u } \end{array} ] ] } \\ { = } & { [ \begin{array} { c c } { 0 } & { 1 / \rho _ { 2 } } \\ { 1 / \mu _ { 1 } } & { - \tau / a _ { 1 } ^ { 2 } } \end{array} ] + \mathbb { E } _ { \alpha } [ \begin{array} { c c } { \eta ^ { \top } ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) \frac { 1 } { ( 1 - A ) ! } \eta } & { u ^ { \top } ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) \frac { 1 } { | \mu _ { 1 } + 1 , h - w - 1 | ^ { u } } ] } \\ { 0 } & { u ^ { \top } ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) \frac { 1 } { | \mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \frac { 1 } { | \mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \frac { 1 } { | \mu _ { 1 } + 1 , h - w - 1 | ^ { u } } } \end{array} ] } \\ { = } & [ \begin{array} { c c } { 0 } & { 1 / \rho _ { 2 } } \\ { 1 / \mu _ { 1 } } & { - \tau / a _ { 1 } ^ { 2 } } \end{array} ] + [ \begin{array} { c c } { \mathbb { t r } ( ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) \frac { 1 } { ( 1 - A ) ! } \mathbb { E } _ { \alpha } | \eta \eta ^ { \top } ) } & \mathrm { t r } ( ( \mathring { A } _ { \star } - \xi I _ { N - 1 } ) \frac { 1 } | \end{array} \end{array} +$$ + +And hence as $n , d , h \infty$ , + +$$ +S ^ { - 1 } \to \mathbb { E } _ { a } S ^ { - 1 } = \left[ \begin{array} { c c } { \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } m _ { 1 , n } ( \xi , \rho , \tau ) } & { 1 / a _ { 1 } } \\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \xi , \rho , \tau ) - \tau / a ^ { 2 } } \end{array} \right] . +$$ + +Therefore by combining (78), (80), (82), we arrive at the following expression on $m _ { 2 , n }$ + +$$ +\begin{array} { r l r } & { } & { m _ { 2 , n } ( \xi , \rho , \tau ) \to \mathbb { E } _ { a } \Big [ \Big ( - \xi - u + v ^ { \top } S v \Big ) ^ { - 1 } \Big ] \to \Big ( - \xi - u + v ^ { \top } S v \Big ) ^ { - 1 } } \\ & { \to \Big ( - \xi - r \gamma _ { 2 } m _ { 1 , n } + \Big ( a _ { 1 } \sqrt { \gamma _ { 2 } ^ { 2 } / \gamma _ { 1 } } m _ { 1 , n } \Big ) ^ { 2 } \left[ \begin{array} { c c } { \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } m _ { 1 , n } ( \xi , \rho , \tau ) } & { 1 / a _ { 1 } } \\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \xi , \rho , \tau ) - \tau / a ^ { 2 } } \end{array} \right] _ { [ 1 , 1 ] } ^ { - 1 } \Big ) ^ { - 1 } } \\ & { } & { = \left( - \xi - r \gamma _ { 2 } m _ { 1 , n } + \frac { \gamma _ { 2 } a _ { 1 } ^ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \tau ) } { m _ { 1 , n } \left( a _ { 1 } ^ { 2 } m _ { 2 , n } - \tau \right) - \gamma _ { 1 } \gamma _ { 2 } ^ { - 1 } } \right) ^ { - 1 } . } \end{array} +$$ + +Similarly we can calculate $m _ { 1 , n } ( \xi , \rho , \tau )$ as + +$$ +\begin{array} { r l } & { m _ { 1 , n } ( \xi , \rho , \tau ) \to } \\ & { \left( - \xi - \rho - \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } \tau ^ { 2 } m _ { 1 , n } - r m _ { 2 , n } + \frac { \tau ^ { 2 } \gamma _ { 1 } ^ { - 1 } \gamma _ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 } - \tau ) - 2 \tau a _ { 1 } ^ { 2 } m _ { 1 , n } m _ { 2 , n } + a _ { 1 } ^ { 4 } m _ { 1 , n } m _ { 2 , n } ^ { 2 } } { m _ { 1 , n } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \tau ) - \gamma _ { 1 } \gamma _ { 2 } ^ { - 1 } } \right) ^ { - 1 } } \end{array} +$$ + +Uniqueness in (84)(83) follows from (Hastie et al., 2019, Sec B.1) and is omitted. + +# C.6 PROOF OF COROLLARY 5 + +In this section we take the limit $\gamma _ { 1 } \to \infty$ . In this case (8), (9) simplify to + +$$ +\begin{array} { l } { { m _ { 2 } = \left( - \xi - r \gamma _ { 2 } m _ { 1 } \right) ^ { - 1 } , } } \\ { { \nonumber } } \\ { { m _ { 1 } = \left( - \xi - \rho - \gamma _ { 2 } \tau ^ { 2 } m _ { 1 } - r m _ { 2 } \right) ^ { - 1 } . } } \end{array} +$$ + +Recall that $m ( \xi , \rho , \tau ) = \gamma _ { 2 } m _ { 1 } ( \xi , \rho , \tau ) + m _ { 2 } ( \xi , \rho , \tau )$ . By taking the derivative we have + +$$ +\begin{array} { l } { \displaystyle - q ( \xi ) = \frac { \partial } { \partial x } m ( \xi , r x , t x ) \Big | _ { x = 0 } = \left. r \frac { \partial } { \partial \rho } m ( \xi , \rho , 0 ) \right| _ { \rho = 0 } + \left. t \frac { \partial } { \partial \tau } m ( \xi , 0 , \tau ) \right| _ { \tau = 0 } } \\ { \displaystyle \quad = \left. r \gamma _ { 2 } \frac { \partial } { \partial \rho } m _ { 1 } ( \xi , \rho , 0 ) \right| _ { \rho = 0 } + \left. r \frac { \partial } { \partial \rho } m _ { 2 } ( \xi , \rho , 0 ) \right| _ { \rho = 0 } + \left. t \gamma _ { 2 } \frac { \partial } { \partial \tau } m _ { 1 } ( \xi , 0 , \tau ) \right| _ { \tau = 0 } + \left. t \frac { \partial } { \partial \tau } m _ { 2 } ( \xi , 0 , \tau ) \right| _ { \tau = 0 } . } \end{array} +$$ + +Observe that (85), (86) constitutes a set of implicit functions. Thus differentiating the two functions with respect to $\tau , \rho$ and then substitute by $\rho = \tau = 0$ gives + +$$ +q ( \xi ) = \frac { \left( r \left( \gamma _ { 2 } - 1 \right) - \xi ^ { 2 } \right) \left( \sqrt { \left( r \left( \gamma _ { 2 } - 1 \right) + \xi ^ { 2 } \right) ^ { 2 } - 4 r \gamma _ { 2 } \xi ^ { 2 } } + r \left( \gamma _ { 2 } - 1 \right) + \xi ^ { 2 } \right) } { 2 \xi ^ { 2 } \sqrt { \left( r \left( \gamma _ { 2 } - 1 \right) + \xi ^ { 2 } \right) ^ { 2 } - 4 r \gamma _ { 2 } \xi ^ { 2 } } } . +$$ + +Hence by (62) we obtain the asymptotic variance: + +$$ +V _ { ( \gamma _ { 1 } \to \infty ) } = \operatorname* { l i m } _ { \xi \to 0 } q _ { + } ( \xi ) = \operatorname* { l i m } _ { \xi \to 0 } \left( q ( \xi ) - \frac { \gamma _ { 2 } - 1 } { \xi ^ { 2 } } \right) = \frac { 1 } { \gamma _ { 2 } - 1 } . +$$ + +Combining the case where $\gamma _ { 2 } < 1$ in Theorem 4 completes the proof. + +Remark. For $\phi ( x ) = \mathrm { R e L U } ( x )$ , $c _ { 1 } = 1 / 2 - 1 / ( 2 \pi )$ , $c _ { 2 } = 1 / 4$ . For $\phi ( x ) = \mathrm { S o f t P l u s } ( x ) =$ $\log ( 1 + e ^ { x } )$ , numerical integration yields $c _ { 1 } \approx 0 . 2 7 1 5$ , $c _ { 2 } = 1 / 4$ . + +# C.7 PROOF OF COROLLARY 6 + +# C.7.1 UNBOUNDED BIAS WHEN $h = n$ + +From the bias-variance decomposition (7), the bias $B$ is written as $\begin{array} { r } { B = \frac { r ^ { 2 } } { d } \mathrm { t r } \left( Q _ { 1 } + Q _ { 2 } + I _ { d } \right) } \end{array}$ , where + +$$ +Q _ { 1 } = X [ \phi ( W ^ { \top } X ) ] ^ { \dagger } K _ { W } [ \phi ( X ^ { \top } W ) ] ^ { \dagger } X ^ { \top } ; \quad Q _ { 2 } = X [ \phi ( W ^ { \top } X ) ] ^ { \dagger } W ^ { \top } . +$$ + +When $\textit { h } = \textit { n }$ , due to the nonlinearity of $\phi$ , we have $\phi ( W ^ { \top } X )$ is full rank a.s., and hence $[ \phi ( X ^ { \top } W ) ] ^ { \dagger } = [ \phi ( X ^ { \top } W ) ] ^ { - 1 }$ . We have the following bound for $Q _ { 1 }$ + +$$ +\begin{array} { r } { \overset { ! } { \operatorname { t r } } ( Q _ { 1 } ) = \displaystyle \frac { 1 } { d } \mathrm { t r } \left( X [ \phi ( W ^ { \top } X ) ] ^ { \dagger } K _ { W } [ \phi ( X ^ { \top } W ) ] ^ { \dagger } X ^ { \top } \right) = \displaystyle \frac { 1 } { d } \mathrm { t r } \left( K _ { W } [ \phi ( X ^ { \top } W ) ] ^ { - 1 } X ^ { \top } X [ \phi ( W ^ { \top } X ) ] ^ { - 1 } \right) } \\ { \geq \displaystyle \frac { 1 } { d } \lambda _ { \operatorname* { m i n } } ( K _ { W } ) \mathrm { t r } \left( [ \phi ( X ^ { \top } W ) ] ^ { - 1 } X ^ { \top } X [ \phi ( W ^ { \top } X ) ] ^ { - 1 } \right) = \frac { \lambda _ { \operatorname* { m i n } } ( K _ { W } ) } { n } \mathrm { t r } \left( ( S S ^ { \top } ) ^ { - 1 } \cdot \frac { 1 } { d } X ^ { \top } X [ \phi ( W ^ { \top } X ) ] ^ { - 1 } \right) } \end{array} +$$ + +Since $W$ and $X / { \sqrt { d } }$ are $\mathbb { R } ^ { d \times n } ~ = ~ \mathbb { R } ^ { d \times p }$ follows the same distribution where each entry i.i.d. $\mathcal { N } ( 0 , 1 / d )$ , + +$$ +\begin{array} { r l } & { \displaystyle \frac { 1 } { d \lambda _ { \operatorname* { m i n } } ( K _ { W } ) } \mathrm { t r } ( Q _ { 1 } ) \geq \frac { 1 } { n } \mathrm { t r } ( ( S S ^ { \top } ) ^ { - 1 } \cdot \frac { 1 } { d } X ^ { \top } X ) \sim \frac { 1 } { n } \mathrm { t r } ( ( S ^ { \top } S ) ^ { - 1 } \cdot W ^ { \top } W ) } \\ & { \quad \quad \quad \quad = \displaystyle \frac { 1 } { n } \mathrm { t r } ( ( S ^ { \top } S ) ^ { - 1 } \cdot ( I + Q ) ) = \operatorname* { l i m } _ { \xi 0 } - \frac { \partial } { \partial x } \tilde { m } _ { n } ( \xi , x , 0 , x ) \infty , } \end{array} +$$ + +where in Section C.5 we have showed (91) is unbounded when $n \infty$ . Moreover, by (154) and Weyl’s theorem we have $\lambda _ { \operatorname* { m i n } } ( K _ { W } ) = O ( 1 )$ , and thus $d ^ { - 1 } \mathrm { t r } \left( Q _ { 1 } \right)$ is unbounded. For $d ^ { - 1 } \mathrm { t r } \left( Q _ { 2 } \right)$ , + +$$ +\frac { 1 } { d } \mathrm { t r } ( Q _ { 2 } ) = \frac { 1 } { d } \mathrm { t r } \left( W ^ { \top } X [ \phi ( W ^ { \top } X ) ] ^ { - 1 } \right) \leq \frac { 1 } { d } \lambda _ { \operatorname* { m a x } } ( \phi ( W ^ { \top } X ) ^ { - 1 } ) \mathrm { t r } \left( W ^ { \top } X \right) = O ( 1 ) . +$$ + +To sum up, for $n \to \infty$ and $\gamma _ { 2 } 1$ we have $B \infty$ . + +# C.7.2 BOUNDED BIAS WHEN $\gamma _ { 2 } > 1$ + +Since $h > n$ , the two terms in the expression of the bias can be written as + +$$ +\begin{array} { r l } & { Q _ { 1 } = X \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } \phi ( X ^ { \top } W ) K _ { W } \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } X ^ { \top } , } \\ & { Q _ { 2 } = X \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } \phi ( X ^ { \top } W ) W ^ { \top } . } \end{array} +$$ + +Therefore we have + +$$ +\operatorname { I } ^ { \mathrm { { T } } } ( Q _ { 1 } ) = 2 \mathrm { { t r } } \left( { \frac { X ^ { \top } X } { d } } { \Big ( } \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) { \Big ) } ^ { - 1 } \phi ( X ^ { \top } W ) K _ { W } \phi ( W ^ { \top } X ) { \Big ( } \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) { \Big ) } ^ { - 1 } \right) +$$ + +$$ +\begin{array} { r l r } { { \le 2 \lambda _ { \operatorname* { m a x } } ( \frac { X ^ { \top } X } { d } ) \operatorname { t r } ( \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } \phi ( X ^ { \top } W ) K _ { W } \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } \phi ( X ^ { \top } W ) K _ { W } \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 1 } \phi ( X ^ { \top } W ) K _ { W } \phi ( W ^ { \top } X ) \Big ) } } \\ & { = O ( 1 ) \cdot \operatorname { t r } ( \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 2 } \phi ( X ^ { \top } W ) K _ { W } ) } \\ & { \le O ( 1 ) \cdot \lambda _ { \operatorname* { m a x } } ( \phi ( W ^ { \top } X ) \Big ( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \Big ) ^ { - 2 } \phi ( X ^ { \top } W ) ) \cdot \operatorname { t r } ( K _ { W } ) } \\ & { = O ( 1 ) \cdot \sigma _ { \operatorname* { m i n } } ^ { - 2 } ( \phi ( X ^ { \top } W ) ) \operatorname { t r } ( K _ { W } ) = O ( 1 ) \cdot O ( n ^ { - 1 } ) \cdot O ( n ) = O ( 1 ) , } & { \quad \mathrm { ~ \displaystyle ( 9 5 ) ~ } } \end{array} +$$ + +in which we used Lemma 11 and $\operatorname { t r } \left( A B \right) \leq \lambda _ { \operatorname* { m a x } } ( A ) \operatorname { t r } \left( B \right)$ for positive semi-definite $A , B$ . Similarly + +$$ +\begin{array} { r l } & { \frac { 2 } { d } \mathrm { t r } \left( Q _ { 2 } \right) = 2 \mathrm { t r } \left( \left( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \right) ^ { - 1 } \phi ( X ^ { \top } W ) \cdot \frac { 1 } { d } W ^ { \top } X \right) } \\ & { \qquad \leq \mathrm { t r } \left( \left( \left( \phi ( X ^ { \top } W ) \phi ( W ^ { \top } X ) \right) ^ { - 1 } \phi ( X ^ { \top } W ) \right) ( \ldots ) ^ { \top } \right) + \mathrm { t r } \left( d ^ { - 2 } X ^ { \top } W W ^ { \top } X \right) } \\ & { \qquad \leq n \cdot \sigma _ { \operatorname* { m i n } } \left( \phi ( X ^ { \top } W ) \right) ^ { - 2 } + \frac { 1 } { d } \lambda _ { \operatorname* { m a x } } \left( \frac { 1 } { d } X X ^ { \top } \right) \mathrm { t r } \left( W W ^ { \top } \right) = O ( 1 ) , } \end{array} +$$ + +where we applied $\mathrm { t r } \left( A B \right) \leq ( \mathrm { t r } \left( A ^ { \top } A \right) + \mathrm { t r } \left( B ^ { \top } B \right) ) / 2$ . We therefore conclude that $B$ is bounded when $h > n$ and $h , n \infty$ .  + +Remark. Concurrent to this work, Mei and Montanari (2019) provides a complete characterization of the bias term and confirms our observations above. + +# C.8 PROOF OF THEOREM 7 + +For simplicity we assume $n , d , h$ to be even and let $d _ { 0 } = d / 2$ , $n _ { 0 } = n / 2$ and $h _ { 0 } = h / 2$ . Since the second layer is fixed $a _ { i } \sim \mathrm { U n i f } \{ - 1 / \sqrt { h } , 1 / \sqrt { h } \}$ , we let $a _ { i } = 1 / \sqrt { h }$ and $a _ { i + h _ { 0 } } = - 1 / \sqrt { h }$ for all $1 \leq i \leq h _ { 0 }$ . We therefore write $\pmb { a } ^ { \top } = h ^ { - 1 / 2 } [ \mathbf { 1 } _ { h _ { 0 } } , - \mathbf { 1 } _ { h _ { 0 } } ] ^ { \top }$ and $W = [ W _ { + } , W _ { - } ]$ : + +$$ +f ( \pmb { x } ; W _ { - } , W _ { + } ) = \pmb { a } ^ { \top } \phi ( W ^ { \top } \pmb { x } ) = \frac { 1 } { \sqrt { h } } \pmb { 1 } ^ { \top } \phi ( W _ { + } ^ { \top } \pmb { x } ) - \frac { 1 } { \sqrt { h } } \pmb { 1 } ^ { \top } \phi ( W _ { - } ^ { \top } \pmb { x } ) . +$$ + +The empirical risk can thus be written as + +$$ +L ( X ; W _ { + } , W _ { - } ) = \frac { 1 } { n } \sum _ { \substack { x \in X } } L ( x ; W _ { + } , W _ { - } ) = \frac { 1 } { n } \sum _ { \substack { x \in X } } \left[ y - \frac { 1 } { \sqrt { h } } \mathbf { 1 } ^ { \top } \phi ( W _ { + } ^ { \top } x ) + \frac { 1 } { \sqrt { h } } \mathbf { 1 } ^ { \top } \phi ( W _ { - } ^ { \top } x ) \right] ^ { 2 } . +$$ + +# C.8.1 DEFINING GRADIENT FLOWS + +In this section we define three gradient flows and show that the three flows are similar in some sense. +$\pmb { G F }$ -Original is the original gradient flow (11), i.e. + +$$ +\begin{array} { l } { \displaystyle \frac { \partial W _ { + } ^ { O } } { \partial t } = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \Big [ \frac { 1 } { \sqrt { h } } \Big ( y _ { i } - \frac { 1 } { \sqrt { h } } \mathbf { 1 } ^ { \top } \boldsymbol { \phi } ( W _ { + } ^ { O \top } \mathbf { x } _ { i } ) + \frac { 1 } { \sqrt { h } } \mathbf { 1 } ^ { \top } \boldsymbol { \phi } ( W _ { - } ^ { O \top } \mathbf { x } _ { i } ) \Big ) \mathbf { x } _ { i } \boldsymbol { \phi } ^ { \prime } ( { \mathbf { x } } _ { i } ^ { \top } W _ { + } ^ { O } ) \Big ] } \\ { \displaystyle \qquad = \frac { 1 } { 2 n _ { 0 } } \left[ X ( \boldsymbol { y } - \boldsymbol { y } ^ { O } ( t ) ) \frac { 1 } { \sqrt { h } } \mathbf { 1 } ^ { \top } \circ \boldsymbol { \phi } ^ { \prime } ( X W _ { + } ^ { O } ) \right] , } \end{array} +$$ + +starting from the vanishing initialization ${ \pmb w } _ { i } ^ { O } ( 0 ) \sim \mathcal { N } ( { \bf 0 } , I / d h ^ { 1 + \epsilon } )$ . Note that the gradient for the negative part $W _ { - } ^ { O }$ can be similarly defined. + +We now define the flow under the same objective but from exact zero initialization ${ \pmb w } _ { i } ^ { D } ( 0 ) = { \bf 0 }$ termed $\pmb { G F }$ -Double. Due to zero initialization, a basic observation is that the solution $[ W _ { + } ^ { D } , W _ { - } ^ { D } ]$ is + +at most rank-2, and more precisely, the parameters in the flow takes the form of $W _ { \pm } ^ { D } ( t ) = { \pmb w } _ { \pm } ^ { D } ( t ) { \bf 1 } ^ { \top }$ where ${ \pmb w } _ { \pm } ^ { D } ( t )$ admits the following dynamics: + +$$ +\frac { \partial w _ { + } ^ { D } } { \partial t } = g _ { + } ^ { D } ( \boldsymbol { w } _ { + } ^ { D } ) = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \left[ \frac { 1 } { \sqrt { h } } \Big ( y _ { i } - \sqrt { h } \phi ( \boldsymbol { w } _ { + } ^ { D \top } \boldsymbol { x } _ { i } ) + \sqrt { h } \phi ( \boldsymbol { w } _ { - } ^ { D \top } \boldsymbol { x } _ { i } ) \Big ) \phi ^ { \prime } ( \boldsymbol { w } _ { + } ^ { D \top } \boldsymbol { x } _ { i } ) \boldsymbol { x } _ { i } \right] . +$$ + +Lastly, we define the $\pmb { G F }$ -Single with solution denoted as ${ \pmb w } _ { \pm } = { \pmb w } _ { \pm } ^ { S } ( t )$ : + +$$ +\frac { \partial w _ { + } ^ { S } } { \partial t } = g _ { + } ^ { S } ( w _ { + } ^ { S } ) = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \Big [ \frac { 1 } { \sqrt { h } } \Big ( y _ { i } - \sqrt { h } \phi ^ { \prime } ( 0 ) w _ { + } ^ { S \top } x _ { i } + \sqrt { h } \phi ^ { \prime } ( 0 ) w _ { - } ^ { 1 \top } x _ { i } \Big ) \phi ^ { \prime } ( 0 ) x _ { i } \Big ] . +$$ + +from zero initialization ${ \pmb w } _ { \pm } ^ { D } ( 0 ) = { \bf 0 }$ . This can be seen as replacing the nonlinearity $\phi$ with its first-order Taylor expansion at the origin. + +# C.8.2 FROM GF-DOUBLE TO GF-SINGLE + +Step 1. Solution of GF-single. Among the three flows defined above, only GF-single an explicit form at any time $t$ . Specifically, the solution can be written as: + +$$ +w _ { + } ^ { S } ( t ) = - w _ { - } ^ { S } ( t ) = \frac { 1 } { 2 \sqrt { h } } \left( I - e ^ { - \frac { \phi ^ { \prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \top } t } \right) ( X X ^ { \top } ) ^ { - 1 } X \pmb { y } , +$$ + +when $d < n$ , or otherwise + +$$ +\pmb { w } _ { + } ^ { S } ( t ) = - \pmb { w } _ { - } ^ { S } ( t ) = \frac { 1 } { 2 \sqrt { h } } X \left( I - e ^ { - \frac { \phi ^ { \prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X ^ { \top } X t } \right) ( X ^ { \top } X ) ^ { - 1 } \pmb { y } . +$$ + +when $d > n$ . For simplicity we elaborate the proof only for $d < n$ . We provide a condition under which the difference between the trajectories can be controlled, and we first that this condition holds for the linearized flow GF-single for all $t$ in Lemma 17. + +$$ +\mathrm { C o n d i t i o n \ A : } \ \| w ( t ) \| _ { 2 } = O \left( { \frac { 1 } { \sqrt { d } } } \right) ; \left\| X ^ { \top } w ( t ) \right\| _ { \infty } = O \left( { \frac { \mathrm { p o l y } \log d } { \sqrt { d } } } \right) . +$$ + +Step 2. Bounding the Difference in Gradient Flow Trajectory. Due to low rank property of GF-single and GF-double, in this subsection we slightly abuse the notation and define + +$$ +f ( \pmb { x } ; \pmb { w } _ { \pm } ) = f ( \pmb { x } ; \pmb { w } _ { + } \pmb { 1 } ^ { \top } , \pmb { w } _ { - } \pmb { 1 } ^ { \top } ) = \sqrt { h } \phi ( \pmb { w } _ { + } ^ { \top } \pmb { x } ) - \sqrt { h } \phi ( \pmb { w } _ { - } ^ { \top } \pmb { x } ) . +$$ + +We now show that the difference between the two trajectories defined above is asymptotically vanishing for ${ \pmb w } _ { + }$ $\mathbf { \nabla } w _ { - }$ follows the same argument). Compare the two trajectories up to time $T$ + +$$ +\begin{array} { r l r } { { \| { \boldsymbol w } _ { + } ^ { D } ( T ) - { \boldsymbol w } _ { + } ^ { S } ( T ) } \| _ { 2 } = \| \int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \boldsymbol w } _ { + } ^ { S } ( t ) ) d t \| _ { 2 } } \\ & { \le \| \int _ { 0 } ^ { T } g _ { + } ^ { S } ( { \boldsymbol w } _ { + } ^ { S } ( t ) ) - g _ { + } ^ { S } ( { \boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \mathrm { d } t \| _ { 2 } + \| \int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \mathrm { d } t \| _ { 2 } } \\ & { = \| \int _ { 0 } ^ { T } \frac { 1 } { 2 n _ { 0 } } \frac { 2 n _ { 0 } } { i = 1 } [ ( - \phi ^ { \prime } ( 0 ) ( { \boldsymbol w } _ { + } ^ { D } ( t ) - { \boldsymbol w } _ { + } ^ { S } ( t ) ) ^ { \top } { \boldsymbol x } _ { i } + \phi ^ { \prime } ( 0 ) ( { \boldsymbol w } _ { - } ^ { D } ( t ) - { \boldsymbol w } _ { - } ^ { S } ( t ) ) ^ { \top } { \boldsymbol x } _ { i } ) \phi ^ { \prime } ( 0 ) { \boldsymbol x } _ { i } ] \mathrm { d } t \| _ { 2 } } & \\ & { = O ( 1 ) \int _ { 0 } ^ { T } ( \| { \boldsymbol w } _ { + } ^ { D } ( t ) - { \boldsymbol w } _ { + } ^ { S } ( t ) \| _ { 2 } + \| { \boldsymbol w } _ { - } ^ { D } ( t ) - { \boldsymbol w } _ { - } ^ { S } ( t ) \| _ { 2 } ) \mathrm { d } t + { \boldsymbol E } _ { + } , } & { ( 1 0 6 ) } \end{array} +$$ + +where we have defined the error term as + +$$ +E _ { + } = \left\| \int _ { 0 } ^ { T } \pmb { g } _ { + } ^ { D } ( \pmb { w } _ { + } ^ { D } ( s ) ) - \pmb { g } _ { + } ^ { S } ( \pmb { w } _ { + } ^ { D } ( s ) ) ~ \mathrm { d } t \right\| _ { 2 } . +$$ + +To bound the error term, we note that at $t = 0$ Condition A holds for GF-double. Assume that for some $0 \leq t \leq T$ , Condition A also holds for GF-double, we have + +$$ +\begin{array} { r l } & { E _ { + } = \Bigg \| \int _ { 0 } ^ { T } \frac { \Phi } { u } \int _ { 0 } ^ { u } [ u , \frac { \Phi } { u } ] ( \boldsymbol { \cdot } } ) - \boldsymbol { \cdot } u _ { + } ^ { \Phi } ( \boldsymbol { \cdot } , \boldsymbol { \cdot } ) u _ { + } ^ { \Phi } ( \boldsymbol { \cdot } , \boldsymbol { \cdot } ) \boldsymbol { \cdot } \boldsymbol { } u _ { + } ^ { \Phi } ( \boldsymbol { \cdot } , \boldsymbol { \cdot } ) \boldsymbol { \cdot } \boldsymbol { \cdot } \\ & { \leq \Bigg \| \int _ { 0 } ^ { T } \frac { 1 } { u } \sum _ { \mathrm { i } = 1 } ^ { N } \Bigg [ \frac { 1 } { \sqrt { \delta } } ( \boldsymbol { \cdot } - \sqrt { \delta } \dot { u } ( u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } ) - \boldsymbol { \cdot } u _ { + } ^ { \Phi } ( \boldsymbol { \cdot } , \boldsymbol { \cdot } \boldsymbol { \cdot } ) u _ { - } ^ { \Phi } ( \boldsymbol { \cdot } , \boldsymbol { \cdot } ) ) ( \mathcal { \cdot } \langle u ( u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } u _ { - } ^ { \Phi } ) - \mathcal { \cdot } u ( \boldsymbol { \cdot } ^ { \Phi } , \boldsymbol { \cdot } ) u _ { - } ^ { \Phi } \rangle ) \boldsymbol { \cdot } u _ { - } ^ { \Phi } \Bigg ] \Bigg \| _ { 0 } ^ { 2 } , } \\ & \quad + \Bigg \| \int _ { 0 } ^ { T } \frac { 1 } { u _ { + } ^ { T } } \sum _ { \mathrm { i } = 1 } ^ { N } \Bigg [ ( - \mathcal { \cdot } \langle u ( u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } u _ { + } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } ) - \mathcal { \cdot } ( u _ { + } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } u _ { - } ^ { \Phi } ) ) \boldsymbol { \cdot } u _ { - } ^ { \Phi } \Bigg ] | \int _ { 0 } ^ { T } \ \end{array} +$$ + +where (i) follows from the error of Taylor expansion on $\phi$ and (ii) from Condition A. Therefore, by Equation (106) and Gronwall’s inequality, + +$$ +\left\| w _ { + } ^ { D } ( T ) - w _ { + } ^ { S } ( T ) \right\| _ { 2 } \leq C _ { 1 } \cdot \frac { \log ^ { c } h } { h } e ^ { C _ { 2 } T } = O \left( \frac { \mathrm { p o l y l o g } h } { h } \right) \to 0 . +$$ + +for $T \in O ( \log \log h )$ . This shows that up to time $\mathrm { T } ,$ , the difference between the trajectories of GF-single and GF-double vanishes under the the assumption that Condition A holds for $\mathbf { \Delta } w ^ { D }$ up to T. Importantly, note that ${ \pmb w } ^ { S } ( 0 ) = { \pmb w } ^ { D } ( 0 ) = 0$ , and that Condition A holds for $\pmb { w } ^ { S }$ for all $T > 0$ . Therefore, by a standard contradiction argument (e.g. (Du et al., 2018, Lemma 3.4)), one can show that this closeness between the two trajectories implies that Condition A also holds for $\mathbf { \Delta } w ^ { D }$ up to time T. With Equation (108) we bound the difference in population risk between the two flows. + +Step 3. Bounding the Difference in Risk. Now we consider the difference of the population risk $R ^ { S }$ and $R ^ { D }$ of two models with parameters $\pmb { w } _ { \pm } ^ { S }$ and $\pmb { w } _ { \pm } ^ { D }$ . + +$$ +\begin{array} { r l } & { \quad | ( { \mathcal R } ^ { 3 } - { \mathcal R } ^ { D } ) | = | \mathbb { E } _ { { \mathbf z } } ( x ^ { \top } ) - f ( x ; { \mathcal R } _ { \mathbf z } ^ { \theta } ) | ^ { 2 } - \mathbb { E } _ { { \mathbf z } _ { \mathbf z } } ( x ^ { \top } \beta - f ( x ; { \mathcal R } _ { \mathbf z } ^ { D } ) ) ^ { 2 } | } \\ & { \stackrel { ( i ) } { \le } \sqrt { \mathbb { H } _ { { \mathbf z } } ^ { 5 } [ f ( { \mathbf x } ; { \mathbf x } _ { \mathbf z } ^ { \top } ) - f ( x ; { \mathbf x } _ { \mathbf z } ^ { D } ) ] ^ { 2 } \mathbb { E } _ { { \mathbf z } } [ | { \mathcal R } ^ { 7 } \cdot f ( { \mathbf x } ) - f ( x ; { \mathbf x } _ { \mathbf z } ^ { D } ) + \beta ^ { \top } x - f ( x ; { \mathbf x } _ { \mathbf z } ^ { D } ) ] ^ { 2 } } } \\ & \le \sqrt { \mathbb { H } _ { { \mathbf z } } ^ { 5 } [ | { \mathcal R } | _ { { \mathbf z } } [ \langle \delta | ^ { \mathcal { R } } \rangle - \phi ; \langle { \mathcal R } _ { \mathbf z } ^ { D } \rangle ] ^ { 2 } \mathbb { E } _ { | { \mathbf z } } ] + | ( { \mathcal R } \langle \mathbf x ^ { \top } \mathbf x \rangle - \phi ( { \mathcal R } _ { \mathbf z } ^ { D } \rangle ) [ ^ { 2 } } \\ & { \quad \cdot \sqrt { \mathbb { H } _ { { \mathbf z } } [ | { \mathcal R } | _ { { \mathbf z } } ^ { 7 } - f ( x ; { \mathcal R } _ { \mathbf z } ^ { \theta } ) ] + | \beta ^ { \top } x - f ( x ; { \mathbf x } _ { \mathbf z } ^ { D } ) | ] ^ { 2 } } } \\ & \stackrel { ( i i ) } { \le } 2 \sqrt { \mathbb { H } _ { { \mathbf z } } [ | { \mathcal R } | _ { { \mathbf z } } \mathbb { R } ^ { 5 } ] - \phi ( { \mathcal R } _ { \mathbf z } ^ { D } \mathbb { I } _ { { \mathbf z } } ] ^ { 2 } + | \beta \langle \mathbf w _ { \mathbf z } ^ { \top } \mathbf x \rangle - \phi ( { \mathcal R } _ { \mathbf z } ^ { D } \mathbb { I } _ { { \mathbf z } } \rangle | ^ { 2 } } \\ & \quad \cdot \sqrt \mathbb { E } _ { \mathbf z } [ | { \mathcal R } \end{array} +$$ + +where (i) is due to Cauchy-Schwarz inequality on norm, (ii) from Jenson’s inequality on squares and Young’s inequality, and (iii) from the Lipschitz assumption on the activation, and the observation + +that both $R ^ { S }$ and $R ^ { D }$ are finite for $\gamma _ { 1 } \neq 1$ due to the justified Condition A above. Therefore the difference between $R ^ { S }$ and $R ^ { D }$ vanishes for $T = O ( \log \log h )$ . + +Step 4. Bounding the Difference from Stationarity We compute the difference in risk between the model at some finite time $t$ and the stationary point i.e. $t = \infty$ , + +$$ +\begin{array} { r l } & { \left| R ^ { S } ( t ) - R ^ { S } ( \infty ) \right| \leq C \sqrt { h } \cdot \left\| w _ { + } ^ { S } ( t ) - w _ { + } ^ { S } ( \infty ) \right\| = C \sqrt { h } \left\| \frac { 1 } { 2 \sqrt { h } } e ^ { - \frac { \phi ^ { \prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \top } t } ( X X ^ { \top } ) ^ { - 1 } X y \right\| _ { 2 } } \\ & { = C \left\| e ^ { - \frac { \phi ^ { \prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \top } t } ( X X ^ { \top } ) ^ { - 1 } X X ^ { \top } \beta \right\| _ { 2 } \leq C \exp \left( - \phi ^ { \prime } ( 0 ) ^ { 2 } \left\| \frac { 1 } { n } X X ^ { \top } \right\| _ { 2 } t \right) \| \beta \| _ { 2 } = C _ { 3 } e ^ { - C _ { 4 } t } . } \end{array} +$$ + +for constants $C _ { 3 } , C _ { 4 } > 0$ . Combining (110) and (111) yields for $T = \log \log h$ , + +$$ +\begin{array} { r } { | R ^ { D } ( T ) - R ^ { S } ( \infty ) | \leq | R ^ { S } ( T ) - R ^ { D } ( T ) | + | R ^ { S } ( T ) - R ^ { S } ( \infty ) | } \\ { = O ( \frac { \mathrm { p o l y l o g } h } { \sqrt { h } } ) + O ( \frac { 1 } { \mathrm { p o l y l o g } h } ) 0 . } \end{array} +$$ + +The result in (112) for case $d > n$ follows a similar proof and is omitted. + +# C.8.3 FROM GF-ORIGINAL TO GF-DOUBLE + +In this section we compare GF-original with GF-double. Note that the two flows differ only at initialization: vanishing initialization ${ \pmb w } _ { i } ^ { O } ( 0 ) \sim \mathcal { N } ( { \bf 0 } , I / d h ^ { 1 + \epsilon } )$ v.s. zero initialization ${ \pmb w } _ { i } ^ { D } ( 0 ) \dot { = } { \bf 0 }$ . Note that at initialization $\| { \pmb y } - { \pmb f } ( { \pmb X } ) \| _ { 2 } = { \cal O } ( { \sqrt { n } } )$ , and gradient flow decreases the empirical risk; therefore the condition for Lemma 18 is satisfied, and by the Lipschitz condition on the empirical gradient we have + +$$ +\begin{array} { l } { \displaystyle \left\| W ^ { O } ( T ) - W ^ { D } ( T ) \right\| _ { F } ^ { 2 } } \\ { \displaystyle = \left\| W ^ { O } ( 0 ) - W ^ { D } ( 0 ) \right\| _ { F } ^ { 2 } + \int _ { 0 } ^ { T } \frac { \partial \left\| W ^ { O } ( t ) - W ^ { D } ( t ) \right\| _ { F } ^ { 2 } } { \partial t } \mathrm { d } t } \\ { \displaystyle = \left\| W ^ { O } ( 0 ) \right\| _ { F } ^ { 2 } + \int _ { 0 } ^ { T } \left\| W ^ { O } ( t ) - W ^ { D } ( t ) ) \right\| _ { F } \left\| \frac { \partial L \left( X ; W ^ { O } ( t ) \right) } { \partial W } - \frac { \partial L \left( X ; W ^ { D } ( t ) \right) } { \partial W } \right\| _ { F } \mathrm { d } t } \\ { \displaystyle \leq \left\| W ^ { O } ( 0 ) \right\| _ { F } ^ { 2 } + \int _ { 0 } ^ { T } \left\| W ^ { O } ( t ) - W ^ { D } ( t ) \right\| _ { F } ^ { 2 } + \left\| \frac { \partial L \left( X ; W ^ { O } ( t ) \right) } { \partial W } - \frac { \partial L \left( X ; W ^ { D } ( t ) \right) } { \partial W } \right\| _ { F } ^ { 2 } \mathrm { d } t } \\ { \displaystyle \leq \left\| W ^ { O } ( 0 ) \right\| _ { F } ^ { 2 } + \left( 1 + L _ { f } \right) \int _ { 0 } ^ { T } \left\| W ^ { O } ( t ) - W ^ { D } ( t ) \right\| _ { F } ^ { 2 } \mathrm { d } t , } \end{array} +$$ + +And hence by Gronwall’s lemma one obtains: + +$$ +\left\| { W ^ { O } ( T ) - W ^ { D } ( T ) } \right\| _ { F } \leq \left\| { W ^ { O } ( 0 ) } \right\| _ { F } e ^ { ( 1 + L _ { f } ) T / 2 } = O ( d ^ { - ( 1 + \epsilon ) / 2 } ) e ^ { C T } . +$$ + +Therefore the difference in the function output can be bounded as + +$$ +\begin{array} { r l } & { \quad \left\| f ( \pmb { x } , W ^ { D } ( T ) ) - f ( \pmb { x } , W ^ { O } ( T ) ) \right\| _ { 2 } = \left\| \phi ( \pmb { x } ^ { \top } W ^ { D } ( T ) ) \pmb { a } - \phi ( \pmb { x } ^ { \top } W ^ { O } ( T ) ) \pmb { a } \right\| _ { 2 } } \\ & { \leq \left\| \phi ( \pmb { x } ^ { \top } W ^ { D } ( T ) ) - \phi ( \pmb { x } ^ { \top } W ^ { O } ( T ) ) \right\| _ { 2 } \left\| \pmb { a } \right\| _ { 2 } \leq L _ { \phi } \left\| \pmb { x } \right\| _ { 2 } \left\| W ^ { D } ( T ) - W ^ { O } ( T ) \right\| _ { F } \left\| \pmb { a } \right\| _ { 2 } } \\ & { = O ( \sqrt { d } ) \cdot \left\| W ^ { D } ( T ) - W ^ { O } ( T ) \right\| _ { F } = O ( d ^ { - \epsilon / 2 } ) e ^ { C T } . } \end{array} +$$ + +Taking $T = \log \log h$ , together with the same argument in Step 1 yields + +$$ +| R ^ { O } ( T ) - R ^ { D } ( T ) | = O \left( \frac { \mathrm { p o l y l o g } h } { d ^ { \epsilon / 2 } } \right) \to 0 . +$$ + +# C.8.4 PUTTING THINGS TOGETHER + +By (112) and (116), we know that for $T = \log \log h$ , + +$$ +| R ^ { O } ( T ) - R ^ { S } ( \infty ) | \leq | R ^ { O } ( T ) - R ^ { D } ( T ) | + | R ^ { D } ( T ) - R ^ { S } ( \infty ) | \to 0 . +$$ + +Finally the proof is completed by observing that $R ^ { S } ( \infty )$ is the risk of the minimum-norm solution on the input discussed in Section 3. In addition, note that at $T = \log \log h$ , from (109)(114) one obtains that $\left\| \mathbf { \dot { W } } ^ { O } ( t ) - W ^ { S } ( t ) \right\| _ { F } \to 0$ , and therefore by Lemma 18 we have $\left\| \partial L ( X ; W ^ { \mathcal { O } } ( t ) ) / \partial W \right\| _ { F } \to 0$ , i.e. the flow on the original objective reaches a $o ( 1 )$ first-order stationary point.  + +# C.9 PROOF OF THEOREM 8 + +Denote $\omega = \mathrm { v e c } ( W ) = \mathrm { v e c } ( [ W _ { + } , W _ { - } ] )$ , and $\omega _ { 0 } = \mathrm { v e c } ( W ^ { \mathrm { i n i t } } )$ . Define + +$$ +K ( t ) = \frac { \partial \pmb { f } ( X ; \omega ( t ) ) } { \partial \omega ( t ) } ^ { \top } \frac { \partial \pmb { f } ( X ; \omega ( t ) ) } { \partial \omega ( t ) } , +$$ + +which is the kernel matrix of the neural tangent kernel Jacot et al. (2018); Du et al. (2018). In the following sections we show that under the non-vanishing initialization, the trained two-layer network is well-approximated by the regression model on the NTK, for which we derive the population risk. + +# C.9.1 THE KERNEL LINEARIZATION + +Write $\pmb { y } _ { N N } ( t ) = f _ { N N } ( \boldsymbol { X } , t ) \in \mathbb { R } ^ { n }$ and its evolution: + +$$ +\mathrm { d } { \pmb y } _ { N N } ( t ) = \frac { 1 } { n } K ( t ) ( { \pmb y } - { \pmb y } _ { N N } ( t ) ) \ \mathrm { d } t , +$$ + +and the corresponding linearized flow: + +$$ +\mathrm { d } { \pmb y } _ { N T K } ( t ) = \frac { 1 } { n } K ( 0 ) ( { \pmb y } - { \pmb y } _ { N T K } ( t ) ) \ \mathrm { d } t , +$$ + +Previous works (e.g. Du et al. (2018); Oymak and Soltanolkotabi (2019)) have proved (nonasymptotically) that the two trajectories (119) and (120) are close if the model is overparameterized, i.e. $h = \mathrm { p o l y } ( n )$ , under no assumptions on the teacher model. In our asymptotic setup (together with assumptions (A1-3)), we argue that similar conclusion holds without significant overparameterization. + +We first show the global convergence of the training of two-layer neural network $f _ { N N }$ . We employ an argument similar to (Du et al., 2018, Theo. 3.2) by first identifying the condition under which training converges at linear rate: + +From Corollary 15 we know that at initialization the lowest eigenvalue of the NTK matrix satisfies $\lambda _ { \operatorname* { m i n } } ( K ( 0 ) ) \stackrel { } { = } O ( d )$ , and thus the kernel regression on the NTK enjoys linear convergence. By Lemma 19, we know that for $\lVert W ( t ) - W ( 0 ) \rVert _ { 2 } = O ( d ^ { 1 - \epsilon } )$ , the order of $\lambda _ { \operatorname* { m i n } } ( K ( t ) ) \stackrel { } { = } O ( d )$ remains unchanged. Therefore, if we assume that up to time $T$ the weights satisfy $\| { \pmb w } _ { i } ( t ) - { \pmb w } _ { i } ( 0 ) \| _ { 2 } =$ $O ( d ^ { - 1 / 2 } )$ , then Condition B is satisfied for ${ \pmb y } _ { N N }$ and from (Chizat and Bach, 2018b, Lemma B1) we have the following linear convergence + +$$ +\begin{array} { r } { \| { \pmb y } _ { N N } ( t ) - { \pmb y } \| _ { 2 } \le C _ { 1 } \| { \pmb y } _ { N N } ( 0 ) - { \pmb y } \| _ { 2 } e ^ { - C _ { 2 } t } . } \end{array} +$$ + +Since $\| { \pmb y } _ { N N } ( 0 ) - { \pmb y } \| _ { 2 } = O ( \sqrt { n } )$ at initialization, setting $T = O ( \log d )$ ensures that the training loss $\begin{array} { r } { \frac { 1 } { n } \| { \pmb y } _ { N N } ( T ) - { \pmb y } \| _ { 2 } ^ { 2 } 0 } \end{array}$ as $n \to \infty$ . Consequently it is easy to check that $\left\| \frac { \partial { \cal L } ( X ; W ( T ) ) } { \partial W ( T ) } \right\| _ { 2 } \to 0$ and thus at time T the gradient flow reaches an $\mathsf { o } ( 1 )$ first order stationary point. + +We now verify that each weight vector ${ \pmb w } _ { i }$ travels at most $O ( d ^ { - 1 / 2 } )$ from initialization. The norm of gradient for ${ \pmb w } _ { i }$ can be bounded as: + +$$ +\begin{array} { r l } & { \left\| \frac { \partial L ( X ; \boldsymbol { w } _ { i } ( t ) ) } { \partial \boldsymbol { w } _ { i } ( t ) } \right\| _ { 2 } = \left\| \frac { 1 } { n } X \left[ \frac { 1 } { \sqrt { h } } ( \boldsymbol { y } - \boldsymbol { y } _ { N N } ( t ) ) \circ \phi ^ { \prime } ( X ^ { \top } \boldsymbol { w } _ { i } ( t ) ) \right] \right\| _ { 2 } } \\ & { \qquad \overset { ( i ) } { \leq } O ( 1 ) \frac { 1 } { d ^ { 1 . 5 } } \left\| X \right\| _ { 2 } \left\| \boldsymbol { y } - \boldsymbol { y } _ { N N } ( t ) \right\| _ { 2 } \leq O ( 1 ) d ^ { - 1 } \left\| \boldsymbol { y } - \boldsymbol { y } _ { N N } ( 0 ) \right\| _ { 2 } e ^ { - t } , } \end{array} +$$ + +where (i) follows from the boundedness of $\phi ^ { \prime }$ . Integrating the gradient yields $\| { \pmb w } _ { i } ( t ) - { \pmb w } _ { i } ( 0 ) \| _ { 2 } =$ $O ( d ^ { - 1 / 2 } )$ , i.e. $\| W ( t ) - W ( 0 ) \| _ { 2 } = O ( 1 )$ . Thus the distance traveled by $W$ indeed satisfies the norm assumption above, and following the same argument as (Du et al., 2018, Lemma 3.4) we conclude that Condition B holds true for the gradient flow of $f _ { N N }$ for $t > 0$ . + +Next we show that for any input $\hat { \pmb x }$ with $\| \hat { \pmb { x } } \| _ { 2 } = O ( \sqrt { d } )$ , the prediction of the neural network is uniformly close to the prediction of the prediction of the kernel model, a result similar to (Arora et al., 2019a, Lemma F.1). Following the notation of Arora et al. (2019a), we write the time derivative of the prediction as + +$\frac { \mathrm { d } } { \mathrm { d } t } f _ { N N } ( \hat { x } , t ) = \frac { 1 } { n } { u _ { N N } ( \hat { x } , t ) } ^ { \top } ( { y - y _ { N N } ( t ) } ) ; \frac { \mathrm { d } } { \mathrm { d } t } f _ { N T K } ( \hat { x } , t ) = \frac { 1 } { n } { u _ { N T K } ( \hat { x } , t ) } ^ { \top } ( { y - y _ { N T K } ( t ) } ) ,$ where $\begin{array} { r } { { \pmb u } _ { N N } ( { \pmb x } , t ) = \frac { \partial { \pmb f } ( X ; \omega ( t ) ) } { \partial { \pmb \omega } ( t ) } ^ { \top } \frac { \partial { \pmb f } ( { \pmb x } ; \omega ( t ) ) } { \partial { \pmb \omega } ( t ) } \in \mathbb { R } ^ { n } } \end{array}$ and ${ \pmb u } _ { N T K } ( { \pmb x } , t )$ similarly defined on the initialized weights $\omega ( 0 )$ . We bound the difference between the predictions on $\hat { \pmb x }$ up to terminal time T as + +$$ +\begin{array} { r l } & { \quad | \int _ { \mathbf { N } ^ { \mathrm { N } } } ( \hat { x } , t ) - \int _ { \mathbf { N } \cap \mathbf { K } } ( \hat { x } , t ) | = | \int _ { 0 } ^ { T } \left[ \frac { \mathrm { d } } { \mathrm { d } t } \int _ { \mathbf { N } \cap \mathbf { K } } ( \hat { x } , t ) - \frac { \mathrm { d } } { \mathrm { d } t } f _ { N \cap \mathbf { K } } ( \hat { x } , t ) \right] \mathrm { d } t | } \\ & { = \frac { 1 } { n } | \int _ { 0 } ^ { T } \left[ \mathbf { a } _ { N \wedge \mathbf { N } } ( \hat { x } , t ) ^ { \top } ( y - y _ { N N } ( t ) ) - u _ { N \cap \mathbf { K } } ( \hat { x } , t ) ^ { \top } ( y - y _ { N T } ( \hat { x } ) ) \right] \mathrm { d } t | } \\ & { \leq \frac { 1 } { n } | \int _ { 0 } ^ { T } u _ { N \wedge \mathbf { N } } ( \hat { x } , t ) ^ { \top } ( y _ { N N } ( t ) - y _ { N \wedge \mathbf { K } } ( t ) ) \mathrm { d } t | } \\ & { \quad + \frac { 1 } { n } | \int _ { 0 } ^ { T } \left( u _ { N \wedge \mathbf { N } } ( \hat { x } , t ) - u _ { N T \mathbf { K } } ( \hat { x } , t ) \right) ^ { \top } ( y - y _ { N N } ( t ) ) \mathrm { d } t | } \\ & { \leq \frac { 1 } { n } \| u _ { N \cap \mathbf { K } } ( \hat { x } , t ) \| _ { 2 } \int _ { 0 } ^ { T } \| y _ { N N } ( t ) - y _ { N T \mathbf { K } } ( t ) \| _ { 2 } \mathrm { d } t } \\ & { \quad + \frac { 1 } { n } \| u _ { N \wedge \mathbf { N } } ( \hat { x } , t ) - u _ { N T \mathbf { K } } ( \hat { x } , t ) \| _ { 2 } \int _ { 0 } ^ { T } \| y - y _ { N N } ( t ) \| _ { 2 } \mathrm { d } t } \\ & { \quad + \frac { 1 } { n } \operatorname* { m a x } \left( u _ { N \wedge \mathbf { N } } ( \hat { x } , t ) - u _ { N T \mathbf { K } } ( \hat { x } , t ) \right) \| _ { 2 } \int _ { 0 } ^ { T } \| y - y _ { N N } ( t ) \| _ { 2 } \mathrm { d } t . } \end{array} +$$ + +For the first term have + +$$ +\begin{array} { r l r } { { \| { \boldsymbol y } _ { N N } ( T ) - { \boldsymbol y } _ { N T K } ( T ) \| _ { 2 } \le \frac { 1 } { n } \int _ { 0 } ^ { T } \| K ( t ) ( { \boldsymbol y } - { \boldsymbol y } _ { N N } ( t ) ) - K ( 0 ) ( { \boldsymbol y } - { \boldsymbol y } _ { N T K } ( t ) ) \| _ { 2 } \mathrm { d } t } } \\ & { } & { \le \frac { 1 } { n 0 < t < T } \| K ( t ) - K ( 0 ) \| _ { 2 } \int _ { 0 } ^ { T } \| { \boldsymbol y } - { \boldsymbol y } _ { N N } ( t ) \| _ { 2 } \mathrm { d } t + \frac { 1 } { n } \| K ( 0 ) \| _ { 2 } \int _ { 0 } ^ { T } \| { \boldsymbol y } _ { N N } ( t ) - { \boldsymbol y } _ { N T K } ( t ) \| _ { 2 } \mathrm { d } t } \\ & { } & { \overset { ( i ) } { \le } \frac { 1 } { n } O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } ) O ( \sqrt { d } ) + O ( 1 ) \int _ { 0 } ^ { T } \| { \boldsymbol y } _ { N N } ( t ) - { \boldsymbol y } _ { N T K } ( t ) \| _ { 2 } \mathrm { d } t \overset { ( i i ) } { \le } O ( d ^ { - \epsilon ^ { \prime } } ) , \quad \quad \quad \quad ( 1 2 4 ) \mathrm { d } { \boldsymbol z } . } \end{array} +$$ + +where (i) is due to Corollary 15, Lemma 19 for some $\epsilon ^ { \prime } > 0$ and the linear convergence of ${ \bf { \it { \mathbf { y } } } } _ { N N }$ , and (ii) is due to Gronwall’s inequality. Note that the log factor in $T$ is omitted. Similarly, for the second term we have + +$$ +\frac { 1 } { n \hbar \epsilon \mathcal { T } } \left. u _ { N N } ( \hat { x } , t ) - u _ { N T K } ( \hat { x } , t ) \right. _ { 2 } \int _ { 0 } ^ { T } \left. y - y _ { N N } ( t ) \right. _ { 2 } \mathrm { d } t \overset { ( i ) } { \leq } \frac { 1 } { n } O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } ) O ( \sqrt { d } ) = O ( d ^ { - \epsilon ^ { \prime } } ) , +$$ + +where we used Lemma 19 and the linear convergence of ${ \pmb y } _ { N N }$ in (i). Combining the two cases yields + +$$ +| f _ { N N } ( \pmb { \hat { x } } , t ) - f _ { N T K } ( \pmb { \hat { x } } , t ) | \leq \frac { 1 } { n } \| \pmb { u } _ { N T K } ( \pmb { \hat { x } } , t ) \| _ { 2 } O ( d ^ { - \epsilon ^ { \prime } } ) + O ( d ^ { - \epsilon ^ { \prime } } ) \overset { ( i ) } { = } O ( d ^ { - \epsilon ^ { \prime } } ) , +$$ + +where we utilized Corollary 14 in (i). Thus we know that the difference between the population risk of $f _ { N N }$ and $f _ { N T K }$ is also asymptotically vanishing (note that the derivation above is independent of√ the target function as long as $\bar { | | \mathbf { y } | | _ { 2 } } = \bar { O ( \sqrt { n } ) } )$ . Therefore, in the following subsection we compute the risk of the linearized (kernel) model $f _ { N T K }$ . + +# C.9.2 COMPUTING THE KERNEL RISK + +Given input $X \in \mathbb { R } ^ { d \times n }$ and label $\pmb { y } = \pmb { \beta } ^ { \top } \pmb { X } + \pmb { \varepsilon }$ , gradient flow on the tangent kernel solves the following equation of the parameters $\omega$ : + +$$ +\pmb { y } = \pmb { f } ( X ; \omega ) = \frac { \partial \pmb { f } ( X ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } ( \pmb { \omega } - \pmb { \omega } _ { 0 } ) , +$$ + +where $\partial f ( X ; \omega _ { 0 } ) / \partial \omega$ is a $d h \times n$ matrix with column $\partial f ( \pmb { x } _ { i } ; \pmb { \omega } _ { 0 } ) / \partial \omega$ . Note that for $n \to \infty$ and $\gamma _ { 1 } , \gamma _ { 2 } \in ( 0 , \infty )$ , $d h > n$ trivially holds, and by Corollary 15 we know that solution is given by + +$$ +\omega _ { 1 } = \omega _ { 0 } + \frac { \partial { f ( X ; \omega _ { 0 } ) } } { \partial { \omega } } \left( \frac { \partial { f ( X ; \omega _ { 0 } ) } } { \partial { \omega } } ^ { \top } \frac { \partial { f ( X ; \omega _ { 0 } ) } } { \partial { \omega } } \right) ^ { - 1 } ( X ^ { \top } \beta + \varepsilon ) . +$$ + +And the population risk can be written as (note that there is a factor of 2 due to the "doubling trick" at initialization to ensure $f ^ { \mathrm { i n i t } } ( \cdot ) = 0 .$ ): + +$$ +\begin{array} { r l } & { 2 R = \mathbb { E } _ { \alpha , \epsilon } [ ( x ^ { \top } \beta - f ( x ; \omega _ { 1 } ) ) ^ { 2 } ] } \\ & { \quad = \mathbb { E } _ { \alpha , \epsilon } [ ( x ^ { \top } \beta - \frac { \partial f ( x ; \omega _ { 0 } ) ^ { \top } } { \partial \omega } ( \omega _ { 1 } - \omega _ { 0 } ) ) ^ { 2 } ] } \\ & { \quad = \mathbb { E } _ { \alpha , \epsilon } [ ( x ^ { \top } \beta - \frac { \partial f ( x ; \omega _ { 0 } ) ^ { \top } } { \partial \omega } \frac { \partial f ( X ; \omega _ { 0 } ) } { \partial \omega } ( \frac { \partial f ( X ; \omega _ { 0 } ) ^ { \top } } { \partial \omega } \frac { \partial f ( X ; \omega _ { 0 } ) } { \partial \omega } ) ^ { - 1 } ( X ^ { \top } \beta + \varepsilon ) ) ^ { 2 } ] } \\ & { \quad = \mathbb { E } _ { \alpha } [ ( x ^ { \top } \beta - \frac { \partial \tilde { f } ( \kappa ^ { - 1 } X ^ { \top } \beta ) } { \partial \beta } \frac { \partial \tilde { f } ( X ; \omega _ { 0 } ) } { \partial \omega } ( \frac { \partial \tilde { f } ( \tilde { X } ^ { - 1 } \tilde { \alpha } ^ { - 1 } \tilde { \alpha } ^ { - 1 } \tilde { \alpha } ^ { - 1 } \tilde { \alpha } ^ { - 1 } ) } { 2 V } \frac { \partial ^ { 2 } } { \partial \omega } , \qquad ( 1 2 9 ) ) } \end{array} +$$ + +where a bias-variance decomposition is made here, and for simplicity we define + +$$ +\hat { \pmb { u } } = \frac { \partial { \pmb { f } } ( X ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial { \pmb { f } } ( { \pmb { x } } ; \omega _ { 0 } ) } { \partial \omega } , \quad \hat { K } _ { X } = \frac { \partial { \pmb { f } } ( X ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial { \pmb { f } } ( X ; \omega _ { 0 } ) } { \partial \omega } . +$$ + +# C.9.3 APPROXIMATING THE KERNEL MATRIX + +In this section we drop the negligible $\epsilon$ in the initialization. Following Cheng and Singer (2013) we utilize the orthonormal decomposition of $\phi ^ { \prime } ( x )$ in $L ^ { 2 } ( \mathbb { R } , \mu _ { G } )$ . Denote $b _ { 0 } = \mathbb { E } [ \phi ^ { \prime } ( G ) ]$ , and $b _ { 1 } ^ { 2 } = \mathbb { E } [ \phi ^ { \prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }$ . We have the orthogonal decomposition of $\phi ^ { \prime }$ + +$$ +\phi ^ { \prime } ( x ) = b _ { 0 } + \phi _ { \perp } ^ { \prime } ( x ) , +$$ + +where $\mathbb { E } [ \phi _ { \perp } ^ { \prime } ( G ) ] = 0$ . We develop the following lemmas to approximate the kernel matrix. + +Lemma 13 (Approximation of $( \hat { K } _ { X } ) _ { i j } .$ ). There exist constants $c , c ^ { \prime } > 0$ such that for $i \neq j$ with probability $1 - e ^ { - c n \varepsilon ^ { 2 } }$ we have + +$$ +\left| \frac { 1 } { d } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega _ { 0 } ) } { \partial \omega } - \frac { 1 } { d } b _ { 0 } ^ { 2 } { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \right| < \varepsilon ^ { 2 } , +$$ + +and with probability 1 − e−c0nε2 , + +$$ +\left| \frac { 1 } { d } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } - ( b _ { 0 } ^ { 2 } + b _ { 1 } ^ { 2 } ) \right| < \varepsilon . +$$ + +Proof. When $i \neq j$ (i.e. Equation (132)), we have + +$$ +\begin{array} { r l } & { \quad \displaystyle \frac 1 d [ \hat { K } _ { X } ] _ { i j } = \frac 1 d \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega _ { 0 } ) } { \partial \omega } } \\ & { = \displaystyle \frac 1 d { \sum _ { k = 1 } ^ { h } } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial { \pmb w } _ { k } } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega _ { 0 } ) } { \partial { \pmb w } _ { k } } = \frac 1 { d h } { \displaystyle \sum _ { k = 1 } ^ { h } } { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \phi ^ { \prime } ( { \pmb w } _ { k } ^ { \top } { \pmb x } _ { i } ) \phi ^ { \prime } ( { \pmb w } _ { k } ^ { \top } { \pmb x } _ { j } ) } \\ & { \to \displaystyle \frac 1 d { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \mathbb { E } _ { \pmb w } \Big [ \phi ^ { \prime } ( { \pmb w } ^ { \top } { \pmb x } _ { i } ) \phi ^ { \prime } ( { \pmb w } ^ { \top } { \pmb x } _ { j } ) \Big ] = \frac 1 d H ( { \pmb x } _ { i } , { \pmb x } _ { j } ) . } \end{array} +$$ + +The matrix $H ( \pmb { x } _ { i } , \pmb { x } _ { j } ) = \pmb { x } _ { i } ^ { \top } \pmb { x } _ { j } \mathbb { E } _ { \pmb { w } } \Big [ \phi ^ { \prime } ( \pmb { w } ^ { \top } \pmb { x } _ { i } ) \phi ^ { \prime } ( \pmb { w } ^ { \top } \pmb { x } _ { j } ) \Big ]$ can be seen as the expected tangent kernel of nonlinear activation function studied in Du et al. (2018); Arora et al. (2019b). Moreover, due to the assumed boundedness of $\phi ^ { \prime } ( x )$ (A3), by Hoeffding’s inequality we have + +$$ +\operatorname* { P r } \left| \frac { 1 } { d } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega _ { 0 } ) } { \partial \omega } - \frac { 1 } { d } H ( { \pmb x } _ { i } , { \pmb x } _ { j } ) \right| < \frac { 1 } { d } { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \varepsilon > 1 - e ^ { - c _ { 1 } h \varepsilon ^ { 2 } } . +$$ + +In addition, by the concentration of $\pmb { x } _ { i } ^ { \top } \pmb { x } _ { j }$ and $\| \pmb { x } _ { i } \| _ { 2 } ^ { 2 }$ , i.e. $\mathrm { P r } \pmb { x } _ { i } ^ { \top } \pmb { x } _ { j } / d > \varepsilon < 1 - e ^ { - c _ { 2 } d \varepsilon ^ { 2 } }$ and $\operatorname* { P r } | \pmb { x } _ { i } ^ { \top } \pmb { x } _ { i } / d - 1 | < \varepsilon > 1 - e ^ { - c _ { 3 } d \varepsilon ^ { 2 } }$ , the orthonormal decomposition $\phi ^ { \prime } ( x ) = b _ { 0 } x + \phi _ { \perp } ^ { \prime } ( x )$ leads to the following linear approximation of the matrix $H$ + +$$ +\frac { 1 } { d } H ( \pmb { x } _ { i } , \pmb { x } _ { j } ) = b _ { 0 } ^ { 2 } \frac { 1 } { d } \pmb { x } _ { i } ^ { \top } \pmb { x } _ { j } + O \big ( ( \pmb { x } _ { i } ^ { \top } \pmb { x } _ { j } / d ) ^ { 2 } \big ) . +$$ + +and by taking $\varepsilon = \pmb { x } _ { i } ^ { \top } \pmb { x } _ { j } / d$ under the joint event we can show that + +$$ +\left| \frac { 1 } { d } \frac { \partial f ( { \pmb x } _ { i } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega _ { 0 } ) } { \partial \omega } - \frac { 1 } { d } b _ { 0 } ^ { 2 } { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \right| < \varepsilon ^ { 2 } +$$ + +with probability $1 - e ^ { - c d \varepsilon ^ { 2 } }$ . The same argument follows for the case where $i = j$ + +Corollary 14 (Approximation of $\hat { \textbf { \textit { u } } }$ ). For large enough $l > 0$ , with probability $1 - d e ^ { - c \log ^ { l } d }$ + +$$ +\frac { 1 } { d } \left\| \hat { \pmb { u } } - \tilde { \pmb { u } } \right\| _ { 2 } = \left\| \frac { 1 } { d } \frac { \partial f ( \pmb { x } ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial \pmb { f } ( X ; \omega _ { 0 } ) } { \partial \omega } - \frac { 1 } { d } \tilde { \pmb { u } } \right\| _ { 2 } < \frac { \log ^ { l } d } { d } , +$$ + +where $\tilde { \pmb { u } } = b _ { 0 } ^ { 2 } \pmb { x } ^ { \top } \boldsymbol { X }$ . + +Proof. Taking $\varepsilon = \log ^ { l } d / d$ together with Lemma 13 yields the desired result. + +Corollary 15 (Approximation of ${ \hat { K } } _ { X } { \mathrm { . } }$ ). With probability $1 - d e ^ { - c \log ^ { l } d }$ + +$$ +\frac { 1 } { d } \left\| \hat { K } _ { X } - \tilde { K } _ { X } \right\| _ { F } = \left\| \frac { 1 } { d } \frac { \partial f ( X ; \omega _ { 0 } ) } { \partial \omega } ^ { \top } \frac { \partial f ( X ; \omega _ { 0 } ) } { \partial \omega } - \frac { 1 } { d } \tilde { K } _ { X } \right\| _ { F } < \log ^ { l } d , +$$ + +where $\tilde { K } _ { X } = b _ { 0 } ^ { 2 } X ^ { \top } X + b _ { 1 } ^ { 2 } d I$ . + +Proof. Also by directly applying Lemma 13. + +Remark. For initialization larger or equal to ${ \pmb w } _ { i } ( 0 ) \sim N ( 0 , I _ { d } / d )$ , the above approximation does not depend on the scale of initialization. + +Remark. For $\phi ( x ) = \mathrm { S o f t P l u s } ( x ) .$ , $b _ { 0 } ^ { 2 } = 1 / 4$ , $b _ { 1 } ^ { 2 } = 0 . 0 4 3 3 7 9$ . For $\phi ( x ) = \mathrm { s i g m o i d } ( x ) = ( 1 +$ $e ^ { - x } ) ^ { - 1 }$ , $b _ { 0 } ^ { 2 } = 0 . 0 4 2 6 9 2$ , $b _ { 1 } ^ { 2 } = 0 . 0 0 2 1 4 4$ . Note that $b _ { 1 } \geq 0$ for all smooth activations $\phi$ , and the equality holds (i.e. $b _ { 1 } = 0$ ) if and only if $\phi$ is linear. We comment that smaller $b _ { 1 }$ entails larger variance as $\gamma _ { 1 } 1$ , and vice versa, as shown in the following section. + +# C.9.4 THE BIAS TERM + +With these approximation above we proceed to calculating (129) + +$$ +2 B = \mathbb { E } _ { \pmb { x } } \left[ \left( \pmb { x } ^ { \top } \pmb { \beta } - \hat { \pmb { u } } ^ { \top } \hat { K } _ { X } ^ { - 1 } X ^ { \top } \pmb { \beta } \right) ^ { 2 } \right] . +$$ + +We first bound the error in substituting $\hat { \textbf { \textit { u } } }$ with $\tilde { \mathbf { \pmb { u } } }$ : + +$$ +\begin{array} { r l } & { \left\| \hat { \boldsymbol u } ^ { \top } \hat { \boldsymbol K } _ { X } ^ { - 1 } \boldsymbol X ^ { \top } \boldsymbol \beta - \tilde { \boldsymbol u } ^ { \top } \hat { \boldsymbol K } _ { X } ^ { - 1 } \boldsymbol X ^ { \top } \boldsymbol \beta \right\| _ { 2 } \leq \| \hat { \boldsymbol u } - \tilde { \boldsymbol u } \| _ { 2 } \left\| \hat { \boldsymbol K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \boldsymbol X \right\| _ { 2 } \left\| \boldsymbol \beta \right\| _ { 2 } } \\ & { \quad \quad \quad \quad \stackrel { ( i ) } { = } O \left( \log ^ { l } d \cdot d ^ { - 1 } \cdot \sqrt { d } \cdot 1 \right) = O \left( \frac { \log ^ { l } d } { \sqrt { d } } \right) , } \end{array} +$$ + +where (i) is due to the fact that $\left\| \hat { K } _ { X } ^ { - 1 } \right\| _ { 2 } = \lambda _ { \operatorname* { m i n } } ^ { - 1 } ( \hat { K } _ { X } ) = O ( 1 / d )$ and $\| X \| _ { 2 } = O ( { \sqrt { d } } )$ . Therefore we have as $n , d , h \infty$ + +$$ +\begin{array} { r l } & { 2 B = \mathbb { E } _ { x } [ ( { \pmb x } ^ { \top } \beta - \hat { \pmb u } ^ { \top } \hat { K } _ { X } ^ { - 1 } X ^ { \top } \beta ) ^ { 2 } ] \mathbb { E } _ { \pmb { x } } [ ( { \pmb x } ^ { \top } \beta - \tilde { \pmb u } ^ { \top } \hat { K } _ { X } ^ { - 1 } X ^ { \top } \beta ) ^ { 2 } ] } \\ & { \qquad = \mathbb { E } _ { \pmb { x } } [ ( { \pmb x } ^ { \top } \beta - b _ { 0 } ^ { 2 } { \pmb x } ^ { \top } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \beta ) ^ { 2 } ] . } \end{array} +$$ + +By taking expectation over $_ { \textbf { \em x } }$ and the rotational invariance argument similar to Hastie et al. (2019), + +$$ +\begin{array} { r l } { \mathbb { E } _ { x } \left[ \left( x ^ { \top } \beta - b _ { 0 } ^ { 2 } x ^ { \top } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \beta \right) ^ { 2 } \right] } & { = \mathbb { E } _ { x } \left[ \beta ^ { \top } \left( I - b _ { 0 } ^ { 2 } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \right) ^ { 2 } \beta \right] } \\ & { = \frac { \beta ^ { \top } \beta } { d } \mathrm { t r } \left( \left( I - b _ { 0 } ^ { 2 } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \right) \left( I - b _ { 0 } ^ { 2 } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \right) \right) . } \end{array} +$$ + +In addition, we bound the error in substituting $\hat { K } _ { X }$ by $\tilde { K } _ { X }$ defined in (139): + +$$ +\begin{array} { r l } & { \left| \frac { 1 } { d } \mathrm { t r } \left( X \hat { K } _ { X } ^ { - 1 } X ^ { \top } - X \tilde { K } _ { X } ^ { - 1 } X ^ { \top } \right) \right| = \left| \frac { 1 } { d } \mathrm { t r } \left( X ^ { \top } X \hat { K } _ { X } ^ { - 1 } ( \hat { K } _ { X } - \tilde { K } _ { X } ) \tilde { K } _ { X } ^ { - 1 } \right) \right| } \\ & { \qquad < \displaystyle \frac { 1 } { d } \left\| X ^ { \top } X \right\| _ { 2 } \left\| \hat { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \hat { K } _ { X } - \tilde { K } _ { X } \right\| _ { F } \left\| \tilde { K } _ { X } ^ { - 1 } \right\| _ { 2 } } \\ & { \qquad = O \left( d ^ { - 1 } \cdot d \cdot d ^ { - 1 } \cdot d \log ^ { l } d \cdot d ^ { - 1 } \right) = O \left( \frac { \log ^ { l } d } { d } \right) , } \end{array} +$$ + +and similarly, + +$$ +\begin{array} { r l } & { ~ \left| \frac { 1 } { d } \mathrm { t r } \left( X \hat { K } _ { X } ^ { - 1 } X ^ { \top } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } - X \hat { K } _ { X } ^ { - 1 } X ^ { \top } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \right) \right| } \\ & { = \left| \frac { 1 } { d } \mathrm { t r } \left( X ^ { \top } X ( \hat { K } _ { X } ^ { - 1 } - \tilde { K } _ { X } ^ { - 1 } ) X ^ { \top } X ( \hat { K } _ { X } ^ { - 1 } + \tilde { K } _ { X } ^ { - 1 } ) \right) \right| } \\ & { < \frac { 1 } { d } \left\| X ^ { \top } X \right\| _ { 2 } \left\| \hat { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \hat { K } _ { X } - \tilde { K } _ { X } \right\| _ { F } \left\| \tilde { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| X ^ { \top } X \right\| _ { 2 } \left\| \hat { K } _ { X } ^ { - 1 } + \tilde { K } _ { X } ^ { - 1 } \right\| _ { 2 } } \\ & { = O \left( d ^ { - 1 } \cdot d \cdot d ^ { - 1 } \cdot d \log ^ { l } d \cdot d ^ { - 1 } \cdot d \cdot d ^ { - 1 } \right) = O \left( \frac { \log ^ { l } d } { d } \right) . } \end{array} +$$ + +Combining these two formulas in (143) yields + +$$ +\begin{array} { r l } & { 2 B \to \mathbb { E } _ { x } [ ( x ^ { \top } \beta - b _ { 0 } ^ { 2 } x ^ { \top } X \hat { K } _ { X } ^ { - 1 } X ^ { \top } \beta ) ^ { 2 } ] } \\ & { \quad \cfrac { \beta ^ { \top } \beta } { d } \mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X \tilde { K } _ { X } ^ { - 1 } X ^ { \top } ) ( I - b _ { 0 } ^ { 2 } X \tilde { K } _ { X } ^ { - 1 } X ^ { \top } ) ) } \\ & { \quad = \cfrac { \beta ^ { \top } \beta } { d } \mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X ( b _ { 0 } ^ { 2 } X ^ { \top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 1 } X ^ { \top } ) ^ { 2 } ) . } \end{array} +$$ + +Utilizing the Marcenko–Pastur law from Section ˇ B.2 we obtain + +$$ +B = \beta ^ { \top } \beta \left( \frac { \gamma _ { 1 } - 1 } { 2 \gamma _ { 1 } } + \frac { \gamma _ { 1 } ( \gamma _ { 1 } + \gamma _ { 1 } m + m - 2 ) + 1 } { 2 \gamma _ { 1 } \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } \right) , +$$ + +where $m = { b _ { 0 } } ^ { - 2 } { b _ { 1 } } ^ { 2 }$ . + +# C.9.5 THE VARIANCE TERM + +Similarly, for the variance we utilize the approximation + +$$ +2 V = \mathbb { E } _ { \pmb { x } } \left[ \hat { \pmb { u } } \hat { K } _ { \scriptscriptstyle X } ^ { - 1 } \hat { K } _ { \scriptscriptstyle X } ^ { - 1 } \hat { \pmb { u } } ^ { \top } \right] \sigma ^ { 2 } +$$ + +Specifically, we bound the approximation error + +$$ +\begin{array} { r } { \left| \hat { \boldsymbol u } \hat { K } _ { X } ^ { - 1 } \hat { \boldsymbol K } _ { X } ^ { - 1 } \hat { \boldsymbol u } ^ { \top } - \tilde { \boldsymbol u } \hat { \boldsymbol K } _ { X } ^ { - 1 } \hat { \boldsymbol K } _ { X } ^ { - 1 } \tilde { \boldsymbol u } ^ { \top } \right| \leq \left\| \hat { \boldsymbol u } - \tilde { \boldsymbol u } \right\| _ { 2 } \left\| \hat { \boldsymbol K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \hat { \boldsymbol u } + \tilde { \boldsymbol u } \right\| _ { 2 } \left\| \hat { \boldsymbol K } _ { X } ^ { - 1 } \right\| _ { 2 } } \\ { = O \left( \log ^ { l } d \cdot \frac { 1 } { d } \cdot d \cdot \frac { 1 } { d } \right) = O \left( \frac { \log ^ { l } d } { d } \right) , } \end{array} +$$ + +and similarly + +$$ +\begin{array} { r l } & { \quad \left| \mathbb { E } _ { \mathbf { x } } \left[ \tilde { u } \hat { K } _ { X } ^ { - 1 } \hat { K } _ { X } ^ { - 1 } \tilde { u } ^ { \top } - \tilde { u } \tilde { K } _ { X } ^ { - 1 } \tilde { K } _ { X } ^ { - 1 } \tilde { u } ^ { \top } \right] \right| } \\ & { = \mathrm { t r } \left( \left( \hat { K } _ { X } ^ { - 1 } - \tilde { K } _ { X } ^ { - 1 } \right) \left( \hat { K } _ { X } ^ { - 1 } + \tilde { K } _ { X } ^ { - 1 } \right) \mathbb { E } _ { \alpha \tilde { u } \tilde { u } ^ { \top } } \right) } \\ & { = \mathrm { t r } \left( \hat { K } _ { X } ^ { - 1 } \left( \hat { K } _ { X } - \tilde { K } _ { X } \right) \tilde { K } _ { X } ^ { - 1 } \left( \hat { K } _ { X } ^ { - 1 } + \tilde { K } _ { X } ^ { - 1 } \right) X ^ { T } X \right) } \\ & { \leq \left\| \hat { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \hat { K } _ { X } - \tilde { K } _ { X } \right\| _ { F } \left\| \tilde { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| \hat { K } _ { X } ^ { - 1 } + \tilde { K } _ { X } ^ { - 1 } \right\| _ { 2 } \left\| X ^ { T } X \right\| _ { 2 } } \\ & { = O ( d ^ { - 1 } \cdot d \log ^ { l } d \cdot d ^ { - 1 } \cdot d ^ { - 1 } \cdot d ) = O \left( \frac { \log ^ { l } d } { d } \right) , } \end{array} +$$ + +By combining the two approximations above we know that as $n , d , p \to \infty$ + +$$ +| 2 V - \mathbb { E } _ { \pmb { x } } [ \tilde { \pmb { u } } \tilde { K } _ { X } ^ { - 1 } \tilde { K } _ { X } ^ { - 1 } \tilde { \pmb { u } } ^ { \top } ] \sigma ^ { 2 } | = O ( \frac { \log ^ { l } d } { d } ) 0 . +$$ + +Therefore the variance is given as + +$$ +\begin{array} { r l } & { 2 V \sigma ^ { 2 } \mathbb { E } _ { x } \Big [ \tilde { u } \tilde { K } _ { X } ^ { - 1 } \tilde { K } _ { X } ^ { - 1 } \tilde { u } ^ { \top } \Big ] } \\ & { \quad = \sigma ^ { 2 } \mathbb { E } _ { x } \Big [ b _ { 0 } ^ { 4 } { \boldsymbol x } ^ { T } X ( b _ { 0 } ^ { 2 } X ^ { \top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 2 } X ^ { T } { \boldsymbol x } \Big ] } \\ & { \quad = \sigma ^ { 2 } \frac { 1 } { d } \mathrm { t r } ( \frac { 1 } { d } X ^ { T } X \cdot ( \frac { 1 } { d } X ^ { T } X + b _ { 0 } ^ { - 2 } b _ { 1 } ^ { 2 } I ) ^ { - 2 } ) } \\ & { \quad = \sigma ^ { 2 } ( - \frac { 1 } { 2 } + \frac { \gamma _ { 1 } + \gamma _ { 1 } m + 1 } { 2 \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) , } \end{array} +$$ + +where $m$ is defined in the derivation of the bias term. + +# C.9.6 PUTTING THINGS TOGETHER + +Recall the population risk is the sum of the bias and variance + +$$ +\begin{array} { c } { { R r ^ { 2 } ( \frac { \gamma _ { 1 } - 1 } { 2 \gamma _ { 1 } } + \frac { \gamma _ { 1 } ( \gamma _ { 1 } + \gamma _ { 1 } m + m - 2 ) + 1 } { 2 \gamma _ { 1 } \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) } } \\ { { + \sigma ^ { 2 } ( - \frac { 1 } { 4 } + \frac { \gamma _ { 1 } + \gamma _ { 1 } m + 1 } { 4 \sqrt { \gamma _ { 1 } ( \gamma _ { 1 } + m ( \gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) . } } \end{array} +$$ + +Observe that the population risk is independent of $\gamma _ { 2 }$ , i.e. double descent does not occur when the network is overparameterized via changing the width. In addition, the bias is monotonically increasing and upper-bounded by the null risk $r ^ { \bar { 2 } }$ and lower-bounded by the bias of the least squares solution on the input features ${ \hat { \boldsymbol { \beta } } } = X ^ { \dagger } \boldsymbol { y }$ , whereas the variance remains bounded for all $\gamma _ { 1 } \in ( 0 , \infty )$ as long as $m > 0$ , i.e. $\phi$ is nonlinear. + +# D USEFUL LEMMAS + +Lemma 16. Given $K _ { W }$ from (46), define + +$$ +\tilde { K } _ { W } = r I _ { h } + s { \bf 1 } _ { h } { \bf 1 } _ { h } ^ { \top } + t Q , +$$ + +where $Q \in \mathbb { R } ^ { h \times h }$ with $Q _ { i \neq j } \ = \ w _ { i } ^ { \top } w _ { j }$ and $Q _ { i , i } ~ = ~ 0$ , and $r = \mathbb { E } [ \phi ( G ) ^ { 2 } ] - \mathbb { E } [ \phi ( G ) ] ^ { 2 } , s =$ $\mathbb { E } [ \phi ( G ) ] ^ { 2 }$ , $t = \mathbb { E } [ G \phi ( G ) ] ^ { 2 }$ . Then as $d , h \to \infty , \left\| K _ { W } - \tilde { K } _ { W } \right\| _ { F } \leq \log ^ { c } d a . s .$ .. + +Proof. Consider the event where $\mathcal { A } _ { \epsilon } = \big \{ | \| \pmb { w } _ { i } \| _ { 2 } - 1 | < \epsilon , | \pmb { w } _ { i } ^ { \top } \pmb { w } _ { j } | < \epsilon \big \}$ . Under event $\mathcal { A } _ { \epsilon }$ , for the diagonal term of the kernel matrix we have + +$$ +\begin{array} { r l } & { ~ \Big | [ K _ { W } ] _ { i i } - [ \tilde { K } _ { W } ] _ { i i } \Big | = \Big | \mathbb { E } _ { \pmb { x } } \Big [ \phi ( \pmb { w } _ { i } ^ { \top } \pmb { x } ) \phi ( \pmb { w } _ { i } ^ { \top } \pmb { x } ) \Big ] - \mathbb { E } [ \phi ( G ) ^ { 2 } ] \Big | } \\ & { = \big | \mathbb { E } [ \phi ( \| \pmb { w } _ { i } \| _ { 2 } G ) ^ { 2 } ] - \mathbb { E } [ \phi ( G ) ^ { 2 } ] \big | = O ( \epsilon ) . } \end{array} +$$ + +And for off-diagonal term, by the decomposition introduced in Section B.3 we have + +$$ +[ K _ { W } ] _ { i j } = \mathbb { E } _ { \boldsymbol { x } } \Big [ \phi ( \boldsymbol { w } _ { i } ^ { \top } \boldsymbol { x } ) \phi ( \boldsymbol { w } _ { j } ^ { \top } \boldsymbol { x } ) \Big ] = \| \boldsymbol { w } _ { i } \| _ { 2 } \| \boldsymbol { w } _ { j } \| _ { 2 } + \mathbb { E } _ { \boldsymbol { x } } [ \phi _ { \bot } ( \boldsymbol { w } _ { i } ^ { \top } \boldsymbol { x } ) \phi _ { \bot } ( \boldsymbol { w } _ { j } ^ { \top } \boldsymbol { x } ) ] , +$$ + +hence $| [ K _ { W } ] _ { i j } - [ \tilde { K } _ { W } ] _ { i j } | < ( \pmb { w } _ { i } ^ { \top } \pmb { w } _ { j } ) ^ { 2 }$ . Notice that $\mathcal { A } _ { \epsilon }$ holds a.s. for $\epsilon = \log ^ { c } d / \sqrt { d }$ and large enough $c > 0$ ; we therefore have $\left\| K _ { W } - \tilde { K } _ { W } \right\| _ { F } \leq \log ^ { c } d .$ . + +Lemma 17. Let ${ \hat { \boldsymbol { \beta } } } ( t )$ be the solution to the gradient flow at time t defined in (103)(104). Then as $n , d \to \infty$ and $\gamma _ { 1 } \neq 1$ the following holds.: + +$$ +\| \hat { \boldsymbol { \beta } } ( t ) \| _ { 2 } = O _ { P } ( 1 ) ; \quad \| \boldsymbol { X } ^ { \top } \hat { \boldsymbol { \beta } } ( t ) \| _ { \infty } = O _ { P } ( \mathrm { p o l y l o g } d ) . +$$ + +Proof. We consider $d < n$ for simplicity, and result for the other case follows in similar fashion. + +In this case $\begin{array} { r } { \hat { \pmb { \beta } } ( t ) = \left( I - \exp ( - \frac { t } { n } X X ^ { \top } ) \right) ( X X ^ { \top } ) ^ { - 1 } X \pmb { y } } \end{array}$ . From (Hastie et al., 2019, Corollary 1) we know that $\| \hat { \pmb \beta } ( \infty ) \| _ { 2 } = \left\| ( X X ^ { \top } ) ^ { - 1 } X \pmb y \right\| _ { 2 } = O ( 1 )$ for $\gamma _ { 1 } < 1$ . Note that $\begin{array} { r l r } { { \| { \cal I } - \exp ( - \frac { t } { n } X X ^ { \top } ) \| _ { 2 } = } } \end{array}$ $O ( 1 )$ for $t \geq 0$ ; it follows that $\| \hat { \boldsymbol { \beta } } ( t ) \| _ { 2 } = O ( 1 )$ . + +For the second part, we utilize the SVD $\boldsymbol { X } = \boldsymbol { U \Sigma V } ^ { \top }$ , where $\Sigma = [ \hat { \Sigma } ; 0 ]$ , $\hat { \Sigma } \in \mathbb { R } ^ { d \times d }$ and $\hat { \Sigma } _ { i i } = \lambda _ { i }$ . +We have $\begin{array} { r } { \left( I - \exp ( - \frac { t } { n } X X ^ { \top } ) \right) = U \bar { \Sigma } U ^ { \top } } \end{array}$ where $\bar { \Sigma } _ { i , i } = 1 - \exp ( - t \lambda _ { i } ^ { 2 } / n )$ if $i \leq d$ and 0 otherwise. + +$$ +\begin{array} { r l } & { \| X ^ { \top } \hat { \pmb \beta } ( t ) \| _ { \infty } = \left\| X ^ { \top } \left( I - \exp ( - \frac t n X X ^ { \top } ) \right) \left( X X ^ { \top } \right) ^ { - 1 } X \pmb y \right\| _ { \infty } } \\ & { \qquad \leq \left\| V \Sigma ^ { \top } U ^ { \top } U \bar { \Sigma } U ^ { \top } U \hat { \Sigma } ^ { - 2 } U ^ { \top } U \Sigma V ^ { \top } \right\| _ { \infty } \| \pmb y \| _ { \infty } } \\ & { \qquad \leq \left\| V \Sigma ^ { \top } \bar { \Sigma } \hat { \Sigma } ^ { - 2 } \Sigma V ^ { \top } \right\| _ { \infty } \left( \| X ^ { \top } \pmb \beta \| _ { \infty } + \| \varepsilon \| _ { \infty } \right) } \\ & { \qquad \leq \| V \| _ { \infty } \left\| \Sigma ^ { \top } \bar { \Sigma } \hat { \Sigma } ^ { - 2 } \Sigma \right\| _ { \infty } \| V ^ { \top } \| _ { \infty } \left( \| X ^ { \top } \pmb \beta \| _ { \infty } + \| \varepsilon \| _ { \infty } \right) \overset { ( i ) } { \leq } O _ { P } ( \mathrm { p o l y } \log d ) , } \end{array} +$$ + +where (i) follows from the concentration of the Gaussian maxima, and the fact that the law of $V$ is the Haar measure on $S O ( n )$ , and thus for any unit vector $_ z$ independent to $V$ , $V z$ is uniform on sphere and $\| V z \| _ { \infty } = O ( \log d / \sqrt { d } )$ . We wherefore have + +$$ +\left\| V \right\| _ { \infty } = \operatorname* { s u p } _ { z } { \frac { \left\| V z \right\| _ { \infty } } { \left\| z \right\| _ { \infty } } } = O \left( { \frac { \log d } { \sqrt { d } } } \right) { \frac { \left\| z \right\| _ { 2 } } { \left\| z \right\| _ { \infty } } } = O ( \log d ) , +$$ + +Note that this result also implies that $\| \pmb { y } - X ^ { \top } \pmb { \beta } ( t ) \| _ { \infty } = O _ { P } ( \mathrm { p o l y l o g } d )$ . + +Lemma 18. For weight matrices $W , W ^ { \prime }$ satisfying $\| { \pmb y } - { \pmb f } ( { \pmb X } ) \| _ { 2 } \ = \ { \cal O } ( { \sqrt { n } } )$ , where $f ( X ) \ =$ $\phi ( X W ) \mathbf { a }$ with fixed $a _ { i } \sim \mathrm { U n i f } \{ - 1 / \sqrt { h } , 1 / \sqrt { h } \}$ , given (A1)-(A3), the gradient of the empirical risk defined in (11) is Lipschitz w.r.t. $W$ in the Frobenius norm, i.e. + +$$ +\left\| \frac { \partial L ( X ; W ) } { \partial W } - \frac { \partial L ( X ; W ^ { \prime } ) } { \partial W } \right\| _ { F } \leq L \left\| W - W ^ { \prime } \right\| _ { F } . +$$ + +Proof. Denote $\pmb { y } _ { 1 } = \phi ( \pmb { X } ^ { \top } \pmb { W } _ { 1 } ) \pmb { a }$ and $\pmb { y } _ { 2 } = \phi ( \pmb { X } ^ { \top } \pmb { W } _ { 2 } ) \pmb { a }$ for $W _ { 1 } , W _ { 2 }$ satisfying the assumption above (which can be seen as a condition on the magnitude of training loss), we have + +$$ +\begin{array} { r l } & { \quad \displaystyle \left\| \frac { \partial L ( W _ { 1 } ) } { \partial W _ { 1 } } - \frac { \partial L ( W _ { 2 } ) } { \partial W _ { 2 } } \right\| _ { F } } \\ & { = \displaystyle \left\| \frac { 1 } { n } X \left[ ( y - y _ { 1 } ) a ^ { \top } \circ \phi ^ { \prime } ( X ^ { \top } W _ { 1 } ) \right] - \frac { 1 } { n } X \left[ ( y - y _ { 2 } ) a ^ { \top } \circ \phi ^ { \prime } ( X ^ { \top } W _ { 2 } ) \right] \right\| _ { F } } \\ & { \leq \displaystyle \frac { 1 } { n } \| X \| _ { 2 } \left\| ( y - y _ { 1 } ) a ^ { \top } \circ \phi ^ { \prime } ( X ^ { \top } W _ { 1 } ) - ( y - y _ { 2 } ) a ^ { \top } \circ \phi ^ { \prime } ( X ^ { \top } W _ { 2 } ) \right\| _ { F } } \\ & { \leq \ O \left( \frac { 1 } { \sqrt { d } } \right) \left\| ( y _ { 2 } - y _ { 1 } ) a ^ { \top } \circ \phi ^ { \prime } ( X ^ { \top } W _ { 1 } ) \right\| _ { F } } \\ & { \quad + O \left( \frac { 1 } { \sqrt { d } } \right) \left\| ( y - y _ { 2 } ) a ^ { \top } \circ ( \phi ^ { \prime } ( X ^ { \top } W _ { 1 } ) - \phi ^ { \prime } ( X ^ { \top } W _ { 2 } ) ) \right\| _ { F } . } \end{array} +$$ + +We upper bound the two terms separately: + +$$ +\begin{array} { r l } & { \left\| ( y _ { 2 } - y _ { 1 } ) { \boldsymbol a } ^ { \top } \circ \phi ^ { \prime } ( \boldsymbol { X } ^ { \top } \boldsymbol { W } _ { 1 } ) \right\| _ { 2 } \overset { ( i ) } { \leq } \operatorname* { m a x } \{ \phi ^ { \prime } ( \boldsymbol { X } ^ { \top } \boldsymbol { W } _ { 1 } ) _ { i j } \} \left\| y _ { 2 } - y _ { 1 } \right\| _ { 2 } \left\| { \boldsymbol a } \right\| _ { 2 } } \\ & { \qquad \overset { ( i i ) } { \leq } O ( 1 ) \left\| \phi ^ { \prime } ( \boldsymbol { X } ^ { \top } \boldsymbol { W } _ { 1 } ) { \boldsymbol a } - \phi ^ { \prime } ( \boldsymbol { X } ^ { \top } \boldsymbol { W } _ { 2 } ) { \boldsymbol a } \right\| _ { F } } \\ & { \qquad \overset { ( i i i ) } { \leq } O ( 1 ) \left\| \boldsymbol { X } \right\| _ { 2 } \left\| \boldsymbol { W } _ { 1 } - \boldsymbol { W } _ { 2 } \right\| _ { F } = O ( \sqrt { d } ) \left\| \boldsymbol { W } _ { 1 } - \boldsymbol { W } _ { 2 } \right\| _ { F } , } \end{array} +$$ + +where we applied the inequality $\left\| A \circ B \right\| _ { F } \leq \operatorname* { m a x } \{ | A _ { i j } | \} \left\| B \right\| _ { F }$ in (i), boundedness of $\phi ^ { \prime }$ in (ii) and Lipschitzity of $\phi$ in (iii). Similarly, for the second term + +$$ +\begin{array} { r l } & { \qquad \left\| ( \pmb { y } - \pmb { y } _ { 2 } ) \pmb { a } ^ { \top } \circ ( \phi ^ { \prime } ( X ^ { \top } \pmb { W } _ { 1 } ) - \phi ^ { \prime } ( X ^ { \top } \pmb { W } _ { 2 } ) ) \right\| _ { F } } \\ & { \leq \operatorname* { m a x } \{ \left| a _ { i } \right| \} \left\| \pmb { y } - \pmb { y } _ { 2 } \right\| _ { 2 } \left\| \phi ^ { \prime } ( X ^ { \top } \pmb { W } _ { 1 } ) - \phi ^ { \prime } ( X ^ { \top } \pmb { W } _ { 2 } ) \right\| _ { F } } \\ & { \overset { ( i ) } { \leq } O ( 1 ) \left\| X \right\| _ { 2 } \left\| \pmb { W } _ { 1 } - \pmb { W } _ { 2 } \right\| _ { F } = O ( \sqrt { d } ) \left\| \pmb { W } _ { 1 } - \pmb { W } _ { 2 } \right\| _ { F } , } \end{array} +$$ + +where we used the assumption on the training loss and the Lipscthizity of $\phi ^ { \prime }$ in (i). Combining the two terms yields the desired result. + +Lemma 19. Under assumptions (A1-3) and the non-vanishing initialization, given that $\parallel { \pmb w } _ { i } ( t ) -$ ${ \pmb w } _ { i } ( 0 ) \| _ { 2 } = O ( d ^ { - 1 / 2 } )$ for all $i$ , then we have $\| K ( t ) - K ( 0 ) \| _ { 2 } = O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } )$ for some positive $\epsilon ^ { \prime } \in \Theta ( 1 )$ . + +Proof. Recall the definition of the NTK: + +$$ +K _ { i j } ( t ) = \frac { \partial f ( { \pmb x } _ { i } ; \omega ( t ) ) } { \partial \omega ( t ) } ^ { \top } \frac { \partial f ( { \pmb x } _ { j } ; \omega ( t ) ) } { \partial \omega ( t ) } = { \pmb x } _ { i } ^ { \top } { \pmb x } _ { j } \frac { 1 } { h } \sum _ { k = 1 } ^ { h } \phi ^ { \prime } ( { \pmb w } _ { k } ( t ) ^ { \top } { \pmb x } _ { i } ) \phi ^ { \prime } ( { \pmb w } _ { k } ( t ) ^ { \top } { \pmb x } _ { j } ) , +$$ + +or equivalently the matrix form + +$$ +K ( t ) = X ^ { \top } X \circ \frac { 1 } { h } [ \phi ^ { \prime } ( X ^ { \top } W ( t ) ) \phi ^ { \prime } ( W ( t ) ^ { \top } X ) ] . +$$ + +At initialization, $\mathbf { { x } } _ { i } ~ \sim ~ N ( 0 , I _ { d } )$ and ${ \pmb w } _ { k } ( 0 ) ~ \sim ~ N ( 0 , d ^ { \epsilon } I _ { d } )$ . Thus for fixed $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { i } }$ , by Gaussian anti-concentration we have $\operatorname* { P r } | \pmb { x } _ { i } ^ { \top } \pmb { w } _ { k } | < \log d \ \leq \ O ( 1 / d ^ { 1 / 2 + \epsilon _ { 1 } } )$ for some $\epsilon _ { 1 } ~ > ~ 0$ . In addition, note that $\lVert \pmb { w } _ { k } ( t ) - \pmb { w } _ { k } ( 0 ) \rVert _ { 2 } ~ = ~ O ( d ^ { - 1 / 2 } )$ for all $k$ , and therefore for $i , j , k$ such that + +$| { \pmb x } _ { i } ^ { \top } { \pmb w } _ { k } ( 0 ) | > O ( \log d )$ and $| x _ { j } ^ { \top } w _ { k } ( 0 ) | > O ( \log d )$ , we know that $| \phi ^ { \prime } ( \pmb { x } _ { i } ^ { \top } \pmb { w } _ { k } ( t ) ) \phi ^ { \prime } ( \pmb { x } _ { j } ^ { \top } \pmb { w } _ { k } ( t ) ) -$ $\phi ^ { \prime } ( { \pmb x } _ { i } ^ { \top } { \pmb w } _ { k } ( 0 ) ) \phi ^ { \prime } ( { \pmb x } _ { j } ^ { \top } { \pmb w } _ { k } ( 0 ) ) | = = O ( d ^ { - 2 } )$ . + +Given fixed $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { i } }$ , define $y _ { k } = \mathbf { 1 } \{ | x _ { i } ^ { \top } w _ { k } | < \log d \}$ as the indicator variable that the $k$ -th neuron does not saturate. We know that $\mathbb { E } [ y _ { k } ] = O ( 1 / d ^ { 1 / 2 + \epsilon _ { 1 } } )$ , and $\mathrm { V a r } [ y _ { k } ] = \mathbb { E } [ y _ { k } ^ { 2 } ] - \mathbb { E } [ y _ { k } ] ^ { 2 } = O ( 1 / d ^ { 1 / 2 + \epsilon _ { 1 } } )$ . By Bernstein’s inequality + +$$ +\operatorname* { P r } \left| \frac { 1 } { h } \sum _ { k = 1 } ^ { h } y _ { k } - \mathbb { E } [ y _ { k } ] \right| > \varepsilon \leq 2 \exp \left( - \frac { h \varepsilon ^ { 2 } } { 2 \sigma ^ { 2 } + 2 \varepsilon / 3 } \right) . +$$ + +Setting ε = q $\begin{array} { r } { \varepsilon = \sqrt { \frac { c \log h } { h ^ { 1 + \epsilon _ { 2 } } } } } \end{array}$ , we know that with probability at least $1 - h ^ { - c }$ , + +$$ +{ \frac { 1 } { h } } \sum _ { k = 1 } ^ { h } y _ { k } \leq \varepsilon + \mathbb { E } [ y _ { k } ] = O \left( { \frac { \mathrm { p o l y l o g } h } { h ^ { 1 / 2 + \epsilon _ { 3 } } } } \right) . +$$ + +Therefore, given $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { i } }$ and $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { j } }$ , for large enough $c _ { 1 }$ with probability at least $1 - h ^ { - 3 }$ we have + +$$ +\left| \sum _ { k = 1 } ^ { h } \phi ^ { \prime } ( \pmb { w } _ { k } ( t ) ^ { \top } \pmb { x } _ { i } ) \phi ^ { \prime } ( \pmb { w } _ { k } ( t ) ^ { \top } \pmb { x } _ { j } ) - \sum _ { k = 1 } ^ { h } \phi ^ { \prime } ( \pmb { w } _ { k } ( 0 ) ^ { \top } \pmb { x } _ { i } ) \phi ^ { \prime } ( \pmb { w } _ { k } ( 0 ) ^ { \top } \pmb { x } _ { j } ) \right| = O ( h ^ { 1 / 2 - \epsilon _ { 4 } } ) , +$$ + +in which we utilized the boundedness of $\phi ^ { \prime }$ . Taking union bound over $d ^ { 2 }$ elements in the random feature matrix yields + +$$ +\begin{array} { r l } & { ~ \| K ( t ) - K ( 0 ) \| _ { 2 } } \\ & { = \left\| X ^ { \top } X \circ \frac { 1 } { h } \left[ \phi ^ { \prime } ( X ^ { \top } W ( t ) ) \phi ^ { \prime } ( W ( t ) ^ { \top } X ) - \phi ^ { \prime } ( X ^ { \top } W ( 0 ) ) \phi ^ { \prime } ( W ( 0 ) ^ { \top } X ) \right] \right\| } \\ & { \leq \frac { 1 } { h } \left\| X ^ { \top } X \right\| _ { 2 } \operatorname* { m a x } \left\{ \left| \phi ^ { \prime } ( X ^ { \top } W ( t ) ) \phi ^ { \prime } ( W ( t ) ^ { \top } X ) - \phi ^ { \prime } ( X ^ { \top } W ( 0 ) ) \phi ^ { \prime } ( W ( 0 ) ^ { \top } X ) \right| _ { i j } \right\} } \\ & { \leq \frac { 1 } { h } O ( d ) O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } ) = O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } ) . } \end{array} +$$ + +Using the exact same argument, one can derive that $\| { \pmb u } _ { N N } ( { \hat { \pmb x } } ) - { \pmb u } _ { N T K } ( { \hat { \pmb x } } ) \| _ { 2 } = O ( d ^ { 1 / 2 - \epsilon ^ { \prime } } )$ , the proof of which we omit. + +# E ADDITIONAL RESULTS + +E.1 RISK OF ReLU NETWORK UNDER SYMMETRIC DATA + +If the dataset is symmetric, that is + +then population risk of the gradient flow solution can be given explicitly for certain nonlinearities: + +Proposition 20. Given (A1-3)(A5), if the nonlinearity satisfies $\phi ^ { \prime } ( { \pmb x } ) + \phi ^ { \prime } ( - { \pmb x } ) = C$ for constant $C$ then as $n , d , h \infty$ + +$$ +R _ { ( \gamma _ { 1 } < 0 . 5 ) } ( \hat { f } ) \frac { 2 \gamma _ { 1 } } { 1 - 2 \gamma _ { 1 } } \sigma ^ { 2 } ; \quad R _ { ( \gamma _ { 1 } \geq 0 . 5 ) } ( \hat { f } ) = ( 1 - \frac { 1 } { 2 \gamma _ { 1 } } ) r ^ { 2 } + \frac { 1 } { 2 \gamma _ { 1 } - 1 } \sigma ^ { 2 } . +$$ + +Note that the requirement on the nonlinearity holds for ReLU and SoftPlus. This expression is again independent to $\gamma _ { 2 }$ and aligns with the experimental results in Figure 9 (we only plot the bias component for verification). In addition, the bias is upper-bounded by the null risk for all $\gamma _ { 1 }$ . We remark that the symmetry assumption does not hold for i.i.d. samples from symmetric distributions, and Figure 9 demonstrates that the additional condition alters the risk. + +![](images/21277368c4f422787876a9038b601c0d6dee6f086e04a2580c02aae5b0b456d2.jpg) +Figure 9: Bias of two-layer ReLU networks with optimized first layer under Gaussian data and linear teacher. Individual dotted lines correspond to different $\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. (a) Vanishing initialization. The bias under symmetric data is predicted by Proposition 20. (b) Non-vanishing initialization. The red and blue lines represent models optimized from i.i.d. and symmetric initialization, respectively. The bias for symmetric initialization is predicted by Theorem 8. + +Proof. Without loss of generality assume $X = [ X _ { 0 } , - X _ { 0 } ]$ . Then by (100) we have + +$$ +\frac { \partial \pmb { w } _ { + } } { \partial t } = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \left[ \Big ( y _ { i } - h _ { 0 } \phi ( \pmb { w } _ { + } ^ { \top } \pmb { x } _ { i } ) + h _ { 0 } \phi ( \pmb { w } _ { - } ^ { \top } \pmb { x } _ { i } ) \Big ) \phi ^ { \prime } ( \pmb { w } _ { + } ^ { \top } \pmb { x } _ { i } ) \pmb { x } _ { i } \right] , +$$ + +and the flow for ${ \pmb w } _ { - }$ follows from symmetry. In this case one can show that from exact zero initialization, for nonlinearity satisfying $\phi ( x ) - \phi ( - x ) = x$ , such as ReLU and SoftPlus, + +$$ +\frac { \partial ( { \pmb w } _ { + } ) } { \partial t } + \frac { \partial ( { \pmb w } _ { - } ) } { \partial t } = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \left[ \Big ( y _ { i } - h _ { 0 } \phi ( { \pmb w } _ { + } ^ { \top } { \pmb x } _ { i } ) + h _ { 0 } \phi ( { \pmb w } _ { - } ^ { \top } { \pmb x } _ { i } ) \Big ) ( \phi ^ { \prime } ( { \pmb w } _ { + } ^ { \top } { \pmb x } _ { i } ) - \phi ^ { \prime } ( { \pmb w } _ { - } ^ { \top } { \pmb x } _ { i } ) ) { \pmb x } _ { i } \right] = 0 +$$ + +And therefore the gradient flow of ${ \pmb w } _ { + }$ is + +$$ +\begin{array} { l } { \displaystyle \frac { \partial w _ { + } } { \partial t } = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { 2 n _ { 0 } } \Big [ \Big ( y _ { i } - h _ { 0 } \phi \big ( w _ { + } ^ { \top } x _ { i } \big ) + h _ { 0 } \phi \big ( - w _ { + } ^ { \top } x _ { i } \big ) \Big ) \phi ^ { \prime } \big ( w _ { + } ^ { \top } x _ { i } \big ) x _ { i } \Big ] } \\ { \displaystyle \qquad = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { n _ { 0 } } \Big [ \Big ( y _ { i } - h \phi \big ( w _ { + } ^ { \top } x _ { i } \big ) + h _ { 0 } \phi \big ( - w _ { + } ^ { \top } x _ { i } \big ) \Big ) \big ( \phi ^ { \prime } \big ( w _ { + } ^ { \top } x _ { i } \big ) + \phi ^ { \prime } \big ( - w _ { + } ^ { \top } x _ { i } \big ) \big ) x _ { i } \Big ] } \\ { \displaystyle \qquad = \frac { 1 } { 2 n _ { 0 } } \sum _ { i = 1 } ^ { n _ { 0 } } \Big [ \Big ( y _ { i } - h _ { 0 } w _ { + } ^ { \top } x _ { i } \Big ) x _ { i } \Big ] = \frac { 1 } { 2 n _ { 0 } } X _ { 0 } y _ { 0 } - \frac { 1 } { 2 n _ { 0 } } h _ { 0 } X _ { 0 } X _ { 0 } ^ { \top } w _ { + } . } \end{array} +$$ + +The flow of ${ \pmb w } _ { - }$ follows from symmetry. Solving for the stationary points (i.e. gradient becomes zero), it the clear that + +$$ +\pmb { w } _ { + } ^ { ( t = \infty ) } = - \pmb { w } _ { - } ^ { ( t = \infty ) } = \left\{ \begin{array} { l l } { \displaystyle \frac { 1 } { h _ { 0 } } ( X X ^ { \top } ) ^ { - 1 } X \pmb { y } , } & { \gamma _ { 1 } < 0 . 5 , } \\ { \displaystyle } \\ { \displaystyle \frac { 1 } { h _ { 0 } } X ( X ^ { \top } X ) ^ { - 1 } \pmb { y } , } & { \gamma _ { 1 } > 0 . 5 . } \end{array} \right. +$$ + +And hence the asymptotic risk is + +$$ +R _ { ( \gamma _ { 1 } < 0 . 5 ) } \to \frac { 2 \gamma _ { 1 } } { 1 - 2 \gamma _ { 1 } } \sigma ^ { 2 } ; \quad R _ { ( \gamma _ { 1 } \geq 0 . 5 ) } = \left( 1 - \frac { 1 } { 2 \gamma _ { 1 } } \right) r ^ { 2 } + \frac { 1 } { 2 \gamma _ { 1 } - 1 } \sigma ^ { 2 } . +$$ + +The same conclusion holds for vanishing initialization if we assume that the trajectory stays close to that of exact zero initialization. Note that although the prediction aligns well with the experimental results, the argument in Theorem 7 does not directly apply due to the undefined derivative of ReLU at the origin, and thus this result is not rigorously justified. □ + +# F EXPERIMENT SETUP + +Optimizing the Second Layer. We compute the minimum-norm solution by directly solving the pseudo-inverse. We set $n = 1 0 0 0$ and vary $\gamma _ { 1 } , \gamma _ { 2 }$ from 0.1 to 3. The linear teacher model $F ( { \pmb x } ) = { \pmb x } ^ { \top } \beta$ is fixed as $\beta = - \mathbf { 1 } _ { d } / \sqrt { d }$ . For each $( \gamma _ { 1 } , \gamma _ { 2 } )$ we average across 50 random draws of data. + +Optimizing the First Layer. For both initializations, we use gradient descent with small step size $( \eta = 0 . 1 )$ ) and train the model for minimally 25000 steps and till $\| \nabla _ { W } f ( X , W ) \| _ { F } ^ { 2 } < 1 0 ^ { - 6 }$ . We fix $n = 3 2 0$ and vary $\gamma _ { 1 } , \gamma _ { 2 }$ from 0.1 to 3 with the same linear teacher model $\beta = - \mathbf { 1 } _ { d } / \sqrt { d }$ . The risk is averaged across 20 models trained from different initializations. \ No newline at end of file diff --git a/parse/train/H1gBsgBYwH/H1gBsgBYwH_content_list.json b/parse/train/H1gBsgBYwH/H1gBsgBYwH_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..cb2ec9c271bdb4fee87d112a2a114f25bf55505d --- /dev/null +++ b/parse/train/H1gBsgBYwH/H1gBsgBYwH_content_list.json @@ -0,0 +1,7021 @@ +[ + { + "type": "text", + "text": "GENERALIZATION OF TWO-LAYER NEURAL NET-WORKS: AN ASYMPTOTIC VIEWPOINT", + "text_level": 1, + "bbox": [ + 174, + 98, + 828, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Jimmy $\\mathbf { B a } ^ { 1 , 2 }$ , Murat A. Erdogdu1,2, Taiji Suzuki3,4, Denny $\\mathbf { W _ { u } } 1 , 2 , 4$ , Tianzong Zhang2,5 University of Toronto1, Vector Institute2, University of Tokyo3, RIKEN AIP4, Tsinghua University5 {jba,erdogdu,dennywu}@cs.toronto.edu, taiji@mist.i.u-tokyo.ac.jp, ztz16@mails.tsinghua.edu.cn ", + "bbox": [ + 181, + 167, + 836, + 227 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 263, + 544, + 279 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "This paper investigates the generalization properties of two-layer neural networks in high-dimensions, i.e. when the number of samples $n$ , features $d$ , and neurons $h$ tend to infinity at the same rate. Specifically, we derive the exact population risk of the unregularized least squares regression problem with two-layer neural networks when either the first or the second layer is trained using a gradient flow under different initialization setups. When only the second layer coefficients are optimized, we recover the double descent phenomenon: a cusp in the population risk appears at $h \\approx n$ and further overparameterization decreases the risk. In contrast, when the first layer weights are optimized, we highlight how different scales of initialization lead to different inductive bias, and show that the resulting risk is independent of overparameterization. Our theoretical and experimental results suggest that previously studied model setups that provably give rise to double descent might not translate to optimizing two-layer neural networks. ", + "bbox": [ + 233, + 292, + 764, + 473 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 496, + 336, + 512 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In modern neural networks, the number of parameters can easily exceed the number of training samples, yet in many circumstances, there is little sign of overfitting even in the absence of explicit regularization (Zhang et al., 2016). This phenomenon is usually explained by the interplay between the model architecture and the optimization method. Existing works have analyzed the implicit regularization of gradient descent on simple models (Gunasekar et al., 2018; Ji and Telgarsky, 2018), and provided generalization guarantees (Arora et al., 2018; Bartlett et al., 2017; Dziugaite and Roy, 2017) that align with the empirical observations. ", + "bbox": [ + 174, + 526, + 825, + 625 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Recently, a series of works highlighted the implicit regularization of interpolators in the overparameterized regime (Belkin et al., 2018; Spigler et al., 2018; Geiger et al., 2018; Advani and Saxe, 2017). Specifically, a second decrease in the population risk is observed when the model is further overparameterized beyond the interpolation limit, i.e. when the model achieves zero training error. This phenomenon is known as double descent, and can be precisely quantified for certain linear models (Hastie et al., 2019; Mei and Montanari, 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Among the recent works, Hastie et al. (2019) and Mei and Montanari (2019) explicitly derived the population risk of linear regression and random features regression models in high dimensions using tools from random matrix theory. ", + "bbox": [ + 174, + 631, + 825, + 756 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "However, there is still a gap between the practical benefit of overparameterization and the recently proved double descent phenomenon, which is typically established under models that exhibits the following structure: the trained model solves a linear inverse problem, and the “cusp” in the risk arises from the instability of the inverse at the interpolation threshold. Moreover, given a dataset or fixed $n , d .$ , the number of parameters in the linear regression model is also fixed, i.e. the level of overparameterization cannot be altered. It is therefore unclear if the trend persists in the optimization of more complex models, for instance in two-layer neural networks where overparameterization can be controlled simply by adding more neurons. ", + "bbox": [ + 174, + 763, + 825, + 875 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In this work, we analyze the generalization properties of two-layer neural networks in the unregularized least squares regression setting and examine the presence/absence of the double descent phenomenon. We consider the proportional asymptotic limit where the number of samples $n$ , input features $d$ , and neurons $h$ tend to infinity at the same rate, under which overparameterization corresponds to increasing the limit of $h / n$ (network “width”). This regime is particularly interesting because even though $n \\to \\infty$ , the empirical risk is not equivalent to the population risk. In addition, the joint scaling of $n , d , h$ is parallel to the practical choice of model architectures, where it is common to train a larger network when the number of samples and input features are larger. Following Hastie et al. (2019), we assume unit Gaussian input and noisy linear observations, and analytically derive the population risk of the solution of gradient flow on either the first or the second layer parameters when the flow is initialized close to zero. ", + "bbox": [ + 176, + 882, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 214 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our findings can be summarized as follows (see Figure 1): ", + "bbox": [ + 174, + 222, + 557, + 236 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• When only the second layer is optimized, we derive the risk in its bias-variance decomposition and demonstrate the presence of the double descent phenomenon. • When the first layer is optimized, we compare two solutions of gradient flow from different scales of initialization, which we term as vanishing and non-vanishing initialization, and show in both cases the population risk is independent to overparameterization. • For the vanishing initialization, we show that the risk of the gradient flow solution is asymptotically close to that of a rank-1 model. For non-vanishing initialization, we show that the gradient flow solution is wellapproximated by a kernel model and derive the risk. ", + "bbox": [ + 215, + 250, + 589, + 444 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/fdb2f7294504073f6249cfffccb6993d35f35a22379caec2e8d08b45a4316c5a.jpg", + "image_caption": [ + "Figure 1: Illustration of the double descent risk curve in two-layer linear networks $( \\mathrm { S N R } = 1 6 $ ). Brighter color indicates larger $\\gamma _ { 1 } = d / \\bar { n }$ . Double descent is observed when the second layer coefficients are optimized (main figure), but not when the first layer weights are optimized (subfigure). " + ], + "image_footnote": [], + "bbox": [ + 602, + 223, + 820, + 349 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1.1 RELATED WORKS ", + "text_level": 1, + "bbox": [ + 176, + 463, + 338, + 477 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Global Convergence of Two-layer Networks. A plethora of recent works have explored the global convergence of shallow neural networks. Mei et al. (2018; 2019); Chizat and Bach (2018a); Rotskoff and Vanden-Eijnden (2018); Sirignano and Spiliopoulos (2018); Nitanda and Suzuki (2017) studied the mean-field limit where the number of neurons $h \\to \\infty$ and the second layer scaled by $1 / h$ , and established correspondence between the main-particle limit of gradient descent and Wasserstein gradient flow to demonsrate global convergence. On the other hand, Jacot et al. (2018); Du et al. (2018); Oymak and Soltanolkotabi (2019); Allen-Zhu et al. (2018b); Song and Yang (2019) considered a different scaling and showed that gradient descent on overparameterized models converges to global minimizer at a linear rate; key to these results is an observation that optimization via gradient descent is asymptotically equivalent to kernel regression with respect to the neural tangent kernel. ", + "bbox": [ + 173, + 489, + 825, + 628 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Active vs. Lazy Training. Following Chizat and Bach (2018b), we refer to the two aforementioned scalings as the active and lazy (kernel) regime. It has been observed that different regimes lead to contrasting inductive biases. Williams et al. (2019); Woodworth et al. (2019); Li et al. (2017) showed that for certain two-layer network or overparameterized linear model, the scale of initialization controls the implicit regularization of gradient descent (from sparse to smooth solution). In the student-teacher setup (Tian, 2017; Zhong et al., 2017), Ghorbani et al. (2019b;a) showed that kernel models in high dimensions perform no better than low-degree polynomials on the input or fullytrained two-layer network. Additionally, Suzuki (2018); Allen-Zhu and Li (2019); Yehudai and Shamir (2019); Wei et al. (2018) demonstrated that neural network outperforms linear estimators (including kernel method) in learning various target functions. The difference between fixed bases and adaptive bases mirrors the difference in optimizing the first or second layer in our setup. ", + "bbox": [ + 174, + 638, + 825, + 792 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Generalization of Overparameterized Models. It is often observed that overparameterization does not result in overfitting (Neyshabur et al., 2014). In the lazy regime, generalization guarantees can be derived from the distance traveled by the parameters (Neyshabur et al., 2018; Nagarajan and Kolter, 2019), which becomes small if the model is sufficiently overparameterized (Arora et al., 2019b; Li and Liang, 2018; Allen-Zhu et al., 2018a; Cao and Gu, 2019). Compared to these guarantees that usually require significant overparameterization, our result relies on stronger data assumptions, but consequently we obtain the exact population risk instead of a vacuous upper-bound. Beyond the kernel regime, Advani and Saxe (2017); Goldt et al. (2019) analyzed the generalization dynamics of overparameterized models in the student-teacher setup. ", + "bbox": [ + 174, + 803, + 825, + 928 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Double Descent. The term double descent refers to the phenomenon that the population risk of an empirical risk minimizer manifests a \"cusp\" at the interpolation threshold, and further overparameterization decreases the risk. First observed in Krogh and Hertz (1992), the phenomenon has been recently connected to the benefit of overparameterization (Belkin et al., 2018; Geiger et al., 2018; Spigler et al., 2018; Advani and Saxe, 2017), and can be precisely characterized for certain simple models (Hastie et al., 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Our work is inspired by Hastie et al. (2019) which uses random matrix theory to derive the asymptotic risk for linear and random feature models. Concurrent to our work, Mei and Montanari (2019) analyzed the random features model and derived its population risk for which double descent occurs both in bias and variance. This aligns with our results on optimizing the second layer in Section 4 although we do not derive the bias component explicitly. Compared to Hastie et al. (2019); Mei and Montanari (2019), the focus of this work is to highlight the different generalization property of models obtained from optimizing different layers of the network and from different initialization. ", + "bbox": [ + 174, + 103, + 825, + 284 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Random Matrix Theory. High-dimensional models, including kernel models and neural networks, can be analyzed by studying the properties of random matrices. El Karoui et al. (2010); Cheng and Singer (2013); Fan and Montanari (2019) studied the spectral properties of kernel matrix via decomposing the nonlinearity with Taylor series or Hermite polynomials, which in turn explains the generalization of high-dimensional kernel ridgeless interpolators (Liang and Rakhlin, 2018). In addition, similar tools have been used to study two-layer neural networks (Louart et al., 2018; Pennington and Worah, 2017) and related quantities such as the Fisher information matrix (Karakida et al., 2018; Pennington and Worah, 2018). ", + "bbox": [ + 173, + 291, + 826, + 404 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 PRELIMINARIES: TWO-LAYER NEURAL NETWORK ", + "text_level": 1, + "bbox": [ + 174, + 422, + 625, + 440 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Consider the following bias-free two-layer neural network $f : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }$ with $h$ hidden units ", + "bbox": [ + 173, + 453, + 772, + 469 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/8b955db9905f8d4e4578b7d8f6087b36cbba2fd74a040c0a3d27e182c997902e.jpg", + "text": "$$\nf ( \\pmb { x } ) = \\sum _ { i = 1 } ^ { h } a _ { i } \\phi ( \\langle \\pmb { x } , \\pmb { w } _ { i } \\rangle ) ,\n$$", + "text_format": "latex", + "bbox": [ + 411, + 473, + 584, + 516 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\pmb { x } \\in \\mathbb { R } ^ { d }$ is the input, $\\boldsymbol { w } _ { i } \\in \\mathbb { R } ^ { d }$ is the weights corresponding to neuron $i$ , $a _ { i } \\in \\mathbb { R }$ is the $i$ -th coefficient of the second layer, and $\\phi : \\mathbb { R } \\mathbb { R }$ is a Lipschitz continuous activation function with bounded Gaussian moments, i.e. $\\mathbb { E } [ \\phi ( G ) ^ { k } ] < \\infty$ , $\\forall k \\in \\mathbb { Z } _ { + }$ for $G \\sim \\mathcal { N } ( 0 , 1 )$ . For concise notation, we write $W = [ { \\pmb w } _ { 1 } , . . . { \\pmb w } _ { h } ] \\in \\mathbb { R } ^ { d \\times h }$ for the weight matrix, $\\pmb { a } = [ a _ { 1 } , . . . a _ { h } ] \\in \\mathbb { R } ^ { h }$ for the coefficient vector, $X = [ { \\pmb x } _ { 1 } , . . . { \\pmb x } _ { n } ] \\in \\mathbb { R } ^ { d \\times n }$ for the data matrix, $\\ b { y } \\in \\mathbb { R } ^ { n }$ for the corresponding vector of labels, and $\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }$ for the feature matrix at the first layer. We omit arguments of $f$ when they are clear from the context. ", + "bbox": [ + 173, + 520, + 826, + 619 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We consider a student-teacher setup, in which data is generated by a teacher model $F : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }$ with additive noise, and the student model aims to minimize the squared loss: ", + "bbox": [ + 173, + 626, + 823, + 655 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/7391a72a42304ee6c7c207df5a4bbffed4639cbd4c44a5b4510acfb46bad444c.jpg", + "text": "$$\n( { \\pmb x } _ { i } , \\varepsilon _ { i } ) \\overset { \\mathrm { i . i . d . } } { \\sim } P _ { { \\pmb x } } \\times P _ { \\varepsilon } , \\quad y _ { i } = F ( { \\pmb x } _ { i } ) + \\varepsilon _ { i } , \\quad L ( { \\boldsymbol X } ; f ) = \\frac { 1 } { 2 n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - f ( { \\pmb x } _ { i } ) \\right) ^ { 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 233, + 659, + 763, + 700 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\mathbb { E } [ { \\pmb x } _ { i } ] = 0$ , $\\mathrm { C o v } ( { \\pmb x } _ { i } ) = \\Sigma$ , $\\mathbb { E } [ \\varepsilon _ { i } ] = 0$ , $\\mathrm { V a r } ( \\varepsilon _ { i } ) = \\sigma ^ { 2 }$ . We are interested in the population risk $R ( f ) = \\mathbb { E } _ { P _ { x } } [ ( F ( { \\pmb x } ) - f ( { \\pmb x } ) ) ^ { 2 } ]$ . Our analysis will be made under the proportional asymptotics: ", + "bbox": [ + 171, + 703, + 823, + 733 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/c5b31a0e06ef7e1df6ee2341304694fefeca77f8d51d43542792fa49fc18d39c.jpg", + "text": "$$\nn , d , h \\infty ; \\quad d / n \\gamma _ { 1 } , h / n \\gamma _ { 2 } ; \\quad \\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty ) ,\n$$", + "text_format": "latex", + "bbox": [ + 303, + 737, + 691, + 755 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "in which overparameterization corresponds to increasing $\\gamma _ { 2 }$ . Thus the characteristics of double descent considered in this work are: 1) large population risk as $\\gamma _ { 2 } 1 ; 2$ ) decrease in the risk for $\\gamma _ { 2 } > 1$ . While the empirical risk can be minimized in various ways, we analyze the solution of gradient flow, in which we update either the first layer $W$ or the second layer $\\textbf { \\em a }$ : ", + "bbox": [ + 173, + 758, + 825, + 814 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/f595ce94eebb2a30d8aaf4b9ab2bd9b5d7785c15231383757712d4b23e1b2765.jpg", + "text": "$$\n\\mathrm { d } W ( t ) = - \\nabla _ { W } L ( X ; f ) \\mathrm { d } t \\quad \\mathrm { o r } \\quad \\mathrm { d } a ( t ) = - \\nabla _ { a } L ( X ; f ) \\mathrm { d } t ,\n$$", + "text_format": "latex", + "bbox": [ + 290, + 819, + 705, + 835 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "from small initialization. The rest of the paper is organized as follows. In Section 3, we start with a simple example of two-layer linear network as warm-up. In Section 4, we consider optimizing the second layer coefficients (flow over $\\textbf { \\em a }$ ) of a non-linear two-layer neural network under fixed Gaussian first layer, which is a random feature model. Section 5 considers optimizing the first layer weights (flow over $W$ ) of such network under fixed Rademacher second layer. We defer all proofs and details on experiments to appendix. ", + "bbox": [ + 173, + 839, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 WARM-UP: LINEAR NETWORK ", + "text_level": 1, + "bbox": [ + 174, + 102, + 462, + 118 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We begin with a simple linear model with $\\phi ( { \\pmb x } ) = { \\pmb x }$ , i.e. $\\Phi = W ^ { \\top } X$ . We remark that although the model is linear, the solution obtained by gradient flow on the two-layer model can be different than that from directly solving the linear regression problem on input features. ", + "bbox": [ + 173, + 131, + 825, + 175 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Training the Second Layer. Following Hastie et al. (2019), we fix the first layer parameters to be randomly drawn from a unit Gaussian and optimize the coefficients $\\textbf { \\em a }$ by minimizing $\\left| \\left| \\pmb { a } ^ { \\top } \\Phi - \\pmb { y } \\right| \\right| _ { 2 } ^ { 2 }$ . The following lemma characterizes the solution of the gradient flow. ", + "bbox": [ + 174, + 188, + 826, + 234 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Lemma 1 (Least squares solution). Given data matrix $X _ { i }$ , response vector $\\textbf { { y } }$ and model $f ( { \\pmb x } ) =$ $\\langle \\phi ( \\pmb { x } ^ { \\top } W ) , \\hat { \\pmb { a } } \\rangle$ with fixed first layer coefficients $W$ , gradient flow on the coefficients $\\textbf { \\em a }$ starting from zero initialization converges to $\\mathbf { \\bar { a } } = \\Phi ^ { \\dagger } \\mathbf { y }$ , where $\\dagger$ stands for the Moore-Penrose inverse. ", + "bbox": [ + 174, + 237, + 825, + 280 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We make two assumptions on the data and the teacher model to simplify the computation. ", + "bbox": [ + 183, + 289, + 761, + 304 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "(A1) Gaussian Features: $\\pmb { x } _ { i } \\sim \\mathcal { N } ( 0 , I _ { d } )$ ; (A2) Linear Teacher: $F ( { \\pmb x } ) = \\langle { \\pmb x } , { \\pmb \\beta } \\rangle$ , $\\| { \\boldsymbol { \\beta } } \\| = r$ ", + "bbox": [ + 194, + 305, + 792, + 321 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Denote the linear student network as $f ( \\pmb { x } ) = \\langle \\pmb { x } , \\hat { \\beta } \\rangle$ , where $\\hat { \\boldsymbol { \\beta } } = W \\hat { \\mathbf { a } }$ and $\\hat { \\textbf { \\textit a } }$ is the least-square solution defined by Lemma 1. We write the population risk in its bias-variance decomposition. ", + "bbox": [ + 176, + 325, + 823, + 356 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/d18643c3281713eae817cf95ea0a2aa7076ffbe532bc9cb5d02f70cfadd69742.jpg", + "text": "$$\nR = \\mathbb { E } _ { { \\mathbf { x } } \\sim P _ { \\mathbf { x } } } [ \\Vert \\hat { \\beta } - \\beta \\Vert _ { \\Sigma } ^ { 2 } \\vert { \\cal X } , { \\cal W } ] = \\underbrace { \\Vert \\mathbb { E } [ \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ] - \\beta \\Vert _ { 2 } ^ { 2 } } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathrm { t r } \\left( \\mathrm { C o v } ( \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ) \\right) } _ { { \\cal V } = \\mathrm { v a r i a n c e } } ,\n$$", + "text_format": "latex", + "bbox": [ + 243, + 357, + 753, + 401 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\left\\| \\pmb { x } \\right\\| _ { \\Sigma } ^ { 2 } = \\pmb { x } ^ { \\top } \\Sigma \\pmb { x }$ . We compute the bias and the variance separately to obtain the risk. ", + "bbox": [ + 169, + 405, + 763, + 421 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Theorem 2. Given (A1)(A2) and let ${ \\pmb w } _ { i }$ i.i.d. ∼ $\\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )$ , at $n , d , h \\infty$ we have ", + "bbox": [ + 173, + 425, + 720, + 444 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/8b38e4e0f9807763f6f0a6585220ff7376399aa2616d84e51b73efd6e2e1732c.jpg", + "text": "$$\nR _ { ( \\gamma _ { 1 } < 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < \\gamma _ { 1 } , } \\\\ { \\frac { \\gamma _ { 1 } } { g _ { 1 } } \\sigma ^ { 2 } , } & R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 2 } g _ { 1 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { g _ { 1 } + g _ { 2 } } { g _ { 1 } g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } > 1 . } \\end{array} } \\end{array} } \\end{array} \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 445, + 813, + 522 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $d / n \\to \\gamma _ { 1 } , h / n \\to \\gamma _ { 2 } , g _ { 1 } = | \\gamma _ { 1 } - 1 | ,$ , and $g _ { 2 } = | \\gamma _ { 2 } - 1 |$ . ", + "bbox": [ + 174, + 535, + 591, + 553 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We observe that when $d > n$ (i.e. $\\gamma _ { 1 } > 1$ ), we obtain the double descent risk curve, i.e., the population risk achieves its maximum at $\\gamma _ { 2 } 1$ and further overparameterization $( \\gamma _ { 2 } > 1 )$ ) reduces both the bias and the variance. Conversely when $n > d$ and $h > d$ (i.e. $\\gamma _ { 1 } < \\operatorname* { m i n } ( 1 , \\gamma _ { 2 } ) )$ , the population risk becomes constant and equals to that of the minimum-norm solution ${ \\hat { \\boldsymbol { \\beta } } } _ { \\operatorname* { m i n } } = { \\boldsymbol { X } } ^ { \\dagger } { \\boldsymbol { y } }$ on the input features. ", + "bbox": [ + 173, + 560, + 825, + 633 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Training the First Layer. When the first layer of a linear network is optimized via gradient flow and the second layer is fixed, the following holds for zero-initialization of $W$ . ", + "bbox": [ + 174, + 647, + 823, + 676 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Proposition 3. Given $W ( 0 ) = 0$ and fixed $\\pm \\ : 0$ , at any time $t > 0$ of the gradient flow on $W$ , $W ( t )$ is rank-1. Further, ${ \\hat { \\boldsymbol { \\beta } } } = { \\widehat { \\boldsymbol { W } } } \\mathbf { a }$ converges to the least squares solution of $\\boldsymbol { y } = \\boldsymbol { X } ^ { \\intercal } \\hat { \\boldsymbol { \\beta } }$ , the population risk of which is given in (Hastie et al., 2019, Thm. 1 & 3)) as ", + "bbox": [ + 173, + 678, + 825, + 723 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/f5a4682cd92d7d3b10ff18ef37be5a460f8f02025a4402ea4f6cce0978f6f081.jpg", + "text": "$$\nR _ { ( \\gamma _ { 1 } < 1 ) } \\to \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } r ^ { 2 } + \\frac { 1 } { \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 297, + 724, + 699, + 757 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this case, overparameterization by increasing $\\gamma _ { 2 }$ does not influence the population risk. In addition, since the obtained two-layer linear model is equivalent to the minimum-norm solution on the input features $\\hat { \\beta } _ { \\mathrm { m i n } }$ , optimizing the first layer always results in smaller or equal population risk compared to optimizing the second layer. ", + "bbox": [ + 173, + 773, + 825, + 834 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this simple scenario for two-layer linear networks, double descent is observed only when the second layer is optimized, which reduces the objective to least squares regression on the intermediate features. On the other hand, training the first layer from zero-initialization always yields the same solution that is independent to overparameterization. One natural question to ask is: does this phenomenon generalize to nonlinear two-layer neural networks? The following sections answer this question in the affirmative under certain conditions. ", + "bbox": [ + 173, + 839, + 825, + 924 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/277c0189331694fef2e6d8835188d44b8e63f3a5b84f2bea72662ce57820d839.jpg", + "image_caption": [ + "Figure 2: Population risk of two-layer neural networks with optimized second layer under (A1)(A2). Brighter color indicates larger $\\gamma _ { 1 }$ . (a) risk of linear network with $r ^ { 2 } / \\sigma ^ { 2 } \\overset { \\cdot } { = } 1 6$ and $\\gamma _ { 1 } < 1$ . $( \\gamma _ { 1 } > 1$ is shown in Figure 1) (b) variance of network with ReLU activation. Black line corresponds to $\\gamma _ { 1 } \\to \\infty$ predicted by Corollary 5. (c) bias of network with ReLU activation. Black line corresponds to $\\gamma _ { 1 } \\to \\infty$ for linear network, which is empirically observed as an upper-bound. Note that as $\\gamma _ { 2 } 1$ both bias and variance becomes unbounded. " + ], + "image_footnote": [], + "bbox": [ + 183, + 101, + 810, + 234 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 NONLINEAR MODEL: OPTIMIZING THE SECOND LAYER ", + "text_level": 1, + "bbox": [ + 174, + 333, + 669, + 349 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we analyze the case when the second layer $\\textbf { \\em a }$ is learned under fixed $W$ and a nonlinear activation function $\\phi$ (a random feature model). We first observe that by Lemma 1, the gradient flow finds the solution $\\hat { \\mathbf { a } } = \\Phi ^ { \\dagger } \\mathbf { y }$ . We again consider the following bias-variance decomposition. ", + "bbox": [ + 174, + 364, + 825, + 409 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/79e7a6630230c8b5d7c361b0fb4680b51a56f6463dff229f6bdafaf826b79294.jpg", + "text": "$$\n\\begin{array} { r l } & { R = \\mathbb { E } _ { \\boldsymbol { x } \\sim P _ { \\boldsymbol { x } } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } | \\boldsymbol { X } , \\boldsymbol { W } ] } \\\\ & { \\quad = \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } | \\boldsymbol { X } , \\boldsymbol { W } ] - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } ] } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } ] \\| _ { 2 } ^ { 2 } \\big | \\boldsymbol { X } , \\boldsymbol { W } ] } _ { \\boldsymbol { V } = \\mathrm { v a r i a n c e } } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 191, + 414, + 805, + 489 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We highlight that the variance term does not depend on the target function. The following result characterizes the variance of the random feature model. ", + "bbox": [ + 173, + 500, + 823, + 529 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Theorem 4. Given (A1) and ${ \\pmb w } _ { i } \\overset { \\mathrm { i . i . d . } } { \\sim } \\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )$ , when $n , d , h \\infty$ , we have ", + "bbox": [ + 173, + 531, + 700, + 551 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/f60367fc1748700e44737fc2d426db24d12634332d11a3cbb9a7290314a8d432.jpg", + "text": "$$\nV = \\{ \\begin{array} { l l } { \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } & { \\gamma _ { 2 } < 1 , } \\\\ { \\quad } & { \\gamma _ { 2 } < 1 \\leq r \\leq r , } \\\\ { \\sigma ^ { 2 } \\displaystyle \\operatorname* { l i m } _ { \\xi 0 } - [ \\gamma _ { 2 } \\frac { \\partial } { \\partial x } m _ { 1 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\partial } { \\partial x } m _ { 2 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } ] } & { \\gamma _ { 2 } > 1 . } \\end{array} \n$$", + "text_format": "latex", + "bbox": [ + 186, + 559, + 808, + 635 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "in which $m _ { 1 } ( \\xi , \\rho , \\tau )$ , $m _ { 2 } ( \\xi , \\rho , \\tau )$ is the unique solution in $\\{ | m _ { 1 } | , | m _ { 2 } | < 1 / { \\mathfrak { T } } \\xi \\}$ of ", + "bbox": [ + 173, + 640, + 722, + 656 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/b5a38e0b58ff533d613c40382db42c25a20f5c774b0213de7c35e0a32bfe7e6d.jpg", + "text": "$$\n\\begin{array} { r l } & { m _ { 1 } ^ { - 1 } = - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - c _ { 1 } m _ { 2 } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - 2 \\tau c _ { 2 } m _ { 1 } m _ { 2 } + c _ { 2 } ^ { 2 } m _ { 1 } m _ { 2 } ^ { 2 } } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\\\ & { m _ { 2 } ^ { - 1 } = - \\xi - r \\gamma _ { 2 } m _ { 1 } + \\frac { \\gamma _ { 2 } c _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 194, + 661, + 781, + 738 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where variables $\\xi , \\rho , \\tau$ satisfies $\\Im \\xi > 0$ or $\\xi < 0$ , $\\rho > \\tau > 0$ , and constants $c _ { 1 } , c _ { 2 }$ defined as, ", + "bbox": [ + 176, + 742, + 784, + 758 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/3b3fa5be4207d4f5ea8dff7f6542e69ad92e9a815e63fce4b8a7ab24e6234fba.jpg", + "text": "$$\nc _ { 1 } = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , \\quad c _ { 2 } = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 336, + 763, + 660, + 782 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "for $G \\sim \\mathcal { N } ( 0 , 1 )$ and $\\Im \\xi$ denoting the imaginary part of $\\xi$ . ", + "bbox": [ + 171, + 789, + 558, + 804 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Remark. $c _ { 1 } \\geq c _ { 2 }$ and the equality holds iff $\\phi$ is linear. ", + "bbox": [ + 173, + 808, + 532, + 823 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Corollary 5. If we let $\\gamma _ { 1 } \\to \\infty$ , the variance is equal to the lowest value of the variance of the linear model $V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\sigma ^ { 2 } \\mathrm { m i n } \\{ \\gamma _ { 2 } , 1 \\} / | 1 - \\gamma _ { 2 } |$ . ", + "bbox": [ + 173, + 827, + 823, + 858 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The proof of Theorem 4 largely follows from Hastie et al. (2019) with techniques similar to Cheng and Singer (2013), but with modifications in otder to handle unnormalized and uncentered activation functions. The above theorem holds irrespective of the underlying teacher model, and is consistent with the double descent risk curve as it suggests that for all $\\gamma _ { 1 }$ , variance of the random feature model peaks at $h = n$ then drops as $\\gamma _ { 2 }$ further increases. Note that as $\\gamma _ { 1 } \\to \\infty$ , a linear and nonlinear network would have the same asymptotic variance. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Since double descent is observed in the variance term, we do not derive the bias for all $\\gamma _ { 1 } , \\gamma _ { 2 }$ . Instead, we show that for linear teacher, the bias also becomes unbounded as $\\gamma _ { 2 } 1$ . ", + "bbox": [ + 171, + 138, + 825, + 167 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Proposition 6. Given $( A I ) ( A 2 )$ and ${ \\pmb w } _ { i }$ i.i.d. ∼ $\\mathcal { N } ( 0 , I _ { d } )$ , then $B \\to \\infty$ as $\\gamma _ { 2 } 1$ . Furthermore, $B$ is finite when $\\gamma _ { 2 } > 1$ . ", + "bbox": [ + 173, + 172, + 823, + 205 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Thus we have shown that a “cusp” in the population risk appears at $h = n$ , which aligns with the double descent phenomenon. Empirically as $\\gamma _ { 1 } \\to \\infty$ the nonlinear model also shares the same asymptotic bias with the linear model, as shown in Figure 2. We note that (Mei and Montanari, 2019, Thm. 1 & 3) analytically solved the risk of random feature model for a larger class of target functions than ours and confirmed that double descent appears in both the bias and the variance. ", + "bbox": [ + 173, + 217, + 825, + 287 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5 NONLINEAR MODEL: OPTIMIZING THE FIRST LAYER ", + "text_level": 1, + "bbox": [ + 174, + 308, + 648, + 325 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Having observed the double descent phenomenon in optimizing the second layer, in the sequel we consider a two-layer neural network with fixed second layer coefficients initialized from a Rademacher√ √ distribution $a _ { i } \\sim \\operatorname { U n i f } \\{ - 1 / \\sqrt { h } , 1 \\sqrt { h } \\}$ , and the first layer $W$ is optimized with the following update ", + "bbox": [ + 174, + 340, + 825, + 386 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/fb33ae8432099b540d82f7bb6276429643c466a862ef7cc060ce27adab37408b.jpg", + "text": "$$\n\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { \\partial L ( X ; W ) } { \\partial W } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left[ y _ { i } - \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } _ { i } ) \\right] \\pmb { x } _ { i } [ \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } W ) \\circ \\pmb { a } ] ,\n$$", + "text_format": "latex", + "bbox": [ + 254, + 392, + 743, + 434 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "which potentially has different stationary solutions with no explicit form, depending on the initialization. We denote the solution of this flow at time $t$ started from designated initialization by $W ^ { \\mathrm { i n i t } } ( t )$ , its stationary solution by $\\widehat { W }$ , and the corresponding network by $\\hat { f }$ . ", + "bbox": [ + 173, + 440, + 826, + 486 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Remark. Although we let $n \\to \\infty$ , this dynamics does not corresponds to the population gradient flow considered in Tian (2017). For instance when $\\pmb { x } \\sim \\mathcal { N } ( 0 , I / d )$ , the spectrum of the data covariance is Marcenko–Pastur, whereas the population covariance is identity. ˇ ", + "bbox": [ + 176, + 488, + 821, + 531 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "As we cannot characterize the gradient flow solution from all possible initializations, we consider two specific scales of initialization: ", + "bbox": [ + 173, + 540, + 823, + 568 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Note that neither of the two initializations correspond to the “mean-field” regime (e.g. analyzed in√ Mei et al. (2018)) due to the $1 / { \\sqrt { h } }$ scaling of the second layer. In other words, as $h$ increases, we expect the distance traveled by each parameter to decrease under both initializations. The difference, however, is the “relative” amount the parameters traveled compared to their initialized magnitude, which leads to solutions with contrasting properties. As we will see, under (A1)(A2) and vanishing initialization we have $\\lVert W ( 0 ) - \\widehat { W } \\rVert _ { F } / \\lVert W ( 0 ) \\rVert _ { F } \\gg 1$ , i.e. the contribution of initialization vanishes at the end of training, whereas for non-vanishing initialization the inequality is in the opposite direction, i.e. $\\widehat { W }$ “barely moves” and resembles the initialization $W ( 0 )$ . ", + "bbox": [ + 173, + 607, + 826, + 728 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5.1 VANISHING INITIALIZATION ", + "text_level": 1, + "bbox": [ + 176, + 746, + 411, + 760 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "As $d , h \\infty$ , the vanishing initialization becomes arbitrarily close to zero-initialization. We thus expect the gradient flow under vanishing initialization to \"resemble\" that of starting from exactly zero if the flow converges sufficiently fast and the gradient being Lipschitz. The Lipschitz condition (Lemma 18) can be established under the following assumption on the activation. ", + "bbox": [ + 173, + 771, + 825, + 828 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "(A3): $\\phi$ is smooth, Lipschitz and monotone with $\\phi ^ { \\prime } ( 0 ) \\neq 0 ; | \\phi ^ { \\prime } ( \\pm x ) - \\phi ^ { \\prime } ( \\pm \\infty ) | = O ( e ^ { - x } ) .$ ", + "bbox": [ + 186, + 835, + 805, + 853 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The above assumption requires that the derivative of the nonlinearity $\\phi$ saturates beyond a O(1) region, which holds true for the commonly-used smooth activations such as sigmoid and SoftPlus. In addition, the choice of scaling ensures that the gradient flow converges sufficiently fast. We thus have the following characterization of the population risk: ", + "bbox": [ + 174, + 867, + 826, + 924 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Theorem 7. Given (A1-3). Let $T = O ( \\log \\log h )$ and $\\hat { f } ( \\cdot ) = f ^ { \\nu a n } ( \\cdot , W ( T ) )$ , then as $n , d , h \\infty$ , the gradient flow reaches a $o ( 1 )$ first-order stationary point at time $T ,$ i.e. $\\| \\partial W ( T ) / \\partial t \\| _ { F } \\in o ( 1 ) ,$ , at which point the population risk is given as ", + "bbox": [ + 173, + 102, + 826, + 146 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/6149c6ff40f51527e17ac9a862dcf6f79d63aacf6bbd1f2ed3d0b379f0f723a6.jpg", + "text": "$$\nR ( \\hat { f } ) \\operatorname* { m a x } \\{ 0 , \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } \\} r ^ { 2 } + \\frac { \\operatorname* { m i n } \\{ \\gamma _ { 1 } , 1 \\} } { | 1 - \\gamma _ { 1 } | } \\sigma ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 338, + 146, + 660, + 181 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "The expression above is the same as the risk of the least squares solution on input ${ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }$ ; therefore the risk is independent to overparameterization (increasing $\\gamma _ { 2 }$ ). The intuition is that when the weights are initialized sufficiently small and travel infinitesimally, then the activation can be linearized around 0 and thus the model is equivalent to a two-layer linear network. Note that this result does not apply to the non-smooth ReLU activation. Instead, in Appendix $\\mathrm { E }$ we heuristically show that under the additional assumption that the data is symmetric, the risk of ReLU network is also independent to $\\gamma _ { 2 }$ ", + "bbox": [ + 173, + 196, + 825, + 281 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.2 NON-VANISHING INITIALIZATION ", + "text_level": 1, + "bbox": [ + 176, + 296, + 447, + 311 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "When initialization is sufficiently large, the amount each parameter travels to minimize the empirical risk becomes asymptotically negligible compared to the magnitude of initialization. In this case we establish under (A1-3) that (11) is asymptotically equivalent to the kernel gradient flow on the tangent kernel: $k ( { \\pmb x } , { \\pmb y } ) = \\langle \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb x } ) , \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb y } ) \\rangle$ . The converged parameters under this linearized dynamics has the following closed-form: ", + "bbox": [ + 173, + 323, + 825, + 392 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/1c4303b859f171566dc09e6e10d5245a8e9ec21b71d979b638716501829868b7.jpg", + "text": "$$\n\\mathrm { v e c } ( W ^ { * } ) \\approx \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) + \\Delta ; \\quad \\Delta = J ^ { \\dagger } ( { \\pmb y } - f ^ { \\mathrm { i n i t } } ( { \\pmb X } ) ) ; \\quad J _ { [ i , j ] } = \\nabla _ { \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) _ { j } } f ^ { \\mathrm { i n i t } } ( { \\pmb x } _ { i } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 197, + 393, + 769, + 414 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $J \\in \\mathbb { R } ^ { n \\times ( d \\times h ) }$ is the Jacobian matrix w.r.t. to the model parameters. We remark that in contrast to most NTK-type global convergence results that require the width of the model to grow faster than the number of data points (e.g. Du et al. (2018)), our result is not built upon the overparameterization on width, but instead an anti-concentration that relies on the scale of initialization. Consequently the above initialization is larger than the scale that is commonly used in practice. ", + "bbox": [ + 173, + 417, + 825, + 488 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "One may naturally expect the double descent phenomenon to appear in this kernel solution, as $\\Delta$ exhibits the form of a least squares solution which contains a pseudo-inverse. However, we show that this is not the case under the same assumptions in Section 4; in fact, the risk is also independent to $\\gamma _ { 2 }$ ", + "bbox": [ + 173, + 494, + 825, + 536 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "An obstacle in computing the risk of the kernel model is the potentially non-zero $f ^ { \\mathrm { i n i t } } ( X )$ . We thus adopt the \"doubling-trick\" from Chizat and Bach (2018b) to ensure $f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0$ , i.e. we assume the following symmetric property on the initialized weights: ", + "bbox": [ + 174, + 542, + 825, + 585 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "(A4) Symmetric Initialization: $\\forall i \\in [ 1 , h ]$ , ∃! $j \\in [ 1 , h ]$ s.t. $a _ { i } \\mathbf { w } _ { i } ^ { \\mathrm { i n i t } } = - a _ { j } \\mathbf { w } _ { j } ^ { \\mathrm { i n i t } }$ ", + "bbox": [ + 228, + 588, + 763, + 606 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Theorem 8. Given (A1-4) and let $n , d , h \\infty$ , the stationary solution $\\hat { f }$ has the following risk ", + "bbox": [ + 169, + 609, + 797, + 625 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/524b0bbaa826651e5a48f9d5ad59b478de80ebfc70999f8e0225196119624bbe.jpg", + "text": "$$\n\\begin{array} { c } { { R ( \\hat { f } ) ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) r ^ { 2 } } } \\\\ { { + ( \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } - \\frac { 1 } { 4 } ) \\sigma ^ { 2 } , } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 266, + 627, + 732, + 712 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $m = b _ { 1 } ^ { 2 } / b _ { 0 } ^ { 2 }$ , $b _ { 0 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ] ^ { 2 }$ , and $b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }$ , $G \\sim \\mathcal { N } ( 0 , 1 )$ . ", + "bbox": [ + 176, + 712, + 674, + 729 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Note that the population risk is again independent to $\\gamma _ { 2 }$ , and thus double descent does not appear for this initialization. Roughly speaking, the reason that the risk does not become unbounded at some point is that in the asymptotic limit the pseudo-inverse $( J J ^ { \\top } ) ^ { \\dagger }$ is stable due to the nonlinearity and $d h \\gg n$ . We make two additional observations: ", + "bbox": [ + 174, + 736, + 825, + 792 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "• the stability of the inverse at $\\gamma _ { 1 } 1$ depends on the lowest eigenvalue of the tangent kernel matrix (smaller $\\lambda _ { \\operatorname* { m i n } } ( K )$ entails larger variance), which is determined by the nonlinearity; While our result only holds for network with zero initial output, for non-symmetric (i.i.d.) initialization we also observe that the risk is independent to $\\gamma _ { 2 }$ , but the bias is higher than that in symmetric initialization as shown in Figure 8. We comment that the non-zero $f ^ { \\mathrm { i n i t } } ( X )$ in (13) behaves as zero-mean Gaussian due to central limit theorem; therefore, in the kernel regime the function output at initialization is equivalent to additive noise to the labels $\\textbf { { y } }$ , and we thus expect the magnitude of $f ^ { \\mathrm { i n i t } } ( X )$ to negatively influence the model generalization. ", + "bbox": [ + 215, + 801, + 825, + 919 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/389923c17d029c0a38688561f57f18607a9ed4055a5409c2180d94b356e6ec61.jpg", + "image_caption": [ + "Figure 3: Bias and variance of two-layer sigmoid network with optimized first layer under (A1)(A2). Individual dotted lines correspond to different $\\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. The bias and variance for both initializations is well-aligned with Theorem 7 and Theorem 8. " + ], + "image_footnote": [], + "bbox": [ + 184, + 102, + 807, + 289 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.3 COMPARING THE INITIALIZATIONS ", + "text_level": 1, + "bbox": [ + 176, + 363, + 457, + 377 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Figure 3 shows the agreement between theoretical prediction and experimental results. Although in both cases the risk is independent to overparameterization $( \\gamma _ { 2 } )$ , the two initializations lead to models with contrasting properties, as demonstrated by the following comparison on the risk. ", + "bbox": [ + 174, + 388, + 823, + 433 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Corollary 9. For any $\\gamma _ { 1 } \\in ( 0 , \\infty )$ and nonlinearity $\\phi$ , $B ( \\hat { f } ^ { V a n } ) \\le B ( \\hat { f } ^ { N V } ) \\le 1 .$ . On the other hand, for all $m > 0$ , $V ( \\hat { f } ^ { N V } ) = { \\cal { O } } ( 1 )$ , whereas $V ( \\hat { f } ^ { V a n } )$ can be arbitrarily large as $\\gamma _ { 1 } 1$ . ", + "bbox": [ + 174, + 435, + 823, + 469 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Remark. $m \\geq 0$ for all smooth activations $\\phi$ , and the equality holds if $\\phi$ is linear. ", + "bbox": [ + 174, + 472, + 707, + 488 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Intuitively, small initialization enables the model \"evolve\" more during optimization and better align with the data and target function. This potentially results in a lower bias, at the expense of overfitting more to the noise (high variance). In contrast, with sufficiently large initialization the final model becomes close to the initialized model, and thus we may expect it to be less “aligned” to the target (high bias) but is more stable (lower variance). ", + "bbox": [ + 174, + 496, + 825, + 565 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In illustrate the different inductive bias of the two initializations, we plot the trajectory of neurons in Appendix A Figure 4. Observe that for vanishing initialization the neurons stay close to one another throughout the trajectory, which results in a low-rank weight matrix, as predicted by Theorem 7. In contrast, for non-vanishing initialization the neurons stay close to initialization (therefore full-rank), which validates the kernel approximation. Last but not least, although the derived risk is only for learning a linear target function, we empirically observe that when the teacher is also a two-layer network, the population risk follows the same trend, i.e. double descent occurs when only the second layer is optimized, as shown in Figure 7. ", + "bbox": [ + 174, + 571, + 825, + 684 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "6 DISCUSSION AND FUTURE WORKS ", + "text_level": 1, + "bbox": [ + 176, + 704, + 495, + 719 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We derived the exact population risk of high-dimensional two-layer neural networks in learning a linear target function over Gaussian data with additive label noise, and showed that optimizing the first or the second layer via gradient flow results in solutions with contrasting properties. Specifically, double descent is present when the second layer coefficients are optimized, but not when the first layer weights are optimized under certain initializations. Moreover, we highlight that the scale of initialization leads to different inductive bias in optimizing the first layer. ", + "bbox": [ + 174, + 734, + 825, + 819 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "It should be noted that our analysis only applies to the unregularized objective: it has been shown that explicit regularization (such as $\\ell _ { 2 }$ penalty) stabilizes the singularity at $\\gamma _ { 2 } ~ ~ 1$ (Mei and Montanari, 2019), and algorithmic regularization (Li et al., 2019a; Dong et al., 2019) also provides robustness against noisy observations. We further remark that our findings do not directly contradict the experimental double descent phenomenon, nor the practical benefit of overparameterization. In particular, the interpolation limit could occur at $\\gamma _ { 2 } 0$ which is beyond the regime we consider (such as Figure 4 in (Belkin et al., 2018)). Thus what we conclude is that under the studied proportional asymptotics, the mechanism that provably gives rise to double descent from previous works on least squares regression might not translate to neural networks trained with gradient descent. ", + "bbox": [ + 174, + 827, + 823, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 103, + 823, + 132 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To simplify the computation, we rely on a set of strong assumptions similar to those in Hastie et al. (2019), some of which we believe can be relaxed in future works, such as isotropic Gaussian input and linear target. Importantly, the two specific scales of initialization studied in Section 5 are by no means exhaustive, and thus one would expect that under a different initialization of the first layer, or the mean-field $1 / h$ scaling of the second layer, the risk of the trained model can be very different. Changing the loss function may also alter the generalization behavior of the network. Another challenging problem is to extend the current analysis to beyond two layers. Last but not least, our result characterizes gradient flow which resembles gradient descent with small stepsize, and thus it would be interesting to study the effect of learning rate schedule, which has a known impact on generalization (Smith and Le, 2017; Li et al., 2019b). ", + "bbox": [ + 174, + 140, + 825, + 277 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "ACKNOWLEDGEMENT ", + "text_level": 1, + "bbox": [ + 176, + 296, + 326, + 309 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We thank Xiuyuan Cheng, Xuechen Li, Yiping Lu, Atsushi Nitanda, Shengyang Sun and anonymous reviewers for helpful comments and feedback. JB and DW were partially funded by LG Electronics and NSERC. JB and MAE were supported by the CIFAR AI Chairs program. TS was partially supported by JSPS Kakenhi (26280009, 15H05707 and 18H03201), Japan Digital Design and JST-CREST. ", + "bbox": [ + 174, + 321, + 825, + 391 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 176, + 429, + 285, + 445 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Madhu S Advani and Andrew M Saxe. High-dimensional dynamics of generalization error in neural networks. arXiv preprint arXiv:1710.03667, 2017. \nZeyuan Allen-Zhu and Yuanzhi Li. What can resnet learn efficiently, going beyond kernels? arXiv preprint arXiv:1905.10337, 2019. \nZeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized neural networks, going beyond two layers. arXiv preprint arXiv:1811.04918, 2018a. \nZeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via overparameterization. arXiv preprint arXiv:1811.03962, 2018b. \nSanjeev Arora, Rong Ge, Behnam Neyshabur, and Yi Zhang. Stronger generalization bounds for deep nets via a compression approach. arXiv preprint arXiv:1802.05296, 2018. \nSanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Ruslan Salakhutdinov, and Ruosong Wang. On exact computation with an infinitely wide neural net. arXiv preprint arXiv:1904.11955, 2019a. \nSanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. arXiv preprint arXiv:1901.08584, 2019b. \nZhidong Bai and Jack W Silverstein. Spectral analysis of large dimensional random matrices, volume 20. Springer, 2010. \nPeter L Bartlett, Dylan J Foster, and Matus J Telgarsky. Spectrally-normalized margin bounds for neural networks. In Advances in Neural Information Processing Systems, pages 6240–6249, 2017. \nPeter L Bartlett, Philip M Long, Gábor Lugosi, and Alexander Tsigler. Benign overfitting in linear regression. arXiv preprint arXiv:1906.11300, 2019. \nMikhail Belkin, Daniel Hsu, Siyuan Ma, and Soumik Mandal. Reconciling modern machine learning and the bias-variance trade-off. arXiv preprint arXiv:1812.11118, 2018. \nMikhail Belkin, Daniel Hsu, and Ji Xu. Two models of double descent for weak features. arXiv preprint arXiv:1903.07571, 2019. ", + "bbox": [ + 171, + 448, + 826, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning overparameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019. ", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Xiuyuan Cheng and Amit Singer. The spectrum of random inner-product kernel matrices. Random Matrices: Theory and Applications, 2(04):1350010, 2013. ", + "bbox": [ + 171, + 142, + 823, + 171 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Lenaic Chizat and Francis Bach. On the global convergence of gradient descent for over-parameterized models using optimal transport. In Advances in neural information processing systems, pages 3036–3046, 2018a. ", + "bbox": [ + 174, + 180, + 823, + 223 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Lenaic Chizat and Francis Bach. A note on lazy training in supervised differentiable programming. arXiv preprint arXiv:1812.07956, 2018b. ", + "bbox": [ + 173, + 232, + 823, + 262 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Bin Dong, Jikai Hou, Yiping Lu, and Zhihua Zhang. Distillation $\\approx$ early stopping? harvesting dark knowledge utilizing anisotropic information retrieval for overparameterized neural network. arXiv preprint arXiv:1910.01255, 2019. ", + "bbox": [ + 173, + 271, + 826, + 314 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Simon S Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes over-parameterized neural networks. arXiv preprint arXiv:1810.02054, 2018. ", + "bbox": [ + 173, + 324, + 823, + 353 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Gintare Karolina Dziugaite and Daniel M Roy. Computing nonvacuous generalization bounds for deep (stochastic) neural networks with many more parameters than training data. arXiv preprint arXiv:1703.11008, 2017. ", + "bbox": [ + 173, + 362, + 823, + 405 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Noureddine El Karoui et al. The spectrum of kernel random matrices. The Annals of Statistics, 38(1): 1–50, 2010. ", + "bbox": [ + 171, + 415, + 823, + 444 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Zhou Fan and Andrea Montanari. The spectral norm of random inner-product kernel matrices. Probability Theory and Related Fields, 173(1-2):27–85, 2019. ", + "bbox": [ + 169, + 454, + 825, + 483 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mario Geiger, Stefano Spigler, Stéphane d’Ascoli, Levent Sagun, Marco Baity-Jesi, Giulio Biroli, and Matthieu Wyart. The jamming transition as a paradigm to understand the loss landscape of deep neural networks. arXiv preprint arXiv:1809.09349, 2018. ", + "bbox": [ + 174, + 492, + 823, + 535 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Limitations of lazy training of two-layers neural networks. arXiv preprint arXiv:1906.08899, 2019a. ", + "bbox": [ + 173, + 544, + 821, + 574 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Linearized two-layers neural networks in high dimension. arXiv preprint arXiv:1904.12191, 2019b. ", + "bbox": [ + 173, + 583, + 821, + 613 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Sebastian Goldt, Madhu S Advani, Andrew M Saxe, Florent Krzakala, and Lenka Zdeborová. Generalisation dynamics of online learning in over-parameterised neural networks. arXiv preprint arXiv:1901.09085, 2019. ", + "bbox": [ + 174, + 621, + 826, + 665 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Suriya Gunasekar, Jason D Lee, Daniel Soudry, and Nati Srebro. Implicit bias of gradient descent on linear convolutional networks. In Advances in Neural Information Processing Systems, pages 9461–9471, 2018. ", + "bbox": [ + 173, + 674, + 826, + 717 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Trevor Hastie, Andrea Montanari, Saharon Rosset, and Ryan J Tibshirani. Surprises in highdimensional ridgeless least squares interpolation. arXiv preprint arXiv:1903.08560, 2019. ", + "bbox": [ + 171, + 727, + 823, + 756 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and generalization in neural networks. In Advances in neural information processing systems, pages 8571–8580, 2018. ", + "bbox": [ + 173, + 765, + 825, + 809 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ziwei Ji and Matus Telgarsky. Gradient descent aligns the layers of deep linear networks. arXiv preprint arXiv:1810.02032, 2018. ", + "bbox": [ + 171, + 818, + 825, + 847 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ryo Karakida, Shotaro Akaho, and Shun-ichi Amari. Universal statistics of fisher information in deep neural networks: Mean field approach. arXiv preprint arXiv:1806.01316, 2018. ", + "bbox": [ + 173, + 856, + 823, + 886 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Anders Krogh and John A. Hertz. A simple weight decay can improve generalization. pages 950–957, 1992. ", + "bbox": [ + 174, + 895, + 823, + 922 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mingchen Li, Mahdi Soltanolkotabi, and Samet Oymak. Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks. arXiv preprint arXiv:1903.11680, 2019a. ", + "bbox": [ + 173, + 103, + 825, + 145 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yuanzhi Li and Yingyu Liang. Learning overparameterized neural networks via stochastic gradient descent on structured data. In Advances in Neural Information Processing Systems, pages 8157– 8166, 2018. ", + "bbox": [ + 174, + 155, + 825, + 198 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yuanzhi Li, Tengyu Ma, and Hongyang Zhang. Algorithmic regularization in over-parameterized matrix sensing and neural networks with quadratic activations. arXiv preprint arXiv:1712.09203, 2017. ", + "bbox": [ + 174, + 205, + 826, + 248 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yuanzhi Li, Colin Wei, and Tengyu Ma. Towards explaining the regularization effect of initial large learning rate in training neural networks. arXiv preprint arXiv:1907.04595, 2019b. ", + "bbox": [ + 173, + 258, + 821, + 287 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tengyuan Liang and Alexander Rakhlin. Just interpolate: Kernel\" ridgeless\" regression can generalize. arXiv preprint arXiv:1808.00387, 2018. ", + "bbox": [ + 173, + 295, + 823, + 325 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Zhenyu Liao and Romain Couillet. On the spectrum of random features maps of high dimensional data. arXiv preprint arXiv:1805.11916, 2018. ", + "bbox": [ + 173, + 334, + 823, + 363 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Cosme Louart, Zhenyu Liao, Romain Couillet, et al. A random matrix approach to neural networks. The Annals of Applied Probability, 28(2):1190–1248, 2018. ", + "bbox": [ + 174, + 372, + 825, + 401 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "V.A. Marcenko and Leonid Pastur. Distribution of eigenvalues for some sets of random matrices. ˇ Math USSR Sb, 1:457–483, 01 1967. ", + "bbox": [ + 176, + 409, + 823, + 439 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Song Mei and Andrea Montanari. The generalization error of random features regression: Precise asymptotics and double descent curve. arXiv preprint arXiv:1908.05355, 2019. ", + "bbox": [ + 173, + 446, + 823, + 477 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Song Mei, Andrea Montanari, and Phan-Minh Nguyen. A mean field view of the landscape of twolayer neural networks. Proceedings of the National Academy of Sciences, 115(33):E7665–E7671, 2018. ", + "bbox": [ + 174, + 484, + 826, + 527 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Mean-field theory of two-layers neural networks: dimension-free bounds and kernel limit. arXiv preprint arXiv:1902.06015, 2019. ", + "bbox": [ + 174, + 536, + 823, + 566 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Vaishnavh Nagarajan and J Zico Kolter. Generalization in deep networks: The role of distance from initialization. arXiv preprint arXiv:1901.01672, 2019. ", + "bbox": [ + 174, + 574, + 823, + 604 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the role of implicit regularization in deep learning. arXiv preprint arXiv:1412.6614, 2014. ", + "bbox": [ + 173, + 612, + 825, + 642 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Behnam Neyshabur, Zhiyuan Li, Srinadh Bhojanapalli, Yann LeCun, and Nathan Srebro. Towards understanding the role of over-parametrization in generalization of neural networks. arXiv preprint arXiv:1805.12076, 2018. ", + "bbox": [ + 173, + 650, + 823, + 693 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Atsushi Nitanda and Taiji Suzuki. Stochastic particle gradient descent for infinite ensembles. arXiv preprint arXiv:1712.05438, 2017. ", + "bbox": [ + 168, + 702, + 825, + 731 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global convergence guarantees for training shallow neural networks. arXiv preprint arXiv:1902.04674, 2019. ", + "bbox": [ + 174, + 739, + 825, + 782 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jeffrey Pennington and Pratik Worah. Nonlinear random matrix theory for deep learning. In Advances in Neural Information Processing Systems, pages 2637–2646, 2017. ", + "bbox": [ + 169, + 791, + 825, + 821 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jeffrey Pennington and Pratik Worah. The spectrum of the fisher information matrix of a singlehidden-layer neural network. In Advances in Neural Information Processing Systems, pages 5410–5419, 2018. ", + "bbox": [ + 176, + 829, + 823, + 872 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Grant M Rotskoff and Eric Vanden-Eijnden. Neural networks as interacting particle systems: Asymptotic convexity of the loss landscape and universal scaling of the approximation error. arXiv preprint arXiv:1805.00915, 2018. ", + "bbox": [ + 174, + 881, + 825, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Justin Sirignano and Konstantinos Spiliopoulos. Mean field analysis of neural networks: A central limit theorem. arXiv preprint arXiv:1808.09372, 2018. ", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Samuel L Smith and Quoc V Le. A bayesian perspective on generalization and stochastic gradient descent. arXiv preprint arXiv:1710.06451, 2017. ", + "bbox": [ + 171, + 141, + 823, + 170 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Zhao Song and Xin Yang. Quadratic suffices for over-parametrization via matrix chernoff bound. arXiv preprint arXiv:1906.03593, 2019. ", + "bbox": [ + 171, + 178, + 825, + 208 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Stefano Spigler, Mario Geiger, Stéphane d’Ascoli, Levent Sagun, Giulio Biroli, and Matthieu Wyart. A jamming transition from under-to over-parametrization affects loss landscape and generalization. arXiv preprint arXiv:1810.09665, 2018. ", + "bbox": [ + 176, + 215, + 825, + 258 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Taiji Suzuki. Adaptivity of deep relu network for learning in besov and mixed smooth besov spaces: optimal rate and curse of dimensionality. arXiv preprint arXiv:1810.08033, 2018. ", + "bbox": [ + 171, + 268, + 825, + 297 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012. ", + "bbox": [ + 171, + 305, + 803, + 321 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yuandong Tian. An analytical formula of population gradient for two-layered relu network and its applications in convergence and critical point analysis. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 3404–3413. JMLR. org, 2017. ", + "bbox": [ + 178, + 329, + 823, + 372 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Colin Wei, Jason D Lee, Qiang Liu, and Tengyu Ma. On the margin theory of feedforward neural networks. arXiv preprint arXiv:1810.05369, 2018. ", + "bbox": [ + 176, + 381, + 823, + 410 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Francis Williams, Matthew Trager, Claudio Silva, Daniele Panozzo, Denis Zorin, and Joan Bruna. Gradient dynamics of shallow univariate relu networks. arXiv preprint arXiv:1906.07842, 2019. ", + "bbox": [ + 174, + 419, + 823, + 448 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Blake Woodworth, Suriya Gunasekar, Jason Lee, Daniel Soudry, and Nathan Srebro. Kernel and deep regimes in overparametrized models. arXiv preprint arXiv:1906.05827, 2019. ", + "bbox": [ + 176, + 457, + 821, + 486 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ji Xu and Daniel Hsu. On the number of variables to use in principal component regression. 2019. ", + "bbox": [ + 173, + 494, + 820, + 511 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Gilad Yehudai and Ohad Shamir. On the power and limitations of random features for understanding neural networks. arXiv preprint arXiv:1904.00687, 2019. ", + "bbox": [ + 171, + 517, + 823, + 547 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016. ", + "bbox": [ + 174, + 555, + 823, + 585 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Kai Zhong, Zhao Song, Prateek Jain, Peter L Bartlett, and Inderjit S Dhillon. Recovery guarantees for one-hidden-layer neural networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 4140–4149. JMLR. org, 2017. ", + "bbox": [ + 174, + 593, + 825, + 637 + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/8be263e99069d2ff2756929d4f59b8c08885627a89e302d107583385553bcd06.jpg", + "image_caption": [ + "Figure 4: trajectory of neurons from initialization (dark blue) to optimum (orange) on the first two dimensions (two-layer SoftPlus student and linear teacher; $\\mathrm { S N R } { = } 1 / 4 \\mathrm { \\Omega }$ . For vanishing initialization the neurons stay close to one another throughout the trajectory, whereas for non-vanishing initialization the neurons stay close to initialization. " + ], + "image_footnote": [], + "bbox": [ + 189, + 140, + 807, + 313 + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/67d5f63ec5bc3ba3b56af6a95e487287d5f924b0f1e30f617bcb3ec06520d044.jpg", + "image_caption": [ + "Figure 5: Population risk of two-layer linear network with fixed random 1st layer with $S _ { \\mathrm { { N R = } } 2 5 / 1 6 }$ under Gaussian input and linear teacher. Brighter color indicates larger $\\gamma _ { 1 }$ . " + ], + "image_footnote": [], + "bbox": [ + 194, + 397, + 797, + 587 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/daf90c88b28403b003b2e689cb5162fcef4c8474104f827875680e451e9cedd4.jpg", + "table_caption": [ + "SUMMARY OF THE PRESENCE / ABSENCE OF DOUBLE DESCENT " + ], + "table_footnote": [], + "table_body": "
Singularity in2nd Layer Trained (RF)Vanishing Init.Non-vanishing Init.
Bias1: No; Y2: Yesγ1: No; γ2: Noγ1: No; 2: No
VarianceY1: No; 2: Yes1: Yes; Y2: NoY1: No; 2: No
", + "bbox": [ + 173, + 676, + 735, + 723 + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/487235692f241523abf1600424158205f71f19e5f44d0aed9c5fb66d2d8f7cb9.jpg", + "image_caption": [ + "Figure 6: Bias and variance of two-layer SoftPlus network with optimized first layer under (A1)(A2). Individual dotted lines correspond to different $\\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. The bias and variance for both initializations is well-aligned with Theorem 7 and Theorem 8, respectively. " + ], + "image_footnote": [], + "bbox": [ + 200, + 101, + 794, + 273 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/0a41e6cc5c4bd20c793bbe18e10c8b870f315c2459db916ab633ef96972f668a.jpg", + "image_caption": [ + "Figure 7: Population risk (scaled by $1 / d )$ of two-layer ReLU network trained to fit a two-layer ReLU teacher model with $h = d$ neurons. Brighter color corresponds to larger $\\gamma _ { 1 }$ . Similar to the linear teacher case, double descent is observed when the second layer is optimized (a) but not when the first layer is optimized (b). " + ], + "image_footnote": [], + "bbox": [ + 199, + 334, + 794, + 508 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/c366b578fe991ae26f542e54255e90708f233ee5480d69cce84eed842879b63d.jpg", + "image_caption": [ + "Figure 8: Bias of (a) SoftPlus and (b) sigmoid two-layer network with optimized first layer under (A1)(A2). Note that bias under i.i.d. initialization is also independent to overparameterization $\\left( \\gamma _ { 2 } \\right)$ , but is higher than the bias under symmetric initialization (“doubling trick”) and not always upper-bounded by the null risk $r ^ { 2 }$ . " + ], + "image_footnote": [], + "bbox": [ + 199, + 570, + 794, + 742 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "B BACKGROUND ", + "text_level": 1, + "bbox": [ + 176, + 815, + 330, + 830 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "B.1 ROTATIONAL INVARIANCE ", + "text_level": 1, + "bbox": [ + 176, + 851, + 400, + 866 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "The rotational invariance of Gaussian distribution is crucial in our analysis throughout this paper. A basic observation is that for a random Gaussian matrix $X$ and any fixed unitary matrix $U$ , the distribution of $X$ and $U X$ are the same. ", + "bbox": [ + 174, + 880, + 825, + 922 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Lemma 10 (Rotational Invariance). Denote $A ( X ) \\in \\mathbb { R } ^ { d \\times d }$ a matrix function of $X \\in \\mathbb { R } ^ { d \\times n }$ . If $A ( X )$ satisfies that $A ( U X ) = U A ( X ) U ^ { T }$ for all unitary $U$ , then ", + "bbox": [ + 171, + 102, + 823, + 132 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/833071b965fe2b8b3823963911062089ca34d7e99f1a2b40fc6e6d4f65e91370.jpg", + "text": "$$\n\\operatorname { \\mathbb { E } } _ { X } [ \\beta ^ { T } A ( X ) \\beta ] = { \\frac { 1 } { d } } \\beta ^ { T } \\beta \\operatorname { \\mathbb { E } } _ { X } [ \\operatorname { t r } \\left( A ( X ) \\right) ] .\n$$", + "text_format": "latex", + "bbox": [ + 357, + 135, + 638, + 165 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "for any fixed nonzero $\\beta \\in \\mathbb { R } ^ { d }$ and random matrix $X$ with each entry i.i.d. $X _ { i j } \\sim \\mathcal { N } ( 0 , \\sigma ^ { 2 } )$ ", + "bbox": [ + 168, + 167, + 767, + 184 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Proof. Choose a set of Unitary matrices $\\{ U _ { i } \\} _ { i = 1 } ^ { d }$ such that $U _ { i } ^ { \\top } \\beta = \\| \\beta \\| e _ { i }$ , where $e _ { i }$ is the $i$ -th canonical vector in $\\mathbb { R } ^ { d }$ . Since $U _ { i } X \\sim X$ , we have $\\mathbb { E } [ A ( X ) ] = \\mathbb { E } [ A ( U _ { i } X ) ] = U _ { i } \\mathbb { E } [ A ( X ) ] U _ { i } ^ { \\top }$ and hence ", + "bbox": [ + 173, + 193, + 826, + 238 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/0f0a8bdb9f31ba01e5ce82b1e4ed0b7f59808e7937c7a2f8da7c0eba8f7617dc.jpg", + "text": "$$\n\\mathbb { E } [ \\beta ^ { T } A ( X ) \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } \\mathbb { E } [ \\beta ^ { T } U _ { i } A ( X ) U _ { i } ^ { \\top } \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } e _ { i } ^ { T } \\mathbb { E } [ A ( X ) ] e _ { i } = \\frac { \\beta ^ { T } \\beta } { d } \\mathbb { E } [ \\mathrm { t r } ( A ( X ) ) ] .\n$$", + "text_format": "latex", + "bbox": [ + 186, + 238, + 785, + 282 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Note that the property also holds for matrix function $A ( X )$ that satisfies $A ( X U ) = U A ( X ) U ^ { T }$ , and can be extended to matrix function $A$ that takes multiple matrices as input. □ ", + "bbox": [ + 174, + 285, + 828, + 315 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For brevity we will refer to rotational invariance instead of equation (15). ", + "bbox": [ + 173, + 329, + 653, + 344 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "B.2 MARCENKO ˇ –PASTUR LAW ", + "text_level": 1, + "bbox": [ + 176, + 359, + 401, + 375 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For a real symmetric random matrix $A \\in \\mathbb { R } ^ { p \\times p }$ , define its empirical spectral density as ", + "bbox": [ + 174, + 385, + 740, + 401 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/bd285da91081ea0ebf82d1825ac45c0adacc1c9e692c65d7e2093dcd9bafee6d.jpg", + "text": "$$\n\\mu _ { A } ( d \\lambda ) = \\frac { 1 } { p } \\sum _ { i = 1 } ^ { p } \\delta _ { \\lambda _ { i } ( A ) } ( \\lambda ) d \\lambda ,\n$$", + "text_format": "latex", + "bbox": [ + 397, + 404, + 599, + 445 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "where $\\delta _ { a } ( x ) = \\delta ( x - a )$ is the Dirac delta function. Assume $A = W _ { p } \\sim W _ { p } ( I , n )$ is a Wishart matrix, i.e. $W _ { p } = X ^ { \\top } X / n$ and $\\ b { X } \\in \\mathbb { R } ^ { n \\times p }$ is random Gaussian matrix with each column i.i.d. $X _ { i } \\sim \\mathcal { N } ( \\mathbf { 0 } , I )$ . Marcenko and Pastur ˇ (1967) showed that as $n , p \\to \\infty$ and $p / n = \\gamma \\in ( 0 , \\infty )$ , the empirical spectral density $\\mu _ { W _ { p } } ( d \\lambda )$ converges weakly to a limiting density $\\mu _ { \\mathrm { M P } ( \\gamma ) } ( \\lambda )$ : ", + "bbox": [ + 173, + 446, + 826, + 507 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/982a95475ba530700c71603522d670cc909fbc0f9a160aa61cf0f58c7ff4736f.jpg", + "text": "$$\n\\mu _ { \\mathrm { M P } ( \\gamma ) } ( d \\lambda ) = [ 1 - \\gamma ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\frac { 1 } { 2 \\pi \\gamma \\lambda } \\sqrt { ( ( 1 + \\sqrt \\gamma ) ^ { 2 } - \\lambda ) ( \\lambda - ( 1 - \\sqrt \\gamma ) ^ { 2 } ) } d \\lambda .\n$$", + "text_format": "latex", + "bbox": [ + 207, + 510, + 764, + 542 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We say $\\mu _ { \\mathrm { M P } ( \\gamma ) }$ is the density of the Marˇcenko–Pastur distribution with support $S = [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 +$ $\\sqrt { \\gamma } ) ^ { 2 } ]$ (for $0 < \\gamma < 1$ ) or $S = \\{ 0 \\} \\cup [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 + \\sqrt { \\gamma } ) ^ { 2 } ]$ (for $\\gamma \\geq 1$ ). Note that this implies that the smallest non-zero eigenvalue of $A$ is bounded away from 0 a.s. for $\\gamma \\neq 1$ . ", + "bbox": [ + 173, + 545, + 826, + 590 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The explicit form of Marcenko–Pastur distribution allows us to investigate the asymptotic properties ˇ of random matrices. Generally speaking, by Pormanteau theorem one can translate any bounded continuous function on the empirical spectral density to the one on Marcenko–Pastur distribution, i.e. ˇ for any bounded $f ( \\lambda ) \\in C ( S )$ , as $n , p \\to \\infty$ with $p / n = \\gamma$ , almost surely ", + "bbox": [ + 173, + 597, + 826, + 654 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/b5b5c50caa17b9bae2674f02aa659aac7984332e27c225a9b1c289dc84c7a551.jpg", + "text": "$$\n\\int _ { S } f ( \\lambda ) \\cdot \\mu _ { W } ( d \\lambda ) \\to \\int f _ { S } ( \\lambda ) \\cdot \\mu _ { M P } ( d \\lambda ) .\n$$", + "text_format": "latex", + "bbox": [ + 352, + 656, + 643, + 689 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "One implication is the following trace concentration on the inverse Wishart matrix for $\\gamma < 1$ , ", + "bbox": [ + 174, + 698, + 782, + 713 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/2d0f780ecfa10baaa2208e09863b4e1e023a7dda5466a5cd80525455e3c92adb.jpg", + "text": "$$\n\\operatorname { t r } ( X ^ { \\top } X ) = \\operatorname { t r } ( { \\frac { 1 } { p } } W _ { p } ^ { - 1 } ) = { \\frac { 1 } { p } } \\sum _ { i = 1 } ^ { p } { \\frac { 1 } { \\lambda _ { i } ( W _ { p } ) } } = \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { W _ { p } } ( d \\lambda ) \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { \\operatorname { M P } ( \\gamma ) } ( d \\lambda ) = { \\frac { 1 } { 1 - \\gamma } } .\n$$", + "text_format": "latex", + "bbox": [ + 187, + 715, + 807, + 757 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We remark that instability of the trace of the invert Wishart matrix as $\\gamma 1$ plays an important role in the double descent phenomenon. ", + "bbox": [ + 173, + 773, + 826, + 803 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "B.3 ORTHOGONAL POLYNOMIALS ", + "text_level": 1, + "bbox": [ + 176, + 818, + 424, + 833 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Orthogonal polynomials are useful in the analysis of nonlinear random matrices. Suppose $\\phi$ is a function in $L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )$ , where $G \\sim \\mathcal { N } ( 0 , 1 )$ , $\\mu _ { G } ( d x ) = ( \\sqrt { 2 \\pi } ) ^ { - 1 } e ^ { - x ^ { 2 } / 2 } d x$ is the Gaussian measure. For $n \\geq 0$ , define the Hermite polynomials ", + "bbox": [ + 173, + 843, + 826, + 888 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/b7dfb5d7f01e6357a4b55bb87789baf72f245abf1cc48ff117c370fea81ed43c.jpg", + "text": "$$\nH _ { n } ( x ) = ( - 1 ) ^ { n } e ^ { - x ^ { 2 } / 2 } \\frac { \\partial ^ { n } } { \\partial x ^ { n } } e ^ { - x ^ { 2 } / 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 379, + 892, + 617, + 922 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Note that orthogonality can be easily verified: ", + "bbox": [ + 176, + 103, + 475, + 118 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/2e9d9c0c84ec4d7edfddbeca264979db59bb9c3273641cfe0a87f5b12ec5abcf.jpg", + "text": "$$\n\\mathbb { E } [ H _ { j } ( G ) H _ { k } ( G ) ] = \\int _ { \\mathbb { R } } H _ { j } ( x ) H _ { k } ( x ) \\mu _ { G } ( d x ) = j ! \\cdot \\delta _ { j k } .\n$$", + "text_format": "latex", + "bbox": [ + 313, + 126, + 683, + 160 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Since $\\{ H _ { i } ( x ) \\} _ { i = 0 } ^ { \\infty }$ forms a set of orthogonal basis in $L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )$ , the function $\\phi ( x )$ can be expanded under the Hermite basis as ", + "bbox": [ + 176, + 169, + 823, + 199 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/cf3e2ec794a62ef6df6913e0a506cd1108f960db43f48819f91342e73f19bdbb.jpg", + "text": "$$\n\\phi ( x ) = \\sum _ { i = 0 } ^ { \\infty } c _ { i } H _ { i } ( x ) = \\sum _ { i = 0 } ^ { \\infty } \\left( \\frac { 1 } { k ! } \\int _ { \\mathbb { R } } \\phi ( a ) H _ { i } ( a ) \\mu _ { G } ( d a ) \\right) H _ { i } ( x ) .\n$$", + "text_format": "latex", + "bbox": [ + 289, + 204, + 709, + 247 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Following Cheng and Singer (2013), we mainly focus on the first few terms of the Hermite expansion. One can check that $H _ { 0 } ( x ) = 1$ , $H _ { 1 } ( x ) = x$ , and $\\phi ( x )$ can be expanded as ", + "bbox": [ + 173, + 261, + 826, + 291 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/226531a5330bd604d496dcddb6c029eb33b8f83d7b3f79483fb402113d0504ba.jpg", + "text": "$$\n\\phi ( x ) = c _ { 0 } + c _ { 1 } x + \\phi _ { \\perp } ( x ) ,\n$$", + "text_format": "latex", + "bbox": [ + 406, + 299, + 589, + 318 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "where $c _ { 0 } = \\mathbb { E } [ \\phi ( G ) ]$ , $c _ { 1 } = \\mathbb { E } [ G \\phi ( G ) ]$ , and terms in the RHS are orthogonal to one another in $L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )$ . Taking square and expectation over both sides yields ", + "bbox": [ + 173, + 325, + 826, + 354 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/12684a4b97f9202c539740eb08f6762f801dfc54acbb2ad6395521a01bc8ca6b.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] = \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 323, + 363, + 668, + 382 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "which indicates $\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } - \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } = \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] \\ge 0 } \\end{array}$ . Note that the equality holds if and only if $\\mathbb { E } [ \\phi _ { \\bot } ( G ) ^ { 2 } ] = 0$ , i.e. $\\phi ( x ) = c _ { 0 } + c _ { 1 } x$ is linear. One application of this decomposition is to “linearize” a non-linear matrix in high dimensions, which will be useful for the following sections. ", + "bbox": [ + 173, + 390, + 826, + 435 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "C PROOF OF MAIN RESULTS ", + "text_level": 1, + "bbox": [ + 176, + 457, + 428, + 474 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "C.1 PROOF OF LEMMA 1 ", + "text_level": 1, + "bbox": [ + 174, + 491, + 357, + 506 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Given features $X \\in \\mathbb { R } ^ { d \\times n }$ , labels $\\pmb { y } \\in \\mathbb { R } ^ { d }$ and model parameters $\\pmb \\theta$ , the gradient flow of $\\pmb \\theta$ on the squared loss $\\left\\| \\pmb { y } - X ^ { \\top } \\pmb { \\theta } \\right\\| _ { 2 } ^ { 2 }$ can be written as ", + "bbox": [ + 176, + 517, + 823, + 551 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/dbdad43c1e0beba1bc12f91a6bd87045cd6f5dcdf6c7b08528a7ae7dcfaaf568.jpg", + "text": "$$\n\\frac { \\partial \\pmb { \\theta } ( t ) } { \\partial t } = \\frac { 1 } { n } X ( y - X ^ { \\top } \\pmb { \\theta } ( t ) ) .\n$$", + "text_format": "latex", + "bbox": [ + 401, + 560, + 598, + 593 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Thus with initialization $\\pmb { \\theta } _ { 0 }$ , the solution of this ODE at time $t$ can be written in explicit form ", + "bbox": [ + 173, + 599, + 777, + 616 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/2d757828b13e9155af54df8b08c0c8971ae620a6a40afc9f48108c39ecdde395.jpg", + "text": "$$\n\\pmb { \\theta } ( t ) = e ^ { - \\frac { t } { n } X X ^ { \\top } } \\pmb { \\theta } _ { 0 } + ( X X ^ { \\top } ) ^ { \\dagger } \\left( I - e ^ { - \\frac { t } { n } X X ^ { \\top } } \\right) X \\pmb { y } .\n$$", + "text_format": "latex", + "bbox": [ + 318, + 623, + 678, + 651 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Since $\\pmb { \\theta } _ { 0 } = 0$ , taking $t \\to \\infty$ yields the desired result. ", + "bbox": [ + 174, + 659, + 526, + 675 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "C.2 PROOF OF THEOREM 2 ", + "text_level": 1, + "bbox": [ + 174, + 700, + 375, + 715 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We compute the bias and variance for different cases of $\\gamma _ { 1 } , \\gamma _ { 2 }$ . We first discuss the case where the random feature $\\Phi _ { X }$ is not full rank (Case I). Otherwise when $\\Phi _ { X }$ is full rank, we discuss whether it is full column rank (Case II) or full row rank (Case III). ", + "bbox": [ + 174, + 727, + 825, + 771 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Case I: $W ^ { \\top } X$ is not full rank, i.e. $\\gamma _ { 1 } < 1 , \\gamma _ { 2 } > \\gamma _ { 1 }$ . In this case rank $( \\Phi _ { X } ) = d < \\operatorname* { m i n } ( n , h )$ , and thus by taking the Moore-Penrose inverse we obtain ", + "bbox": [ + 173, + 787, + 825, + 818 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/5ef80b31e9945ee11acf8d4a05a6697c17d69130393268c46722d4b3c1e888d6.jpg", + "text": "$$\n{ \\hat { \\boldsymbol { \\beta } } } = W ( X ^ { \\top } W ) ^ { \\dagger } { \\boldsymbol { y } } = ( X X ^ { \\top } ) ^ { - 1 } X { \\boldsymbol { y } } .\n$$", + "text_format": "latex", + "bbox": [ + 372, + 825, + 625, + 845 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "It is clear that the mean and variance is identical to the underparameterized regime in Hastie et al. (2019), i.e. when $n , d , h \\infty$ , ", + "bbox": [ + 171, + 861, + 826, + 890 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/474990e4ccfc71621442e1ac6495e4e3d2d2543a9b05f97006a93d40244500cd.jpg", + "text": "$$\nB 0 ; \\quad V \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 408, + 898, + 589, + 929 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Case II: $W ^ { \\top } X$ has full column rank, i.e. $\\gamma _ { 2 } < 1 , \\gamma _ { 1 } > \\gamma _ { 2 }$ . By Lemma 1, the solution of the second layer coefficients is ", + "bbox": [ + 171, + 102, + 825, + 133 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/db6a7ab2e5d37354a1a7e2bfefe5b814e4b18a174d7d8cf315f09918a89cd80c.jpg", + "text": "$$\n\\begin{array} { r } { \\hat { \\beta } = W ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X y . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 382, + 138, + 614, + 159 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Denote $W = U \\Sigma V ^ { \\top }$ the singular value decomposition. We perform the block decomposition: ", + "bbox": [ + 173, + 166, + 794, + 183 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/8e724e50ca9424395b9283a4a5889c12e652bf61e7a42bfd5db110cf4a12fed3.jpg", + "text": "$$\n\\Sigma = \\left[ \\Sigma _ { 0 } \\right] , X = \\left[ \\Sigma _ { 1 } \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 413, + 188, + 581, + 224 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where $\\Sigma _ { 0 } \\in \\mathbb { R } ^ { h \\times h } , X _ { 0 } \\in \\mathbb { R } ^ { h \\times n } , X _ { 1 } \\in \\mathbb { R } ^ { ( d - h ) \\times n }$ , and notice that $X _ { 0 } , X _ { 1 }$ are independent. By a concentration of measure argument (e.g. Tao (2012); Hastie et al. (2019)) one can show that the quantity below tightly concentrates at its expectation. For the variance we have ", + "bbox": [ + 174, + 229, + 825, + 275 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/a478871fcdaef4bc974ffb67e8b80c1135afc95829ea44be74511ef48dcc3d72.jpg", + "text": "$$\n\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } X ^ { \\top } \\sigma ^ { 2 } X W \\left( \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } \\right) \\right. } \\\\ & { \\left. = \\sigma ^ { 2 } \\mathrm { t r } \\left( W ^ { \\top } W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } U ^ { \\top } X X ^ { \\top } U \\Sigma \\right) ^ { - 1 } \\right) \\right. } \\\\ & { \\left. \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\left( X _ { 0 } X _ { 0 } ^ { \\top } \\right) ^ { - 1 } \\right) \\right. \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 240, + 280, + 758, + 367 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where the last equality follows from Appendix B.2. Similarly for the bias term we have ", + "bbox": [ + 171, + 371, + 746, + 387 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/515891f2ef048c678fbfca2d4a5db9c09d60df581a642c374b48b8a5384a2777.jpg", + "text": "$$\n\\begin{array} { r l } & { B = \\| { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } \\beta - \\beta \\| _ { 2 } ^ { 2 } } \\\\ & { \\quad = \\beta ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) \\beta } \\\\ & { \\stackrel { ( \\psi ) } { = } \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma X ^ { \\top } ( V \\Sigma X ^ { \\top } X X ^ { \\top } U \\Sigma \\Sigma { \\cal { W } } ^ { \\top } ) ^ { - 1 } V \\Sigma \\Sigma ^ { \\top } { \\cal { U } } ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( \\cdots ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma ^ { \\top } \\Sigma \\sigma \\Sigma ^ { \\top } \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( \\Sigma \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\Sigma \\Big ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( ( \\Sigma ^ { \\top } X ^ { \\top } \\Sigma ) ^ { - 1 } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 186, + 395, + 812, + 598 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where symmetric arguments are omitted as $( \\cdot \\cdot \\cdot )$ , and (i) follows from the rotational invariance argument introduced in Lemma 10 and that $\\beta ^ { \\top } \\beta = r ^ { 2 }$ . By the block decomposition (30), ", + "bbox": [ + 169, + 603, + 823, + 633 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/75b68ce49c070f2e797f82c3695a9dbb8c5ffcaddcd04477af0fd67b3b103535.jpg", + "text": "$$\n\\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } = \\left[ \\begin{array} { l l } { 0 } & { ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } } \\\\ { 0 } & { - I _ { d - h } } \\end{array} \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 279, + 640, + 717, + 675 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Therefore the bias term simplifies to ", + "bbox": [ + 173, + 680, + 413, + 695 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/d5f2f2ce220c66d82124355105f5bd55076362bad3c56e92df6e32ab92705ff6.jpg", + "text": "$$\n\\begin{array} { l } { { \\displaystyle B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) ^ { \\top } \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) \\right) } } \\\\ { { \\mathrm { } = \\frac { r ^ { 2 } } { d } \\left( \\mathrm { t r } \\left( ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } X _ { 1 } X _ { 0 } ^ { \\top } ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } \\right) + ( d - h ) \\right) } } \\\\ { { \\mathrm { } \\to \\frac { r ^ { 2 } } { d } \\left( \\frac { ( d - h ) h } { n - h - 1 } + d - h \\right) \\to \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } \\left( 1 - \\gamma _ { 2 } \\right) } r ^ { 2 } . } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 209, + 702, + 787, + 806 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Thus we have obtained that as $n , d , h \\infty$ ", + "bbox": [ + 174, + 811, + 460, + 825 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/b44679d17b6fe7ed1fadb754db3d0f4639acf3b59a5c7f8f6e646d0ca2791004.jpg", + "text": "$$\nB \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } ( 1 - \\gamma _ { 2 } ) } r ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 429, + 833, + 568, + 864 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Case III: $W ^ { \\top } X$ has full row rank, i.e. $\\gamma _ { 1 } > 1 , \\gamma _ { 2 } > 1$ . Similarly, the least squares solution is ", + "bbox": [ + 169, + 878, + 808, + 896 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/06daaa58c150186f04888890add9047d6e277956a9d567db4bf4a049dbeaf530.jpg", + "text": "$$\n\\hat { \\beta } = W W ^ { \\top } X ( X ^ { \\top } W W ^ { \\top } X ) ^ { - 1 } \\pmb { y } ,\n$$", + "text_format": "latex", + "bbox": [ + 383, + 901, + 612, + 921 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Simplifying the variance: ", + "bbox": [ + 173, + 103, + 341, + 118 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/fb55650dbf8ef0fa3382b662b0fa85e7de89f0fd1ffb2fb001ed04409ef08846.jpg", + "text": "$$\n\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W W ^ { \\top } X \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } \\sigma ^ { 2 } \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } X ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } U \\Sigma V ^ { \\top } \\left( V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } U \\Sigma V ^ { \\top } \\right) ^ { - 2 } V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma \\right) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } \\right) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 254, + 127, + 743, + 212 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "where we applied the SVD of $X = U \\Sigma V ^ { \\top }$ and the rotational invariance argument. Using a similar block decomposition on $W$ : ", + "bbox": [ + 173, + 219, + 826, + 250 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/23adeaa39b60c9788b5a183b8ad08c7b80dab595e2cb883c50a3b0aef6a65dd6.jpg", + "text": "$$\n\\Sigma = \\left[ \\stackrel { \\Sigma _ { 0 } } { 0 } \\right] , W = \\left[ \\stackrel { W _ { 0 } } { W _ { 1 } } \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 411, + 258, + 584, + 295 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "where $\\Sigma _ { 0 } \\in \\mathbb { R } ^ { n \\times n } , W _ { 0 } \\in \\mathbb { R } ^ { n \\times h } , W _ { 1 } \\in \\mathbb { R } ^ { ( d - n ) \\times h }$ , and $W _ { 0 } , W _ { 1 }$ independent. We thus simplify $V$ as ", + "bbox": [ + 171, + 304, + 825, + 323 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/af984b7aada2fb32194ea1fbcbe8852714b6279464135d08e84fcc1015befb4a.jpg", + "text": "$$\n\\begin{array} { r l } & { V = \\sigma ^ { 2 } \\mathrm { t r } ( W W ^ { \\top } \\Sigma ( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma ) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } ) } \\\\ & \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( [ \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { \\cdots } & { \\cdots } \\\\ { ( \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { W _ { 1 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 1 } ^ { \\top } } \\end{array} ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( \\mathrm { t r } ( \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ) + \\mathrm { t r } ( W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } W _ { 1 } ^ { \\top } ) ) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( ( X ^ { \\top } X ) ^ { - 1 } ) + \\sigma ^ { 2 } \\mathrm { t r } ( W _ { 1 } ^ { \\top } W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } ) \\cdot \\ ( 3 9 ) } \\end{array} \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 184, + 329, + 839, + 444 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Hence we obtain the following expression on the variance ", + "bbox": [ + 173, + 450, + 555, + 465 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/874afd9111c23f1cff8c896a34da4f7616adad78aecef68652ac8bc445969cd3.jpg", + "text": "$$\nV \\sigma ^ { 2 } \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\sigma ^ { 2 } ( d - n ) \\mathbb { E } _ { W , X } V \\mathrm { t r } ( ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ) \\sigma ^ { 2 } ( \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\frac { 1 } { \\gamma _ { 2 } - 1 } ) .\n$$", + "text_format": "latex", + "bbox": [ + 179, + 474, + 815, + 510 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We omit the derivation of bias, which follows a similar derivation: ", + "bbox": [ + 173, + 534, + 607, + 549 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/2288a3ebd351e85f1503cb89a8d9d6caf04b57eb29d22f1ff4b5a50629a20461.jpg", + "text": "$$\nB \\frac { \\gamma _ { 2 } ( \\gamma _ { 1 } - 1 ) } { \\gamma _ { 1 } ( \\gamma _ { 2 } - 1 ) } r ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 429, + 558, + 568, + 592 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Combining Case I, II, III yields theorem 2. ", + "bbox": [ + 174, + 608, + 454, + 623 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "C.3 PROOF OF PROPOSITION 3 ", + "text_level": 1, + "bbox": [ + 174, + 650, + 400, + 665 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Given the squared loss, one can derive the dynamics of $W$ with fixed second layer $^ { a }$ w.r.t the loss: ", + "bbox": [ + 171, + 678, + 815, + 694 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/0ca5d134c0e3bb46bde6687e000ff933f332606ecfe415aa7dd013398706531c.jpg", + "text": "$$\n\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - X ^ { \\top } W ( t ) \\pmb { a } ) \\pmb { a } ^ { \\top } .\n$$", + "text_format": "latex", + "bbox": [ + 370, + 702, + 627, + 734 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Note that the update of $W$ can be written as a linear combination of $^ { a }$ . Since $W ( 0 ) = 0$ , we can write $W ( t ) = \\bar { \\hat { w } } ( t ) \\mathbf { { a } } ^ { \\top }$ for some $\\hat { \\textbf { \\textit { w } } }$ . The corresponding flow on $\\hat { w }$ is ", + "bbox": [ + 173, + 751, + 825, + 781 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/b992fc90e218018b684b44a84a932474a897f282e9308707633ba3c4fb4c0fab.jpg", + "text": "$$\n\\frac { \\partial \\pmb { \\hat { w } } ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - \\boldsymbol { X } ^ { \\top } \\pmb { \\hat { w } } ( t ) \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 790, + 625, + 821 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "which gives the following solution ", + "bbox": [ + 173, + 830, + 401, + 845 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/81713cb939b19c512a2893faf164a0bb87be96797d7b1b1e201e30a5a003185b.jpg", + "text": "$$\n{ \\hat { \\pmb { w } } } ^ { * } = { \\frac { 1 } { \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } } } { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } \\Rightarrow { \\hat { \\beta } } = { \\boldsymbol { W } } ^ { * } \\pmb { a } = { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } .\n$$", + "text_format": "latex", + "bbox": [ + 362, + 854, + 635, + 890 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Thus gradient flow on the first layer leads to the minimum-norm solution on the input features. ", + "bbox": [ + 171, + 898, + 794, + 915 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "C.4 PROOF OF THEOREM 4 ", + "text_level": 1, + "bbox": [ + 174, + 103, + 377, + 118 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Following the bias-variance decomposition (7), the variance term can be written as ( $\\cdot \\sigma ^ { 2 }$ omitted) ", + "bbox": [ + 169, + 128, + 803, + 145 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/2cbbd84b03cbf4922e3d6e17f5f9ef4c7ad64a61dccd27f8a5bbfea7a801866a.jpg", + "text": "$$\n\\begin{array} { r l } & { V = \\mathbb { E } _ { \\pmb { x } , \\pmb { \\varepsilon } } \\Big [ \\| \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) - \\mathbb { E } _ { \\pmb { \\varepsilon } } \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\Big \\| \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\phi ( W ^ { \\top } \\pmb { x } ) \\Big \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] \\right) } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } K _ { W } \\right) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 289, + 146, + 705, + 261 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "where we define the expected non-linear Gram matrix $K _ { W } \\in \\mathbb { R } ^ { h \\times h }$ as ", + "bbox": [ + 173, + 262, + 637, + 279 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/4150c3e50a9017d8f65e32db158b3b719961d1c95653b8261f1a875cfcf0d895.jpg", + "text": "$$\nK _ { W } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] .\n$$", + "text_format": "latex", + "bbox": [ + 383, + 280, + 612, + 306 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "and for each entry we have $( K _ { W } ) _ { [ i , j ] } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { j } ^ { \\top } \\pmb { x } ) \\Big ] .$ ", + "bbox": [ + 173, + 309, + 598, + 334 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Random matrix in the form of covariance matrix of nonlinear features has been studied in many works Hastie et al. (2019); Mei and Montanari (2019); Liao and Couillet (2018); Louart et al. (2018); Pennington and Worah (2017). We note that our setup for the variance term is very similar to that for nonlinear features in Hastie et al. (2019) with modifications mentioned below. ", + "bbox": [ + 173, + 339, + 826, + 396 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "In contrast to the linear network in Section C.2, the Gram matrix of a nonlinear activation is almost surely full-rank, as specified in the following lemma from Pennington and Worah (2017): ", + "bbox": [ + 171, + 402, + 823, + 431 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Lemma 11. Suppose √ $\\phi$ is not linear. Then the smallest singular value of $\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }$ is of order $O ( { \\sqrt { n } } )$ . To be precise, consider the empirical spectral density $\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )$ , when $n , d , h \\infty$ with $d / n \\gamma _ { 1 }$ and $h / n \\gamma _ { 2 }$ , $\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )$ converges weakly to ", + "bbox": [ + 173, + 434, + 825, + 478 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/30105f33e88d205694fab25035f259f3b7cc7cada4bf9e48291fa6222d7b70c1.jpg", + "text": "$$\n\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda ) [ 1 - \\gamma _ { 2 } ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\mu ^ { + } ( d \\lambda ) ,\n$$", + "text_format": "latex", + "bbox": [ + 339, + 482, + 655, + 502 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "where $\\mu ^ { + } ( d \\lambda )$ has non-negative support $\\lbrack \\rho , \\infty )$ with $\\rho > 0$ . ", + "bbox": [ + 173, + 503, + 568, + 520 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "We therefore consider two scenarios: $\\Phi$ is full column rank (Case I) or full row rank (Case II). ", + "bbox": [ + 171, + 529, + 789, + 545 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Case 1. $h < n$ . In this case (45) simplifies into ", + "bbox": [ + 173, + 558, + 501, + 573 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/3668294f9bd1d547e0fbc6005c68eed1228bfc1f53d35acd53c1b9fcb29b594f.jpg", + "text": "$$\nV = \\operatorname { t r } \\left( \\left( \\phi ( W ^ { \\top } X ) \\phi ( X ^ { \\top } W ) \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } \\operatorname { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } V _ { \\xi } ,\n$$", + "text_format": "latex", + "bbox": [ + 205, + 575, + 789, + 609 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "where the continuity and boundness of $V _ { \\xi }$ at $\\xi = 0 ^ { - }$ is guaranteed by Lemma 11 when $\\gamma _ { 2 } = h / n \\neq 1$ . A ridge $\\xi$ is added make use of (Louart et al., 2018, Theorem 1), which derived the asymptotic equivalent of the resolvent $\\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 }$ . It follows that as $n , d , h \\infty$ , ", + "bbox": [ + 173, + 613, + 826, + 659 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/526904641963f8ba69ef39d63c4d674173d9230b06c35a9dd75ff16689448313.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathrm { t r } \\left( h ^ { - 1 } \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) - h ^ { - 1 } I \\right) } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } \\left\\| \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } - \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) ^ { - 1 } \\right\\| _ { F } } \\\\ & { \\quad \\cdot \\left\\| \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right\\| _ { 2 } } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } O ( n ^ { - 1 / 2 + \\varepsilon } ) O ( n ^ { 1 / 2 } ) \\to 0 . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 230, + 662, + 759, + 852 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "where we have used the inequality $\\operatorname { t r } \\left( A B \\right) \\leq \\left\\| A \\right\\| _ { F } \\left\\| B \\right\\| _ { 2 }$ , and equivalently by taking $\\xi 0$ we get ", + "bbox": [ + 173, + 856, + 825, + 873 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/47918eab7f209ed0d7ec24e3972688136f5931475e7d0360095f3688a840a5b9.jpg", + "text": "$$\n\\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } \\frac { n } { h } \\frac { \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } { 1 + \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } = 1 .\n$$", + "text_format": "latex", + "bbox": [ + 328, + 876, + 668, + 929 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Therefore $\\begin{array} { r } { \\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } n / h \\cdot \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\gamma _ { 2 } / ( 1 - \\gamma _ { 2 } ) . } \\end{array}$ . Note that $\\partial V _ { \\xi } / \\partial \\xi =$ $\\mathrm { t r } \\left( \\xi ( \\Phi \\Phi ^ { \\top } - \\xi I ) ^ { - 2 } K _ { W } \\right)$ is bounded around the neighbourhood of $\\xi = 0$ , hence following the same argument as (Hastie et al., 2019, Theorem 4), we exchange the limit of $\\xi 0$ and $n , d , h 0$ : ", + "bbox": [ + 173, + 101, + 825, + 155 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/c92aa2600d8044011547634f0b384701c1f6c8ff7d58733fb78e4194f2952399.jpg", + "text": "$$\nV \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } .\n$$", + "text_format": "latex", + "bbox": [ + 450, + 160, + 545, + 190 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Case 2. $h > n$ . Techniques used in the current proof are largely borrowed from Hastie et al. (2019); Cheng and Singer (2013), and we include the full proof for completeness. It should be noted that compared to Hastie et al. (2019) we handle the non-zero expectation of the nonlinearity under Gaussian distribution, i.e. the off-diagonal entries of the kernel matrix is no longer zero-centered. For simplicity we mainly adhere to the notations in Hastie et al. (2019). ", + "bbox": [ + 173, + 202, + 826, + 273 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We briefly summarizes the procedure for deriving $V$ . Instead of calculating the variance directly, we analyze a modified quantity $V _ { \\xi }$ and then take $\\xi 0$ , which can be connected to the trace of the resolvent of matrix $\\tilde { A }$ defined in (56); this translates the calculation of $V _ { \\xi }$ into the calculation of the Stieltjes transform of $\\tilde { A }$ (57), (61), (62). ", + "bbox": [ + 173, + 279, + 825, + 342 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "C.5 DERIVING THE VARIANCE $V$ FOR $h > n$ ", + "bbox": [ + 173, + 357, + 493, + 372 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Step 1. An equivalent expression. For notational simplicity we omit the magnitude $\\sigma$ : ", + "bbox": [ + 169, + 383, + 756, + 400 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/deb1415ea32e2554feb854247738d950e9a7cc3b2c9c5d080c17a325eb6e09ad.jpg", + "text": "$$\n\\begin{array} { r } { V = \\mathrm { t r } \\bigg ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } \\bigg ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 290, + 405, + 707, + 440 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "and due to the same continuity argument as in Case 1 we have ", + "bbox": [ + 174, + 445, + 581, + 460 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/8b2d551025e06da7434a2b9f5271a75e04ab44cce52bdc7d9e4404774b445e77.jpg", + "text": "$$\nV = \\operatorname * { l i m } _ { \\xi 0 } \\frac { 1 } { n } \\Big [ \\mathrm { t r } ( S ( S ^ { \\top } S - \\xi I _ { n } ) ^ { - 2 } S ^ { \\top } K _ { W } ) \\Big ] = \\operatorname * { l i m } _ { \\xi 0 } V _ { \\xi } .\n$$", + "text_format": "latex", + "bbox": [ + 313, + 465, + 683, + 497 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where $S = \\phi ( W ^ { \\top } X ) / \\sqrt { n } = \\Phi / \\sqrt { n } , \\xi \\in \\mathbb { C }$ and $\\Im \\xi > 0$ or $\\xi < 0$ . ", + "bbox": [ + 173, + 505, + 607, + 522 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We decompose the normalized feature matrix $S = \\phi ( W ^ { \\top } X ) / \\sqrt { n }$ as $S = U \\Sigma V ^ { \\top }$ , where $\\Sigma =$ $\\mathrm { d i a g } _ { h \\times n } \\big ( \\mathring { \\phi _ { 1 } } , \\cdot \\cdot \\cdot , \\phi _ { n } \\big ) \\in \\mathbb { R } ^ { h \\times n }$ is a tall diagonal matrix, and $U \\overset { ^ { \\prime } } { = } [ \\pmb { u } _ { 1 } , \\cdots , \\pmb { u } _ { h } ] \\in \\mathbb { R } ^ { h \\times h }$ is the set of orthogonal eigenvectors of $S \\underline { { S } } ^ { \\top } = \\phi ( \\underline { { W } } ^ { \\top } X ) \\phi ( X ^ { \\top } W ) / n$ , and $V \\in \\mathbb { R } ^ { n \\times n }$ is the set of orthogonal eigenvectors of $S ^ { \\top } S = \\phi ( X ^ { \\top } W ) \\phi ( \\dot { W } ^ { \\top } X ) / n$ . Now the variance can be written as ", + "bbox": [ + 173, + 526, + 826, + 587 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/1e601b2a761065abacfb294a61d5e371d9d05fe2d3cc49ea946ce1d251bd3f17.jpg", + "text": "$$\nV _ { \\xi } = \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 352, + 593, + 645, + 622 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "By the same argument as in Lemma 13 of Hastie et al. (2019) one can show that when $n , d , h \\infty$ ", + "bbox": [ + 173, + 627, + 821, + 643 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/1d70b869882ea158d5128f5c765fadd3fbb0a6758b7069413ff1fe35e856aec0.jpg", + "text": "$$\nV _ { \\xi } = \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) \\Big ] \\to \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) \\Big ] ,\n$$", + "text_format": "latex", + "bbox": [ + 187, + 648, + 782, + 679 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "in which $\\tilde { K } _ { W }$ is the approxmation of $K _ { W }$ defined in Lemma 16. Writing the trace explicitly (denote eigenvalues $\\lambda _ { i } = \\phi _ { i } ^ { 2 }$ , and $\\phi _ { n + 1 } = \\cdot \\cdot \\cdot = \\phi _ { h } = 0 )$ ), we have ", + "bbox": [ + 173, + 686, + 828, + 717 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/5f7703358b5aaabc04bb952542ba6f16e46959915b54dd57af799b2688729815.jpg", + "text": "$$\nV _ { \\xi } \\to \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) = \\gamma _ { 2 } \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } { \\pmb { u } } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\pmb { u } } _ { i } .\n$$", + "text_format": "latex", + "bbox": [ + 243, + 723, + 754, + 767 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Since the positive support of spectrum $\\lambda$ is lower bounded and the density at 0 is $1 - \\gamma _ { 2 } ^ { - 1 }$ , we have ", + "bbox": [ + 173, + 773, + 821, + 790 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/98350499ada2aed4c3ced6b99bb30d22702b96ffa89753c2282abce0bdfe94ba.jpg", + "text": "$$\n\\begin{array} { r } { \\gamma _ { \\xi } \\to \\gamma _ { 2 } \\displaystyle \\frac { 1 } { h } \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } = \\gamma _ { 2 } \\displaystyle \\int \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\gamma _ { 2 } \\displaystyle \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 796, + 838, + 840 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where we define $\\begin{array} { r } { \\mu _ { n } ( x ) = \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\delta _ { \\lambda _ { i } } ( x ) \\pmb { u } _ { i } ^ { \\top } \\tilde { K } _ { W } \\pmb { u } _ { i } } \\end{array}$ and its positive part $\\mu _ { n } ^ { + } ( x )$ . Hence we have ", + "bbox": [ + 174, + 861, + 803, + 882 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/d2b089387b970da85b422f5e2b23f06cc9985d07a89aa42c5bb61dbfb5b0b019.jpg", + "text": "$$\nV = \\operatorname* { l i m } _ { \\xi \\to 0 } V _ { \\xi } = \\operatorname* { l i m } _ { \\xi \\to 0 } \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda } \\mu _ { \\infty } ^ { + } ( d \\lambda ) .\n$$", + "text_format": "latex", + "bbox": [ + 269, + 887, + 727, + 922 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We define the following matrix $\\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) \\in \\mathbb { R } ^ { N \\times N }$ where $N = n + h$ : ", + "bbox": [ + 173, + 101, + 650, + 119 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/7098dd044cf0ce45d2d339dc153c061a34188a1721556182854930ef68fe8937.jpg", + "text": "$$\n\\boldsymbol { \\tilde { A } _ { n } } ( \\rho , \\varsigma , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\varsigma \\mathbf { 1 } _ { h } \\mathbf { 1 } _ { h } ^ { \\top } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 346, + 123, + 650, + 159 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "And denote the Stieltjes transform of ${ \\tilde { A } } _ { n }$ as ", + "bbox": [ + 174, + 166, + 460, + 183 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/ca8963aa61947f8a7a7e07c2f564b1831096eb0b4cdefd30ae14d19cde68b56a.jpg", + "text": "$$\n\\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 338, + 186, + 660, + 218 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Then following the definition of $\\tilde { K } _ { W }$ one can show that ", + "bbox": [ + 174, + 223, + 539, + 239 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/d7cfa9ab41dcac8f1118b47b56f863d3a6d347394000145d246e81194df0a232.jpg", + "text": "$$\n\\tilde { m } _ { n } ( \\xi , r x , s x , t x ) = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } x - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) ,\n$$", + "text_format": "latex", + "bbox": [ + 308, + 244, + 689, + 287 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "and taking matrix derivative gives ", + "bbox": [ + 173, + 292, + 398, + 308 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/af41f24faadcea66d2bba2d1806e83d830fc6bd694112deb886e27e03a9b1ced.jpg", + "text": "$$\n\\begin{array} { r l } & { - \\displaystyle \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\xi ^ { 2 } I _ { h } + S S ^ { \\top } } & { 0 } \\\\ { 0 } & { I _ { n } + S ^ { \\top } S } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( U ( \\Sigma \\Sigma ^ { \\top } + \\xi ^ { 2 } I _ { h } ) ^ { - 1 } U ^ { \\top } \\tilde { K } _ { W } \\right) = \\frac { 1 } { n } \\displaystyle \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 186, + 313, + 808, + 444 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Denote the limit $\\begin{array} { r } { \\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau ) = \\operatorname* { l i m } _ { n , h , d \\infty } \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) } \\end{array}$ , and the derivative is given as ", + "bbox": [ + 173, + 462, + 758, + 479 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/e0099db1440f89576a9471d07dd9d9a25693c783b2187c1c40d990dbfa28b03f.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle \\left. - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\right. _ { x = 0 } = \\displaystyle \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\frac { 1 } { n } \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } { \\boldsymbol u } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\boldsymbol u } _ { i } } \\\\ { \\displaystyle = \\gamma _ { 2 } \\int _ { \\lambda \\geq 0 } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } + \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 178, + 484, + 789, + 565 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "For simplicity we define the following function on $\\xi$ : ", + "bbox": [ + 173, + 568, + 522, + 583 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/8378bc5c34eda2f679b92343c138dedffe13146e09d609d8147748dbb5a9ff2a.jpg", + "text": "$$\nq ( \\xi ) = - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } ,\n$$", + "text_format": "latex", + "bbox": [ + 383, + 588, + 612, + 619 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "$\\begin{array} { r } { q _ { + } ( \\xi ) = q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac 1 { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) } \\end{array}$ ", + "bbox": [ + 173, + 625, + 754, + 645 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/96522d192b8d179a6fe57aff2ebd3476d5d0a88738dfd0126e5d7acd9cedef32.jpg", + "text": "$$\nV = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) .\n$$", + "text_format": "latex", + "bbox": [ + 442, + 650, + 553, + 675 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Step 2. Calculating $q ( \\xi )$ and $m _ { n } ( \\xi , \\rho , \\varsigma , \\tau )$ This subsection aims to calculate $\\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau )$ and $q ( \\xi ) = - \\tilde { m } _ { x } ^ { \\prime } ( \\xi , r x , s x , t x ) | _ { x = 0 }$ , from which the variance can be computed from (57)(61)(62). ", + "bbox": [ + 169, + 689, + 825, + 719 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "We define $A _ { n }$ by subtracting the off-diagonal entries of the upper-left block of ${ \\tilde { A } } _ { n }$ : ", + "bbox": [ + 173, + 727, + 717, + 742 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/ccbb6c63072a3e4058671b0e8a6e601a81aac9ee807b7a8860cc04cdf5762e7c.jpg", + "text": "$$\nA _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 385, + 747, + 611, + 782 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "where $S = \\tilde { S } - a _ { 0 } I _ { p \\times n }$ , i.e. $S _ { i k } = \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } ) - a _ { 0 } = \\varphi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } )$ , $a _ { 0 } = \\mathbb { E } [ \\phi ( x ) ]$ . The Stieltjes transform of $A _ { n }$ given by $\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) } \\end{array}$ . The following Lemma shows that $\\tilde { m } _ { n }$ and $m _ { n }$ have the same limit: ", + "bbox": [ + 173, + 787, + 825, + 834 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Lemma 12. when $n \\to \\infty$ and for $\\Im \\xi > 0 o r \\xi < 0 $ , we have $m _ { n } ( \\xi , \\rho , \\tau ) \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) .$ ", + "bbox": [ + 173, + 837, + 776, + 854 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Proof. By definition of ${ \\bar { A } } _ { n }$ and $A _ { n }$ , ", + "bbox": [ + 174, + 863, + 413, + 878 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/8af219b2bfb34b777549f260c1474405eea56646dceb8fd9b6e82824732dfa15.jpg", + "text": "$$\n\\begin{array} { r } { \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] \\left[ \\begin{array} { c c } { \\xi } & { a _ { 0 } } \\\\ { a _ { 0 } } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] ^ { \\top } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 277, + 883, + 718, + 922 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "which is a rank-2 matrix. By theorem A.43 from Bai and Silverstein (2010), which characterizes the effect of finite-rank perturbation on the e.s.d. of random matrices: ", + "bbox": [ + 169, + 103, + 825, + 132 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/3aeaf525edaa1ae902f9fdb47acff8d4dad67e9db825334c4a19d53080e11e4c.jpg", + "text": "$$\n\\operatorname* { s u p } _ { x } | F ^ { \\tilde { A } _ { n } } ( x ) - F ^ { A _ { n } } ( x ) | \\leq O \\left( n ^ { - 1 } \\right) ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 138, + 624, + 169 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "where $F ^ { M }$ is the empirical spectral distribution of $M \\in \\mathbb { R } ^ { n \\times n }$ . The claim follows from the Stieltjes continuity theorem (e.g. Section 2.4 in Tao (2012)). □ ", + "bbox": [ + 169, + 176, + 825, + 207 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "To calculate the Stieltjes transform $m _ { n }$ , we take advantage of the block structure of $A _ { n }$ ", + "bbox": [ + 173, + 219, + 746, + 236 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/9ebf10c37453fa053d2e979503fda8f10419487e6f42fd71f42c6c7d2f799d00.jpg", + "text": "$$\n\\begin{array} { l } { { m _ { 1 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { p } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ 1 . . p , 1 . . p ] } ^ { - 1 } \\right) , } } \\\\ { { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ p + 1 . . p + n , p + 1 . . p + n ] } ^ { - 1 } \\right) . } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 294, + 242, + 704, + 306 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "One can observe that $\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) + m _ { 2 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}$ ", + "bbox": [ + 174, + 311, + 627, + 329 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "In the following equations we omit the subscript $n$ , as well as dependency on $\\rho , \\varsigma , \\tau$ . Following Hastie et al. (2019), we rewrite the matrix $A _ { n } = A = { \\left[ \\begin{array} { l l } { A _ { * } } & { a } \\\\ { \\mathbf { 1 } } & { 0 } \\end{array} \\right] }$ , where $A ^ { * }$ is a $\\left( N - 1 \\right) \\times \\left( N - 1 \\right)$ matrix with last column and row of $A$ removed and $\\pmb { s }$ the activation vector: ", + "bbox": [ + 173, + 333, + 826, + 392 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/b34f7a5369954cf11554bfe26ab8230e571a2fe0fc7b1912585e7c823e2d5555.jpg", + "text": "$$\n\\begin{array} { r } { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S _ { * } } \\\\ { S _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] ; \\quad \\pmb { a } ^ { \\top } = [ \\phi ( \\boldsymbol { W } ^ { \\top } \\mathbf { x } _ { n } ) ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] = [ \\boldsymbol { s } ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 243, + 396, + 756, + 431 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Hence by the block matrix inverse formula ", + "bbox": [ + 174, + 438, + 457, + 453 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/a7bf8105f5cfca014049958264580ebadc742124d6f5b90b1327ed4d973b3844.jpg", + "text": "$$\n\\begin{array} { r } { ( A - \\xi I _ { N } ) ^ { - 1 } = \\left[ \\begin{array} { l l } { * } & { * } \\\\ { * } & { [ - \\xi - { \\pmb a } ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } { \\pmb a } ] ^ { - 1 } } \\end{array} \\right] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 302, + 459, + 696, + 494 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Plugging this back in the Stieltjes transform we have ", + "bbox": [ + 173, + 501, + 519, + 516 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/f176ad0980439202a213c14bd6a4d34054a238f6980d3ef3c754459f486fb731.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) _ { [ h + 1 , N ] } ^ { - 1 } \\right) = \\mathbb { E } _ { a } \\left[ ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right] _ { N N } } \\\\ { \\displaystyle = \\mathbb { E } _ { a } \\bigg [ \\Big ( - \\xi - a ^ { \\top } ( A ^ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ) ^ { - 1 } \\bigg ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 218, + 522, + 779, + 585 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "To obtain an asymptotic description of $m _ { 2 , n }$ , we perform the orthonormal decomposition on the nonlinearity $\\varphi$ introduced in Cheng and Singer (2013): $\\varphi ( x ) \\ = \\ a _ { 1 } x + \\varphi _ { \\perp } ( x )$ , where $a _ { 1 } = \\mathbb { E } _ { x \\sim \\mathcal { N } ( 0 , 1 ) } [ x \\varphi ( x ) ]$ . For each ${ \\pmb w } _ { i } ( \\bar { 1 } \\leq i \\leq \\bar { h } )$ , we perform the following orthonormal decomposition (along the direction of ${ \\bf { x } } _ { n }$ and the direction of $\\tilde { \\mathbf { \\pmb { w } } } _ { i }$ perpendicular to ${ \\mathbf { \\mathcal { x } } } _ { n }$ ): ", + "bbox": [ + 173, + 597, + 826, + 654 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/cee414213d3b40bac825a65641fde07e4ad883d5b7cb0bba8c495de72d91b4ea.jpg", + "text": "$$\n\\pmb { w } _ { i } = \\underbrace { \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { n } } _ { \\eta _ { i } } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } = \\eta _ { i } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } .\n$$", + "text_format": "latex", + "bbox": [ + 352, + 660, + 643, + 700 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "We can thus simplify the activation vector $\\textbf { \\em a }$ as ", + "bbox": [ + 173, + 707, + 480, + 722 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/dfd9392f1a095abc0522233c533911ce1fef4318c4702efd92bf8bde56cf40af.jpg", + "text": "$$\n\\begin{array} { r l } { a ^ { \\top } = \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { 1 } ) } & { \\cdots \\cdot \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { = \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { 1 } } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { h } \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { + \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { 1 } ) } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdot \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 264, + 728, + 733, + 925 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "and for $1 \\leq i \\neq j \\leq h , 1 \\leq k \\leq n - 1$ , ", + "bbox": [ + 173, + 103, + 434, + 119 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/181ef276eb865e4cd29ca48f27514054f218948bd7346c449c4145ffd515e926.jpg", + "text": "$$\nQ _ { i j } = \\left( \\eta _ { i } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { i } \\right) ^ { \\top } \\left( \\eta _ { j } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { j } \\right) = \\eta _ { i } \\eta _ { j } + \\underbrace { \\tilde { \\pmb { w } } _ { i } ^ { \\top } \\tilde { \\pmb { w } } _ { j } } _ { \\tilde { Q } _ { i j } } .\n$$", + "text_format": "latex", + "bbox": [ + 285, + 121, + 710, + 172 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Similarly we decompose the activation function in $S$ , ", + "bbox": [ + 173, + 176, + 524, + 191 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/618584fd5f5e9c29878a555c52c0ac3adb13e0962d66ea03df40eb535d077628.jpg", + "text": "$$\n\\begin{array} { l } { { \\displaystyle { S _ { i k } = \\frac { 1 } { \\sqrt n } \\varphi \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) = \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } a _ { 1 } \\tilde { w } _ { i } ^ { \\top } x _ { k } + \\frac { 1 } { \\sqrt n } \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) } } \\\\ { { \\displaystyle { \\quad = \\frac { 1 } { \\sqrt n } \\varphi ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) + \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } \\left[ \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) - \\varphi _ { \\bot } ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) \\right] } } \\cdot { \\displaystyle ( 7 4 ) } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 183, + 194, + 839, + 286 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "We thus have an equivalent expression of matrix $A _ { * }$ ", + "bbox": [ + 174, + 291, + 511, + 306 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/7f605ae8ef8e552482a0d07a320a468883d23f6f7287b64cee9511c5716f210a.jpg", + "text": "$$\n\\begin{array} { r l } & { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { t \\eta \\eta ^ { \\top } } & { a _ { 1 } \\eta u ^ { \\top } } \\\\ { a _ { 1 } u \\eta ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } \\\\ & { \\quad = \\underbrace { \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } _ { \\tilde { A } _ { * } } + \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] } _ { U } \\underbrace { \\left[ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} \\right] } _ { C } \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] ^ { \\top } } _ { U ^ { \\top } } + \\underbrace { \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { \\mathbf { 0 } _ { n - 1 } } \\end{array} \\right] } _ { E } } \\\\ & { \\quad = \\tilde { A } _ { * } + U C U ^ { \\top } + E . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 176, + 308, + 816, + 424 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "By argument similar to (Hastie et al., 2019, B.1.2), $E$ diminishes to 0 as $n \\to \\infty$ with respect to the Frobenius norm, therefore by the Woodbury’s identity and the expression of $m _ { 2 , n }$ in (70) ", + "bbox": [ + 171, + 428, + 825, + 458 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/4bb68b057174ebc232ea9b5b5c2a800e4f5a9317e52bb2d552e4465104d95399.jpg", + "text": "$$\n\\begin{array} { r l } & { n _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } + U C U ^ { \\top } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } } _ { u } + } \\\\ & { \\qquad \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U } _ { v ^ { \\top } } \\underbrace { ( C ^ { - 1 } + U ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U ) ^ { - 1 } } _ { S } \\underbrace { U ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a } _ { v } \\bigg ) ^ { - 1 } \\bigg ] } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 460, + 839, + 611 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "We bound each term $u , v , S$ to compute $m _ { 2 , n }$ . For $u$ ", + "bbox": [ + 173, + 627, + 519, + 642 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/6465813f446dd9de9e8b3f200a891ab0a68ab06f69e90248d875f54672e8fcf2.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbb { E } _ { a } u = \\mathbb { E } _ { a } \\Big [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ] = \\mathrm { t r } \\left( \\mathbb { E } _ { s } \\big [ s s ^ { \\top } \\big ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . { h } , 1 . { h } ] } ^ { - 1 } \\right) = b \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 178, + 647, + 813, + 674 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "where $b = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ \\varphi ( \\boldsymbol { x } ) ^ { 2 } ] = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ ( \\phi ( \\boldsymbol { x } ) - \\mathbb { E } \\phi ( \\boldsymbol { x } ) ) ^ { 2 } ] = r _ { ! }$ . From a standard concentration of measure argument we have that as $n , h , d \\infty$ ", + "bbox": [ + 169, + 691, + 825, + 722 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/3de955f14284b3bc42d4fd78c3cbe1cbe6c04cd4be13063cb696def49ea7fdd1.jpg", + "text": "$$\nu \\mathbb { E } _ { \\pmb { a } } u = r \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) .\n$$", + "text_format": "latex", + "bbox": [ + 397, + 727, + 601, + 744 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "And for $\\textbf { { v } }$ (note that $U$ is dependent on $\\textbf { \\em a }$ ) ", + "bbox": [ + 173, + 747, + 450, + 762 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/c83a48b452e463c3496ecb37ceb21e3ebf147dda5286c7d6effeb35823e84ee6.jpg", + "text": "$$\n\\begin{array} { r l } { \\mathbb { E } _ { a } \\boldsymbol { v } ^ { \\top } = \\mathbb { E } _ { a } [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } \\boldsymbol { U } ] = \\mathbb { E } _ { a } [ [ \\boldsymbol { s } ^ { \\top } , \\boldsymbol { 0 } _ { n - 1 } ^ { \\top } ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\boldsymbol { \\eta } } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { \\boldsymbol { u } } \\end{array} ] ] } & { } \\\\ { = [ \\mathbb { E } _ { s } [ \\boldsymbol { s } ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\boldsymbol { \\eta } ] ] } & { 0 ] = [ \\underbrace { \\mathrm { t r } ( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\boldsymbol { \\eta } \\boldsymbol { s } ^ { \\top } ] ) } _ { \\boldsymbol { v } } } & { 0 ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 179, + 767, + 813, + 847 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { v = \\mathrm { t r } \\left( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . . h , 1 . . h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\eta s ^ { \\top } ] \\right) = a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}$ We thus have ", + "bbox": [ + 169, + 864, + 771, + 890 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/be198bd9a9c589f5ffdfb83ba80d6f26c337394810f47059d0839259a99f3a4f.jpg", + "text": "$$\nv \\mathbb { E } _ { a } v = [ a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\quad 0 ] .\n$$", + "text_format": "latex", + "bbox": [ + 351, + 895, + 647, + 929 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "as $n , d , p \\to \\infty$ . And finally for $S$ , ", + "bbox": [ + 173, + 103, + 403, + 119 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/e0317d45bcbd655a5a3ab8fc2a92a602f4d095bf3f002fe76c8b9ee801e910ac.jpg", + "text": "$$\n\\begin{array} { r l } { \\iota _ { \\alpha } S ^ { - 1 } = \\mathbb { E } _ { \\alpha } [ C ^ { - 1 } + U ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } U ] } \\\\ { = } & { [ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} ] ^ { - 1 } + \\mathbb { E } _ { \\alpha } [ [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ] } \\\\ { = } & { [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + \\mathbb { E } _ { \\alpha } [ \\begin{array} { c c } { \\eta ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\eta } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } ] } \\\\ { 0 } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } } \\end{array} ] } \\\\ { = } & [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + [ \\begin{array} { c c } { \\mathbb { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\mathbb { E } _ { \\alpha } | \\eta \\eta ^ { \\top } ) } & \\mathrm { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } | \\end{array} \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 125, + 839, + 362 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "And hence as $n , d , h \\infty$ , ", + "bbox": [ + 174, + 366, + 356, + 381 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/77d228e451b8149b74f1076da3e7f58b07ed4bf3f12d8722b505970c17cd28d9.jpg", + "text": "$$\nS ^ { - 1 } \\to \\mathbb { E } _ { a } S ^ { - 1 } = \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 272, + 387, + 725, + 422 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Therefore by combining (78), (80), (82), we arrive at the following expression on $m _ { 2 , n }$ ", + "bbox": [ + 173, + 429, + 748, + 444 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/2e463b72ecdc08c758158ed164610061d27ce21a56c00e98a0c5e918bcb14259.jpg", + "text": "$$\n\\begin{array} { r l r } & { } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) \\to \\mathbb { E } _ { a } \\Big [ \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } \\Big ] \\to \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } } \\\\ & { \\to \\Big ( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\Big ( a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } \\Big ) ^ { 2 } \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] _ { [ 1 , 1 ] } ^ { - 1 } \\Big ) ^ { - 1 } } \\\\ & { } & { = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\frac { \\gamma _ { 2 } a _ { 1 } ^ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) } { m _ { 1 , n } \\left( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 452, + 834, + 569 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Similarly we can calculate $m _ { 1 , n } ( \\xi , \\rho , \\tau )$ as ", + "bbox": [ + 173, + 575, + 457, + 590 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/7168d04b9d261062afc1d4d9ed42de994590f68307a13b5ac077e17e14cd5241.jpg", + "text": "$$\n\\begin{array} { r l } & { m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\to } \\\\ & { \\left( - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 , n } - r m _ { 2 , n } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 } - \\tau ) - 2 \\tau a _ { 1 } ^ { 2 } m _ { 1 , n } m _ { 2 , n } + a _ { 1 } ^ { 4 } m _ { 1 , n } m _ { 2 , n } ^ { 2 } } { m _ { 1 , n } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 169, + 599, + 816, + 660 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Uniqueness in (84)(83) follows from (Hastie et al., 2019, Sec B.1) and is omitted. ", + "bbox": [ + 173, + 683, + 707, + 699 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "C.6 PROOF OF COROLLARY 5 ", + "text_level": 1, + "bbox": [ + 174, + 722, + 393, + 738 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "In this section we take the limit $\\gamma _ { 1 } \\to \\infty$ . In this case (8), (9) simplify to ", + "bbox": [ + 173, + 748, + 651, + 765 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/9cfbc26107344339a87cccee67503863e23a242db4b518e43119b45a91cd3d13.jpg", + "text": "$$\n\\begin{array} { l } { { m _ { 2 } = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 } \\right) ^ { - 1 } , } } \\\\ { { \\nonumber } } \\\\ { { m _ { 1 } = \\left( - \\xi - \\rho - \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - r m _ { 2 } \\right) ^ { - 1 } . } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 369, + 770, + 629, + 815 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Recall that $m ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 } ( \\xi , \\rho , \\tau ) + m _ { 2 } ( \\xi , \\rho , \\tau )$ . By taking the derivative we have ", + "bbox": [ + 171, + 821, + 746, + 839 + ], + "page_idx": 23 + }, + { + "type": "equation", + "img_path": "images/48ede76d93106e807ad8bb633e222d0ffb6ee52649e86b9606a60815bb7ef8d4.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle - q ( \\xi ) = \\frac { \\partial } { \\partial x } m ( \\xi , r x , t x ) \\Big | _ { x = 0 } = \\left. r \\frac { \\partial } { \\partial \\rho } m ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } } \\\\ { \\displaystyle \\quad = \\left. r \\gamma _ { 2 } \\frac { \\partial } { \\partial \\rho } m _ { 1 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. r \\frac { \\partial } { \\partial \\rho } m _ { 2 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\gamma _ { 2 } \\frac { \\partial } { \\partial \\tau } m _ { 1 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m _ { 2 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 844, + 839, + 915 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Observe that (85), (86) constitutes a set of implicit functions. Thus differentiating the two functions with respect to $\\tau , \\rho$ and then substitute by $\\rho = \\tau = 0$ gives ", + "bbox": [ + 171, + 103, + 825, + 133 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/f3a30b56497382899bcdbcb36652db9ead678eaf172d9441fad1f710afe7a1c4.jpg", + "text": "$$\nq ( \\xi ) = \\frac { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) - \\xi ^ { 2 } \\right) \\left( \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } + r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) } { 2 \\xi ^ { 2 } \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } } .\n$$", + "text_format": "latex", + "bbox": [ + 235, + 138, + 763, + 200 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Hence by (62) we obtain the asymptotic variance: ", + "bbox": [ + 174, + 205, + 500, + 220 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/86eeef0b3abd7ad987a9ea07a9ae6f290701e7389f424b75e6dfa7996e256ba1.jpg", + "text": "$$\nV _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) = \\operatorname* { l i m } _ { \\xi \\to 0 } \\left( q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } \\right) = \\frac { 1 } { \\gamma _ { 2 } - 1 } .\n$$", + "text_format": "latex", + "bbox": [ + 300, + 227, + 696, + 262 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Combining the case where $\\gamma _ { 2 } < 1$ in Theorem 4 completes the proof. ", + "bbox": [ + 173, + 267, + 629, + 282 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Remark. For $\\phi ( x ) = \\mathrm { R e L U } ( x )$ , $c _ { 1 } = 1 / 2 - 1 / ( 2 \\pi )$ , $c _ { 2 } = 1 / 4$ . For $\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) =$ $\\log ( 1 + e ^ { x } )$ , numerical integration yields $c _ { 1 } \\approx 0 . 2 7 1 5$ , $c _ { 2 } = 1 / 4$ . ", + "bbox": [ + 173, + 290, + 825, + 320 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "C.7 PROOF OF COROLLARY 6 ", + "text_level": 1, + "bbox": [ + 174, + 335, + 393, + 351 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "C.7.1 UNBOUNDED BIAS WHEN $h = n$ ", + "text_level": 1, + "bbox": [ + 176, + 361, + 455, + 377 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "From the bias-variance decomposition (7), the bias $B$ is written as $\\begin{array} { r } { B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( Q _ { 1 } + Q _ { 2 } + I _ { d } \\right) } \\end{array}$ , where ", + "bbox": [ + 173, + 383, + 823, + 402 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/d7c0e84d892b7cc2080851eaa9255a8cffe4c12a695558b64413bdd223c190de.jpg", + "text": "$$\nQ _ { 1 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } ; \\quad Q _ { 2 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } W ^ { \\top } .\n$$", + "text_format": "latex", + "bbox": [ + 261, + 409, + 736, + 428 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "When $\\textit { h } = \\textit { n }$ , due to the nonlinearity of $\\phi$ , we have $\\phi ( W ^ { \\top } X )$ is full rank a.s., and hence $[ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } = [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 }$ . We have the following bound for $Q _ { 1 }$ ", + "bbox": [ + 173, + 435, + 828, + 465 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/5766ed0a2fd875e0e35d4aa77091b0681c5f63f7e856db1527c5fff9a055611d.jpg", + "text": "$$\n\\begin{array} { r } { \\overset { ! } { \\operatorname { t r } } ( Q _ { 1 } ) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } \\right) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\\\ { \\geq \\displaystyle \\frac { 1 } { d } \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) \\mathrm { t r } \\left( [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) = \\frac { \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } { n } \\mathrm { t r } \\left( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 470, + 839, + 537 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Since $W$ and $X / { \\sqrt { d } }$ are $\\mathbb { R } ^ { d \\times n } ~ = ~ \\mathbb { R } ^ { d \\times p }$ follows the same distribution where each entry i.i.d. $\\mathcal { N } ( 0 , 1 / d )$ , ", + "bbox": [ + 173, + 544, + 826, + 575 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/9800d639b0ce67d163d79772a44df6e0175a04082a36777a81e4a888953acb7e.jpg", + "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { d \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } \\mathrm { t r } ( Q _ { 1 } ) \\geq \\frac { 1 } { n } \\mathrm { t r } ( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X ) \\sim \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot W ^ { \\top } W ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot ( I + Q ) ) = \\operatorname* { l i m } _ { \\xi 0 } - \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , x , 0 , x ) \\infty , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 204, + 580, + 766, + 650 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "where in Section C.5 we have showed (91) is unbounded when $n \\infty$ . Moreover, by (154) and Weyl’s theorem we have $\\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) = O ( 1 )$ , and thus $d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 1 } \\right)$ is unbounded. For $d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 2 } \\right)$ , ", + "bbox": [ + 169, + 655, + 826, + 685 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/f5847230779d421975019dfa8d98de3ee3f410a4e17b78e4f242cba552ba1bdf.jpg", + "text": "$$\n\\frac { 1 } { d } \\mathrm { t r } ( Q _ { 2 } ) = \\frac { 1 } { d } \\mathrm { t r } \\left( W ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) \\leq \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) ^ { - 1 } ) \\mathrm { t r } \\left( W ^ { \\top } X \\right) = O ( 1 ) .\n$$", + "text_format": "latex", + "bbox": [ + 202, + 690, + 766, + 720 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "To sum up, for $n \\to \\infty$ and $\\gamma _ { 2 } 1$ we have $B \\infty$ . ", + "bbox": [ + 173, + 726, + 527, + 742 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "C.7.2 BOUNDED BIAS WHEN $\\gamma _ { 2 } > 1$ ", + "text_level": 1, + "bbox": [ + 174, + 762, + 437, + 779 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Since $h > n$ , the two terms in the expression of the bias can be written as ", + "bbox": [ + 176, + 786, + 655, + 803 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/b65bbc4a7bf15d47244e002c33f957f922700ec73c4a07f13628851548b9fa01.jpg", + "text": "$$\n\\begin{array} { r l } & { Q _ { 1 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } X ^ { \\top } , } \\\\ & { Q _ { 2 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) W ^ { \\top } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 189, + 808, + 779, + 869 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Therefore we have ", + "bbox": [ + 173, + 875, + 299, + 888 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/802e384616b8ee310544d57e8871b3da8cb31d01dc153c6778a231feb4f86630.jpg", + "text": "$$\n\\operatorname { I } ^ { \\mathrm { { T } } } ( Q _ { 1 } ) = 2 \\mathrm { { t r } } \\left( { \\frac { X ^ { \\top } X } { d } } { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 181, + 892, + 839, + 928 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/7742145ad610797abc0b4215eb1e361c0f199ca161702f321c5e6d9aba9c8ca2.jpg", + "text": "$$\n\\begin{array} { r l r } { { \\le 2 \\lambda _ { \\operatorname* { m a x } } ( \\frac { X ^ { \\top } X } { d } ) \\operatorname { t r } ( \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ) } } \\\\ & { = O ( 1 ) \\cdot \\operatorname { t r } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } ) } \\\\ & { \\le O ( 1 ) \\cdot \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) ) \\cdot \\operatorname { t r } ( K _ { W } ) } \\\\ & { = O ( 1 ) \\cdot \\sigma _ { \\operatorname* { m i n } } ^ { - 2 } ( \\phi ( X ^ { \\top } W ) ) \\operatorname { t r } ( K _ { W } ) = O ( 1 ) \\cdot O ( n ^ { - 1 } ) \\cdot O ( n ) = O ( 1 ) , } & { \\quad \\mathrm { ~ \\displaystyle ( 9 5 ) ~ } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 235, + 99, + 882, + 228 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "in which we used Lemma 11 and $\\operatorname { t r } \\left( A B \\right) \\leq \\lambda _ { \\operatorname* { m a x } } ( A ) \\operatorname { t r } \\left( B \\right)$ for positive semi-definite $A , B$ . Similarly ", + "bbox": [ + 173, + 232, + 826, + 262 + ], + "page_idx": 25 + }, + { + "type": "equation", + "img_path": "images/d88b10f67d37e9ed023388efdd27bc012f383dde71ca24414fd2e1bfce283fdb.jpg", + "text": "$$\n\\begin{array} { r l } & { \\frac { 2 } { d } \\mathrm { t r } \\left( Q _ { 2 } \\right) = 2 \\mathrm { t r } \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\cdot \\frac { 1 } { d } W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq \\mathrm { t r } \\left( \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\right) ( \\ldots ) ^ { \\top } \\right) + \\mathrm { t r } \\left( d ^ { - 2 } X ^ { \\top } W W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq n \\cdot \\sigma _ { \\operatorname* { m i n } } \\left( \\phi ( X ^ { \\top } W ) \\right) ^ { - 2 } + \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } \\left( \\frac { 1 } { d } X X ^ { \\top } \\right) \\mathrm { t r } \\left( W W ^ { \\top } \\right) = O ( 1 ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 204, + 268, + 790, + 376 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "where we applied $\\mathrm { t r } \\left( A B \\right) \\leq ( \\mathrm { t r } \\left( A ^ { \\top } A \\right) + \\mathrm { t r } \\left( B ^ { \\top } B \\right) ) / 2$ . We therefore conclude that $B$ is bounded when $h > n$ and $h , n \\infty$ . \u0003 ", + "bbox": [ + 173, + 382, + 823, + 414 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Remark. Concurrent to this work, Mei and Montanari (2019) provides a complete characterization of the bias term and confirms our observations above. ", + "bbox": [ + 171, + 422, + 825, + 452 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "C.8 PROOF OF THEOREM 7 ", + "text_level": 1, + "bbox": [ + 174, + 491, + 375, + 506 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "For simplicity we assume $n , d , h$ to be even and let $d _ { 0 } = d / 2$ , $n _ { 0 } = n / 2$ and $h _ { 0 } = h / 2$ . Since the second layer is fixed $a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}$ , we let $a _ { i } = 1 / \\sqrt { h }$ and $a _ { i + h _ { 0 } } = - 1 / \\sqrt { h }$ for all $1 \\leq i \\leq h _ { 0 }$ . We therefore write $\\pmb { a } ^ { \\top } = h ^ { - 1 / 2 } [ \\mathbf { 1 } _ { h _ { 0 } } , - \\mathbf { 1 } _ { h _ { 0 } } ] ^ { \\top }$ and $W = [ W _ { + } , W _ { - } ]$ : ", + "bbox": [ + 173, + 517, + 826, + 566 + ], + "page_idx": 25 + }, + { + "type": "equation", + "img_path": "images/fb080e2c6e923d48b65f31336b269d422b9ee4359ae926993eb1d278d413c92d.jpg", + "text": "$$\nf ( \\pmb { x } ; W _ { - } , W _ { + } ) = \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) = \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } \\pmb { x } ) - \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } \\pmb { x } ) .\n$$", + "text_format": "latex", + "bbox": [ + 259, + 573, + 736, + 607 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "The empirical risk can thus be written as ", + "bbox": [ + 176, + 613, + 441, + 628 + ], + "page_idx": 25 + }, + { + "type": "equation", + "img_path": "images/29f8c6e1d0bc276ee218a695699ddee6dbe27eec007019c7008a79ded3d7bc9b.jpg", + "text": "$$\nL ( X ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } L ( x ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } \\left[ y - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } x ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } x ) \\right] ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 181, + 637, + 821, + 675 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "C.8.1 DEFINING GRADIENT FLOWS ", + "text_level": 1, + "bbox": [ + 174, + 704, + 434, + 719 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "In this section we define three gradient flows and show that the three flows are similar in some sense. \n$\\pmb { G F }$ -Original is the original gradient flow (11), i.e. ", + "bbox": [ + 171, + 729, + 825, + 760 + ], + "page_idx": 25 + }, + { + "type": "equation", + "img_path": "images/43969bee080eb8e8ee088b02384d3ebb35bd85c910439835e45cd5f0708a2377.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle \\frac { \\partial W _ { + } ^ { O } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { + } ^ { O \\top } \\mathbf { x } _ { i } ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { - } ^ { O \\top } \\mathbf { x } _ { i } ) \\Big ) \\mathbf { x } _ { i } \\boldsymbol { \\phi } ^ { \\prime } ( { \\mathbf { x } } _ { i } ^ { \\top } W _ { + } ^ { O } ) \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\left[ X ( \\boldsymbol { y } - \\boldsymbol { y } ^ { O } ( t ) ) \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\circ \\boldsymbol { \\phi } ^ { \\prime } ( X W _ { + } ^ { O } ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 192, + 767, + 777, + 847 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "starting from the vanishing initialization ${ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )$ . Note that the gradient for the negative part $W _ { - } ^ { O }$ can be similarly defined. ", + "bbox": [ + 174, + 853, + 823, + 886 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "We now define the flow under the same objective but from exact zero initialization ${ \\pmb w } _ { i } ^ { D } ( 0 ) = { \\bf 0 }$ termed $\\pmb { G F }$ -Double. Due to zero initialization, a basic observation is that the solution $[ W _ { + } ^ { D } , W _ { - } ^ { D } ]$ is ", + "bbox": [ + 174, + 893, + 823, + 924 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "at most rank-2, and more precisely, the parameters in the flow takes the form of $W _ { \\pm } ^ { D } ( t ) = { \\pmb w } _ { \\pm } ^ { D } ( t ) { \\bf 1 } ^ { \\top }$ where ${ \\pmb w } _ { \\pm } ^ { D } ( t )$ admits the following dynamics: ", + "bbox": [ + 173, + 102, + 820, + 135 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/d71db0f99a7bbdce8f380ddc4807b59959cb4b1f89dea238fb5baa5ed83b913e.jpg", + "text": "$$\n\\frac { \\partial w _ { + } ^ { D } } { \\partial t } = g _ { + } ^ { D } ( \\boldsymbol { w } _ { + } ^ { D } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) + \\sqrt { h } \\phi ( \\boldsymbol { w } _ { - } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\boldsymbol { x } _ { i } \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 191, + 143, + 805, + 186 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Lastly, we define the $\\pmb { G F }$ -Single with solution denoted as ${ \\pmb w } _ { \\pm } = { \\pmb w } _ { \\pm } ^ { S } ( t )$ : ", + "bbox": [ + 173, + 208, + 651, + 226 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/d46d15fbec94e0029916dec56e8ba99890b28de3d83e6f15a2436677356efc44.jpg", + "text": "$$\n\\frac { \\partial w _ { + } ^ { S } } { \\partial t } = g _ { + } ^ { S } ( w _ { + } ^ { S } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { + } ^ { S \\top } x _ { i } + \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { - } ^ { 1 \\top } x _ { i } \\Big ) \\phi ^ { \\prime } ( 0 ) x _ { i } \\Big ] .\n$$", + "text_format": "latex", + "bbox": [ + 187, + 232, + 774, + 276 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "from zero initialization ${ \\pmb w } _ { \\pm } ^ { D } ( 0 ) = { \\bf 0 }$ . This can be seen as replacing the nonlinearity $\\phi$ with its first-order Taylor expansion at the origin. ", + "bbox": [ + 174, + 284, + 826, + 313 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "C.8.2 FROM GF-DOUBLE TO GF-SINGLE ", + "text_level": 1, + "bbox": [ + 174, + 328, + 470, + 343 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Step 1. Solution of GF-single. Among the three flows defined above, only GF-single an explicit form at any time $t$ . Specifically, the solution can be written as: ", + "bbox": [ + 169, + 352, + 823, + 381 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/b762dc632ef5f0943f1c17c6b6373cfd7d3777aa5ffa60d6850ed9156790ace3.jpg", + "text": "$$\nw _ { + } ^ { S } ( t ) = - w _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } ,\n$$", + "text_format": "latex", + "bbox": [ + 282, + 387, + 715, + 422 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "when $d < n$ , or otherwise ", + "bbox": [ + 173, + 429, + 346, + 444 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/7881650689b109f035de1a165987e743b3b7bd91073898c77af0933159d8d813.jpg", + "text": "$$\n\\pmb { w } _ { + } ^ { S } ( t ) = - \\pmb { w } _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } X \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X ^ { \\top } X t } \\right) ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } .\n$$", + "text_format": "latex", + "bbox": [ + 282, + 449, + 715, + 484 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "when $d > n$ . For simplicity we elaborate the proof only for $d < n$ . We provide a condition under which the difference between the trajectories can be controlled, and we first that this condition holds for the linearized flow GF-single for all $t$ in Lemma 17. ", + "bbox": [ + 174, + 491, + 826, + 534 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/20e90bda0059164dc605b61170a2f4f17ac694686cc4b9ff7c0ca6d776f91e81.jpg", + "text": "$$\n\\mathrm { C o n d i t i o n \\ A : } \\ \\| w ( t ) \\| _ { 2 } = O \\left( { \\frac { 1 } { \\sqrt { d } } } \\right) ; \\left\\| X ^ { \\top } w ( t ) \\right\\| _ { \\infty } = O \\left( { \\frac { \\mathrm { p o l y } \\log d } { \\sqrt { d } } } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 254, + 540, + 743, + 575 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Step 2. Bounding the Difference in Gradient Flow Trajectory. Due to low rank property of GF-single and GF-double, in this subsection we slightly abuse the notation and define ", + "bbox": [ + 173, + 587, + 825, + 617 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/4763bc2bd2dade9cd1d216f11bf583a0917967d2ff51fe88646e3722206e783c.jpg", + "text": "$$\nf ( \\pmb { x } ; \\pmb { w } _ { \\pm } ) = f ( \\pmb { x } ; \\pmb { w } _ { + } \\pmb { 1 } ^ { \\top } , \\pmb { w } _ { - } \\pmb { 1 } ^ { \\top } ) = \\sqrt { h } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } ) - \\sqrt { h } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } ) .\n$$", + "text_format": "latex", + "bbox": [ + 281, + 625, + 717, + 645 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "We now show that the difference between the two trajectories defined above is asymptotically vanishing for ${ \\pmb w } _ { + }$ $\\mathbf { \\nabla } w _ { - }$ follows the same argument). Compare the two trajectories up to time $T$ ", + "bbox": [ + 173, + 651, + 825, + 680 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/fd7c61c891b9274eed7e2f7c64f4f67079523a911d7dfdbb65843358a8774117.jpg", + "text": "$$\n\\begin{array} { r l r } { { \\| { \\boldsymbol w } _ { + } ^ { D } ( T ) - { \\boldsymbol w } _ { + } ^ { S } ( T ) } \\| _ { 2 } = \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) d t \\| _ { 2 } } \\\\ & { \\le \\| \\int _ { 0 } ^ { T } g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } + \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } } \\\\ & { = \\| \\int _ { 0 } ^ { T } \\frac { 1 } { 2 n _ { 0 } } \\frac { 2 n _ { 0 } } { i = 1 } [ ( - \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } + \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } ) \\phi ^ { \\prime } ( 0 ) { \\boldsymbol x } _ { i } ] \\mathrm { d } t \\| _ { 2 } } & \\\\ & { = O ( 1 ) \\int _ { 0 } ^ { T } ( \\| { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) \\| _ { 2 } + \\| { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) \\| _ { 2 } ) \\mathrm { d } t + { \\boldsymbol E } _ { + } , } & { ( 1 0 6 ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 171, + 685, + 816, + 856 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "where we have defined the error term as ", + "bbox": [ + 173, + 861, + 436, + 875 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/a5a4052e4440e0d387c9ef45daad6cf8330a1f9d303bf27b008b3b06d2e63429.jpg", + "text": "$$\nE _ { + } = \\left\\| \\int _ { 0 } ^ { T } \\pmb { g } _ { + } ^ { D } ( \\pmb { w } _ { + } ^ { D } ( s ) ) - \\pmb { g } _ { + } ^ { S } ( \\pmb { w } _ { + } ^ { D } ( s ) ) ~ \\mathrm { d } t \\right\\| _ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 343, + 878, + 653, + 921 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "To bound the error term, we note that at $t = 0$ Condition A holds for GF-double. Assume that for some $0 \\leq t \\leq T$ , Condition A also holds for GF-double, we have ", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 27 + }, + { + "type": "equation", + "img_path": "images/75f36a782e901fa29ab4f484d3f34c84e620969b34c8f808daa14f7f72a17e26.jpg", + "text": "$$\n\\begin{array} { r l } & { E _ { + } = \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { \\Phi } { u } \\int _ { 0 } ^ { u } [ u , \\frac { \\Phi } { u } ] ( \\boldsymbol { \\cdot } } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } \\\\ & { \\leq \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ \\frac { 1 } { \\sqrt { \\delta } } ( \\boldsymbol { \\cdot } - \\sqrt { \\delta } \\dot { u } ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) ) ( \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } u ( \\boldsymbol { \\cdot } ^ { \\Phi } , \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } \\rangle ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] \\Bigg \\| _ { 0 } ^ { 2 } , } \\\\ & \\quad + \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u _ { + } ^ { T } } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ ( - \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } ( u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] | \\int _ { 0 } ^ { T } \\ \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 138, + 839, + 431 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "where (i) follows from the error of Taylor expansion on $\\phi$ and (ii) from Condition A. Therefore, by Equation (106) and Gronwall’s inequality, ", + "bbox": [ + 174, + 445, + 828, + 474 + ], + "page_idx": 27 + }, + { + "type": "equation", + "img_path": "images/8502fa697ef4301e9370dc17fd11929d08c7c891a2f59fab8fde50582f9d06a8.jpg", + "text": "$$\n\\left\\| w _ { + } ^ { D } ( T ) - w _ { + } ^ { S } ( T ) \\right\\| _ { 2 } \\leq C _ { 1 } \\cdot \\frac { \\log ^ { c } h } { h } e ^ { C _ { 2 } T } = O \\left( \\frac { \\mathrm { p o l y l o g } h } { h } \\right) \\to 0 .\n$$", + "text_format": "latex", + "bbox": [ + 276, + 478, + 720, + 513 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "for $T \\in O ( \\log \\log h )$ . This shows that up to time $\\mathrm { T } ,$ , the difference between the trajectories of GF-single and GF-double vanishes under the the assumption that Condition A holds for $\\mathbf { \\Delta } w ^ { D }$ up to T. Importantly, note that ${ \\pmb w } ^ { S } ( 0 ) = { \\pmb w } ^ { D } ( 0 ) = 0$ , and that Condition A holds for $\\pmb { w } ^ { S }$ for all $T > 0$ . Therefore, by a standard contradiction argument (e.g. (Du et al., 2018, Lemma 3.4)), one can show that this closeness between the two trajectories implies that Condition A also holds for $\\mathbf { \\Delta } w ^ { D }$ up to time T. With Equation (108) we bound the difference in population risk between the two flows. ", + "bbox": [ + 173, + 517, + 826, + 603 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "Step 3. Bounding the Difference in Risk. Now we consider the difference of the population risk $R ^ { S }$ and $R ^ { D }$ of two models with parameters $\\pmb { w } _ { \\pm } ^ { S }$ and $\\pmb { w } _ { \\pm } ^ { D }$ . ", + "bbox": [ + 173, + 616, + 823, + 646 + ], + "page_idx": 27 + }, + { + "type": "equation", + "img_path": "images/8a06f95d31d3591849fac9bfce144a9b74f36b2b33c0ea0daa0eabf0899cc45d.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad | ( { \\mathcal R } ^ { 3 } - { \\mathcal R } ^ { D } ) | = | \\mathbb { E } _ { { \\mathbf z } } ( x ^ { \\top } ) - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) | ^ { 2 } - \\mathbb { E } _ { { \\mathbf z } _ { \\mathbf z } } ( x ^ { \\top } \\beta - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { D } ) ) ^ { 2 } | } \\\\ & { \\stackrel { ( i ) } { \\le } \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ f ( { \\mathbf x } ; { \\mathbf x } _ { \\mathbf z } ^ { \\top } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } \\mathbb { E } _ { { \\mathbf z } } [ | { \\mathcal R } ^ { 7 } \\cdot f ( { \\mathbf x } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) + \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } } } \\\\ & \\le \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ | { \\mathcal R } | _ { { \\mathbf z } } [ \\langle \\delta | ^ { \\mathcal { R } } \\rangle - \\phi ; \\langle { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ] ^ { 2 } \\mathbb { E } _ { | { \\mathbf z } } ] + | ( { \\mathcal R } \\langle \\mathbf x ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ) [ ^ { 2 } } \\\\ & { \\quad \\cdot \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } ^ { 7 } - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) ] + | \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) | ] ^ { 2 } } } \\\\ & \\stackrel { ( i i ) } { \\le } 2 \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } \\mathbb { R } ^ { 5 } ] - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } ] ^ { 2 } + | \\beta \\langle \\mathbf w _ { \\mathbf z } ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } \\rangle | ^ { 2 } } \\\\ & \\quad \\cdot \\sqrt \\mathbb { E } _ { \\mathbf z } [ | { \\mathcal R } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 233, + 650, + 759, + 893 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "where (i) is due to Cauchy-Schwarz inequality on norm, (ii) from Jenson’s inequality on squares and Young’s inequality, and (iii) from the Lipschitz assumption on the activation, and the observation ", + "bbox": [ + 173, + 895, + 825, + 924 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "that both $R ^ { S }$ and $R ^ { D }$ are finite for $\\gamma _ { 1 } \\neq 1$ due to the justified Condition A above. Therefore the difference between $R ^ { S }$ and $R ^ { D }$ vanishes for $T = O ( \\log \\log h )$ . ", + "bbox": [ + 171, + 102, + 825, + 133 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Step 4. Bounding the Difference from Stationarity We compute the difference in risk between the model at some finite time $t$ and the stationary point i.e. $t = \\infty$ , ", + "bbox": [ + 173, + 146, + 823, + 176 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/34524586fd5ffffc2b924b6ec60eec773e70f4360d1f22d5ce90c5996acb901e.jpg", + "text": "$$\n\\begin{array} { r l } & { \\left| R ^ { S } ( t ) - R ^ { S } ( \\infty ) \\right| \\leq C \\sqrt { h } \\cdot \\left\\| w _ { + } ^ { S } ( t ) - w _ { + } ^ { S } ( \\infty ) \\right\\| = C \\sqrt { h } \\left\\| \\frac { 1 } { 2 \\sqrt { h } } e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X y \\right\\| _ { 2 } } \\\\ & { = C \\left\\| e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X X ^ { \\top } \\beta \\right\\| _ { 2 } \\leq C \\exp \\left( - \\phi ^ { \\prime } ( 0 ) ^ { 2 } \\left\\| \\frac { 1 } { n } X X ^ { \\top } \\right\\| _ { 2 } t \\right) \\| \\beta \\| _ { 2 } = C _ { 3 } e ^ { - C _ { 4 } t } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 180, + 825, + 253 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "for constants $C _ { 3 } , C _ { 4 } > 0$ . Combining (110) and (111) yields for $T = \\log \\log h$ , ", + "bbox": [ + 174, + 270, + 692, + 286 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/b76e3a6581adc46b42523ed137e642416aa0148e83feb1cd041022bcd34562e3.jpg", + "text": "$$\n\\begin{array} { r } { | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { S } ( T ) - R ^ { D } ( T ) | + | R ^ { S } ( T ) - R ^ { S } ( \\infty ) | } \\\\ { = O ( \\frac { \\mathrm { p o l y l o g } h } { \\sqrt { h } } ) + O ( \\frac { 1 } { \\mathrm { p o l y l o g } h } ) 0 . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 281, + 290, + 715, + 348 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "The result in (112) for case $d > n$ follows a similar proof and is omitted. ", + "bbox": [ + 176, + 349, + 648, + 366 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "C.8.3 FROM GF-ORIGINAL TO GF-DOUBLE ", + "text_level": 1, + "bbox": [ + 176, + 380, + 488, + 395 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "In this section we compare GF-original with GF-double. Note that the two flows differ only at initialization: vanishing initialization ${ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )$ v.s. zero initialization ${ \\pmb w } _ { i } ^ { D } ( 0 ) \\dot { = } { \\bf 0 }$ . Note that at initialization $\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } = { \\cal O } ( { \\sqrt { n } } )$ , and gradient flow decreases the empirical risk; therefore the condition for Lemma 18 is satisfied, and by the Lipschitz condition on the empirical gradient we have ", + "bbox": [ + 173, + 404, + 826, + 474 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/e70ebf1301b03dada7aa6d1c2a09385f7980c4eb177878ec754e81c383cf9be1.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle \\left\\| W ^ { O } ( T ) - W ^ { D } ( T ) \\right\\| _ { F } ^ { 2 } } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) - W ^ { D } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\frac { \\partial \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } } { \\partial t } \\mathrm { d } t } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) ) \\right\\| _ { F } \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } + \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } ^ { 2 } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\left( 1 + L _ { f } \\right) \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } \\mathrm { d } t , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 200, + 478, + 797, + 657 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "And hence by Gronwall’s lemma one obtains: ", + "bbox": [ + 176, + 660, + 475, + 675 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/9d3d723174790ed8f4e38ffa642641ceec70dfa1b95882eb6d01482d58efe5b5.jpg", + "text": "$$\n\\left\\| { W ^ { O } ( T ) - W ^ { D } ( T ) } \\right\\| _ { F } \\leq \\left\\| { W ^ { O } ( 0 ) } \\right\\| _ { F } e ^ { ( 1 + L _ { f } ) T / 2 } = O ( d ^ { - ( 1 + \\epsilon ) / 2 } ) e ^ { C T } .\n$$", + "text_format": "latex", + "bbox": [ + 258, + 679, + 738, + 702 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Therefore the difference in the function output can be bounded as ", + "bbox": [ + 174, + 705, + 604, + 719 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/9b91cac8e5da3b0178e57a4079146859214f86ab98ae1bdc8d5e1d73bed50592.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad \\left\\| f ( \\pmb { x } , W ^ { D } ( T ) ) - f ( \\pmb { x } , W ^ { O } ( T ) ) \\right\\| _ { 2 } = \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) \\pmb { a } - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\pmb { a } \\right\\| _ { 2 } } \\\\ & { \\leq \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\right\\| _ { 2 } \\left\\| \\pmb { a } \\right\\| _ { 2 } \\leq L _ { \\phi } \\left\\| \\pmb { x } \\right\\| _ { 2 } \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } \\left\\| \\pmb { a } \\right\\| _ { 2 } } \\\\ & { = O ( \\sqrt { d } ) \\cdot \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } = O ( d ^ { - \\epsilon / 2 } ) e ^ { C T } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 210, + 723, + 784, + 790 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Taking $T = \\log \\log h$ , together with the same argument in Step 1 yields ", + "bbox": [ + 173, + 792, + 642, + 808 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/807813dbde2f926a6b2fbc4cd2d6f038cc133a4c11978bc875be0d4da12fd46c.jpg", + "text": "$$\n| R ^ { O } ( T ) - R ^ { D } ( T ) | = O \\left( \\frac { \\mathrm { p o l y l o g } h } { d ^ { \\epsilon / 2 } } \\right) \\to 0 .\n$$", + "text_format": "latex", + "bbox": [ + 348, + 811, + 650, + 847 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "C.8.4 PUTTING THINGS TOGETHER ", + "text_level": 1, + "bbox": [ + 174, + 857, + 431, + 872 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "By (112) and (116), we know that for $T = \\log \\log h$ , ", + "bbox": [ + 174, + 882, + 519, + 898 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/4049008e35d94ec4525b927c36aa8a0b5c776990c8b8619e261eb269b4f5e87a.jpg", + "text": "$$\n| R ^ { O } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { O } ( T ) - R ^ { D } ( T ) | + | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\to 0 .\n$$", + "text_format": "latex", + "bbox": [ + 266, + 901, + 730, + 921 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Finally the proof is completed by observing that $R ^ { S } ( \\infty )$ is the risk of the minimum-norm solution on the input discussed in Section 3. In addition, note that at $T = \\log \\log h$ , from (109)(114) one obtains that $\\left\\| \\mathbf { \\dot { W } } ^ { O } ( t ) - W ^ { S } ( t ) \\right\\| _ { F } \\to 0$ , and therefore by Lemma 18 we have $\\left\\| \\partial L ( X ; W ^ { \\mathcal { O } } ( t ) ) / \\partial W \\right\\| _ { F } \\to 0$ , i.e. the flow on the original objective reaches a $o ( 1 )$ first-order stationary point. \u0003 ", + "bbox": [ + 173, + 102, + 826, + 162 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "C.9 PROOF OF THEOREM 8 ", + "text_level": 1, + "bbox": [ + 174, + 184, + 375, + 199 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Denote $\\omega = \\mathrm { v e c } ( W ) = \\mathrm { v e c } ( [ W _ { + } , W _ { - } ] )$ , and $\\omega _ { 0 } = \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } )$ . Define ", + "bbox": [ + 173, + 209, + 638, + 227 + ], + "page_idx": 29 + }, + { + "type": "equation", + "img_path": "images/56f7456247d698ede4a0225c28ab28b09060851eec4e2f6c89a7530c305b555a.jpg", + "text": "$$\nK ( t ) = \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 232, + 625, + 270 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "which is the kernel matrix of the neural tangent kernel Jacot et al. (2018); Du et al. (2018). In the following sections we show that under the non-vanishing initialization, the trained two-layer network is well-approximated by the regression model on the NTK, for which we derive the population risk. ", + "bbox": [ + 173, + 273, + 826, + 316 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "C.9.1 THE KERNEL LINEARIZATION ", + "text_level": 1, + "bbox": [ + 174, + 330, + 439, + 345 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Write $\\pmb { y } _ { N N } ( t ) = f _ { N N } ( \\boldsymbol { X } , t ) \\in \\mathbb { R } ^ { n }$ and its evolution: ", + "bbox": [ + 173, + 354, + 519, + 371 + ], + "page_idx": 29 + }, + { + "type": "equation", + "img_path": "images/4435a090e161937c1ce992abecbecce3622bbf95c43cfcf615a4ea9b64cd589d.jpg", + "text": "$$\n\\mathrm { d } { \\pmb y } _ { N N } ( t ) = \\frac { 1 } { n } K ( t ) ( { \\pmb y } - { \\pmb y } _ { N N } ( t ) ) \\ \\mathrm { d } t ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 376, + 625, + 406 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "and the corresponding linearized flow: ", + "bbox": [ + 174, + 410, + 428, + 425 + ], + "page_idx": 29 + }, + { + "type": "equation", + "img_path": "images/2f42e3df666f164d85771eac3a45efe612b52feea3878e2008794d014e4b4753.jpg", + "text": "$$\n\\mathrm { d } { \\pmb y } _ { N T K } ( t ) = \\frac { 1 } { n } K ( 0 ) ( { \\pmb y } - { \\pmb y } _ { N T K } ( t ) ) \\ \\mathrm { d } t ,\n$$", + "text_format": "latex", + "bbox": [ + 361, + 430, + 635, + 460 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Previous works (e.g. Du et al. (2018); Oymak and Soltanolkotabi (2019)) have proved (nonasymptotically) that the two trajectories (119) and (120) are close if the model is overparameterized, i.e. $h = \\mathrm { p o l y } ( n )$ , under no assumptions on the teacher model. In our asymptotic setup (together with assumptions (A1-3)), we argue that similar conclusion holds without significant overparameterization. ", + "bbox": [ + 173, + 464, + 826, + 522 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "We first show the global convergence of the training of two-layer neural network $f _ { N N }$ . We employ an argument similar to (Du et al., 2018, Theo. 3.2) by first identifying the condition under which training converges at linear rate: ", + "bbox": [ + 173, + 527, + 823, + 570 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "From Corollary 15 we know that at initialization the lowest eigenvalue of the NTK matrix satisfies $\\lambda _ { \\operatorname* { m i n } } ( K ( 0 ) ) \\stackrel { } { = } O ( d )$ , and thus the kernel regression on the NTK enjoys linear convergence. By Lemma 19, we know that for $\\lVert W ( t ) - W ( 0 ) \\rVert _ { 2 } = O ( d ^ { 1 - \\epsilon } )$ , the order of $\\lambda _ { \\operatorname* { m i n } } ( K ( t ) ) \\stackrel { } { = } O ( d )$ remains unchanged. Therefore, if we assume that up to time $T$ the weights satisfy $\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =$ $O ( d ^ { - 1 / 2 } )$ , then Condition B is satisfied for ${ \\pmb y } _ { N N }$ and from (Chizat and Bach, 2018b, Lemma B1) we have the following linear convergence ", + "bbox": [ + 173, + 604, + 826, + 693 + ], + "page_idx": 29 + }, + { + "type": "equation", + "img_path": "images/df0758e4be5c97a12f736387cbe744a7e30743976b9df9d3e125c98d50b4d4e4.jpg", + "text": "$$\n\\begin{array} { r } { \\| { \\pmb y } _ { N N } ( t ) - { \\pmb y } \\| _ { 2 } \\le C _ { 1 } \\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } e ^ { - C _ { 2 } t } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 343, + 698, + 653, + 718 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Since $\\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } = O ( \\sqrt { n } )$ at initialization, setting $T = O ( \\log d )$ ensures that the training loss $\\begin{array} { r } { \\frac { 1 } { n } \\| { \\pmb y } _ { N N } ( T ) - { \\pmb y } \\| _ { 2 } ^ { 2 } 0 } \\end{array}$ as $n \\to \\infty$ . Consequently it is easy to check that $\\left\\| \\frac { \\partial { \\cal L } ( X ; W ( T ) ) } { \\partial W ( T ) } \\right\\| _ { 2 } \\to 0$ and thus at time T the gradient flow reaches an $\\mathsf { o } ( 1 )$ first order stationary point. ", + "bbox": [ + 173, + 723, + 826, + 777 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "We now verify that each weight vector ${ \\pmb w } _ { i }$ travels at most $O ( d ^ { - 1 / 2 } )$ from initialization. The norm of gradient for ${ \\pmb w } _ { i }$ can be bounded as: ", + "bbox": [ + 174, + 784, + 825, + 814 + ], + "page_idx": 29 + }, + { + "type": "equation", + "img_path": "images/26784e8ec9088b862d2278ac4206066251b182d8ee5c17533cc7a3a6f6917f62.jpg", + "text": "$$\n\\begin{array} { r l } & { \\left\\| \\frac { \\partial L ( X ; \\boldsymbol { w } _ { i } ( t ) ) } { \\partial \\boldsymbol { w } _ { i } ( t ) } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { n } X \\left[ \\frac { 1 } { \\sqrt { h } } ( \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) ) \\circ \\phi ^ { \\prime } ( X ^ { \\top } \\boldsymbol { w } _ { i } ( t ) ) \\right] \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i ) } { \\leq } O ( 1 ) \\frac { 1 } { d ^ { 1 . 5 } } \\left\\| X \\right\\| _ { 2 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) \\right\\| _ { 2 } \\leq O ( 1 ) d ^ { - 1 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( 0 ) \\right\\| _ { 2 } e ^ { - t } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 189, + 818, + 771, + 888 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "where (i) follows from the boundedness of $\\phi ^ { \\prime }$ . Integrating the gradient yields $\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =$ $O ( d ^ { - 1 / 2 } )$ , i.e. $\\| W ( t ) - W ( 0 ) \\| _ { 2 } = O ( 1 )$ . Thus the distance traveled by $W$ indeed satisfies the norm assumption above, and following the same argument as (Du et al., 2018, Lemma 3.4) we conclude that Condition B holds true for the gradient flow of $f _ { N N }$ for $t > 0$ . ", + "bbox": [ + 173, + 892, + 826, + 925 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 103, + 823, + 133 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Next we show that for any input $\\hat { \\pmb x }$ with $\\| \\hat { \\pmb { x } } \\| _ { 2 } = O ( \\sqrt { d } )$ , the prediction of the neural network is uniformly close to the prediction of the prediction of the kernel model, a result similar to (Arora et al., 2019a, Lemma F.1). Following the notation of Arora et al. (2019a), we write the time derivative of the prediction as ", + "bbox": [ + 173, + 138, + 828, + 196 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "$\\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N N } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N N } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N N } ( t ) } ) ; \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N T K } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N T K } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N T K } ( t ) } ) ,$ where $\\begin{array} { r } { { \\pmb u } _ { N N } ( { \\pmb x } , t ) = \\frac { \\partial { \\pmb f } ( X ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } ^ { \\top } \\frac { \\partial { \\pmb f } ( { \\pmb x } ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } \\in \\mathbb { R } ^ { n } } \\end{array}$ and ${ \\pmb u } _ { N T K } ( { \\pmb x } , t )$ similarly defined on the initialized weights $\\omega ( 0 )$ . We bound the difference between the predictions on $\\hat { \\pmb x }$ up to terminal time T as ", + "bbox": [ + 173, + 198, + 826, + 281 + ], + "page_idx": 30 + }, + { + "type": "equation", + "img_path": "images/7db2b15e344d4697b163e0c2a6c9aedc6405d407e15cde0710d3770ed070662a.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad | \\int _ { \\mathbf { N } ^ { \\mathrm { N } } } ( \\hat { x } , t ) - \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) | = | \\int _ { 0 } ^ { T } \\left[ \\frac { \\mathrm { d } } { \\mathrm { d } t } \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) - \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\right] \\mathrm { d } t | } \\\\ & { = \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left[ \\mathbf { a } _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N N } ( t ) ) - u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N T } ( \\hat { x } ) ) \\right] \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } | \\int _ { 0 } ^ { T } u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y _ { N N } ( t ) - y _ { N \\wedge \\mathbf { K } } ( t ) ) \\mathrm { d } t | } \\\\ & { \\quad + \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) ^ { \\top } ( y - y _ { N N } ( t ) ) \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } \\| u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y _ { N N } ( t ) - y _ { N T \\mathbf { K } } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\| u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\operatorname* { m a x } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 245, + 280, + 751, + 502 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "For the first term have ", + "bbox": [ + 173, + 507, + 321, + 522 + ], + "page_idx": 30 + }, + { + "type": "equation", + "img_path": "images/f6bf1504965a4e587dabfba74524500e87a4800efd3b2b27f2fef2157cbef0bd.jpg", + "text": "$$\n\\begin{array} { r l r } { { \\| { \\boldsymbol y } _ { N N } ( T ) - { \\boldsymbol y } _ { N T K } ( T ) \\| _ { 2 } \\le \\frac { 1 } { n } \\int _ { 0 } ^ { T } \\| K ( t ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) ) - K ( 0 ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N T K } ( t ) ) \\| _ { 2 } \\mathrm { d } t } } \\\\ & { } & { \\le \\frac { 1 } { n 0 < t < T } \\| K ( t ) - K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t + \\frac { 1 } { n } \\| K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { } & { \\overset { ( i ) } { \\le } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) + O ( 1 ) \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t \\overset { ( i i ) } { \\le } O ( d ^ { - \\epsilon ^ { \\prime } } ) , \\quad \\quad \\quad \\quad ( 1 2 4 ) \\mathrm { d } { \\boldsymbol z } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 169, + 521, + 816, + 631 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "where (i) is due to Corollary 15, Lemma 19 for some $\\epsilon ^ { \\prime } > 0$ and the linear convergence of ${ \\bf { \\it { \\mathbf { y } } } } _ { N N }$ , and (ii) is due to Gronwall’s inequality. Note that the log factor in $T$ is omitted. Similarly, for the second term we have ", + "bbox": [ + 176, + 631, + 825, + 672 + ], + "page_idx": 30 + }, + { + "type": "equation", + "img_path": "images/e069e54b2709bf413fdd3d4367aa285ae74d3f8552429f93a5279ea72e114207.jpg", + "text": "$$\n\\frac { 1 } { n \\hbar \\epsilon \\mathcal { T } } \\left. u _ { N N } ( \\hat { x } , t ) - u _ { N T K } ( \\hat { x } , t ) \\right. _ { 2 } \\int _ { 0 } ^ { T } \\left. y - y _ { N N } ( t ) \\right. _ { 2 } \\mathrm { d } t \\overset { ( i ) } { \\leq } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) = O ( d ^ { - \\epsilon ^ { \\prime } } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 176, + 671, + 816, + 708 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "where we used Lemma 19 and the linear convergence of ${ \\pmb y } _ { N N }$ in (i). Combining the two cases yields ", + "bbox": [ + 171, + 723, + 825, + 739 + ], + "page_idx": 30 + }, + { + "type": "equation", + "img_path": "images/b92e60630cc091fa04b42cc875a750c4c27f29ff8864975ac2bd6e3e283e65ec.jpg", + "text": "$$\n| f _ { N N } ( \\pmb { \\hat { x } } , t ) - f _ { N T K } ( \\pmb { \\hat { x } } , t ) | \\leq \\frac { 1 } { n } \\| \\pmb { u } _ { N T K } ( \\pmb { \\hat { x } } , t ) \\| _ { 2 } O ( d ^ { - \\epsilon ^ { \\prime } } ) + O ( d ^ { - \\epsilon ^ { \\prime } } ) \\overset { ( i ) } { = } O ( d ^ { - \\epsilon ^ { \\prime } } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 212, + 739, + 748, + 770 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "where we utilized Corollary 14 in (i). Thus we know that the difference between the population risk of $f _ { N N }$ and $f _ { N T K }$ is also asymptotically vanishing (note that the derivation above is independent of√ the target function as long as $\\bar { | | \\mathbf { y } | | _ { 2 } } = \\bar { O ( \\sqrt { n } ) } )$ . Therefore, in the following subsection we compute the risk of the linearized (kernel) model $f _ { N T K }$ . ", + "bbox": [ + 173, + 770, + 825, + 827 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "C.9.2 COMPUTING THE KERNEL RISK ", + "text_level": 1, + "bbox": [ + 176, + 840, + 452, + 854 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Given input $X \\in \\mathbb { R } ^ { d \\times n }$ and label $\\pmb { y } = \\pmb { \\beta } ^ { \\top } \\pmb { X } + \\pmb { \\varepsilon }$ , gradient flow on the tangent kernel solves the following equation of the parameters $\\omega$ : ", + "bbox": [ + 174, + 862, + 821, + 893 + ], + "page_idx": 30 + }, + { + "type": "equation", + "img_path": "images/52ed3c953ae3dba6d02df549f4eb4f4f76a89a2dabfe0cf2aca9e621bd12721f.jpg", + "text": "$$\n\\pmb { y } = \\pmb { f } ( X ; \\omega ) = \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } ( \\pmb { \\omega } - \\pmb { \\omega } _ { 0 } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 361, + 893, + 635, + 929 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "where $\\partial f ( X ; \\omega _ { 0 } ) / \\partial \\omega$ is a $d h \\times n$ matrix with column $\\partial f ( \\pmb { x } _ { i } ; \\pmb { \\omega } _ { 0 } ) / \\partial \\omega$ . Note that for $n \\to \\infty$ and $\\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty )$ , $d h > n$ trivially holds, and by Corollary 15 we know that solution is given by ", + "bbox": [ + 169, + 102, + 826, + 133 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/a593f61a44d8cf9d15adefc40e6776cd65d466e4235ec038ef7974f64a9b4196.jpg", + "text": "$$\n\\omega _ { 1 } = \\omega _ { 0 } + \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\left( \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } ^ { \\top } \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\right) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) .\n$$", + "text_format": "latex", + "bbox": [ + 266, + 140, + 732, + 185 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "And the population risk can be written as (note that there is a factor of 2 due to the \"doubling trick\" at initialization to ensure $f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0 .$ ): ", + "bbox": [ + 173, + 191, + 825, + 222 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/72e1cf33d940d4afe182e3be2eaed1daa330fd38df051ecb9aedf2f1ef1f7d95.jpg", + "text": "$$\n\\begin{array} { r l } & { 2 R = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - f ( x ; \\omega _ { 1 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } ( \\omega _ { 1 } - \\omega _ { 0 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial f ( X ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha } [ ( x ^ { \\top } \\beta - \\frac { \\partial \\tilde { f } ( \\kappa ^ { - 1 } X ^ { \\top } \\beta ) } { \\partial \\beta } \\frac { \\partial \\tilde { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial \\tilde { f } ( \\tilde { X } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } ) } { 2 V } \\frac { \\partial ^ { 2 } } { \\partial \\omega } , \\qquad ( 1 2 9 ) ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 169, + 228, + 816, + 419 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "where a bias-variance decomposition is made here, and for simplicity we define ", + "bbox": [ + 173, + 422, + 696, + 439 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/c1ca5ef4f293f868c59af874b6e441350f64df81c280c8ef9f848f2fb3ee7da6.jpg", + "text": "$$\n\\hat { \\pmb { u } } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( { \\pmb { x } } ; \\omega _ { 0 } ) } { \\partial \\omega } , \\quad \\hat { K } _ { X } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } .\n$$", + "text_format": "latex", + "bbox": [ + 276, + 445, + 720, + 482 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "C.9.3 APPROXIMATING THE KERNEL MATRIX ", + "text_level": 1, + "bbox": [ + 173, + 494, + 508, + 511 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "In this section we drop the negligible $\\epsilon$ in the initialization. Following Cheng and Singer (2013) we utilize the orthonormal decomposition of $\\phi ^ { \\prime } ( x )$ in $L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )$ . Denote $b _ { 0 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ]$ , and $b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }$ . We have the orthogonal decomposition of $\\phi ^ { \\prime }$ ", + "bbox": [ + 173, + 520, + 825, + 563 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/b7dfa8c1de8224d762d70789c4e42b6b546f00eb42b72c16707b34236bcb4e19.jpg", + "text": "$$\n\\phi ^ { \\prime } ( x ) = b _ { 0 } + \\phi _ { \\perp } ^ { \\prime } ( x ) ,\n$$", + "text_format": "latex", + "bbox": [ + 428, + 570, + 570, + 588 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "where $\\mathbb { E } [ \\phi _ { \\perp } ^ { \\prime } ( G ) ] = 0$ . We develop the following lemmas to approximate the kernel matrix. ", + "bbox": [ + 169, + 595, + 767, + 611 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Lemma 13 (Approximation of $( \\hat { K } _ { X } ) _ { i j } .$ ). There exist constants $c , c ^ { \\prime } > 0$ such that for $i \\neq j$ with probability $1 - e ^ { - c n \\varepsilon ^ { 2 } }$ we have ", + "bbox": [ + 173, + 617, + 823, + 650 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/ba124e0959c6d29fde9defcbbb7b95836cd045b48b500e30cc88c9bed64fca92.jpg", + "text": "$$\n\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 334, + 657, + 660, + 700 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "and with probability 1 − e−c0nε2 , ", + "bbox": [ + 173, + 710, + 393, + 726 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/341f89c3c6df5266c8ac725f841033b9c780ac530e50e7a42fb4ba7de26d0e59.jpg", + "text": "$$\n\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } - ( b _ { 0 } ^ { 2 } + b _ { 1 } ^ { 2 } ) \\right| < \\varepsilon .\n$$", + "text_format": "latex", + "bbox": [ + 341, + 733, + 655, + 776 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Proof. When $i \\neq j$ (i.e. Equation (132)), we have ", + "bbox": [ + 171, + 790, + 508, + 806 + ], + "page_idx": 31 + }, + { + "type": "equation", + "img_path": "images/3bace22a3d6842b9608a2d205d9d0976acd8db36707e2b16efbf5549b176bf6c.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad \\displaystyle \\frac 1 d [ \\hat { K } _ { X } ] _ { i j } = \\frac 1 d \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } } \\\\ & { = \\displaystyle \\frac 1 d { \\sum _ { k = 1 } ^ { h } } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } = \\frac 1 { d h } { \\displaystyle \\sum _ { k = 1 } ^ { h } } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { j } ) } \\\\ & { \\to \\displaystyle \\frac 1 d { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\mathbb { E } _ { \\pmb w } \\Big [ \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { j } ) \\Big ] = \\frac 1 d H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 258, + 813, + 740, + 926 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "The matrix $H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } \\mathbb { E } _ { \\pmb { w } } \\Big [ \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { j } ) \\Big ]$ can be seen as the expected tangent kernel of nonlinear activation function studied in Du et al. (2018); Arora et al. (2019b). Moreover, due to the assumed boundedness of $\\phi ^ { \\prime } ( x )$ (A3), by Hoeffding’s inequality we have ", + "bbox": [ + 173, + 101, + 826, + 152 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/a6a6772afb39175d22e5112efd10259f74ce8b844391960549bb99377de3b746.jpg", + "text": "$$\n\\operatorname* { P r } \\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) \\right| < \\frac { 1 } { d } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\varepsilon > 1 - e ^ { - c _ { 1 } h \\varepsilon ^ { 2 } } .\n$$", + "text_format": "latex", + "bbox": [ + 246, + 159, + 751, + 200 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "In addition, by the concentration of $\\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j }$ and $\\| \\pmb { x } _ { i } \\| _ { 2 } ^ { 2 }$ , i.e. $\\mathrm { P r } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d > \\varepsilon < 1 - e ^ { - c _ { 2 } d \\varepsilon ^ { 2 } }$ and $\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { i } / d - 1 | < \\varepsilon > 1 - e ^ { - c _ { 3 } d \\varepsilon ^ { 2 } }$ , the orthonormal decomposition $\\phi ^ { \\prime } ( x ) = b _ { 0 } x + \\phi _ { \\perp } ^ { \\prime } ( x )$ leads to the following linear approximation of the matrix $H$ ", + "bbox": [ + 173, + 207, + 825, + 256 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/cb26e808d6df3fe8d39c20cd87a656067ebe146a2121b73064dd7d03c832ee6e.jpg", + "text": "$$\n\\frac { 1 } { d } H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = b _ { 0 } ^ { 2 } \\frac { 1 } { d } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } + O \\big ( ( \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d ) ^ { 2 } \\big ) .\n$$", + "text_format": "latex", + "bbox": [ + 348, + 260, + 648, + 291 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "and by taking $\\varepsilon = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d$ under the joint event we can show that ", + "bbox": [ + 173, + 296, + 611, + 313 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/a282ba77a940f70a732d35630effa395391b9966ceb1ea7657b0617c6bb673c5.jpg", + "text": "$$\n\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 336, + 319, + 658, + 362 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "with probability $1 - e ^ { - c d \\varepsilon ^ { 2 } }$ . The same argument follows for the case where $i = j$ ", + "bbox": [ + 173, + 368, + 712, + 386 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Corollary 14 (Approximation of $\\hat { \\textbf { \\textit { u } } }$ ). For large enough $l > 0$ , with probability $1 - d e ^ { - c \\log ^ { l } d }$ ", + "bbox": [ + 174, + 396, + 785, + 412 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/4baf9635a33c5c755f86d270576005fb42499035d328fb1c3388a5ac9b07a62e.jpg", + "text": "$$\n\\frac { 1 } { d } \\left\\| \\hat { \\pmb { u } } - \\tilde { \\pmb { u } } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( \\pmb { x } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { \\pmb { u } } \\right\\| _ { 2 } < \\frac { \\log ^ { l } d } { d } ,\n$$", + "text_format": "latex", + "bbox": [ + 290, + 416, + 707, + 460 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "where $\\tilde { \\pmb { u } } = b _ { 0 } ^ { 2 } \\pmb { x } ^ { \\top } \\boldsymbol { X }$ . ", + "bbox": [ + 173, + 467, + 307, + 483 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Proof. Taking $\\varepsilon = \\log ^ { l } d / d$ together with Lemma 13 yields the desired result. ", + "bbox": [ + 174, + 494, + 686, + 511 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Corollary 15 (Approximation of ${ \\hat { K } } _ { X } { \\mathrm { . } }$ ). With probability $1 - d e ^ { - c \\log ^ { l } d }$ ", + "bbox": [ + 176, + 520, + 643, + 539 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/20ccb876ecc17f1c942ed29b2ed64f828b8f11b2e3a94731533694df96a89c6a.jpg", + "text": "$$\n\\frac { 1 } { d } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { K } _ { X } \\right\\| _ { F } < \\log ^ { l } d ,\n$$", + "text_format": "latex", + "bbox": [ + 261, + 542, + 735, + 587 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "where $\\tilde { K } _ { X } = b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I$ . ", + "bbox": [ + 174, + 592, + 377, + 611 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Proof. Also by directly applying Lemma 13. ", + "bbox": [ + 173, + 619, + 472, + 636 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Remark. For initialization larger or equal to ${ \\pmb w } _ { i } ( 0 ) \\sim N ( 0 , I _ { d } / d )$ , the above approximation does not depend on the scale of initialization. ", + "bbox": [ + 173, + 643, + 825, + 672 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Remark. For $\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) .$ , $b _ { 0 } ^ { 2 } = 1 / 4$ , $b _ { 1 } ^ { 2 } = 0 . 0 4 3 3 7 9$ . For $\\phi ( x ) = \\mathrm { s i g m o i d } ( x ) = ( 1 +$ $e ^ { - x } ) ^ { - 1 }$ , $b _ { 0 } ^ { 2 } = 0 . 0 4 2 6 9 2$ , $b _ { 1 } ^ { 2 } = 0 . 0 0 2 1 4 4$ . Note that $b _ { 1 } \\geq 0$ for all smooth activations $\\phi$ , and the equality holds (i.e. $b _ { 1 } = 0$ ) if and only if $\\phi$ is linear. We comment that smaller $b _ { 1 }$ entails larger variance as $\\gamma _ { 1 } 1$ , and vice versa, as shown in the following section. ", + "bbox": [ + 173, + 674, + 826, + 732 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "C.9.4 THE BIAS TERM", + "text_level": 1, + "bbox": [ + 174, + 746, + 348, + 761 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "With these approximation above we proceed to calculating (129) ", + "bbox": [ + 173, + 770, + 598, + 786 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/644d9f31c0a7b3276946727dcfd2dbe64d9e76f80c7ec098763aeb36aea46c38.jpg", + "text": "$$\n2 B = \\mathbb { E } _ { \\pmb { x } } \\left[ \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - \\hat { \\pmb { u } } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\pmb { \\beta } \\right) ^ { 2 } \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 366, + 791, + 632, + 825 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "We first bound the error in substituting $\\hat { \\textbf { \\textit { u } } }$ with $\\tilde { \\mathbf { \\pmb { u } } }$ : ", + "bbox": [ + 173, + 830, + 495, + 847 + ], + "page_idx": 32 + }, + { + "type": "equation", + "img_path": "images/8827867e5a0b0f22c443292f9987b46fb2daebe9ceaebb0473f431c87e781245.jpg", + "text": "$$\n\\begin{array} { r l } & { \\left\\| \\hat { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta - \\tilde { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta \\right\\| _ { 2 } \\leq \\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\boldsymbol X \\right\\| _ { 2 } \\left\\| \\boldsymbol \\beta \\right\\| _ { 2 } } \\\\ & { \\quad \\quad \\quad \\quad \\stackrel { ( i ) } { = } O \\left( \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot \\sqrt { d } \\cdot 1 \\right) = O \\left( \\frac { \\log ^ { l } d } { \\sqrt { d } } \\right) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 214, + 851, + 746, + 922 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "where (i) is due to the fact that $\\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } = \\lambda _ { \\operatorname* { m i n } } ^ { - 1 } ( \\hat { K } _ { X } ) = O ( 1 / d )$ and $\\| X \\| _ { 2 } = O ( { \\sqrt { d } } )$ . Therefore we have as $n , d , h \\infty$ ", + "bbox": [ + 174, + 99, + 825, + 140 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/c46331efc24df13f917fdbc06e8e67c167e3d3a1c36ca7854f077f2c1962415c.jpg", + "text": "$$\n\\begin{array} { r l } & { 2 B = \\mathbb { E } _ { x } [ ( { \\pmb x } ^ { \\top } \\beta - \\hat { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - \\tilde { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\qquad = \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - b _ { 0 } ^ { 2 } { \\pmb x } ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 215, + 143, + 745, + 215 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "By taking expectation over $_ { \\textbf { \\em x } }$ and the rotational invariance argument similar to Hastie et al. (2019), ", + "bbox": [ + 173, + 218, + 820, + 234 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/c7cd0607ab6d3cbace39e38a775f5b4d928f82e32ea776a044c5a9b27956d490.jpg", + "text": "$$\n\\begin{array} { r l } { \\mathbb { E } _ { x } \\left[ \\left( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta \\right) ^ { 2 } \\right] } & { = \\mathbb { E } _ { x } \\left[ \\beta ^ { \\top } \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) ^ { 2 } \\beta \\right] } \\\\ & { = \\frac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } \\left( \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 191, + 238, + 807, + 309 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "In addition, we bound the error in substituting $\\hat { K } _ { X }$ by $\\tilde { K } _ { X }$ defined in (139): ", + "bbox": [ + 173, + 328, + 669, + 347 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/a9ba1caf037d2a81b9fd907732404bfbb489da398eeab5125b2164448ec185a1.jpg", + "text": "$$\n\\begin{array} { r l } & { \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } ( \\hat { K } _ { X } - \\tilde { K } _ { X } ) \\tilde { K } _ { X } ^ { - 1 } \\right) \\right| } \\\\ & { \\qquad < \\displaystyle \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { \\qquad = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 187, + 349, + 769, + 459 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "and similarly, ", + "bbox": [ + 173, + 463, + 266, + 478 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/f72f4a59ecfdbef314b4adb1f28a71d1511f198caf239c57254239278a48941b.jpg", + "text": "$$\n\\begin{array} { r l } & { ~ \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| } \\\\ & { = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } ) X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } ) \\right) \\right| } \\\\ & { < \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 245, + 483, + 750, + 627 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Combining these two formulas in (143) yields ", + "bbox": [ + 173, + 631, + 477, + 646 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/77a8c4bafb16bbda9850c5d84fd4faecc20df8ada8490b1af26cd21bbf1df9d0.jpg", + "text": "$$\n\\begin{array} { r l } & { 2 B \\to \\mathbb { E } _ { x } [ ( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\quad \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ) } \\\\ & { \\quad = \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 1 } X ^ { \\top } ) ^ { 2 } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 299, + 650, + 697, + 755 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Utilizing the Marcenko–Pastur law from Section ˇ B.2 we obtain ", + "bbox": [ + 173, + 757, + 589, + 772 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/2fe9501a13b4ebe6aa84d2e71d0eb4933529d5f20e19673ae512ed5ba8b0d83a.jpg", + "text": "$$\nB = \\beta ^ { \\top } \\beta \\left( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } \\right) ,\n$$", + "text_format": "latex", + "bbox": [ + 266, + 777, + 732, + 820 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "where $m = { b _ { 0 } } ^ { - 2 } { b _ { 1 } } ^ { 2 }$ . ", + "bbox": [ + 173, + 825, + 299, + 843 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "C.9.5 THE VARIANCE TERM", + "text_level": 1, + "bbox": [ + 174, + 856, + 387, + 872 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Similarly, for the variance we utilize the approximation ", + "bbox": [ + 173, + 881, + 539, + 897 + ], + "page_idx": 33 + }, + { + "type": "equation", + "img_path": "images/528a09dc73ebbe81a976158c9006931939c2d35257e3b5d24c5f1aaca5e31b46.jpg", + "text": "$$\n2 V = \\mathbb { E } _ { \\pmb { x } } \\left[ \\hat { \\pmb { u } } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { \\pmb { u } } ^ { \\top } \\right] \\sigma ^ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 397, + 901, + 601, + 929 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Specifically, we bound the approximation error ", + "bbox": [ + 173, + 103, + 483, + 119 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/2a3055c96e8a78a1a11d0516f07e7a9d908f2c246fb0e917f13e9d5fed3eb326.jpg", + "text": "$$\n\\begin{array} { r } { \\left| \\hat { \\boldsymbol u } \\hat { K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol u } ^ { \\top } - \\tilde { \\boldsymbol u } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\tilde { \\boldsymbol u } ^ { \\top } \\right| \\leq \\left\\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol u } + \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ { = O \\left( \\log ^ { l } d \\cdot \\frac { 1 } { d } \\cdot d \\cdot \\frac { 1 } { d } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 243, + 126, + 751, + 199 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "and similarly ", + "bbox": [ + 173, + 207, + 261, + 222 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/9da21504c8e439b6b5d1544adef1105e0ef1f7169fbe031bb04963489a29671e.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad \\left| \\mathbb { E } _ { \\mathbf { x } } \\left[ \\tilde { u } \\hat { K } _ { X } ^ { - 1 } \\hat { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } - \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\right] \\right| } \\\\ & { = \\mathrm { t r } \\left( \\left( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } \\right) \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) \\mathbb { E } _ { \\alpha \\tilde { u } \\tilde { u } ^ { \\top } } \\right) } \\\\ & { = \\mathrm { t r } \\left( \\hat { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } - \\tilde { K } _ { X } \\right) \\tilde { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) X ^ { T } X \\right) } \\\\ & { \\leq \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { T } X \\right\\| _ { 2 } } \\\\ & { = O ( d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d ^ { - 1 } \\cdot d ) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 284, + 229, + 710, + 386 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "By combining the two approximations above we know that as $n , d , p \\to \\infty$ ", + "bbox": [ + 174, + 392, + 665, + 409 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/846c96f34183bd92dc3a78c781ffadc6b8b879cd9cd9cb9d73975cb99118a392.jpg", + "text": "$$\n| 2 V - \\mathbb { E } _ { \\pmb { x } } [ \\tilde { \\pmb { u } } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { \\pmb { u } } ^ { \\top } ] \\sigma ^ { 2 } | = O ( \\frac { \\log ^ { l } d } { d } ) 0 .\n$$", + "text_format": "latex", + "bbox": [ + 320, + 416, + 676, + 459 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Therefore the variance is given as ", + "bbox": [ + 173, + 467, + 397, + 482 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/8f32fa4b189ad2c8aa784797458621d67fe9031221b58ed73fef2d6f0b3593a3.jpg", + "text": "$$\n\\begin{array} { r l } & { 2 V \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ b _ { 0 } ^ { 4 } { \\boldsymbol x } ^ { T } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 2 } X ^ { T } { \\boldsymbol x } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\frac { 1 } { d } \\mathrm { t r } ( \\frac { 1 } { d } X ^ { T } X \\cdot ( \\frac { 1 } { d } X ^ { T } X + b _ { 0 } ^ { - 2 } b _ { 1 } ^ { 2 } I ) ^ { - 2 } ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( - \\frac { 1 } { 2 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 2 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 287, + 489, + 709, + 633 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "where $m$ is defined in the derivation of the bias term. ", + "bbox": [ + 173, + 640, + 521, + 655 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "C.9.6 PUTTING THINGS TOGETHER ", + "text_level": 1, + "bbox": [ + 174, + 672, + 436, + 688 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Recall the population risk is the sum of the bias and variance ", + "bbox": [ + 173, + 696, + 575, + 713 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/3facaf2ed650fba8a18513b67d5c668860efd0ce5075b96d09aac330886e0bde.jpg", + "text": "$$\n\\begin{array} { c } { { R r ^ { 2 } ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) } } \\\\ { { + \\sigma ^ { 2 } ( - \\frac { 1 } { 4 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) . } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 277, + 720, + 720, + 809 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Observe that the population risk is independent of $\\gamma _ { 2 }$ , i.e. double descent does not occur when the network is overparameterized via changing the width. In addition, the bias is monotonically increasing and upper-bounded by the null risk $r ^ { \\bar { 2 } }$ and lower-bounded by the bias of the least squares solution on the input features ${ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }$ , whereas the variance remains bounded for all $\\gamma _ { 1 } \\in ( 0 , \\infty )$ as long as $m > 0$ , i.e. $\\phi$ is nonlinear. ", + "bbox": [ + 173, + 821, + 826, + 896 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "D USEFUL LEMMAS ", + "text_level": 1, + "bbox": [ + 174, + 101, + 359, + 118 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Lemma 16. Given $K _ { W }$ from (46), define ", + "bbox": [ + 173, + 132, + 447, + 148 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/ae034d2ccf49a28b5f14f5cc55bb7efb8f3d80fb70632a984a4783f2fe0f0252.jpg", + "text": "$$\n\\tilde { K } _ { W } = r I _ { h } + s { \\bf 1 } _ { h } { \\bf 1 } _ { h } ^ { \\top } + t Q ,\n$$", + "text_format": "latex", + "bbox": [ + 405, + 155, + 591, + 174 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "where $Q \\in \\mathbb { R } ^ { h \\times h }$ with $Q _ { i \\neq j } \\ = \\ w _ { i } ^ { \\top } w _ { j }$ and $Q _ { i , i } ~ = ~ 0$ , and $r = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , s =$ $\\mathbb { E } [ \\phi ( G ) ] ^ { 2 }$ , $t = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 }$ . Then as $d , h \\to \\infty , \\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d a . s .$ .. ", + "bbox": [ + 173, + 181, + 826, + 223 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Proof. Consider the event where $\\mathcal { A } _ { \\epsilon } = \\big \\{ | \\| \\pmb { w } _ { i } \\| _ { 2 } - 1 | < \\epsilon , | \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } | < \\epsilon \\big \\}$ . Under event $\\mathcal { A } _ { \\epsilon }$ , for the diagonal term of the kernel matrix we have ", + "bbox": [ + 173, + 234, + 823, + 265 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/093ace88c911b29b6694e360e7207d76f52f55a0da3df8ff86c0b4ae705fe0df.jpg", + "text": "$$\n\\begin{array} { r l } & { ~ \\Big | [ K _ { W } ] _ { i i } - [ \\tilde { K } _ { W } ] _ { i i } \\Big | = \\Big | \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\Big ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\Big | } \\\\ & { = \\big | \\mathbb { E } [ \\phi ( \\| \\pmb { w } _ { i } \\| _ { 2 } G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\big | = O ( \\epsilon ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 289, + 270, + 705, + 320 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "And for off-diagonal term, by the decomposition introduced in Section B.3 we have ", + "bbox": [ + 173, + 324, + 722, + 339 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/654bfbbe8b3c7dbb706af222a370df5aefbddcc439911a23d215b55210c35966.jpg", + "text": "$$\n[ K _ { W } ] _ { i j } = \\mathbb { E } _ { \\boldsymbol { x } } \\Big [ \\phi ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) \\Big ] = \\| \\boldsymbol { w } _ { i } \\| _ { 2 } \\| \\boldsymbol { w } _ { j } \\| _ { 2 } + \\mathbb { E } _ { \\boldsymbol { x } } [ \\phi _ { \\bot } ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi _ { \\bot } ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) ] ,\n$$", + "text_format": "latex", + "bbox": [ + 217, + 345, + 743, + 372 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "hence $| [ K _ { W } ] _ { i j } - [ \\tilde { K } _ { W } ] _ { i j } | < ( \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } ) ^ { 2 }$ . Notice that $\\mathcal { A } _ { \\epsilon }$ holds a.s. for $\\epsilon = \\log ^ { c } d / \\sqrt { d }$ and large enough $c > 0$ ; we therefore have $\\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d .$ . ", + "bbox": [ + 173, + 380, + 825, + 422 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Lemma 17. Let ${ \\hat { \\boldsymbol { \\beta } } } ( t )$ be the solution to the gradient flow at time t defined in (103)(104). Then as $n , d \\to \\infty$ and $\\gamma _ { 1 } \\neq 1$ the following holds.: ", + "bbox": [ + 169, + 443, + 823, + 473 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/882d4cb8e4be17bb4e5db8176a16255e6736d70bc8f0aa83a215155c79205713.jpg", + "text": "$$\n\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O _ { P } ( 1 ) ; \\quad \\| \\boldsymbol { X } ^ { \\top } \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d ) .\n$$", + "text_format": "latex", + "bbox": [ + 318, + 479, + 679, + 500 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Proof. We consider $d < n$ for simplicity, and result for the other case follows in similar fashion. ", + "bbox": [ + 169, + 512, + 803, + 529 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "In this case $\\begin{array} { r } { \\hat { \\pmb { \\beta } } ( t ) = \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } } \\end{array}$ . From (Hastie et al., 2019, Corollary 1) we know that $\\| \\hat { \\pmb \\beta } ( \\infty ) \\| _ { 2 } = \\left\\| ( X X ^ { \\top } ) ^ { - 1 } X \\pmb y \\right\\| _ { 2 } = O ( 1 )$ for $\\gamma _ { 1 } < 1$ . Note that $\\begin{array} { r l r } { { \\| { \\cal I } - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\| _ { 2 } = } } \\end{array}$ $O ( 1 )$ for $t \\geq 0$ ; it follows that $\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O ( 1 )$ . ", + "bbox": [ + 173, + 534, + 826, + 589 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "For the second part, we utilize the SVD $\\boldsymbol { X } = \\boldsymbol { U \\Sigma V } ^ { \\top }$ , where $\\Sigma = [ \\hat { \\Sigma } ; 0 ]$ , $\\hat { \\Sigma } \\in \\mathbb { R } ^ { d \\times d }$ and $\\hat { \\Sigma } _ { i i } = \\lambda _ { i }$ . \nWe have $\\begin{array} { r } { \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) = U \\bar { \\Sigma } U ^ { \\top } } \\end{array}$ where $\\bar { \\Sigma } _ { i , i } = 1 - \\exp ( - t \\lambda _ { i } ^ { 2 } / n )$ if $i \\leq d$ and 0 otherwise. ", + "bbox": [ + 174, + 594, + 826, + 631 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/ebcdf9a75a1a71b9f2ae7aeb56343d6144c170cf1d700b8903231f442418092f.jpg", + "text": "$$\n\\begin{array} { r l } & { \\| X ^ { \\top } \\hat { \\pmb \\beta } ( t ) \\| _ { \\infty } = \\left\\| X ^ { \\top } \\left( I - \\exp ( - \\frac t n X X ^ { \\top } ) \\right) \\left( X X ^ { \\top } \\right) ^ { - 1 } X \\pmb y \\right\\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } U ^ { \\top } U \\bar { \\Sigma } U ^ { \\top } U \\hat { \\Sigma } ^ { - 2 } U ^ { \\top } U \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\| \\pmb y \\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) } \\\\ & { \\qquad \\leq \\| V \\| _ { \\infty } \\left\\| \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma \\right\\| _ { \\infty } \\| V ^ { \\top } \\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) \\overset { ( i ) } { \\leq } O _ { P } ( \\mathrm { p o l y } \\log d ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 184, + 637, + 779, + 765 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "where (i) follows from the concentration of the Gaussian maxima, and the fact that the law of $V$ is the Haar measure on $S O ( n )$ , and thus for any unit vector $_ z$ independent to $V$ , $V z$ is uniform on sphere and $\\| V z \\| _ { \\infty } = O ( \\log d / \\sqrt { d } )$ . We wherefore have ", + "bbox": [ + 174, + 767, + 825, + 814 + ], + "page_idx": 35 + }, + { + "type": "equation", + "img_path": "images/cd7f286c737bf8d4242cb3974f2c3c9b4041f5a47e25fddd0d8b59682b61ae56.jpg", + "text": "$$\n\\left\\| V \\right\\| _ { \\infty } = \\operatorname* { s u p } _ { z } { \\frac { \\left\\| V z \\right\\| _ { \\infty } } { \\left\\| z \\right\\| _ { \\infty } } } = O \\left( { \\frac { \\log d } { \\sqrt { d } } } \\right) { \\frac { \\left\\| z \\right\\| _ { 2 } } { \\left\\| z \\right\\| _ { \\infty } } } = O ( \\log d ) ,\n$$", + "text_format": "latex", + "bbox": [ + 310, + 820, + 687, + 856 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Note that this result also implies that $\\| \\pmb { y } - X ^ { \\top } \\pmb { \\beta } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d )$ . ", + "bbox": [ + 174, + 862, + 663, + 880 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Lemma 18. For weight matrices $W , W ^ { \\prime }$ satisfying $\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } \\ = \\ { \\cal O } ( { \\sqrt { n } } )$ , where $f ( X ) \\ =$ $\\phi ( X W ) \\mathbf { a }$ with fixed $a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}$ , given (A1)-(A3), the gradient of the empirical risk defined in (11) is Lipschitz w.r.t. $W$ in the Frobenius norm, i.e. ", + "bbox": [ + 173, + 102, + 826, + 148 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/49ab82c2bf97ce3826d2691396016681f6da33b1cc29cd818f4ed12d49305ab5.jpg", + "text": "$$\n\\left\\| \\frac { \\partial L ( X ; W ) } { \\partial W } - \\frac { \\partial L ( X ; W ^ { \\prime } ) } { \\partial W } \\right\\| _ { F } \\leq L \\left\\| W - W ^ { \\prime } \\right\\| _ { F } .\n$$", + "text_format": "latex", + "bbox": [ + 334, + 152, + 678, + 189 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "Proof. Denote $\\pmb { y } _ { 1 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 1 } ) \\pmb { a }$ and $\\pmb { y } _ { 2 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 2 } ) \\pmb { a }$ for $W _ { 1 } , W _ { 2 }$ satisfying the assumption above (which can be seen as a condition on the magnitude of training loss), we have ", + "bbox": [ + 173, + 199, + 825, + 231 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/675e07d8654bb7d135b3468c762772600803e9e009cc73105b701ba83d129d84.jpg", + "text": "$$\n\\begin{array} { r l } & { \\quad \\displaystyle \\left\\| \\frac { \\partial L ( W _ { 1 } ) } { \\partial W _ { 1 } } - \\frac { \\partial L ( W _ { 2 } ) } { \\partial W _ { 2 } } \\right\\| _ { F } } \\\\ & { = \\displaystyle \\left\\| \\frac { 1 } { n } X \\left[ ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right] - \\frac { 1 } { n } X \\left[ ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right] \\right\\| _ { F } } \\\\ & { \\leq \\displaystyle \\frac { 1 } { n } \\| X \\| _ { 2 } \\left\\| ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\leq \\ O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y _ { 2 } - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right\\| _ { F } } \\\\ & { \\quad + O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y - y _ { 2 } ) a ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) ) \\right\\| _ { F } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 243, + 233, + 751, + 407 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "We upper bound the two terms separately: ", + "bbox": [ + 174, + 416, + 450, + 431 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/822a260f5249dc605d01ded18fe6acd153cea9bb6810ecd8d302d194d47afb06.jpg", + "text": "$$\n\\begin{array} { r l } & { \\left\\| ( y _ { 2 } - y _ { 1 } ) { \\boldsymbol a } ^ { \\top } \\circ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) \\right\\| _ { 2 } \\overset { ( i ) } { \\leq } \\operatorname* { m a x } \\{ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) _ { i j } \\} \\left\\| y _ { 2 } - y _ { 1 } \\right\\| _ { 2 } \\left\\| { \\boldsymbol a } \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i i ) } { \\leq } O ( 1 ) \\left\\| \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) { \\boldsymbol a } - \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 2 } ) { \\boldsymbol a } \\right\\| _ { F } } \\\\ & { \\qquad \\overset { ( i i i ) } { \\leq } O ( 1 ) \\left\\| \\boldsymbol { X } \\right\\| _ { 2 } \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 194, + 435, + 766, + 521 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "where we applied the inequality $\\left\\| A \\circ B \\right\\| _ { F } \\leq \\operatorname* { m a x } \\{ | A _ { i j } | \\} \\left\\| B \\right\\| _ { F }$ in (i), boundedness of $\\phi ^ { \\prime }$ in (ii) and Lipschitzity of $\\phi$ in (iii). Similarly, for the second term ", + "bbox": [ + 174, + 525, + 828, + 554 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/25ddc124b105b561a4c9b4d1924e2305540087c765333d036e7014e8397d5684.jpg", + "text": "$$\n\\begin{array} { r l } & { \\qquad \\left\\| ( \\pmb { y } - \\pmb { y } _ { 2 } ) \\pmb { a } ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) ) \\right\\| _ { F } } \\\\ & { \\leq \\operatorname* { m a x } \\{ \\left| a _ { i } \\right| \\} \\left\\| \\pmb { y } - \\pmb { y } _ { 2 } \\right\\| _ { 2 } \\left\\| \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\overset { ( i ) } { \\leq } O ( 1 ) \\left\\| X \\right\\| _ { 2 } \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 313, + 556, + 683, + 630 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "where we used the assumption on the training loss and the Lipscthizity of $\\phi ^ { \\prime }$ in (i). Combining the two terms yields the desired result. ", + "bbox": [ + 174, + 631, + 823, + 660 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "Lemma 19. Under assumptions (A1-3) and the non-vanishing initialization, given that $\\parallel { \\pmb w } _ { i } ( t ) -$ ${ \\pmb w } _ { i } ( 0 ) \\| _ { 2 } = O ( d ^ { - 1 / 2 } )$ for all $i$ , then we have $\\| K ( t ) - K ( 0 ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )$ for some positive $\\epsilon ^ { \\prime } \\in \\Theta ( 1 )$ . ", + "bbox": [ + 173, + 705, + 825, + 752 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "Proof. Recall the definition of the NTK: ", + "bbox": [ + 173, + 762, + 442, + 776 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/359d9704ed4800e00673e164f155f1f948b69e32172bedc675cbb56736e59bf7.jpg", + "text": "$$\nK _ { i j } ( t ) = \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } = { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { j } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 199, + 779, + 761, + 821 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "or equivalently the matrix form ", + "bbox": [ + 173, + 825, + 382, + 840 + ], + "page_idx": 36 + }, + { + "type": "equation", + "img_path": "images/f1f4ccd7896642f7f848804fd8d50b3b87ee514c8d4f7696293622181fc7f372.jpg", + "text": "$$\nK ( t ) = X ^ { \\top } X \\circ \\frac { 1 } { h } [ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) ] .\n$$", + "text_format": "latex", + "bbox": [ + 338, + 843, + 658, + 873 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "At initialization, $\\mathbf { { x } } _ { i } ~ \\sim ~ N ( 0 , I _ { d } )$ and ${ \\pmb w } _ { k } ( 0 ) ~ \\sim ~ N ( 0 , d ^ { \\epsilon } I _ { d } )$ . Thus for fixed $\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }$ , by Gaussian anti-concentration we have $\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } | < \\log d \\ \\leq \\ O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )$ for some $\\epsilon _ { 1 } ~ > ~ 0$ . In addition, note that $\\lVert \\pmb { w } _ { k } ( t ) - \\pmb { w } _ { k } ( 0 ) \\rVert _ { 2 } ~ = ~ O ( d ^ { - 1 / 2 } )$ for all $k$ , and therefore for $i , j , k$ such that ", + "bbox": [ + 173, + 877, + 826, + 926 + ], + "page_idx": 36 + }, + { + "type": "text", + "text": "$| { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) | > O ( \\log d )$ and $| x _ { j } ^ { \\top } w _ { k } ( 0 ) | > O ( \\log d )$ , we know that $| \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } ( t ) ) \\phi ^ { \\prime } ( \\pmb { x } _ { j } ^ { \\top } \\pmb { w } _ { k } ( t ) ) -$ $\\phi ^ { \\prime } ( { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) \\phi ^ { \\prime } ( { \\pmb x } _ { j } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) | = = O ( d ^ { - 2 } )$ . ", + "bbox": [ + 174, + 101, + 825, + 137 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Given fixed $\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }$ , define $y _ { k } = \\mathbf { 1 } \\{ | x _ { i } ^ { \\top } w _ { k } | < \\log d \\}$ as the indicator variable that the $k$ -th neuron does not saturate. We know that $\\mathbb { E } [ y _ { k } ] = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )$ , and $\\mathrm { V a r } [ y _ { k } ] = \\mathbb { E } [ y _ { k } ^ { 2 } ] - \\mathbb { E } [ y _ { k } ] ^ { 2 } = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )$ . By Bernstein’s inequality ", + "bbox": [ + 174, + 143, + 826, + 189 + ], + "page_idx": 37 + }, + { + "type": "equation", + "img_path": "images/558150d5c3bf874f26be75b207fa7c9b3fb7025e9859b3afd27effc941de806c.jpg", + "text": "$$\n\\operatorname* { P r } \\left| \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } y _ { k } - \\mathbb { E } [ y _ { k } ] \\right| > \\varepsilon \\leq 2 \\exp \\left( - \\frac { h \\varepsilon ^ { 2 } } { 2 \\sigma ^ { 2 } + 2 \\varepsilon / 3 } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 316, + 195, + 681, + 239 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Setting ε = q $\\begin{array} { r } { \\varepsilon = \\sqrt { \\frac { c \\log h } { h ^ { 1 + \\epsilon _ { 2 } } } } } \\end{array}$ , we know that with probability at least $1 - h ^ { - c }$ , ", + "bbox": [ + 173, + 247, + 627, + 271 + ], + "page_idx": 37 + }, + { + "type": "equation", + "img_path": "images/b0f2e0ceafc2dbb7e687deebb7c3d3751653f0ad472c4afce719624c23246db7.jpg", + "text": "$$\n{ \\frac { 1 } { h } } \\sum _ { k = 1 } ^ { h } y _ { k } \\leq \\varepsilon + \\mathbb { E } [ y _ { k } ] = O \\left( { \\frac { \\mathrm { p o l y l o g } h } { h ^ { 1 / 2 + \\epsilon _ { 3 } } } } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 356, + 279, + 643, + 321 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Therefore, given $\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }$ and $\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { j } }$ , for large enough $c _ { 1 }$ with probability at least $1 - h ^ { - 3 }$ we have ", + "bbox": [ + 173, + 329, + 763, + 345 + ], + "page_idx": 37 + }, + { + "type": "equation", + "img_path": "images/4cfd4b7c8dbfa482781d8cb7d8385fcf6097f75a9a74645fbe5cc3bbd1c56379.jpg", + "text": "$$\n\\left| \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { j } ) - \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { j } ) \\right| = O ( h ^ { 1 / 2 - \\epsilon _ { 4 } } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 191, + 352, + 767, + 396 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "in which we utilized the boundedness of $\\phi ^ { \\prime }$ . Taking union bound over $d ^ { 2 }$ elements in the random feature matrix yields ", + "bbox": [ + 169, + 402, + 825, + 433 + ], + "page_idx": 37 + }, + { + "type": "equation", + "img_path": "images/4a7ba69b32b73ae7335de1fd33f0fff428bed75d62d2e5bb24831a8a512d0fa8.jpg", + "text": "$$\n\\begin{array} { r l } & { ~ \\| K ( t ) - K ( 0 ) \\| _ { 2 } } \\\\ & { = \\left\\| X ^ { \\top } X \\circ \\frac { 1 } { h } \\left[ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right] \\right\\| } \\\\ & { \\leq \\frac { 1 } { h } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\operatorname* { m a x } \\left\\{ \\left| \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right| _ { i j } \\right\\} } \\\\ & { \\leq \\frac { 1 } { h } O ( d ) O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 217, + 436, + 781, + 554 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Using the exact same argument, one can derive that $\\| { \\pmb u } _ { N N } ( { \\hat { \\pmb x } } ) - { \\pmb u } _ { N T K } ( { \\hat { \\pmb x } } ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )$ , the proof of which we omit. ", + "bbox": [ + 174, + 568, + 825, + 598 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "E ADDITIONAL RESULTS ", + "text_level": 1, + "bbox": [ + 174, + 647, + 398, + 664 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "E.1 RISK OF ReLU NETWORK UNDER SYMMETRIC DATA ", + "bbox": [ + 173, + 678, + 583, + 694 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "If the dataset is symmetric, that is ", + "bbox": [ + 174, + 704, + 397, + 719 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "then population risk of the gradient flow solution can be given explicitly for certain nonlinearities: ", + "bbox": [ + 171, + 750, + 813, + 765 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Proposition 20. Given (A1-3)(A5), if the nonlinearity satisfies $\\phi ^ { \\prime } ( { \\pmb x } ) + \\phi ^ { \\prime } ( - { \\pmb x } ) = C$ for constant $C$ then as $n , d , h \\infty$ ", + "bbox": [ + 173, + 768, + 823, + 797 + ], + "page_idx": 37 + }, + { + "type": "equation", + "img_path": "images/73b5af9535895247f602303123b3033fbae9c99e5e9d7bd02edc05404f67e048.jpg", + "text": "$$\nR _ { ( \\gamma _ { 1 } < 0 . 5 ) } ( \\hat { f } ) \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } ( \\hat { f } ) = ( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } ) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 223, + 804, + 740, + 838 + ], + "page_idx": 37 + }, + { + "type": "text", + "text": "Note that the requirement on the nonlinearity holds for ReLU and SoftPlus. This expression is again independent to $\\gamma _ { 2 }$ and aligns with the experimental results in Figure 9 (we only plot the bias component for verification). In addition, the bias is upper-bounded by the null risk for all $\\gamma _ { 1 }$ . We remark that the symmetry assumption does not hold for i.i.d. samples from symmetric distributions, and Figure 9 demonstrates that the additional condition alters the risk. ", + "bbox": [ + 173, + 853, + 826, + 924 + ], + "page_idx": 37 + }, + { + "type": "image", + "img_path": "images/21277368c4f422787876a9038b601c0d6dee6f086e04a2580c02aae5b0b456d2.jpg", + "image_caption": [ + "Figure 9: Bias of two-layer ReLU networks with optimized first layer under Gaussian data and linear teacher. Individual dotted lines correspond to different $\\gamma _ { 2 }$ (from 0.2 to 2) which is independent to the risk. (a) Vanishing initialization. The bias under symmetric data is predicted by Proposition 20. (b) Non-vanishing initialization. The red and blue lines represent models optimized from i.i.d. and symmetric initialization, respectively. The bias for symmetric initialization is predicted by Theorem 8. " + ], + "image_footnote": [], + "bbox": [ + 191, + 101, + 803, + 281 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "Proof. Without loss of generality assume $X = [ X _ { 0 } , - X _ { 0 } ]$ . Then by (100) we have ", + "bbox": [ + 174, + 380, + 720, + 397 + ], + "page_idx": 38 + }, + { + "type": "equation", + "img_path": "images/33ac8cdaea85e29ecbb901d809d84e70a09bb3e446d75ca742d8bece5bc87125.jpg", + "text": "$$\n\\frac { \\partial \\pmb { w } _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) + h _ { 0 } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) \\pmb { x } _ { i } \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 269, + 400, + 727, + 443 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "and the flow for ${ \\pmb w } _ { - }$ follows from symmetry. In this case one can show that from exact zero initialization, for nonlinearity satisfying $\\phi ( x ) - \\phi ( - x ) = x$ , such as ReLU and SoftPlus, ", + "bbox": [ + 174, + 445, + 825, + 474 + ], + "page_idx": 38 + }, + { + "type": "equation", + "img_path": "images/7c767067d6d62bb4c5d252f691456852e01d21455de7b6eac1d89c8d69a2256b.jpg", + "text": "$$\n\\frac { \\partial ( { \\pmb w } _ { + } ) } { \\partial t } + \\frac { \\partial ( { \\pmb w } _ { - } ) } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) + h _ { 0 } \\phi ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) \\Big ) ( \\phi ^ { \\prime } ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) - \\phi ^ { \\prime } ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) ) { \\pmb x } _ { i } \\right] = 0\n$$", + "text_format": "latex", + "bbox": [ + 181, + 478, + 839, + 522 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "And therefore the gradient flow of ${ \\pmb w } _ { + }$ is ", + "bbox": [ + 176, + 539, + 442, + 554 + ], + "page_idx": 38 + }, + { + "type": "equation", + "img_path": "images/b7c8baedb48e72bc23843067a4fb0641ddca959153ac7b35a4dc5f49c656340d.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle \\frac { \\partial w _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\big ( \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + \\phi ^ { \\prime } \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } w _ { + } ^ { \\top } x _ { i } \\Big ) x _ { i } \\Big ] = \\frac { 1 } { 2 n _ { 0 } } X _ { 0 } y _ { 0 } - \\frac { 1 } { 2 n _ { 0 } } h _ { 0 } X _ { 0 } X _ { 0 } ^ { \\top } w _ { + } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 210, + 558, + 787, + 685 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "The flow of ${ \\pmb w } _ { - }$ follows from symmetry. Solving for the stationary points (i.e. gradient becomes zero), it the clear that ", + "bbox": [ + 174, + 686, + 823, + 715 + ], + "page_idx": 38 + }, + { + "type": "equation", + "img_path": "images/e60181587ccd6e28ebe9de98be2cab5499ff2872bea90a945f533c3a143331e7.jpg", + "text": "$$\n\\pmb { w } _ { + } ^ { ( t = \\infty ) } = - \\pmb { w } _ { - } ^ { ( t = \\infty ) } = \\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { 1 } { h _ { 0 } } ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } , } & { \\gamma _ { 1 } < 0 . 5 , } \\\\ { \\displaystyle } \\\\ { \\displaystyle \\frac { 1 } { h _ { 0 } } X ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } , } & { \\gamma _ { 1 } > 0 . 5 . } \\end{array} \\right.\n$$", + "text_format": "latex", + "bbox": [ + 308, + 717, + 689, + 795 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "And hence the asymptotic risk is ", + "bbox": [ + 173, + 796, + 390, + 810 + ], + "page_idx": 38 + }, + { + "type": "equation", + "img_path": "images/a984cf11779b8fa79f7dec70b5480784869009784eaab7ffb7690d4dc8b6472f.jpg", + "text": "$$\nR _ { ( \\gamma _ { 1 } < 0 . 5 ) } \\to \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } = \\left( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } \\right) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 263, + 814, + 735, + 849 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "The same conclusion holds for vanishing initialization if we assume that the trajectory stays close to that of exact zero initialization. Note that although the prediction aligns well with the experimental results, the argument in Theorem 7 does not directly apply due to the undefined derivative of ReLU at the origin, and thus this result is not rigorously justified. □ ", + "bbox": [ + 174, + 858, + 825, + 916 + ], + "page_idx": 38 + }, + { + "type": "text", + "text": "F EXPERIMENT SETUP ", + "text_level": 1, + "bbox": [ + 176, + 102, + 379, + 118 + ], + "page_idx": 39 + }, + { + "type": "text", + "text": "Optimizing the Second Layer. We compute the minimum-norm solution by directly solving the pseudo-inverse. We set $n = 1 0 0 0$ and vary $\\gamma _ { 1 } , \\gamma _ { 2 }$ from 0.1 to 3. The linear teacher model $F ( { \\pmb x } ) = { \\pmb x } ^ { \\top } \\beta$ is fixed as $\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }$ . For each $( \\gamma _ { 1 } , \\gamma _ { 2 } )$ we average across 50 random draws of data. ", + "bbox": [ + 174, + 133, + 825, + 191 + ], + "page_idx": 39 + }, + { + "type": "text", + "text": "Optimizing the First Layer. For both initializations, we use gradient descent with small step size $( \\eta = 0 . 1 )$ ) and train the model for minimally 25000 steps and till $\\| \\nabla _ { W } f ( X , W ) \\| _ { F } ^ { 2 } < 1 0 ^ { - 6 }$ . We fix $n = 3 2 0$ and vary $\\gamma _ { 1 } , \\gamma _ { 2 }$ from 0.1 to 3 with the same linear teacher model $\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }$ . The risk is averaged across 20 models trained from different initializations. ", + "bbox": [ + 173, + 207, + 825, + 267 + ], + "page_idx": 39 + } +] \ No newline at end of file diff --git a/parse/train/H1gBsgBYwH/H1gBsgBYwH_middle.json b/parse/train/H1gBsgBYwH/H1gBsgBYwH_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..e0eae1a7766e47e25b597aa51f052d1ed4cb3304 --- /dev/null +++ b/parse/train/H1gBsgBYwH/H1gBsgBYwH_middle.json @@ -0,0 +1,112224 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 507, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 508, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 508, + 97 + ], + "score": 1.0, + "content": "GENERALIZATION OF TWO-LAYER NEURAL NET-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 389, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 389, + 118 + ], + "score": 1.0, + "content": "WORKS: AN ASYMPTOTIC VIEWPOINT", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 111, + 133, + 512, + 180 + ], + "lines": [ + { + "bbox": [ + 113, + 133, + 479, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 133, + 146, + 147 + ], + "score": 1.0, + "content": "Jimmy", + "type": "text" + }, + { + "bbox": [ + 146, + 134, + 168, + 146 + ], + "score": 0.85, + "content": "\\mathbf { B a } ^ { 1 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 133, + 362, + 147 + ], + "score": 1.0, + "content": ", Murat A. Erdogdu1,2, Taiji Suzuki3,4, Denny", + "type": "text" + }, + { + "bbox": [ + 362, + 134, + 393, + 146 + ], + "score": 0.84, + "content": "\\mathbf { W _ { u } } 1 , 2 , 4", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 133, + 479, + 147 + ], + "score": 1.0, + "content": ", Tianzong Zhang2,5", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 511, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 511, + 159 + ], + "score": 1.0, + "content": "University of Toronto1, Vector Institute2, University of Tokyo3, RIKEN AIP4, Tsinghua University5", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 113, + 158, + 332, + 170 + ], + "spans": [ + { + "bbox": [ + 113, + 158, + 332, + 170 + ], + "score": 1.0, + "content": "{jba,erdogdu,dennywu}@cs.toronto.edu,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 111, + 168, + 437, + 182 + ], + "spans": [ + { + "bbox": [ + 111, + 168, + 437, + 182 + ], + "score": 1.0, + "content": "taiji@mist.i.u-tokyo.ac.jp, ztz16@mails.tsinghua.edu.cn", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 278, + 209, + 333, + 221 + ], + "lines": [ + { + "bbox": [ + 276, + 209, + 335, + 222 + ], + "spans": [ + { + "bbox": [ + 276, + 209, + 335, + 222 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 232, + 468, + 375 + ], + "lines": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "spans": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "score": 1.0, + "content": "This paper investigates the generalization properties of two-layer neural networks", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 361, + 255 + ], + "score": 1.0, + "content": "in high-dimensions, i.e. when the number of samples", + "type": "text" + }, + { + "bbox": [ + 361, + 245, + 368, + 253 + ], + "score": 0.66, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 243, + 407, + 255 + ], + "score": 1.0, + "content": ", features", + "type": "text" + }, + { + "bbox": [ + 407, + 244, + 414, + 253 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 243, + 470, + 255 + ], + "score": 1.0, + "content": ", and neurons", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 254, + 469, + 267 + ], + "spans": [ + { + "bbox": [ + 142, + 254, + 149, + 264 + ], + "score": 0.77, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 254, + 469, + 267 + ], + "score": 1.0, + "content": "tend to infinity at the same rate. Specifically, we derive the exact population", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 265, + 470, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 470, + 278 + ], + "score": 1.0, + "content": "risk of the unregularized least squares regression problem with two-layer neural", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "networks when either the first or the second layer is trained using a gradient flow", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 288, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 288, + 469, + 299 + ], + "score": 1.0, + "content": "under different initialization setups. When only the second layer coefficients are", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "optimized, we recover the double descent phenomenon: a cusp in the population", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 206, + 321 + ], + "score": 1.0, + "content": "risk appears at", + "type": "text" + }, + { + "bbox": [ + 206, + 309, + 236, + 319 + ], + "score": 0.9, + "content": "h \\approx n", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "and further overparameterization decreases the risk. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 470, + 332 + ], + "score": 1.0, + "content": "contrast, when the first layer weights are optimized, we highlight how different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 330, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 469, + 343 + ], + "score": 1.0, + "content": "scales of initialization lead to different inductive bias, and show that the resulting", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "score": 1.0, + "content": "risk is independent of overparameterization. Our theoretical and experimental", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "results suggest that previously studied model setups that provably give rise to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 364, + 447, + 376 + ], + "spans": [ + { + "bbox": [ + 141, + 364, + 447, + 376 + ], + "score": 1.0, + "content": "double descent might not translate to optimizing two-layer neural networks.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 206, + 406 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 208, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 208, + 408 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 505, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "In modern neural networks, the number of parameters can easily exceed the number of training", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "samples, yet in many circumstances, there is little sign of overfitting even in the absence of explicit", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "regularization (Zhang et al., 2016). This phenomenon is usually explained by the interplay between", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "the model architecture and the optimization method. Existing works have analyzed the implicit", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "regularization of gradient descent on simple models (Gunasekar et al., 2018; Ji and Telgarsky, 2018),", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 471, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 507, + 486 + ], + "score": 1.0, + "content": "and provided generalization guarantees (Arora et al., 2018; Bartlett et al., 2017; Dziugaite and Roy,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 483, + 302, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 302, + 495 + ], + "score": 1.0, + "content": "2017) that align with the empirical observations.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "Recently, a series of works highlighted the implicit regularization of interpolators in the overpa-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "rameterized regime (Belkin et al., 2018; Spigler et al., 2018; Geiger et al., 2018; Advani and Saxe,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "2017). Specifically, a second decrease in the population risk is observed when the model is further", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 545 + ], + "score": 1.0, + "content": "overparameterized beyond the interpolation limit, i.e. when the model achieves zero training error.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 544, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 506, + 556 + ], + "score": 1.0, + "content": "This phenomenon is known as double descent, and can be precisely quantified for certain linear", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "models (Hastie et al., 2019; Mei and Montanari, 2019; Belkin et al., 2019; Bartlett et al., 2019;", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 566, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 577 + ], + "score": 1.0, + "content": "Xu and Hsu, 2019). Among the recent works, Hastie et al. (2019) and Mei and Montanari (2019)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "explicitly derived the population risk of linear regression and random features regression models in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 587, + 333, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 333, + 600 + ], + "score": 1.0, + "content": "high dimensions using tools from random matrix theory.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 605, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "However, there is still a gap between the practical benefit of overparameterization and the recently", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "proved double descent phenomenon, which is typically established under models that exhibits the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "following structure: the trained model solves a linear inverse problem, and the “cusp” in the risk", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "score": 1.0, + "content": "arises from the instability of the inverse at the interpolation threshold. Moreover, given a dataset", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 140, + 660 + ], + "score": 1.0, + "content": "or fixed", + "type": "text" + }, + { + "bbox": [ + 140, + 649, + 157, + 660 + ], + "score": 0.84, + "content": "n , d .", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 648, + 505, + 660 + ], + "score": 1.0, + "content": ", the number of parameters in the linear regression model is also fixed, i.e. the level of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "overparameterization cannot be altered. It is therefore unclear if the trend persists in the optimization", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "of more complex models, for instance in two-layer neural networks where overparameterization can", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 682, + 293, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 293, + 694 + ], + "score": 1.0, + "content": "be controlled simply by adding more neurons.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "In this work, we analyze the generalization properties of two-layer neural networks in the unreg-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "ularized least squares regression setting and examine the presence/absence of the double descent", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 471, + 733 + ], + "score": 1.0, + "content": "phenomenon. We consider the proportional asymptotic limit where the number of samples", + "type": "text" + }, + { + "bbox": [ + 471, + 723, + 478, + 731 + ], + "score": 0.69, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 721, + 505, + 733 + ], + "score": 1.0, + "content": ", input", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 507, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 508, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 508, + 97 + ], + "score": 1.0, + "content": "GENERALIZATION OF TWO-LAYER NEURAL NET-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 389, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 389, + 118 + ], + "score": 1.0, + "content": "WORKS: AN ASYMPTOTIC VIEWPOINT", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 111, + 133, + 512, + 180 + ], + "lines": [ + { + "bbox": [ + 113, + 133, + 479, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 133, + 146, + 147 + ], + "score": 1.0, + "content": "Jimmy", + "type": "text" + }, + { + "bbox": [ + 146, + 134, + 168, + 146 + ], + "score": 0.85, + "content": "\\mathbf { B a } ^ { 1 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 133, + 362, + 147 + ], + "score": 1.0, + "content": ", Murat A. Erdogdu1,2, Taiji Suzuki3,4, Denny", + "type": "text" + }, + { + "bbox": [ + 362, + 134, + 393, + 146 + ], + "score": 0.84, + "content": "\\mathbf { W _ { u } } 1 , 2 , 4", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 133, + 479, + 147 + ], + "score": 1.0, + "content": ", Tianzong Zhang2,5", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 511, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 511, + 159 + ], + "score": 1.0, + "content": "University of Toronto1, Vector Institute2, University of Tokyo3, RIKEN AIP4, Tsinghua University5", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 113, + 158, + 332, + 170 + ], + "spans": [ + { + "bbox": [ + 113, + 158, + 332, + 170 + ], + "score": 1.0, + "content": "{jba,erdogdu,dennywu}@cs.toronto.edu,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 111, + 168, + 437, + 182 + ], + "spans": [ + { + "bbox": [ + 111, + 168, + 437, + 182 + ], + "score": 1.0, + "content": "taiji@mist.i.u-tokyo.ac.jp, ztz16@mails.tsinghua.edu.cn", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 111, + 133, + 511, + 182 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 209, + 333, + 221 + ], + "lines": [ + { + "bbox": [ + 276, + 209, + 335, + 222 + ], + "spans": [ + { + "bbox": [ + 276, + 209, + 335, + 222 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 232, + 468, + 375 + ], + "lines": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "spans": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "score": 1.0, + "content": "This paper investigates the generalization properties of two-layer neural networks", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 361, + 255 + ], + "score": 1.0, + "content": "in high-dimensions, i.e. when the number of samples", + "type": "text" + }, + { + "bbox": [ + 361, + 245, + 368, + 253 + ], + "score": 0.66, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 243, + 407, + 255 + ], + "score": 1.0, + "content": ", features", + "type": "text" + }, + { + "bbox": [ + 407, + 244, + 414, + 253 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 243, + 470, + 255 + ], + "score": 1.0, + "content": ", and neurons", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 254, + 469, + 267 + ], + "spans": [ + { + "bbox": [ + 142, + 254, + 149, + 264 + ], + "score": 0.77, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 254, + 469, + 267 + ], + "score": 1.0, + "content": "tend to infinity at the same rate. Specifically, we derive the exact population", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 265, + 470, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 470, + 278 + ], + "score": 1.0, + "content": "risk of the unregularized least squares regression problem with two-layer neural", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "networks when either the first or the second layer is trained using a gradient flow", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 288, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 288, + 469, + 299 + ], + "score": 1.0, + "content": "under different initialization setups. When only the second layer coefficients are", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "optimized, we recover the double descent phenomenon: a cusp in the population", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 206, + 321 + ], + "score": 1.0, + "content": "risk appears at", + "type": "text" + }, + { + "bbox": [ + 206, + 309, + 236, + 319 + ], + "score": 0.9, + "content": "h \\approx n", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "and further overparameterization decreases the risk. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 470, + 332 + ], + "score": 1.0, + "content": "contrast, when the first layer weights are optimized, we highlight how different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 330, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 469, + 343 + ], + "score": 1.0, + "content": "scales of initialization lead to different inductive bias, and show that the resulting", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "score": 1.0, + "content": "risk is independent of overparameterization. Our theoretical and experimental", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "results suggest that previously studied model setups that provably give rise to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 364, + 447, + 376 + ], + "spans": [ + { + "bbox": [ + 141, + 364, + 447, + 376 + ], + "score": 1.0, + "content": "double descent might not translate to optimizing two-layer neural networks.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 13, + "bbox_fs": [ + 141, + 232, + 470, + 376 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 206, + 406 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 208, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 208, + 408 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 505, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "In modern neural networks, the number of parameters can easily exceed the number of training", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "samples, yet in many circumstances, there is little sign of overfitting even in the absence of explicit", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "regularization (Zhang et al., 2016). This phenomenon is usually explained by the interplay between", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "the model architecture and the optimization method. Existing works have analyzed the implicit", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "regularization of gradient descent on simple models (Gunasekar et al., 2018; Ji and Telgarsky, 2018),", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 471, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 507, + 486 + ], + "score": 1.0, + "content": "and provided generalization guarantees (Arora et al., 2018; Bartlett et al., 2017; Dziugaite and Roy,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 483, + 302, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 302, + 495 + ], + "score": 1.0, + "content": "2017) that align with the empirical observations.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 416, + 507, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "Recently, a series of works highlighted the implicit regularization of interpolators in the overpa-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "rameterized regime (Belkin et al., 2018; Spigler et al., 2018; Geiger et al., 2018; Advani and Saxe,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "2017). Specifically, a second decrease in the population risk is observed when the model is further", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 545 + ], + "score": 1.0, + "content": "overparameterized beyond the interpolation limit, i.e. when the model achieves zero training error.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 544, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 506, + 556 + ], + "score": 1.0, + "content": "This phenomenon is known as double descent, and can be precisely quantified for certain linear", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "models (Hastie et al., 2019; Mei and Montanari, 2019; Belkin et al., 2019; Bartlett et al., 2019;", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 566, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 577 + ], + "score": 1.0, + "content": "Xu and Hsu, 2019). Among the recent works, Hastie et al. (2019) and Mei and Montanari (2019)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "explicitly derived the population risk of linear regression and random features regression models in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 587, + 333, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 333, + 600 + ], + "score": 1.0, + "content": "high dimensions using tools from random matrix theory.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 500, + 506, + 600 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 605, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "However, there is still a gap between the practical benefit of overparameterization and the recently", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "proved double descent phenomenon, which is typically established under models that exhibits the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "following structure: the trained model solves a linear inverse problem, and the “cusp” in the risk", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "score": 1.0, + "content": "arises from the instability of the inverse at the interpolation threshold. Moreover, given a dataset", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 140, + 660 + ], + "score": 1.0, + "content": "or fixed", + "type": "text" + }, + { + "bbox": [ + 140, + 649, + 157, + 660 + ], + "score": 0.84, + "content": "n , d .", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 648, + 505, + 660 + ], + "score": 1.0, + "content": ", the number of parameters in the linear regression model is also fixed, i.e. the level of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "overparameterization cannot be altered. It is therefore unclear if the trend persists in the optimization", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "of more complex models, for instance in two-layer neural networks where overparameterization can", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 682, + 293, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 293, + 694 + ], + "score": 1.0, + "content": "be controlled simply by adding more neurons.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 605, + 506, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "In this work, we analyze the generalization properties of two-layer neural networks in the unreg-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "ularized least squares regression setting and examine the presence/absence of the double descent", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 471, + 733 + ], + "score": 1.0, + "content": "phenomenon. We consider the proportional asymptotic limit where the number of samples", + "type": "text" + }, + { + "bbox": [ + 471, + 723, + 478, + 731 + ], + "score": 0.69, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 721, + 505, + 733 + ], + "score": 1.0, + "content": ", input", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 141, + 95 + ], + "score": 1.0, + "content": "features", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 142, + 83, + 148, + 93 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 149, + 83, + 204, + 95 + ], + "score": 1.0, + "content": ", and neurons", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 204, + 83, + 212, + 92 + ], + "score": 0.8, + "content": "h", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 212, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "tend to infinity at the same rate, under which overparameterization cor-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 248, + 107 + ], + "score": 1.0, + "content": "responds to increasing the limit of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 249, + 93, + 267, + 106 + ], + "score": 0.91, + "content": "h / n", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 267, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "(network “width”). This regime is particularly interesting", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 191, + 117 + ], + "score": 1.0, + "content": "because even though", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 191, + 106, + 224, + 115 + ], + "score": 0.88, + "content": "n \\to \\infty", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 224, + 104, + 506, + 117 + ], + "score": 1.0, + "content": ", the empirical risk is not equivalent to the population risk. In addition,", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 180, + 127 + ], + "score": 1.0, + "content": "the joint scaling of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 181, + 116, + 207, + 127 + ], + "score": 0.91, + "content": "n , d , h", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 208, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "is parallel to the practical choice of model architectures, where it is common", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "to train a larger network when the number of samples and input features are larger. Following Hastie", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "et al. (2019), we assume unit Gaussian input and noisy linear observations, and analytically derive", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "the population risk of the solution of gradient flow on either the first or the second layer parameters", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 271, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 271, + 171 + ], + "score": 1.0, + "content": "when the flow is initialized close to zero.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 699, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 170 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 141, + 95 + ], + "score": 1.0, + "content": "features", + "type": "text" + }, + { + "bbox": [ + 142, + 83, + 148, + 93 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 83, + 204, + 95 + ], + "score": 1.0, + "content": ", and neurons", + "type": "text" + }, + { + "bbox": [ + 204, + 83, + 212, + 92 + ], + "score": 0.8, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "tend to infinity at the same rate, under which overparameterization cor-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 248, + 107 + ], + "score": 1.0, + "content": "responds to increasing the limit of", + "type": "text" + }, + { + "bbox": [ + 249, + 93, + 267, + 106 + ], + "score": 0.91, + "content": "h / n", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "(network “width”). This regime is particularly interesting", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 191, + 117 + ], + "score": 1.0, + "content": "because even though", + "type": "text" + }, + { + "bbox": [ + 191, + 106, + 224, + 115 + ], + "score": 0.88, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 104, + 506, + 117 + ], + "score": 1.0, + "content": ", the empirical risk is not equivalent to the population risk. In addition,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 180, + 127 + ], + "score": 1.0, + "content": "the joint scaling of", + "type": "text" + }, + { + "bbox": [ + 181, + 116, + 207, + 127 + ], + "score": 0.91, + "content": "n , d , h", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "is parallel to the practical choice of model architectures, where it is common", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "to train a larger network when the number of samples and input features are larger. Following Hastie", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "et al. (2019), we assume unit Gaussian input and noisy linear observations, and analytically derive", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "the population risk of the solution of gradient flow on either the first or the second layer parameters", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 271, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 271, + 171 + ], + "score": 1.0, + "content": "when the flow is initialized close to zero.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 176, + 341, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 342, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 342, + 190 + ], + "score": 1.0, + "content": "Our findings can be summarized as follows (see Figure 1):", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 132, + 198, + 361, + 352 + ], + "lines": [ + { + "bbox": [ + 133, + 198, + 360, + 210 + ], + "spans": [ + { + "bbox": [ + 133, + 198, + 360, + 210 + ], + "score": 1.0, + "content": "• When only the second layer is optimized, we derive the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 209, + 360, + 221 + ], + "spans": [ + { + "bbox": [ + 141, + 209, + 360, + 221 + ], + "score": 1.0, + "content": "risk in its bias-variance decomposition and demonstrate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 219, + 338, + 232 + ], + "spans": [ + { + "bbox": [ + 141, + 219, + 338, + 232 + ], + "score": 1.0, + "content": "the presence of the double descent phenomenon.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 136, + 235, + 362, + 249 + ], + "spans": [ + { + "bbox": [ + 136, + 235, + 362, + 249 + ], + "score": 1.0, + "content": "• When the first layer is optimized, we compare two so-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 248, + 361, + 259 + ], + "spans": [ + { + "bbox": [ + 142, + 248, + 361, + 259 + ], + "score": 1.0, + "content": "lutions of gradient flow from different scales of initial-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 258, + 360, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 360, + 270 + ], + "score": 1.0, + "content": "ization, which we term as vanishing and non-vanishing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 270, + 360, + 281 + ], + "spans": [ + { + "bbox": [ + 141, + 270, + 360, + 281 + ], + "score": 1.0, + "content": "initialization, and show in both cases the population", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 280, + 319, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 319, + 292 + ], + "score": 1.0, + "content": "risk is independent to overparameterization.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 133, + 297, + 360, + 308 + ], + "spans": [ + { + "bbox": [ + 133, + 297, + 360, + 308 + ], + "score": 1.0, + "content": "• For the vanishing initialization, we show that the risk", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 307, + 361, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 307, + 361, + 320 + ], + "score": 1.0, + "content": "of the gradient flow solution is asymptotically close to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 318, + 361, + 330 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 361, + 330 + ], + "score": 1.0, + "content": "that of a rank-1 model. For non-vanishing initializa-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 330, + 362, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 330, + 362, + 340 + ], + "score": 1.0, + "content": "tion, we show that the gradient flow solution is well-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 341, + 352, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 352, + 352 + ], + "score": 1.0, + "content": "approximated by a kernel model and derive the risk.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 369, + 177, + 502, + 277 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 369, + 177, + 502, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 369, + 177, + 502, + 277 + ], + "spans": [ + { + "bbox": [ + 369, + 177, + 502, + 277 + ], + "score": 0.969, + "type": "image", + "image_path": "fdb2f7294504073f6249cfffccb6993d35f35a22379caec2e8d08b45a4316c5a.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 369, + 177, + 502, + 227.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 369, + 227.0, + 502, + 277.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 367, + 284, + 505, + 365 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 367, + 284, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 367, + 284, + 505, + 295 + ], + "score": 1.0, + "content": "Figure 1: Illustration of the double", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 368, + 295, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 368, + 295, + 505, + 305 + ], + "score": 1.0, + "content": "descent risk curve in two-layer linear", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 367, + 304, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 367, + 304, + 405, + 315 + ], + "score": 1.0, + "content": "networks", + "type": "text" + }, + { + "bbox": [ + 405, + 305, + 445, + 314 + ], + "score": 0.79, + "content": "( \\mathrm { S N R } = 1 6 ", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 304, + 505, + 315 + ], + "score": 1.0, + "content": "). Brighter color", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 367, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 367, + 314, + 428, + 325 + ], + "score": 1.0, + "content": "indicates larger", + "type": "text" + }, + { + "bbox": [ + 428, + 314, + 469, + 325 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } = d / \\bar { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 314, + 505, + 325 + ], + "score": 1.0, + "content": ". Double", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 367, + 325, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 367, + 325, + 505, + 335 + ], + "score": 1.0, + "content": "descent is observed when the second", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 368, + 334, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 368, + 334, + 505, + 345 + ], + "score": 1.0, + "content": "layer coefficients are optimized (main", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 367, + 344, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 367, + 344, + 505, + 356 + ], + "score": 1.0, + "content": "figure), but not when the first layer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 367, + 354, + 493, + 366 + ], + "spans": [ + { + "bbox": [ + 367, + 354, + 493, + 366 + ], + "score": 1.0, + "content": "weights are optimized (subfigure).", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27.5 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 108, + 367, + 207, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 209, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 209, + 380 + ], + "score": 1.0, + "content": "1.1 RELATED WORKS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 388, + 505, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "Global Convergence of Two-layer Networks. A plethora of recent works have explored the global", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 399, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 506, + 411 + ], + "score": 1.0, + "content": "convergence of shallow neural networks. Mei et al. (2018; 2019); Chizat and Bach (2018a); Rotskoff", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "and Vanden-Eijnden (2018); Sirignano and Spiliopoulos (2018); Nitanda and Suzuki (2017) studied", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 308, + 433 + ], + "score": 1.0, + "content": "the mean-field limit where the number of neurons", + "type": "text" + }, + { + "bbox": [ + 309, + 422, + 342, + 432 + ], + "score": 0.91, + "content": "h \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 422, + 467, + 433 + ], + "score": 1.0, + "content": "and the second layer scaled by", + "type": "text" + }, + { + "bbox": [ + 468, + 421, + 484, + 433 + ], + "score": 0.89, + "content": "1 / h", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 422, + 505, + 433 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 432, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 445 + ], + "score": 1.0, + "content": "established correspondence between the main-particle limit of gradient descent and Wasserstein", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "gradient flow to demonsrate global convergence. On the other hand, Jacot et al. (2018); Du et al.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "(2018); Oymak and Soltanolkotabi (2019); Allen-Zhu et al. (2018b); Song and Yang (2019) considered", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "a different scaling and showed that gradient descent on overparameterized models converges to global", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "minimizer at a linear rate; key to these results is an observation that optimization via gradient descent", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 487, + 467, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 467, + 499 + ], + "score": 1.0, + "content": "is asymptotically equivalent to kernel regression with respect to the neural tangent kernel.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "Active vs. Lazy Training. Following Chizat and Bach (2018b), we refer to the two aforementioned", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "scalings as the active and lazy (kernel) regime. It has been observed that different regimes lead to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "contrasting inductive biases. Williams et al. (2019); Woodworth et al. (2019); Li et al. (2017) showed", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 540, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 551 + ], + "score": 1.0, + "content": "that for certain two-layer network or overparameterized linear model, the scale of initialization", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "controls the implicit regularization of gradient descent (from sparse to smooth solution). In the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "student-teacher setup (Tian, 2017; Zhong et al., 2017), Ghorbani et al. (2019b;a) showed that kernel", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "models in high dimensions perform no better than low-degree polynomials on the input or fully-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 584, + 504, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 504, + 595 + ], + "score": 1.0, + "content": "trained two-layer network. Additionally, Suzuki (2018); Allen-Zhu and Li (2019); Yehudai and", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "Shamir (2019); Wei et al. (2018) demonstrated that neural network outperforms linear estimators", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "(including kernel method) in learning various target functions. The difference between fixed bases", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 616, + 476, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 476, + 630 + ], + "score": 1.0, + "content": "and adaptive bases mirrors the difference in optimizing the first or second layer in our setup.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "Generalization of Overparameterized Models. It is often observed that overparameterization", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 646, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 660 + ], + "score": 1.0, + "content": "does not result in overfitting (Neyshabur et al., 2014). In the lazy regime, generalization guarantees", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 659, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 671 + ], + "score": 1.0, + "content": "can be derived from the distance traveled by the parameters (Neyshabur et al., 2018; Nagarajan and", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Kolter, 2019), which becomes small if the model is sufficiently overparameterized (Arora et al., 2019b;", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "score": 1.0, + "content": "Li and Liang, 2018; Allen-Zhu et al., 2018a; Cao and Gu, 2019). Compared to these guarantees that", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "usually require significant overparameterization, our result relies on stronger data assumptions, but", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 702, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 106, + 702, + 505, + 714 + ], + "score": 1.0, + "content": "consequently we obtain the exact population risk instead of a vacuous upper-bound. Beyond the", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 712, + 506, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 712, + 506, + 726 + ], + "score": 1.0, + "content": "kernel regime, Advani and Saxe (2017); Goldt et al. (2019) analyzed the generalization dynamics of", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 724, + 328, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 724, + 328, + 737 + ], + "score": 1.0, + "content": "overparameterized models in the student-teacher setup.", + "type": "text" + } + ], + "index": 62 + } + ], + "index": 58 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 170 + ], + "lines": [], + "index": 3.5, + "bbox_fs": [ + 104, + 83, + 506, + 171 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 176, + 341, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 342, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 342, + 190 + ], + "score": 1.0, + "content": "Our findings can be summarized as follows (see Figure 1):", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 174, + 342, + 190 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 198, + 361, + 352 + ], + "lines": [ + { + "bbox": [ + 133, + 198, + 360, + 210 + ], + "spans": [ + { + "bbox": [ + 133, + 198, + 360, + 210 + ], + "score": 1.0, + "content": "• When only the second layer is optimized, we derive the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 209, + 360, + 221 + ], + "spans": [ + { + "bbox": [ + 141, + 209, + 360, + 221 + ], + "score": 1.0, + "content": "risk in its bias-variance decomposition and demonstrate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 219, + 338, + 232 + ], + "spans": [ + { + "bbox": [ + 141, + 219, + 338, + 232 + ], + "score": 1.0, + "content": "the presence of the double descent phenomenon.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 136, + 235, + 362, + 249 + ], + "spans": [ + { + "bbox": [ + 136, + 235, + 362, + 249 + ], + "score": 1.0, + "content": "• When the first layer is optimized, we compare two so-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 248, + 361, + 259 + ], + "spans": [ + { + "bbox": [ + 142, + 248, + 361, + 259 + ], + "score": 1.0, + "content": "lutions of gradient flow from different scales of initial-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 258, + 360, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 360, + 270 + ], + "score": 1.0, + "content": "ization, which we term as vanishing and non-vanishing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 270, + 360, + 281 + ], + "spans": [ + { + "bbox": [ + 141, + 270, + 360, + 281 + ], + "score": 1.0, + "content": "initialization, and show in both cases the population", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 280, + 319, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 319, + 292 + ], + "score": 1.0, + "content": "risk is independent to overparameterization.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 133, + 297, + 360, + 308 + ], + "spans": [ + { + "bbox": [ + 133, + 297, + 360, + 308 + ], + "score": 1.0, + "content": "• For the vanishing initialization, we show that the risk", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 307, + 361, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 307, + 361, + 320 + ], + "score": 1.0, + "content": "of the gradient flow solution is asymptotically close to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 318, + 361, + 330 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 361, + 330 + ], + "score": 1.0, + "content": "that of a rank-1 model. For non-vanishing initializa-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 330, + 362, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 330, + 362, + 340 + ], + "score": 1.0, + "content": "tion, we show that the gradient flow solution is well-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 341, + 352, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 352, + 352 + ], + "score": 1.0, + "content": "approximated by a kernel model and derive the risk.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16, + "bbox_fs": [ + 133, + 198, + 362, + 352 + ] + }, + { + "type": "image", + "bbox": [ + 369, + 177, + 502, + 277 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 369, + 177, + 502, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 369, + 177, + 502, + 277 + ], + "spans": [ + { + "bbox": [ + 369, + 177, + 502, + 277 + ], + "score": 0.969, + "type": "image", + "image_path": "fdb2f7294504073f6249cfffccb6993d35f35a22379caec2e8d08b45a4316c5a.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 369, + 177, + 502, + 227.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 369, + 227.0, + 502, + 277.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 367, + 284, + 505, + 365 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 367, + 284, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 367, + 284, + 505, + 295 + ], + "score": 1.0, + "content": "Figure 1: Illustration of the double", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 368, + 295, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 368, + 295, + 505, + 305 + ], + "score": 1.0, + "content": "descent risk curve in two-layer linear", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 367, + 304, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 367, + 304, + 405, + 315 + ], + "score": 1.0, + "content": "networks", + "type": "text" + }, + { + "bbox": [ + 405, + 305, + 445, + 314 + ], + "score": 0.79, + "content": "( \\mathrm { S N R } = 1 6 ", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 304, + 505, + 315 + ], + "score": 1.0, + "content": "). Brighter color", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 367, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 367, + 314, + 428, + 325 + ], + "score": 1.0, + "content": "indicates larger", + "type": "text" + }, + { + "bbox": [ + 428, + 314, + 469, + 325 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } = d / \\bar { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 314, + 505, + 325 + ], + "score": 1.0, + "content": ". Double", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 367, + 325, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 367, + 325, + 505, + 335 + ], + "score": 1.0, + "content": "descent is observed when the second", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 368, + 334, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 368, + 334, + 505, + 345 + ], + "score": 1.0, + "content": "layer coefficients are optimized (main", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 367, + 344, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 367, + 344, + 505, + 356 + ], + "score": 1.0, + "content": "figure), but not when the first layer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 367, + 354, + 493, + 366 + ], + "spans": [ + { + "bbox": [ + 367, + 354, + 493, + 366 + ], + "score": 1.0, + "content": "weights are optimized (subfigure).", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27.5 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 108, + 367, + 207, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 209, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 209, + 380 + ], + "score": 1.0, + "content": "1.1 RELATED WORKS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 388, + 505, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "Global Convergence of Two-layer Networks. A plethora of recent works have explored the global", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 399, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 506, + 411 + ], + "score": 1.0, + "content": "convergence of shallow neural networks. Mei et al. (2018; 2019); Chizat and Bach (2018a); Rotskoff", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "and Vanden-Eijnden (2018); Sirignano and Spiliopoulos (2018); Nitanda and Suzuki (2017) studied", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 308, + 433 + ], + "score": 1.0, + "content": "the mean-field limit where the number of neurons", + "type": "text" + }, + { + "bbox": [ + 309, + 422, + 342, + 432 + ], + "score": 0.91, + "content": "h \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 422, + 467, + 433 + ], + "score": 1.0, + "content": "and the second layer scaled by", + "type": "text" + }, + { + "bbox": [ + 468, + 421, + 484, + 433 + ], + "score": 0.89, + "content": "1 / h", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 422, + 505, + 433 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 432, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 445 + ], + "score": 1.0, + "content": "established correspondence between the main-particle limit of gradient descent and Wasserstein", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "gradient flow to demonsrate global convergence. On the other hand, Jacot et al. (2018); Du et al.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "(2018); Oymak and Soltanolkotabi (2019); Allen-Zhu et al. (2018b); Song and Yang (2019) considered", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "a different scaling and showed that gradient descent on overparameterized models converges to global", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "minimizer at a linear rate; key to these results is an observation that optimization via gradient descent", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 487, + 467, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 467, + 499 + ], + "score": 1.0, + "content": "is asymptotically equivalent to kernel regression with respect to the neural tangent kernel.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 388, + 506, + 499 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "Active vs. Lazy Training. Following Chizat and Bach (2018b), we refer to the two aforementioned", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "scalings as the active and lazy (kernel) regime. It has been observed that different regimes lead to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "contrasting inductive biases. Williams et al. (2019); Woodworth et al. (2019); Li et al. (2017) showed", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 540, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 551 + ], + "score": 1.0, + "content": "that for certain two-layer network or overparameterized linear model, the scale of initialization", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "controls the implicit regularization of gradient descent (from sparse to smooth solution). In the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "student-teacher setup (Tian, 2017; Zhong et al., 2017), Ghorbani et al. (2019b;a) showed that kernel", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "models in high dimensions perform no better than low-degree polynomials on the input or fully-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 584, + 504, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 504, + 595 + ], + "score": 1.0, + "content": "trained two-layer network. Additionally, Suzuki (2018); Allen-Zhu and Li (2019); Yehudai and", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "Shamir (2019); Wei et al. (2018) demonstrated that neural network outperforms linear estimators", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "(including kernel method) in learning various target functions. The difference between fixed bases", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 616, + 476, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 476, + 630 + ], + "score": 1.0, + "content": "and adaptive bases mirrors the difference in optimizing the first or second layer in our setup.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 506, + 506, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "Generalization of Overparameterized Models. It is often observed that overparameterization", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 646, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 660 + ], + "score": 1.0, + "content": "does not result in overfitting (Neyshabur et al., 2014). In the lazy regime, generalization guarantees", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 659, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 671 + ], + "score": 1.0, + "content": "can be derived from the distance traveled by the parameters (Neyshabur et al., 2018; Nagarajan and", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Kolter, 2019), which becomes small if the model is sufficiently overparameterized (Arora et al., 2019b;", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "score": 1.0, + "content": "Li and Liang, 2018; Allen-Zhu et al., 2018a; Cao and Gu, 2019). Compared to these guarantees that", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "usually require significant overparameterization, our result relies on stronger data assumptions, but", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 702, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 106, + 702, + 505, + 714 + ], + "score": 1.0, + "content": "consequently we obtain the exact population risk instead of a vacuous upper-bound. Beyond the", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 712, + 506, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 712, + 506, + 726 + ], + "score": 1.0, + "content": "kernel regime, Advani and Saxe (2017); Goldt et al. (2019) analyzed the generalization dynamics of", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 105, + 724, + 328, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 724, + 328, + 737 + ], + "score": 1.0, + "content": "overparameterized models in the student-teacher setup.", + "type": "text" + } + ], + "index": 62 + } + ], + "index": 58, + "bbox_fs": [ + 105, + 636, + 506, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Double Descent. The term double descent refers to the phenomenon that the population risk of an", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "empirical risk minimizer manifests a \"cusp\" at the interpolation threshold, and further overparame-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "terization decreases the risk. First observed in Krogh and Hertz (1992), the phenomenon has been", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "score": 1.0, + "content": "recently connected to the benefit of overparameterization (Belkin et al., 2018; Geiger et al., 2018;", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "Spigler et al., 2018; Advani and Saxe, 2017), and can be precisely characterized for certain simple", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 136, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 136, + 505, + 149 + ], + "score": 1.0, + "content": "models (Hastie et al., 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Our work is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 505, + 160 + ], + "score": 1.0, + "content": "inspired by Hastie et al. (2019) which uses random matrix theory to derive the asymptotic risk for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "linear and random feature models. Concurrent to our work, Mei and Montanari (2019) analyzed the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "score": 1.0, + "content": "random features model and derived its population risk for which double descent occurs both in bias", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "score": 1.0, + "content": "and variance. This aligns with our results on optimizing the second layer in Section 4 although we", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 191, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 504, + 204 + ], + "score": 1.0, + "content": "do not derive the bias component explicitly. Compared to Hastie et al. (2019); Mei and Montanari", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "(2019), the focus of this work is to highlight the different generalization property of models obtained", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 213, + 427, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 427, + 225 + ], + "score": 1.0, + "content": "from optimizing different layers of the network and from different initialization.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 506, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Random Matrix Theory. High-dimensional models, including kernel models and neural networks,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "can be analyzed by studying the properties of random matrices. El Karoui et al. (2010); Cheng", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "and Singer (2013); Fan and Montanari (2019) studied the spectral properties of kernel matrix via", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 265, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 276 + ], + "score": 1.0, + "content": "decomposing the nonlinearity with Taylor series or Hermite polynomials, which in turn explains", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 275, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 506, + 287 + ], + "score": 1.0, + "content": "the generalization of high-dimensional kernel ridgeless interpolators (Liang and Rakhlin, 2018).", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "score": 1.0, + "content": "In addition, similar tools have been used to study two-layer neural networks (Louart et al., 2018;", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "Pennington and Worah, 2017) and related quantities such as the Fisher information matrix (Karakida", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 308, + 280, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 280, + 320 + ], + "score": 1.0, + "content": "et al., 2018; Pennington and Worah, 2018).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 107, + 335, + 383, + 349 + ], + "lines": [ + { + "bbox": [ + 104, + 334, + 384, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 334, + 384, + 351 + ], + "score": 1.0, + "content": "2 PRELIMINARIES: TWO-LAYER NEURAL NETWORK", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 473, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 473, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 341, + 373 + ], + "score": 1.0, + "content": "Consider the following bias-free two-layer neural network", + "type": "text" + }, + { + "bbox": [ + 342, + 359, + 391, + 372 + ], + "score": 0.92, + "content": "f : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 359, + 412, + 373 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 412, + 361, + 420, + 370 + ], + "score": 0.82, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 359, + 473, + 373 + ], + "score": 1.0, + "content": "hidden units", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 375, + 358, + 409 + ], + "lines": [ + { + "bbox": [ + 252, + 375, + 358, + 409 + ], + "spans": [ + { + "bbox": [ + 252, + 375, + 358, + 409 + ], + "score": 0.95, + "content": "f ( \\pmb { x } ) = \\sum _ { i = 1 } ^ { h } a _ { i } \\phi ( \\langle \\pmb { x } , \\pmb { w } _ { i } \\rangle ) ,", + "type": "interline_equation", + "image_path": "8b955db9905f8d4e4578b7d8f6087b36cbba2fd74a040c0a3d27e182c997902e.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 375, + 358, + 392.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 252, + 392.0, + 358, + 409.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 506, + 491 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 133, + 426 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 413, + 167, + 424 + ], + "score": 0.91, + "content": "\\pmb { x } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 412, + 219, + 426 + ], + "score": 1.0, + "content": "is the input,", + "type": "text" + }, + { + "bbox": [ + 220, + 413, + 258, + 425 + ], + "score": 0.92, + "content": "\\boldsymbol { w } _ { i } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 412, + 422, + 426 + ], + "score": 1.0, + "content": "is the weights corresponding to neuron", + "type": "text" + }, + { + "bbox": [ + 422, + 415, + 426, + 424 + ], + "score": 0.66, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 412, + 431, + 426 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 431, + 414, + 461, + 425 + ], + "score": 0.91, + "content": "a _ { i } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 412, + 488, + 426 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 488, + 415, + 492, + 424 + ], + "score": 0.72, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 412, + 506, + 426 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 425, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 252, + 436 + ], + "score": 1.0, + "content": "coefficient of the second layer, and", + "type": "text" + }, + { + "bbox": [ + 253, + 425, + 299, + 435 + ], + "score": 0.9, + "content": "\\phi : \\mathbb { R } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 425, + 505, + 436 + ], + "score": 1.0, + "content": "is a Lipschitz continuous activation function with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 240, + 449 + ], + "score": 1.0, + "content": "bounded Gaussian moments, i.e.", + "type": "text" + }, + { + "bbox": [ + 240, + 436, + 303, + 448 + ], + "score": 0.83, + "content": "\\mathbb { E } [ \\phi ( G ) ^ { k } ] < \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 434, + 307, + 449 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 308, + 436, + 345, + 447 + ], + "score": 0.84, + "content": "\\forall k \\in \\mathbb { Z } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 434, + 361, + 449 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 361, + 435, + 415, + 447 + ], + "score": 0.89, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 434, + 506, + 449 + ], + "score": 1.0, + "content": ". For concise notation,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 443, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 104, + 443, + 144, + 461 + ], + "score": 1.0, + "content": "we write", + "type": "text" + }, + { + "bbox": [ + 144, + 446, + 250, + 457 + ], + "score": 0.89, + "content": "W = [ { \\pmb w } _ { 1 } , . . . { \\pmb w } _ { h } ] \\in \\mathbb { R } ^ { d \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 443, + 344, + 461 + ], + "score": 1.0, + "content": "for the weight matrix,", + "type": "text" + }, + { + "bbox": [ + 344, + 447, + 428, + 459 + ], + "score": 0.9, + "content": "\\pmb { a } = [ a _ { 1 } , . . . a _ { h } ] \\in \\mathbb { R } ^ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 443, + 507, + 461 + ], + "score": 1.0, + "content": "for the coefficient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 455, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 104, + 455, + 136, + 471 + ], + "score": 1.0, + "content": "vector,", + "type": "text" + }, + { + "bbox": [ + 136, + 458, + 236, + 468 + ], + "score": 0.87, + "content": "X = [ { \\pmb x } _ { 1 } , . . . { \\pmb x } _ { n } ] \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 455, + 317, + 471 + ], + "score": 1.0, + "content": "for the data matrix,", + "type": "text" + }, + { + "bbox": [ + 317, + 458, + 349, + 469 + ], + "score": 0.91, + "content": "\\ b { y } \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 455, + 507, + 471 + ], + "score": 1.0, + "content": "for the corresponding vector of labels,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 124, + 481 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 469, + 221, + 480 + ], + "score": 0.9, + "content": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 466, + 472, + 481 + ], + "score": 1.0, + "content": "for the feature matrix at the first layer. We omit arguments of", + "type": "text" + }, + { + "bbox": [ + 473, + 469, + 479, + 480 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 466, + 506, + 481 + ], + "score": 1.0, + "content": "when", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 479, + 233, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 233, + 492 + ], + "score": 1.0, + "content": "they are clear from the context.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 504, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 432, + 508 + ], + "score": 1.0, + "content": "We consider a student-teacher setup, in which data is generated by a teacher model", + "type": "text" + }, + { + "bbox": [ + 433, + 495, + 484, + 506 + ], + "score": 0.91, + "content": "F : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 397, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 397, + 519 + ], + "score": 1.0, + "content": "additive noise, and the student model aims to minimize the squared loss:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "interline_equation", + "bbox": [ + 143, + 522, + 467, + 555 + ], + "lines": [ + { + "bbox": [ + 143, + 522, + 467, + 555 + ], + "spans": [ + { + "bbox": [ + 143, + 522, + 467, + 555 + ], + "score": 0.92, + "content": "( { \\pmb x } _ { i } , \\varepsilon _ { i } ) \\overset { \\mathrm { i . i . d . } } { \\sim } P _ { { \\pmb x } } \\times P _ { \\varepsilon } , \\quad y _ { i } = F ( { \\pmb x } _ { i } ) + \\varepsilon _ { i } , \\quad L ( { \\boldsymbol X } ; f ) = \\frac { 1 } { 2 n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - f ( { \\pmb x } _ { i } ) \\right) ^ { 2 } ,", + "type": "interline_equation", + "image_path": "7391a72a42304ee6c7c207df5a4bbffed4639cbd4c44a5b4510acfb46bad444c.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 143, + 522, + 467, + 533.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 143, + 533.0, + 467, + 544.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 143, + 544.0, + 467, + 555.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 557, + 504, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 133, + 570 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 558, + 175, + 570 + ], + "score": 0.85, + "content": "\\mathbb { E } [ { \\pmb x } _ { i } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 558, + 178, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 179, + 558, + 235, + 570 + ], + "score": 0.83, + "content": "\\mathrm { C o v } ( { \\pmb x } _ { i } ) = \\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 558, + 239, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 239, + 558, + 279, + 570 + ], + "score": 0.75, + "content": "\\mathbb { E } [ \\varepsilon _ { i } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 558, + 282, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 283, + 558, + 339, + 570 + ], + "score": 0.87, + "content": "\\mathrm { V a r } ( \\varepsilon _ { i } ) = \\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 558, + 505, + 570 + ], + "score": 1.0, + "content": ". We are interested in the population risk", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 569, + 487, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 231, + 582 + ], + "score": 0.87, + "content": "R ( f ) = \\mathbb { E } _ { P _ { x } } [ ( F ( { \\pmb x } ) - f ( { \\pmb x } ) ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 569, + 487, + 583 + ], + "score": 1.0, + "content": ". Our analysis will be made under the proportional asymptotics:", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 584, + 423, + 598 + ], + "lines": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "spans": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "score": 0.89, + "content": "n , d , h \\infty ; \\quad d / n \\gamma _ { 1 } , h / n \\gamma _ { 2 } ; \\quad \\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty ) ,", + "type": "interline_equation", + "image_path": "c5b31a0e06ef7e1df6ee2341304694fefeca77f8d51d43542792fa49fc18d39c.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 344, + 614 + ], + "score": 1.0, + "content": "in which overparameterization corresponds to increasing", + "type": "text" + }, + { + "bbox": [ + 344, + 603, + 355, + 613 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 601, + 506, + 614 + ], + "score": 1.0, + "content": ". Thus the characteristics of double", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 363, + 625 + ], + "score": 1.0, + "content": "descent considered in this work are: 1) large population risk as", + "type": "text" + }, + { + "bbox": [ + 364, + 613, + 406, + 624 + ], + "score": 0.89, + "content": "\\gamma _ { 2 } 1 ; 2", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ") decrease in the risk for", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 137, + 635 + ], + "score": 0.9, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 623, + 506, + 635 + ], + "score": 1.0, + "content": ". While the empirical risk can be minimized in various ways, we analyze the solution of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 633, + 428, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 324, + 648 + ], + "score": 1.0, + "content": "gradient flow, in which we update either the first layer", + "type": "text" + }, + { + "bbox": [ + 324, + 634, + 336, + 644 + ], + "score": 0.74, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 633, + 415, + 648 + ], + "score": 1.0, + "content": "or the second layer", + "type": "text" + }, + { + "bbox": [ + 416, + 636, + 423, + 644 + ], + "score": 0.79, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 633, + 428, + 648 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 649, + 432, + 662 + ], + "lines": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "spans": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "score": 0.88, + "content": "\\mathrm { d } W ( t ) = - \\nabla _ { W } L ( X ; f ) \\mathrm { d } t \\quad \\mathrm { o r } \\quad \\mathrm { d } a ( t ) = - \\nabla _ { a } L ( X ; f ) \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "f595ce94eebb2a30d8aaf4b9ab2bd9b5d7785c15231383757712d4b23e1b2765.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "from small initialization. The rest of the paper is organized as follows. In Section 3, we start with a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "simple example of two-layer linear network as warm-up. In Section 4, we consider optimizing the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 247, + 700 + ], + "score": 1.0, + "content": "second layer coefficients (flow over", + "type": "text" + }, + { + "bbox": [ + 248, + 690, + 255, + 698 + ], + "score": 0.7, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 687, + 506, + 700 + ], + "score": 1.0, + "content": ") of a non-linear two-layer neural network under fixed Gaussian", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "first layer, which is a random feature model. Section 5 considers optimizing the first layer weights", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 149, + 722 + ], + "score": 1.0, + "content": "(flow over", + "type": "text" + }, + { + "bbox": [ + 149, + 710, + 160, + 720 + ], + "score": 0.61, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ") of such network under fixed Rademacher second layer. We defer all proofs and details", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 721, + 221, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 221, + 732 + ], + "score": 1.0, + "content": "on experiments to appendix.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Double Descent. The term double descent refers to the phenomenon that the population risk of an", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "empirical risk minimizer manifests a \"cusp\" at the interpolation threshold, and further overparame-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "terization decreases the risk. First observed in Krogh and Hertz (1992), the phenomenon has been", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "score": 1.0, + "content": "recently connected to the benefit of overparameterization (Belkin et al., 2018; Geiger et al., 2018;", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "Spigler et al., 2018; Advani and Saxe, 2017), and can be precisely characterized for certain simple", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 136, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 136, + 505, + 149 + ], + "score": 1.0, + "content": "models (Hastie et al., 2019; Belkin et al., 2019; Bartlett et al., 2019; Xu and Hsu, 2019). Our work is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 505, + 160 + ], + "score": 1.0, + "content": "inspired by Hastie et al. (2019) which uses random matrix theory to derive the asymptotic risk for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "linear and random feature models. Concurrent to our work, Mei and Montanari (2019) analyzed the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "score": 1.0, + "content": "random features model and derived its population risk for which double descent occurs both in bias", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "score": 1.0, + "content": "and variance. This aligns with our results on optimizing the second layer in Section 4 although we", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 191, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 504, + 204 + ], + "score": 1.0, + "content": "do not derive the bias component explicitly. Compared to Hastie et al. (2019); Mei and Montanari", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "(2019), the focus of this work is to highlight the different generalization property of models obtained", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 213, + 427, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 427, + 225 + ], + "score": 1.0, + "content": "from optimizing different layers of the network and from different initialization.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 82, + 506, + 225 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 506, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Random Matrix Theory. High-dimensional models, including kernel models and neural networks,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "can be analyzed by studying the properties of random matrices. El Karoui et al. (2010); Cheng", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "and Singer (2013); Fan and Montanari (2019) studied the spectral properties of kernel matrix via", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 265, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 276 + ], + "score": 1.0, + "content": "decomposing the nonlinearity with Taylor series or Hermite polynomials, which in turn explains", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 275, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 506, + 287 + ], + "score": 1.0, + "content": "the generalization of high-dimensional kernel ridgeless interpolators (Liang and Rakhlin, 2018).", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "score": 1.0, + "content": "In addition, similar tools have been used to study two-layer neural networks (Louart et al., 2018;", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 309 + ], + "score": 1.0, + "content": "Pennington and Worah, 2017) and related quantities such as the Fisher information matrix (Karakida", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 308, + 280, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 280, + 320 + ], + "score": 1.0, + "content": "et al., 2018; Pennington and Worah, 2018).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 230, + 506, + 320 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 335, + 383, + 349 + ], + "lines": [ + { + "bbox": [ + 104, + 334, + 384, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 334, + 384, + 351 + ], + "score": 1.0, + "content": "2 PRELIMINARIES: TWO-LAYER NEURAL NETWORK", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 473, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 473, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 341, + 373 + ], + "score": 1.0, + "content": "Consider the following bias-free two-layer neural network", + "type": "text" + }, + { + "bbox": [ + 342, + 359, + 391, + 372 + ], + "score": 0.92, + "content": "f : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 359, + 412, + 373 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 412, + 361, + 420, + 370 + ], + "score": 0.82, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 359, + 473, + 373 + ], + "score": 1.0, + "content": "hidden units", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 359, + 473, + 373 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 375, + 358, + 409 + ], + "lines": [ + { + "bbox": [ + 252, + 375, + 358, + 409 + ], + "spans": [ + { + "bbox": [ + 252, + 375, + 358, + 409 + ], + "score": 0.95, + "content": "f ( \\pmb { x } ) = \\sum _ { i = 1 } ^ { h } a _ { i } \\phi ( \\langle \\pmb { x } , \\pmb { w } _ { i } \\rangle ) ,", + "type": "interline_equation", + "image_path": "8b955db9905f8d4e4578b7d8f6087b36cbba2fd74a040c0a3d27e182c997902e.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 375, + 358, + 392.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 252, + 392.0, + 358, + 409.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 506, + 491 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 133, + 426 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 413, + 167, + 424 + ], + "score": 0.91, + "content": "\\pmb { x } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 412, + 219, + 426 + ], + "score": 1.0, + "content": "is the input,", + "type": "text" + }, + { + "bbox": [ + 220, + 413, + 258, + 425 + ], + "score": 0.92, + "content": "\\boldsymbol { w } _ { i } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 412, + 422, + 426 + ], + "score": 1.0, + "content": "is the weights corresponding to neuron", + "type": "text" + }, + { + "bbox": [ + 422, + 415, + 426, + 424 + ], + "score": 0.66, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 412, + 431, + 426 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 431, + 414, + 461, + 425 + ], + "score": 0.91, + "content": "a _ { i } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 412, + 488, + 426 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 488, + 415, + 492, + 424 + ], + "score": 0.72, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 412, + 506, + 426 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 425, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 252, + 436 + ], + "score": 1.0, + "content": "coefficient of the second layer, and", + "type": "text" + }, + { + "bbox": [ + 253, + 425, + 299, + 435 + ], + "score": 0.9, + "content": "\\phi : \\mathbb { R } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 425, + 505, + 436 + ], + "score": 1.0, + "content": "is a Lipschitz continuous activation function with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 240, + 449 + ], + "score": 1.0, + "content": "bounded Gaussian moments, i.e.", + "type": "text" + }, + { + "bbox": [ + 240, + 436, + 303, + 448 + ], + "score": 0.83, + "content": "\\mathbb { E } [ \\phi ( G ) ^ { k } ] < \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 434, + 307, + 449 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 308, + 436, + 345, + 447 + ], + "score": 0.84, + "content": "\\forall k \\in \\mathbb { Z } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 434, + 361, + 449 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 361, + 435, + 415, + 447 + ], + "score": 0.89, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 434, + 506, + 449 + ], + "score": 1.0, + "content": ". For concise notation,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 443, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 104, + 443, + 144, + 461 + ], + "score": 1.0, + "content": "we write", + "type": "text" + }, + { + "bbox": [ + 144, + 446, + 250, + 457 + ], + "score": 0.89, + "content": "W = [ { \\pmb w } _ { 1 } , . . . { \\pmb w } _ { h } ] \\in \\mathbb { R } ^ { d \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 443, + 344, + 461 + ], + "score": 1.0, + "content": "for the weight matrix,", + "type": "text" + }, + { + "bbox": [ + 344, + 447, + 428, + 459 + ], + "score": 0.9, + "content": "\\pmb { a } = [ a _ { 1 } , . . . a _ { h } ] \\in \\mathbb { R } ^ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 443, + 507, + 461 + ], + "score": 1.0, + "content": "for the coefficient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 455, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 104, + 455, + 136, + 471 + ], + "score": 1.0, + "content": "vector,", + "type": "text" + }, + { + "bbox": [ + 136, + 458, + 236, + 468 + ], + "score": 0.87, + "content": "X = [ { \\pmb x } _ { 1 } , . . . { \\pmb x } _ { n } ] \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 455, + 317, + 471 + ], + "score": 1.0, + "content": "for the data matrix,", + "type": "text" + }, + { + "bbox": [ + 317, + 458, + 349, + 469 + ], + "score": 0.91, + "content": "\\ b { y } \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 455, + 507, + 471 + ], + "score": 1.0, + "content": "for the corresponding vector of labels,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 124, + 481 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 469, + 221, + 480 + ], + "score": 0.9, + "content": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 466, + 472, + 481 + ], + "score": 1.0, + "content": "for the feature matrix at the first layer. We omit arguments of", + "type": "text" + }, + { + "bbox": [ + 473, + 469, + 479, + 480 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 466, + 506, + 481 + ], + "score": 1.0, + "content": "when", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 479, + 233, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 233, + 492 + ], + "score": 1.0, + "content": "they are clear from the context.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 412, + 507, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 504, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 432, + 508 + ], + "score": 1.0, + "content": "We consider a student-teacher setup, in which data is generated by a teacher model", + "type": "text" + }, + { + "bbox": [ + 433, + 495, + 484, + 506 + ], + "score": 0.91, + "content": "F : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 397, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 397, + 519 + ], + "score": 1.0, + "content": "additive noise, and the student model aims to minimize the squared loss:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 494, + 505, + 519 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 143, + 522, + 467, + 555 + ], + "lines": [ + { + "bbox": [ + 143, + 522, + 467, + 555 + ], + "spans": [ + { + "bbox": [ + 143, + 522, + 467, + 555 + ], + "score": 0.92, + "content": "( { \\pmb x } _ { i } , \\varepsilon _ { i } ) \\overset { \\mathrm { i . i . d . } } { \\sim } P _ { { \\pmb x } } \\times P _ { \\varepsilon } , \\quad y _ { i } = F ( { \\pmb x } _ { i } ) + \\varepsilon _ { i } , \\quad L ( { \\boldsymbol X } ; f ) = \\frac { 1 } { 2 n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - f ( { \\pmb x } _ { i } ) \\right) ^ { 2 } ,", + "type": "interline_equation", + "image_path": "7391a72a42304ee6c7c207df5a4bbffed4639cbd4c44a5b4510acfb46bad444c.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 143, + 522, + 467, + 533.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 143, + 533.0, + 467, + 544.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 143, + 544.0, + 467, + 555.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 557, + 504, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 133, + 570 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 558, + 175, + 570 + ], + "score": 0.85, + "content": "\\mathbb { E } [ { \\pmb x } _ { i } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 558, + 178, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 179, + 558, + 235, + 570 + ], + "score": 0.83, + "content": "\\mathrm { C o v } ( { \\pmb x } _ { i } ) = \\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 558, + 239, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 239, + 558, + 279, + 570 + ], + "score": 0.75, + "content": "\\mathbb { E } [ \\varepsilon _ { i } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 558, + 282, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 283, + 558, + 339, + 570 + ], + "score": 0.87, + "content": "\\mathrm { V a r } ( \\varepsilon _ { i } ) = \\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 558, + 505, + 570 + ], + "score": 1.0, + "content": ". We are interested in the population risk", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 569, + 487, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 231, + 582 + ], + "score": 0.87, + "content": "R ( f ) = \\mathbb { E } _ { P _ { x } } [ ( F ( { \\pmb x } ) - f ( { \\pmb x } ) ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 569, + 487, + 583 + ], + "score": 1.0, + "content": ". Our analysis will be made under the proportional asymptotics:", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 558, + 505, + 583 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 584, + 423, + 598 + ], + "lines": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "spans": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "score": 0.89, + "content": "n , d , h \\infty ; \\quad d / n \\gamma _ { 1 } , h / n \\gamma _ { 2 } ; \\quad \\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty ) ,", + "type": "interline_equation", + "image_path": "c5b31a0e06ef7e1df6ee2341304694fefeca77f8d51d43542792fa49fc18d39c.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 186, + 584, + 423, + 598 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 344, + 614 + ], + "score": 1.0, + "content": "in which overparameterization corresponds to increasing", + "type": "text" + }, + { + "bbox": [ + 344, + 603, + 355, + 613 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 601, + 506, + 614 + ], + "score": 1.0, + "content": ". Thus the characteristics of double", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 363, + 625 + ], + "score": 1.0, + "content": "descent considered in this work are: 1) large population risk as", + "type": "text" + }, + { + "bbox": [ + 364, + 613, + 406, + 624 + ], + "score": 0.89, + "content": "\\gamma _ { 2 } 1 ; 2", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ") decrease in the risk for", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 137, + 635 + ], + "score": 0.9, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 623, + 506, + 635 + ], + "score": 1.0, + "content": ". While the empirical risk can be minimized in various ways, we analyze the solution of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 633, + 428, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 324, + 648 + ], + "score": 1.0, + "content": "gradient flow, in which we update either the first layer", + "type": "text" + }, + { + "bbox": [ + 324, + 634, + 336, + 644 + ], + "score": 0.74, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 633, + 415, + 648 + ], + "score": 1.0, + "content": "or the second layer", + "type": "text" + }, + { + "bbox": [ + 416, + 636, + 423, + 644 + ], + "score": 0.79, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 633, + 428, + 648 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 601, + 506, + 648 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 649, + 432, + 662 + ], + "lines": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "spans": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "score": 0.88, + "content": "\\mathrm { d } W ( t ) = - \\nabla _ { W } L ( X ; f ) \\mathrm { d } t \\quad \\mathrm { o r } \\quad \\mathrm { d } a ( t ) = - \\nabla _ { a } L ( X ; f ) \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "f595ce94eebb2a30d8aaf4b9ab2bd9b5d7785c15231383757712d4b23e1b2765.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 178, + 649, + 432, + 662 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "from small initialization. The rest of the paper is organized as follows. In Section 3, we start with a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "simple example of two-layer linear network as warm-up. In Section 4, we consider optimizing the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 247, + 700 + ], + "score": 1.0, + "content": "second layer coefficients (flow over", + "type": "text" + }, + { + "bbox": [ + 248, + 690, + 255, + 698 + ], + "score": 0.7, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 687, + 506, + 700 + ], + "score": 1.0, + "content": ") of a non-linear two-layer neural network under fixed Gaussian", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "first layer, which is a random feature model. Section 5 considers optimizing the first layer weights", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 149, + 722 + ], + "score": 1.0, + "content": "(flow over", + "type": "text" + }, + { + "bbox": [ + 149, + 710, + 160, + 720 + ], + "score": 0.61, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ") of such network under fixed Rademacher second layer. We defer all proofs and details", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 721, + 221, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 221, + 732 + ], + "score": 1.0, + "content": "on experiments to appendix.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 665, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 283, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 284, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 284, + 96 + ], + "score": 1.0, + "content": "3 WARM-UP: LINEAR NETWORK", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 104, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 276, + 118 + ], + "score": 1.0, + "content": "We begin with a simple linear model with", + "type": "text" + }, + { + "bbox": [ + 276, + 105, + 317, + 118 + ], + "score": 0.92, + "content": "\\phi ( { \\pmb x } ) = { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 104, + 336, + 118 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 336, + 104, + 384, + 116 + ], + "score": 0.92, + "content": "\\Phi = W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 104, + 505, + 118 + ], + "score": 1.0, + "content": ". We remark that although the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "score": 1.0, + "content": "model is linear, the solution obtained by gradient flow on the two-layer model can be different than", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 402, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 402, + 140 + ], + "score": 1.0, + "content": "that from directly solving the linear regression problem on input features.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 506, + 186 + ], + "lines": [ + { + "bbox": [ + 106, + 150, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 505, + 162 + ], + "score": 1.0, + "content": "Training the Second Layer. Following Hastie et al. (2019), we fix the first layer parameters to be", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 161, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 378, + 180 + ], + "score": 1.0, + "content": "randomly drawn from a unit Gaussian and optimize the coefficients", + "type": "text" + }, + { + "bbox": [ + 378, + 166, + 385, + 174 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 161, + 448, + 180 + ], + "score": 1.0, + "content": "by minimizing", + "type": "text" + }, + { + "bbox": [ + 448, + 161, + 503, + 177 + ], + "score": 0.94, + "content": "\\left| \\left| \\pmb { a } ^ { \\top } \\Phi - \\pmb { y } \\right| \\right| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 161, + 507, + 180 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 380, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 380, + 187 + ], + "score": 1.0, + "content": "The following lemma characterizes the solution of the gradient flow.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 188, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 334, + 201 + ], + "score": 1.0, + "content": "Lemma 1 (Least squares solution). Given data matrix", + "type": "text" + }, + { + "bbox": [ + 335, + 189, + 343, + 198 + ], + "score": 0.62, + "content": "X _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 187, + 414, + 201 + ], + "score": 1.0, + "content": ", response vector", + "type": "text" + }, + { + "bbox": [ + 415, + 191, + 423, + 200 + ], + "score": 0.37, + "content": "\\textbf { { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 187, + 470, + 201 + ], + "score": 1.0, + "content": "and model", + "type": "text" + }, + { + "bbox": [ + 471, + 188, + 505, + 200 + ], + "score": 0.9, + "content": "f ( { \\pmb x } ) =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 107, + 199, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 107, + 199, + 164, + 211 + ], + "score": 0.91, + "content": "\\langle \\phi ( \\pmb { x } ^ { \\top } W ) , \\hat { \\pmb { a } } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 199, + 295, + 213 + ], + "score": 1.0, + "content": "with fixed first layer coefficients", + "type": "text" + }, + { + "bbox": [ + 295, + 200, + 307, + 209 + ], + "score": 0.68, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 199, + 441, + 213 + ], + "score": 1.0, + "content": ", gradient flow on the coefficients", + "type": "text" + }, + { + "bbox": [ + 442, + 202, + 449, + 209 + ], + "score": 0.32, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 199, + 506, + 213 + ], + "score": 1.0, + "content": "starting from", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 209, + 460, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 231, + 223 + ], + "score": 1.0, + "content": "zero initialization converges to", + "type": "text" + }, + { + "bbox": [ + 232, + 209, + 270, + 221 + ], + "score": 0.91, + "content": "\\mathbf { \\bar { a } } = \\Phi ^ { \\dagger } \\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 210, + 300, + 223 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 300, + 211, + 306, + 222 + ], + "score": 0.8, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 210, + 460, + 223 + ], + "score": 1.0, + "content": "stands for the Moore-Penrose inverse.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 112, + 229, + 466, + 241 + ], + "lines": [ + { + "bbox": [ + 109, + 228, + 467, + 243 + ], + "spans": [ + { + "bbox": [ + 109, + 228, + 467, + 243 + ], + "score": 1.0, + "content": "We make two assumptions on the data and the teacher model to simplify the computation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 119, + 242, + 485, + 255 + ], + "lines": [ + { + "bbox": [ + 124, + 242, + 483, + 256 + ], + "spans": [ + { + "bbox": [ + 124, + 242, + 230, + 256 + ], + "score": 1.0, + "content": "(A1) Gaussian Features:", + "type": "text" + }, + { + "bbox": [ + 231, + 243, + 287, + 255 + ], + "score": 0.94, + "content": "\\pmb { x } _ { i } \\sim \\mathcal { N } ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 242, + 387, + 256 + ], + "score": 1.0, + "content": "; (A2) Linear Teacher:", + "type": "text" + }, + { + "bbox": [ + 387, + 243, + 446, + 255 + ], + "score": 0.82, + "content": "F ( { \\pmb x } ) = \\langle { \\pmb x } , { \\pmb \\beta } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 242, + 451, + 256 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 451, + 243, + 483, + 255 + ], + "score": 0.51, + "content": "\\| { \\boldsymbol { \\beta } } \\| = r", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 258, + 504, + 282 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 262, + 273 + ], + "score": 1.0, + "content": "Denote the linear student network as", + "type": "text" + }, + { + "bbox": [ + 262, + 258, + 325, + 272 + ], + "score": 0.95, + "content": "f ( \\pmb { x } ) = \\langle \\pmb { x } , \\hat { \\beta } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 258, + 357, + 273 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 357, + 258, + 399, + 271 + ], + "score": 0.92, + "content": "\\hat { \\boldsymbol { \\beta } } = W \\hat { \\mathbf { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 258, + 418, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 419, + 260, + 426, + 270 + ], + "score": 0.82, + "content": "\\hat { \\textbf { \\textit a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 258, + 505, + 273 + ], + "score": 1.0, + "content": "is the least-square", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 270, + 485, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 485, + 283 + ], + "score": 1.0, + "content": "solution defined by Lemma 1. We write the population risk in its bias-variance decomposition.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 283, + 461, + 318 + ], + "lines": [ + { + "bbox": [ + 149, + 283, + 461, + 318 + ], + "spans": [ + { + "bbox": [ + 149, + 283, + 461, + 318 + ], + "score": 0.91, + "content": "R = \\mathbb { E } _ { { \\mathbf { x } } \\sim P _ { \\mathbf { x } } } [ \\Vert \\hat { \\beta } - \\beta \\Vert _ { \\Sigma } ^ { 2 } \\vert { \\cal X } , { \\cal W } ] = \\underbrace { \\Vert \\mathbb { E } [ \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ] - \\beta \\Vert _ { 2 } ^ { 2 } } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathrm { t r } \\left( \\mathrm { C o v } ( \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ) \\right) } _ { { \\cal V } = \\mathrm { v a r i a n c e } } ,", + "type": "interline_equation", + "image_path": "d18643c3281713eae817cf95ea0a2aa7076ffbe532bc9cb5d02f70cfadd69742.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 149, + 283, + 461, + 294.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 149, + 294.6666666666667, + 461, + 306.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 149, + 306.33333333333337, + 461, + 318.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 321, + 467, + 334 + ], + "lines": [ + { + "bbox": [ + 105, + 319, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 133, + 336 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 320, + 197, + 335 + ], + "score": 0.93, + "content": "\\left\\| \\pmb { x } \\right\\| _ { \\Sigma } ^ { 2 } = \\pmb { x } ^ { \\top } \\Sigma \\pmb { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 319, + 469, + 336 + ], + "score": 1.0, + "content": ". We compute the bias and the variance separately to obtain the risk.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 337, + 441, + 352 + ], + "lines": [ + { + "bbox": [ + 103, + 332, + 446, + 356 + ], + "spans": [ + { + "bbox": [ + 103, + 332, + 254, + 356 + ], + "score": 1.0, + "content": "Theorem 2. Given (A1)(A2) and let", + "type": "text" + }, + { + "bbox": [ + 254, + 340, + 267, + 351 + ], + "score": 0.73, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 332, + 286, + 356 + ], + "score": 1.0, + "content": "i.i.d. ∼", + "type": "text" + }, + { + "bbox": [ + 287, + 338, + 339, + 352 + ], + "score": 0.87, + "content": "\\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 332, + 354, + 356 + ], + "score": 1.0, + "content": ", at", + "type": "text" + }, + { + "bbox": [ + 354, + 340, + 407, + 351 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 332, + 446, + 356 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 353, + 498, + 414 + ], + "lines": [ + { + "bbox": [ + 111, + 353, + 498, + 414 + ], + "spans": [ + { + "bbox": [ + 111, + 353, + 498, + 414 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < \\gamma _ { 1 } , } \\\\ { \\frac { \\gamma _ { 1 } } { g _ { 1 } } \\sigma ^ { 2 } , } & R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 2 } g _ { 1 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { g _ { 1 } + g _ { 2 } } { g _ { 1 } g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } > 1 . } \\end{array} } \\end{array} } \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "8b38e4e0f9807763f6f0a6585220ff7376399aa2616d84e51b73efd6e2e1732c.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 353, + 498, + 373.3333333333333 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 373.3333333333333, + 498, + 393.66666666666663 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 393.66666666666663, + 498, + 413.99999999999994 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 362, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 363, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 133, + 440 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 425, + 282, + 438 + ], + "score": 0.48, + "content": "d / n \\to \\gamma _ { 1 } , h / n \\to \\gamma _ { 2 } , g _ { 1 } = | \\gamma _ { 1 } - 1 | ,", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 423, + 303, + 440 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 303, + 425, + 359, + 438 + ], + "score": 0.91, + "content": "g _ { 2 } = | \\gamma _ { 2 } - 1 |", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 423, + 363, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 444, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 193, + 457 + ], + "score": 1.0, + "content": "We observe that when", + "type": "text" + }, + { + "bbox": [ + 193, + 445, + 219, + 455 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 444, + 237, + 457 + ], + "score": 1.0, + "content": "(i.e.", + "type": "text" + }, + { + "bbox": [ + 237, + 446, + 267, + 456 + ], + "score": 0.87, + "content": "\\gamma _ { 1 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "), we obtain the double descent risk curve, i.e., the population", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 227, + 469 + ], + "score": 1.0, + "content": "risk achieves its maximum at", + "type": "text" + }, + { + "bbox": [ + 228, + 457, + 260, + 467 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 455, + 398, + 469 + ], + "score": 1.0, + "content": "and further overparameterization", + "type": "text" + }, + { + "bbox": [ + 399, + 456, + 432, + 467 + ], + "score": 0.85, + "content": "( \\gamma _ { 2 } > 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 455, + 505, + 469 + ], + "score": 1.0, + "content": ") reduces both the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 272, + 480 + ], + "score": 1.0, + "content": "bias and the variance. Conversely when", + "type": "text" + }, + { + "bbox": [ + 273, + 467, + 300, + 477 + ], + "score": 0.9, + "content": "n > d", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 465, + 318, + 480 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 319, + 467, + 346, + 477 + ], + "score": 0.9, + "content": "h > d", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 465, + 367, + 480 + ], + "score": 1.0, + "content": "(i.e.", + "type": "text" + }, + { + "bbox": [ + 367, + 467, + 438, + 478 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < \\operatorname* { m i n } ( 1 , \\gamma _ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 465, + 505, + 480 + ], + "score": 1.0, + "content": ", the population", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 477, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 398, + 493 + ], + "score": 1.0, + "content": "risk becomes constant and equals to that of the minimum-norm solution", + "type": "text" + }, + { + "bbox": [ + 399, + 478, + 453, + 491 + ], + "score": 0.95, + "content": "{ \\hat { \\boldsymbol { \\beta } } } _ { \\operatorname* { m i n } } = { \\boldsymbol { X } } ^ { \\dagger } { \\boldsymbol { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 477, + 505, + 493 + ], + "score": 1.0, + "content": "on the input", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 490, + 143, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 143, + 502 + ], + "score": 1.0, + "content": "features.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 504, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "Training the First Layer. When the first layer of a linear network is optimized via gradient flow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 524, + 418, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 402, + 536 + ], + "score": 1.0, + "content": "and the second layer is fixed, the following holds for zero-initialization of", + "type": "text" + }, + { + "bbox": [ + 403, + 525, + 414, + 534 + ], + "score": 0.76, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 524, + 418, + 536 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 199, + 551 + ], + "score": 1.0, + "content": "Proposition 3. Given", + "type": "text" + }, + { + "bbox": [ + 199, + 537, + 245, + 550 + ], + "score": 0.91, + "content": "W ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 535, + 288, + 551 + ], + "score": 1.0, + "content": "and fixed", + "type": "text" + }, + { + "bbox": [ + 288, + 537, + 326, + 549 + ], + "score": 0.92, + "content": "\\pm \\ : 0", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 535, + 380, + 551 + ], + "score": 1.0, + "content": ", at any time", + "type": "text" + }, + { + "bbox": [ + 380, + 538, + 406, + 548 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "of the gradient flow on", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 549, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 118, + 561 + ], + "score": 0.4, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 549, + 122, + 565 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 123, + 551, + 146, + 563 + ], + "score": 0.78, + "content": "W ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 549, + 227, + 565 + ], + "score": 1.0, + "content": "is rank-1. Further,", + "type": "text" + }, + { + "bbox": [ + 227, + 549, + 267, + 563 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = { \\widehat { \\boldsymbol { W } } } \\mathbf { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 549, + 441, + 565 + ], + "score": 1.0, + "content": "converges to the least squares solution of", + "type": "text" + }, + { + "bbox": [ + 441, + 549, + 486, + 563 + ], + "score": 0.93, + "content": "\\boldsymbol { y } = \\boldsymbol { X } ^ { \\intercal } \\hat { \\boldsymbol { \\beta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 549, + 506, + 565 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 561, + 399, + 575 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 399, + 575 + ], + "score": 1.0, + "content": "population risk of which is given in (Hastie et al., 2019, Thm. 1 & 3)) as", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 574, + 428, + 600 + ], + "lines": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "spans": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } r ^ { 2 } + \\frac { 1 } { \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "f5a4682cd92d7d3b10ff18ef37be5a460f8f02025a4402ea4f6cce0978f6f081.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 294, + 627 + ], + "score": 1.0, + "content": "In this case, overparameterization by increasing", + "type": "text" + }, + { + "bbox": [ + 295, + 616, + 306, + 626 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "does not influence the population risk. In addition,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "since the obtained two-layer linear model is equivalent to the minimum-norm solution on the input", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 636, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 636, + 140, + 651 + ], + "score": 1.0, + "content": "features", + "type": "text" + }, + { + "bbox": [ + 140, + 636, + 162, + 650 + ], + "score": 0.92, + "content": "\\hat { \\beta } _ { \\mathrm { m i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 636, + 506, + 651 + ], + "score": 1.0, + "content": ", optimizing the first layer always results in smaller or equal population risk compared", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 649, + 231, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 231, + 662 + ], + "score": 1.0, + "content": "to optimizing the second layer.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "In this simple scenario for two-layer linear networks, double descent is observed only when the second", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "layer is optimized, which reduces the objective to least squares regression on the intermediate features.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "On the other hand, training the first layer from zero-initialization always yields the same solution", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "that is independent to overparameterization. One natural question to ask is: does this phenomenon", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "generalize to nonlinear two-layer neural networks? The following sections answer this question in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 268, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 268, + 732 + ], + "score": 1.0, + "content": "the affirmative under certain conditions.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 283, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 284, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 284, + 96 + ], + "score": 1.0, + "content": "3 WARM-UP: LINEAR NETWORK", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 104, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 276, + 118 + ], + "score": 1.0, + "content": "We begin with a simple linear model with", + "type": "text" + }, + { + "bbox": [ + 276, + 105, + 317, + 118 + ], + "score": 0.92, + "content": "\\phi ( { \\pmb x } ) = { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 104, + 336, + 118 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 336, + 104, + 384, + 116 + ], + "score": 0.92, + "content": "\\Phi = W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 104, + 505, + 118 + ], + "score": 1.0, + "content": ". We remark that although the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "score": 1.0, + "content": "model is linear, the solution obtained by gradient flow on the two-layer model can be different than", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 402, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 402, + 140 + ], + "score": 1.0, + "content": "that from directly solving the linear regression problem on input features.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 104, + 505, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 506, + 186 + ], + "lines": [ + { + "bbox": [ + 106, + 150, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 505, + 162 + ], + "score": 1.0, + "content": "Training the Second Layer. Following Hastie et al. (2019), we fix the first layer parameters to be", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 161, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 378, + 180 + ], + "score": 1.0, + "content": "randomly drawn from a unit Gaussian and optimize the coefficients", + "type": "text" + }, + { + "bbox": [ + 378, + 166, + 385, + 174 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 161, + 448, + 180 + ], + "score": 1.0, + "content": "by minimizing", + "type": "text" + }, + { + "bbox": [ + 448, + 161, + 503, + 177 + ], + "score": 0.94, + "content": "\\left| \\left| \\pmb { a } ^ { \\top } \\Phi - \\pmb { y } \\right| \\right| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 161, + 507, + 180 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 380, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 380, + 187 + ], + "score": 1.0, + "content": "The following lemma characterizes the solution of the gradient flow.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 104, + 150, + 507, + 187 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 188, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 334, + 201 + ], + "score": 1.0, + "content": "Lemma 1 (Least squares solution). Given data matrix", + "type": "text" + }, + { + "bbox": [ + 335, + 189, + 343, + 198 + ], + "score": 0.62, + "content": "X _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 187, + 414, + 201 + ], + "score": 1.0, + "content": ", response vector", + "type": "text" + }, + { + "bbox": [ + 415, + 191, + 423, + 200 + ], + "score": 0.37, + "content": "\\textbf { { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 187, + 470, + 201 + ], + "score": 1.0, + "content": "and model", + "type": "text" + }, + { + "bbox": [ + 471, + 188, + 505, + 200 + ], + "score": 0.9, + "content": "f ( { \\pmb x } ) =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 107, + 199, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 107, + 199, + 164, + 211 + ], + "score": 0.91, + "content": "\\langle \\phi ( \\pmb { x } ^ { \\top } W ) , \\hat { \\pmb { a } } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 199, + 295, + 213 + ], + "score": 1.0, + "content": "with fixed first layer coefficients", + "type": "text" + }, + { + "bbox": [ + 295, + 200, + 307, + 209 + ], + "score": 0.68, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 199, + 441, + 213 + ], + "score": 1.0, + "content": ", gradient flow on the coefficients", + "type": "text" + }, + { + "bbox": [ + 442, + 202, + 449, + 209 + ], + "score": 0.32, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 199, + 506, + 213 + ], + "score": 1.0, + "content": "starting from", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 209, + 460, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 231, + 223 + ], + "score": 1.0, + "content": "zero initialization converges to", + "type": "text" + }, + { + "bbox": [ + 232, + 209, + 270, + 221 + ], + "score": 0.91, + "content": "\\mathbf { \\bar { a } } = \\Phi ^ { \\dagger } \\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 210, + 300, + 223 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 300, + 211, + 306, + 222 + ], + "score": 0.8, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 210, + 460, + 223 + ], + "score": 1.0, + "content": "stands for the Moore-Penrose inverse.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 187, + 506, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 229, + 466, + 241 + ], + "lines": [ + { + "bbox": [ + 109, + 228, + 467, + 243 + ], + "spans": [ + { + "bbox": [ + 109, + 228, + 467, + 243 + ], + "score": 1.0, + "content": "We make two assumptions on the data and the teacher model to simplify the computation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 109, + 228, + 467, + 243 + ] + }, + { + "type": "text", + "bbox": [ + 119, + 242, + 485, + 255 + ], + "lines": [ + { + "bbox": [ + 124, + 242, + 483, + 256 + ], + "spans": [ + { + "bbox": [ + 124, + 242, + 230, + 256 + ], + "score": 1.0, + "content": "(A1) Gaussian Features:", + "type": "text" + }, + { + "bbox": [ + 231, + 243, + 287, + 255 + ], + "score": 0.94, + "content": "\\pmb { x } _ { i } \\sim \\mathcal { N } ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 242, + 387, + 256 + ], + "score": 1.0, + "content": "; (A2) Linear Teacher:", + "type": "text" + }, + { + "bbox": [ + 387, + 243, + 446, + 255 + ], + "score": 0.82, + "content": "F ( { \\pmb x } ) = \\langle { \\pmb x } , { \\pmb \\beta } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 242, + 451, + 256 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 451, + 243, + 483, + 255 + ], + "score": 0.51, + "content": "\\| { \\boldsymbol { \\beta } } \\| = r", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 124, + 242, + 483, + 256 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 258, + 504, + 282 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 262, + 273 + ], + "score": 1.0, + "content": "Denote the linear student network as", + "type": "text" + }, + { + "bbox": [ + 262, + 258, + 325, + 272 + ], + "score": 0.95, + "content": "f ( \\pmb { x } ) = \\langle \\pmb { x } , \\hat { \\beta } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 258, + 357, + 273 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 357, + 258, + 399, + 271 + ], + "score": 0.92, + "content": "\\hat { \\boldsymbol { \\beta } } = W \\hat { \\mathbf { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 258, + 418, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 419, + 260, + 426, + 270 + ], + "score": 0.82, + "content": "\\hat { \\textbf { \\textit a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 258, + 505, + 273 + ], + "score": 1.0, + "content": "is the least-square", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 270, + 485, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 485, + 283 + ], + "score": 1.0, + "content": "solution defined by Lemma 1. We write the population risk in its bias-variance decomposition.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 258, + 505, + 283 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 283, + 461, + 318 + ], + "lines": [ + { + "bbox": [ + 149, + 283, + 461, + 318 + ], + "spans": [ + { + "bbox": [ + 149, + 283, + 461, + 318 + ], + "score": 0.91, + "content": "R = \\mathbb { E } _ { { \\mathbf { x } } \\sim P _ { \\mathbf { x } } } [ \\Vert \\hat { \\beta } - \\beta \\Vert _ { \\Sigma } ^ { 2 } \\vert { \\cal X } , { \\cal W } ] = \\underbrace { \\Vert \\mathbb { E } [ \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ] - \\beta \\Vert _ { 2 } ^ { 2 } } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathrm { t r } \\left( \\mathrm { C o v } ( \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ) \\right) } _ { { \\cal V } = \\mathrm { v a r i a n c e } } ,", + "type": "interline_equation", + "image_path": "d18643c3281713eae817cf95ea0a2aa7076ffbe532bc9cb5d02f70cfadd69742.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 149, + 283, + 461, + 294.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 149, + 294.6666666666667, + 461, + 306.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 149, + 306.33333333333337, + 461, + 318.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 321, + 467, + 334 + ], + "lines": [ + { + "bbox": [ + 105, + 319, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 133, + 336 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 320, + 197, + 335 + ], + "score": 0.93, + "content": "\\left\\| \\pmb { x } \\right\\| _ { \\Sigma } ^ { 2 } = \\pmb { x } ^ { \\top } \\Sigma \\pmb { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 319, + 469, + 336 + ], + "score": 1.0, + "content": ". We compute the bias and the variance separately to obtain the risk.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 319, + 469, + 336 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 337, + 441, + 352 + ], + "lines": [ + { + "bbox": [ + 103, + 332, + 446, + 356 + ], + "spans": [ + { + "bbox": [ + 103, + 332, + 254, + 356 + ], + "score": 1.0, + "content": "Theorem 2. Given (A1)(A2) and let", + "type": "text" + }, + { + "bbox": [ + 254, + 340, + 267, + 351 + ], + "score": 0.73, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 332, + 286, + 356 + ], + "score": 1.0, + "content": "i.i.d. ∼", + "type": "text" + }, + { + "bbox": [ + 287, + 338, + 339, + 352 + ], + "score": 0.87, + "content": "\\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 332, + 354, + 356 + ], + "score": 1.0, + "content": ", at", + "type": "text" + }, + { + "bbox": [ + 354, + 340, + 407, + 351 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 332, + 446, + 356 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 103, + 332, + 446, + 356 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 353, + 498, + 414 + ], + "lines": [ + { + "bbox": [ + 111, + 353, + 498, + 414 + ], + "spans": [ + { + "bbox": [ + 111, + 353, + 498, + 414 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < \\gamma _ { 1 } , } \\\\ { \\frac { \\gamma _ { 1 } } { g _ { 1 } } \\sigma ^ { 2 } , } & R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 2 } g _ { 1 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { g _ { 1 } + g _ { 2 } } { g _ { 1 } g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } > 1 . } \\end{array} } \\end{array} } \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "8b38e4e0f9807763f6f0a6585220ff7376399aa2616d84e51b73efd6e2e1732c.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 353, + 498, + 373.3333333333333 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 373.3333333333333, + 498, + 393.66666666666663 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 393.66666666666663, + 498, + 413.99999999999994 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 362, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 363, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 133, + 440 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 425, + 282, + 438 + ], + "score": 0.48, + "content": "d / n \\to \\gamma _ { 1 } , h / n \\to \\gamma _ { 2 } , g _ { 1 } = | \\gamma _ { 1 } - 1 | ,", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 423, + 303, + 440 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 303, + 425, + 359, + 438 + ], + "score": 0.91, + "content": "g _ { 2 } = | \\gamma _ { 2 } - 1 |", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 423, + 363, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 423, + 363, + 440 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 444, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 193, + 457 + ], + "score": 1.0, + "content": "We observe that when", + "type": "text" + }, + { + "bbox": [ + 193, + 445, + 219, + 455 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 444, + 237, + 457 + ], + "score": 1.0, + "content": "(i.e.", + "type": "text" + }, + { + "bbox": [ + 237, + 446, + 267, + 456 + ], + "score": 0.87, + "content": "\\gamma _ { 1 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "), we obtain the double descent risk curve, i.e., the population", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 227, + 469 + ], + "score": 1.0, + "content": "risk achieves its maximum at", + "type": "text" + }, + { + "bbox": [ + 228, + 457, + 260, + 467 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 455, + 398, + 469 + ], + "score": 1.0, + "content": "and further overparameterization", + "type": "text" + }, + { + "bbox": [ + 399, + 456, + 432, + 467 + ], + "score": 0.85, + "content": "( \\gamma _ { 2 } > 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 455, + 505, + 469 + ], + "score": 1.0, + "content": ") reduces both the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 272, + 480 + ], + "score": 1.0, + "content": "bias and the variance. Conversely when", + "type": "text" + }, + { + "bbox": [ + 273, + 467, + 300, + 477 + ], + "score": 0.9, + "content": "n > d", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 465, + 318, + 480 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 319, + 467, + 346, + 477 + ], + "score": 0.9, + "content": "h > d", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 465, + 367, + 480 + ], + "score": 1.0, + "content": "(i.e.", + "type": "text" + }, + { + "bbox": [ + 367, + 467, + 438, + 478 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < \\operatorname* { m i n } ( 1 , \\gamma _ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 465, + 505, + 480 + ], + "score": 1.0, + "content": ", the population", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 477, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 398, + 493 + ], + "score": 1.0, + "content": "risk becomes constant and equals to that of the minimum-norm solution", + "type": "text" + }, + { + "bbox": [ + 399, + 478, + 453, + 491 + ], + "score": 0.95, + "content": "{ \\hat { \\boldsymbol { \\beta } } } _ { \\operatorname* { m i n } } = { \\boldsymbol { X } } ^ { \\dagger } { \\boldsymbol { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 477, + 505, + 493 + ], + "score": 1.0, + "content": "on the input", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 490, + 143, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 143, + 502 + ], + "score": 1.0, + "content": "features.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 104, + 444, + 505, + 502 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 504, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "Training the First Layer. When the first layer of a linear network is optimized via gradient flow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 524, + 418, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 402, + 536 + ], + "score": 1.0, + "content": "and the second layer is fixed, the following holds for zero-initialization of", + "type": "text" + }, + { + "bbox": [ + 403, + 525, + 414, + 534 + ], + "score": 0.76, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 524, + 418, + 536 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 513, + 505, + 536 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 199, + 551 + ], + "score": 1.0, + "content": "Proposition 3. Given", + "type": "text" + }, + { + "bbox": [ + 199, + 537, + 245, + 550 + ], + "score": 0.91, + "content": "W ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 535, + 288, + 551 + ], + "score": 1.0, + "content": "and fixed", + "type": "text" + }, + { + "bbox": [ + 288, + 537, + 326, + 549 + ], + "score": 0.92, + "content": "\\pm \\ : 0", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 535, + 380, + 551 + ], + "score": 1.0, + "content": ", at any time", + "type": "text" + }, + { + "bbox": [ + 380, + 538, + 406, + 548 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "of the gradient flow on", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 549, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 118, + 561 + ], + "score": 0.4, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 549, + 122, + 565 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 123, + 551, + 146, + 563 + ], + "score": 0.78, + "content": "W ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 549, + 227, + 565 + ], + "score": 1.0, + "content": "is rank-1. Further,", + "type": "text" + }, + { + "bbox": [ + 227, + 549, + 267, + 563 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = { \\widehat { \\boldsymbol { W } } } \\mathbf { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 549, + 441, + 565 + ], + "score": 1.0, + "content": "converges to the least squares solution of", + "type": "text" + }, + { + "bbox": [ + 441, + 549, + 486, + 563 + ], + "score": 0.93, + "content": "\\boldsymbol { y } = \\boldsymbol { X } ^ { \\intercal } \\hat { \\boldsymbol { \\beta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 549, + 506, + 565 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 561, + 399, + 575 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 399, + 575 + ], + "score": 1.0, + "content": "population risk of which is given in (Hastie et al., 2019, Thm. 1 & 3)) as", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 104, + 535, + 506, + 575 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 574, + 428, + 600 + ], + "lines": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "spans": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } r ^ { 2 } + \\frac { 1 } { \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "f5a4682cd92d7d3b10ff18ef37be5a460f8f02025a4402ea4f6cce0978f6f081.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 182, + 574, + 428, + 600 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 294, + 627 + ], + "score": 1.0, + "content": "In this case, overparameterization by increasing", + "type": "text" + }, + { + "bbox": [ + 295, + 616, + 306, + 626 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "does not influence the population risk. In addition,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "since the obtained two-layer linear model is equivalent to the minimum-norm solution on the input", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 636, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 636, + 140, + 651 + ], + "score": 1.0, + "content": "features", + "type": "text" + }, + { + "bbox": [ + 140, + 636, + 162, + 650 + ], + "score": 0.92, + "content": "\\hat { \\beta } _ { \\mathrm { m i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 636, + 506, + 651 + ], + "score": 1.0, + "content": ", optimizing the first layer always results in smaller or equal population risk compared", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 649, + 231, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 231, + 662 + ], + "score": 1.0, + "content": "to optimizing the second layer.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 614, + 506, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "In this simple scenario for two-layer linear networks, double descent is observed only when the second", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "layer is optimized, which reduces the objective to least squares regression on the intermediate features.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "On the other hand, training the first layer from zero-initialization always yields the same solution", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "that is independent to overparameterization. One natural question to ask is: does this phenomenon", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "generalize to nonlinear two-layer neural networks? The following sections answer this question in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 268, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 268, + 732 + ], + "score": 1.0, + "content": "the affirmative under certain conditions.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 666, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 80, + 496, + 186 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 80, + 496, + 186 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 496, + 186 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 496, + 186 + ], + "score": 0.97, + "type": "image", + "image_path": "277c0189331694fef2e6d8835188d44b8e63f3a5b84f2bea72662ce57820d839.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 496, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 115.33333333333334, + 496, + 150.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 150.66666666666669, + 496, + 186.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 193, + 506, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "Figure 2: Population risk of two-layer neural networks with optimized second layer under (A1)(A2). Brighter", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 203, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 182, + 215 + ], + "score": 1.0, + "content": "color indicates larger", + "type": "text" + }, + { + "bbox": [ + 183, + 205, + 193, + 214 + ], + "score": 0.83, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 203, + 305, + 215 + ], + "score": 1.0, + "content": ". (a) risk of linear network with", + "type": "text" + }, + { + "bbox": [ + 305, + 203, + 351, + 214 + ], + "score": 0.92, + "content": "r ^ { 2 } / \\sigma ^ { 2 } \\overset { \\cdot } { = } 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 203, + 366, + 215 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 204, + 394, + 214 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 203, + 398, + 215 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 398, + 204, + 428, + 214 + ], + "score": 0.88, + "content": "( \\gamma _ { 1 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 203, + 506, + 215 + ], + "score": 1.0, + "content": "is shown in Figure 1)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 213, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 376, + 226 + ], + "score": 1.0, + "content": "(b) variance of network with ReLU activation. Black line corresponds to", + "type": "text" + }, + { + "bbox": [ + 376, + 215, + 411, + 224 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 213, + 507, + 226 + ], + "score": 1.0, + "content": "predicted by Corollary 5.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 363, + 235 + ], + "score": 1.0, + "content": "(c) bias of network with ReLU activation. Black line corresponds to", + "type": "text" + }, + { + "bbox": [ + 363, + 225, + 398, + 234 + ], + "score": 0.9, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "for linear network, which is", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 233, + 491, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 300, + 244 + ], + "score": 1.0, + "content": "empirically observed as an upper-bound. Note that as", + "type": "text" + }, + { + "bbox": [ + 300, + 234, + 329, + 244 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 233, + 491, + 244 + ], + "score": 1.0, + "content": "both bias and variance becomes unbounded.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 264, + 410, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 411, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 411, + 279 + ], + "score": 1.0, + "content": "4 NONLINEAR MODEL: OPTIMIZING THE SECOND LAYER", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 334, + 302 + ], + "score": 1.0, + "content": "In this section, we analyze the case when the second layer", + "type": "text" + }, + { + "bbox": [ + 335, + 292, + 342, + 300 + ], + "score": 0.75, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 289, + 429, + 302 + ], + "score": 1.0, + "content": "is learned under fixed", + "type": "text" + }, + { + "bbox": [ + 430, + 290, + 442, + 300 + ], + "score": 0.69, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "and a nonlinear", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 301, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 183, + 312 + ], + "score": 1.0, + "content": "activation function", + "type": "text" + }, + { + "bbox": [ + 183, + 301, + 190, + 312 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 302, + 505, + 312 + ], + "score": 1.0, + "content": "(a random feature model). We first observe that by Lemma 1, the gradient flow", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 311, + 471, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 178, + 325 + ], + "score": 1.0, + "content": "finds the solution", + "type": "text" + }, + { + "bbox": [ + 178, + 311, + 216, + 324 + ], + "score": 0.92, + "content": "\\hat { \\mathbf { a } } = \\Phi ^ { \\dagger } \\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 311, + 471, + 325 + ], + "score": 1.0, + "content": ". We again consider the following bias-variance decomposition.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 328, + 493, + 388 + ], + "lines": [ + { + "bbox": [ + 117, + 328, + 493, + 388 + ], + "spans": [ + { + "bbox": [ + 117, + 328, + 493, + 388 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { R = \\mathbb { E } _ { \\boldsymbol { x } \\sim P _ { \\boldsymbol { x } } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } | \\boldsymbol { X } , \\boldsymbol { W } ] } \\\\ & { \\quad = \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } | \\boldsymbol { X } , \\boldsymbol { W } ] - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } ] } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } ] \\| _ { 2 } ^ { 2 } \\big | \\boldsymbol { X } , \\boldsymbol { W } ] } _ { \\boldsymbol { V } = \\mathrm { v a r i a n c e } } . } \\end{array}", + "type": "interline_equation", + "image_path": "79e7a6630230c8b5d7c361b0fb4680b51a56f6463dff229f6bdafaf826b79294.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 117, + 328, + 493, + 348.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 117, + 348.0, + 493, + 368.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 117, + 368.0, + 493, + 388.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 504, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "We highlight that the variance term does not depend on the target function. The following result", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 408, + 329, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 329, + 420 + ], + "score": 1.0, + "content": "characterizes the variance of the random feature model.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 429, + 437 + ], + "lines": [ + { + "bbox": [ + 103, + 417, + 431, + 441 + ], + "spans": [ + { + "bbox": [ + 103, + 417, + 223, + 441 + ], + "score": 1.0, + "content": "Theorem 4. Given (A1) and", + "type": "text" + }, + { + "bbox": [ + 224, + 422, + 310, + 437 + ], + "score": 0.77, + "content": "{ \\pmb w } _ { i } \\overset { \\mathrm { i . i . d . } } { \\sim } \\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 417, + 337, + 441 + ], + "score": 1.0, + "content": ", when", + "type": "text" + }, + { + "bbox": [ + 337, + 425, + 389, + 436 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 417, + 431, + 441 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 443, + 495, + 503 + ], + "lines": [ + { + "bbox": [ + 114, + 443, + 495, + 503 + ], + "spans": [ + { + "bbox": [ + 114, + 443, + 495, + 503 + ], + "score": 0.89, + "content": "V = \\{ \\begin{array} { l l } { \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } & { \\gamma _ { 2 } < 1 , } \\\\ { \\quad } & { \\gamma _ { 2 } < 1 \\leq r \\leq r , } \\\\ { \\sigma ^ { 2 } \\displaystyle \\operatorname* { l i m } _ { \\xi 0 } - [ \\gamma _ { 2 } \\frac { \\partial } { \\partial x } m _ { 1 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\partial } { \\partial x } m _ { 2 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } ] } & { \\gamma _ { 2 } > 1 . } \\end{array} ", + "type": "interline_equation", + "image_path": "f60367fc1748700e44737fc2d426db24d12634332d11a3cbb9a7290314a8d432.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 114, + 443, + 495, + 463.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 114, + 463.0, + 495, + 483.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 114, + 483.0, + 495, + 503.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 507, + 442, + 520 + ], + "lines": [ + { + "bbox": [ + 104, + 505, + 441, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 143, + 523 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 143, + 508, + 189, + 520 + ], + "score": 0.6, + "content": "m _ { 1 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 505, + 193, + 523 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 194, + 508, + 239, + 520 + ], + "score": 0.71, + "content": "m _ { 2 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 505, + 339, + 523 + ], + "score": 1.0, + "content": "is the unique solution in", + "type": "text" + }, + { + "bbox": [ + 340, + 507, + 428, + 520 + ], + "score": 0.92, + "content": "\\{ | m _ { 1 } | , | m _ { 2 } | < 1 / { \\mathfrak { T } } \\xi \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 505, + 441, + 523 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 524, + 478, + 585 + ], + "lines": [ + { + "bbox": [ + 119, + 524, + 478, + 585 + ], + "spans": [ + { + "bbox": [ + 119, + 524, + 478, + 585 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { m _ { 1 } ^ { - 1 } = - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - c _ { 1 } m _ { 2 } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - 2 \\tau c _ { 2 } m _ { 1 } m _ { 2 } + c _ { 2 } ^ { 2 } m _ { 1 } m _ { 2 } ^ { 2 } } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\\\ & { m _ { 2 } ^ { - 1 } = - \\xi - r \\gamma _ { 2 } m _ { 1 } + \\frac { \\gamma _ { 2 } c _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "b5a38e0b58ff533d613c40382db42c25a20f5c774b0213de7c35e0a32bfe7e6d.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 119, + 524, + 478, + 544.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 119, + 544.3333333333334, + 478, + 564.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 119, + 564.6666666666667, + 478, + 585.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 588, + 480, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 587, + 480, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 172, + 603 + ], + "score": 1.0, + "content": "where variables", + "type": "text" + }, + { + "bbox": [ + 173, + 590, + 198, + 601 + ], + "score": 0.91, + "content": "\\xi , \\rho , \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 587, + 233, + 603 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 234, + 589, + 265, + 601 + ], + "score": 0.9, + "content": "\\Im \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 587, + 277, + 603 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 277, + 589, + 302, + 601 + ], + "score": 0.82, + "content": "\\xi < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 587, + 305, + 603 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 305, + 590, + 349, + 601 + ], + "score": 0.78, + "content": "\\rho > \\tau > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 587, + 410, + 603 + ], + "score": 1.0, + "content": ", and constants", + "type": "text" + }, + { + "bbox": [ + 411, + 591, + 433, + 601 + ], + "score": 0.89, + "content": "c _ { 1 } , c _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 587, + 480, + 603 + ], + "score": 1.0, + "content": "defined as,", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 605, + 404, + 620 + ], + "lines": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "spans": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "score": 0.86, + "content": "c _ { 1 } = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , \\quad c _ { 2 } = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } ,", + "type": "interline_equation", + "image_path": "3b3fa5be4207d4f5ea8dff7f6542e69ad92e9a815e63fce4b8a7ab24e6234fba.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 625, + 342, + 637 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 342, + 641 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 120, + 641 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 625, + 174, + 637 + ], + "score": 0.93, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 622, + 194, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 626, + 207, + 637 + ], + "score": 0.86, + "content": "\\Im \\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 622, + 332, + 641 + ], + "score": 1.0, + "content": "denoting the imaginary part of", + "type": "text" + }, + { + "bbox": [ + 333, + 626, + 339, + 637 + ], + "score": 0.81, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 622, + 342, + 641 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 640, + 326, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 326, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 146, + 654 + ], + "score": 1.0, + "content": "Remark.", + "type": "text" + }, + { + "bbox": [ + 146, + 641, + 178, + 652 + ], + "score": 0.9, + "content": "c _ { 1 } \\geq c _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 639, + 281, + 654 + ], + "score": 1.0, + "content": "and the equality holds iff", + "type": "text" + }, + { + "bbox": [ + 282, + 641, + 289, + 652 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 639, + 326, + 654 + ], + "score": 1.0, + "content": "is linear.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 655, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 195, + 668 + ], + "score": 1.0, + "content": "Corollary 5. If we let", + "type": "text" + }, + { + "bbox": [ + 196, + 657, + 232, + 667 + ], + "score": 0.85, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 654, + 506, + 668 + ], + "score": 1.0, + "content": ", the variance is equal to the lowest value of the variance of the linear", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 282, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 133, + 681 + ], + "score": 1.0, + "content": "model", + "type": "text" + }, + { + "bbox": [ + 133, + 667, + 278, + 680 + ], + "score": 0.91, + "content": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\sigma ^ { 2 } \\mathrm { m i n } \\{ \\gamma _ { 2 } , 1 \\} / | 1 - \\gamma _ { 2 } |", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 666, + 282, + 681 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 104, + 685, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 104, + 685, + 507, + 702 + ], + "score": 1.0, + "content": "The proof of Theorem 4 largely follows from Hastie et al. (2019) with techniques similar to Cheng", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "and Singer (2013), but with modifications in otder to handle unnormalized and uncentered activation", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "functions. The above theorem holds irrespective of the underlying teacher model, and is consistent", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "score": 1.0, + "content": "with the double descent risk curve as it suggests that for all", + "type": "text" + }, + { + "bbox": [ + 341, + 722, + 351, + 732 + ], + "score": 0.84, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 720, + 506, + 733 + ], + "score": 1.0, + "content": ", variance of the random feature model", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 80, + 496, + 186 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 80, + 496, + 186 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 496, + 186 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 496, + 186 + ], + "score": 0.97, + "type": "image", + "image_path": "277c0189331694fef2e6d8835188d44b8e63f3a5b84f2bea72662ce57820d839.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 496, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 115.33333333333334, + 496, + 150.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 150.66666666666669, + 496, + 186.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 193, + 506, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "Figure 2: Population risk of two-layer neural networks with optimized second layer under (A1)(A2). Brighter", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 203, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 182, + 215 + ], + "score": 1.0, + "content": "color indicates larger", + "type": "text" + }, + { + "bbox": [ + 183, + 205, + 193, + 214 + ], + "score": 0.83, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 203, + 305, + 215 + ], + "score": 1.0, + "content": ". (a) risk of linear network with", + "type": "text" + }, + { + "bbox": [ + 305, + 203, + 351, + 214 + ], + "score": 0.92, + "content": "r ^ { 2 } / \\sigma ^ { 2 } \\overset { \\cdot } { = } 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 203, + 366, + 215 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 204, + 394, + 214 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 203, + 398, + 215 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 398, + 204, + 428, + 214 + ], + "score": 0.88, + "content": "( \\gamma _ { 1 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 203, + 506, + 215 + ], + "score": 1.0, + "content": "is shown in Figure 1)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 213, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 376, + 226 + ], + "score": 1.0, + "content": "(b) variance of network with ReLU activation. Black line corresponds to", + "type": "text" + }, + { + "bbox": [ + 376, + 215, + 411, + 224 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 213, + 507, + 226 + ], + "score": 1.0, + "content": "predicted by Corollary 5.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 363, + 235 + ], + "score": 1.0, + "content": "(c) bias of network with ReLU activation. Black line corresponds to", + "type": "text" + }, + { + "bbox": [ + 363, + 225, + 398, + 234 + ], + "score": 0.9, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "for linear network, which is", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 233, + 491, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 300, + 244 + ], + "score": 1.0, + "content": "empirically observed as an upper-bound. Note that as", + "type": "text" + }, + { + "bbox": [ + 300, + 234, + 329, + 244 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 233, + 491, + 244 + ], + "score": 1.0, + "content": "both bias and variance becomes unbounded.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 264, + 410, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 411, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 411, + 279 + ], + "score": 1.0, + "content": "4 NONLINEAR MODEL: OPTIMIZING THE SECOND LAYER", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 334, + 302 + ], + "score": 1.0, + "content": "In this section, we analyze the case when the second layer", + "type": "text" + }, + { + "bbox": [ + 335, + 292, + 342, + 300 + ], + "score": 0.75, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 289, + 429, + 302 + ], + "score": 1.0, + "content": "is learned under fixed", + "type": "text" + }, + { + "bbox": [ + 430, + 290, + 442, + 300 + ], + "score": 0.69, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "and a nonlinear", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 301, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 183, + 312 + ], + "score": 1.0, + "content": "activation function", + "type": "text" + }, + { + "bbox": [ + 183, + 301, + 190, + 312 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 302, + 505, + 312 + ], + "score": 1.0, + "content": "(a random feature model). We first observe that by Lemma 1, the gradient flow", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 311, + 471, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 178, + 325 + ], + "score": 1.0, + "content": "finds the solution", + "type": "text" + }, + { + "bbox": [ + 178, + 311, + 216, + 324 + ], + "score": 0.92, + "content": "\\hat { \\mathbf { a } } = \\Phi ^ { \\dagger } \\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 311, + 471, + 325 + ], + "score": 1.0, + "content": ". We again consider the following bias-variance decomposition.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 289, + 505, + 325 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 328, + 493, + 388 + ], + "lines": [ + { + "bbox": [ + 117, + 328, + 493, + 388 + ], + "spans": [ + { + "bbox": [ + 117, + 328, + 493, + 388 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { R = \\mathbb { E } _ { \\boldsymbol { x } \\sim P _ { \\boldsymbol { x } } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } | \\boldsymbol { X } , \\boldsymbol { W } ] } \\\\ & { \\quad = \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } | \\boldsymbol { X } , \\boldsymbol { W } ] - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } ] } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } ] \\| _ { 2 } ^ { 2 } \\big | \\boldsymbol { X } , \\boldsymbol { W } ] } _ { \\boldsymbol { V } = \\mathrm { v a r i a n c e } } . } \\end{array}", + "type": "interline_equation", + "image_path": "79e7a6630230c8b5d7c361b0fb4680b51a56f6463dff229f6bdafaf826b79294.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 117, + 328, + 493, + 348.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 117, + 348.0, + 493, + 368.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 117, + 368.0, + 493, + 388.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 504, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "We highlight that the variance term does not depend on the target function. The following result", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 408, + 329, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 329, + 420 + ], + "score": 1.0, + "content": "characterizes the variance of the random feature model.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 396, + 505, + 420 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 429, + 437 + ], + "lines": [ + { + "bbox": [ + 103, + 417, + 431, + 441 + ], + "spans": [ + { + "bbox": [ + 103, + 417, + 223, + 441 + ], + "score": 1.0, + "content": "Theorem 4. Given (A1) and", + "type": "text" + }, + { + "bbox": [ + 224, + 422, + 310, + 437 + ], + "score": 0.77, + "content": "{ \\pmb w } _ { i } \\overset { \\mathrm { i . i . d . } } { \\sim } \\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 417, + 337, + 441 + ], + "score": 1.0, + "content": ", when", + "type": "text" + }, + { + "bbox": [ + 337, + 425, + 389, + 436 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 417, + 431, + 441 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 103, + 417, + 431, + 441 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 443, + 495, + 503 + ], + "lines": [ + { + "bbox": [ + 114, + 443, + 495, + 503 + ], + "spans": [ + { + "bbox": [ + 114, + 443, + 495, + 503 + ], + "score": 0.89, + "content": "V = \\{ \\begin{array} { l l } { \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } & { \\gamma _ { 2 } < 1 , } \\\\ { \\quad } & { \\gamma _ { 2 } < 1 \\leq r \\leq r , } \\\\ { \\sigma ^ { 2 } \\displaystyle \\operatorname* { l i m } _ { \\xi 0 } - [ \\gamma _ { 2 } \\frac { \\partial } { \\partial x } m _ { 1 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\partial } { \\partial x } m _ { 2 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } ] } & { \\gamma _ { 2 } > 1 . } \\end{array} ", + "type": "interline_equation", + "image_path": "f60367fc1748700e44737fc2d426db24d12634332d11a3cbb9a7290314a8d432.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 114, + 443, + 495, + 463.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 114, + 463.0, + 495, + 483.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 114, + 483.0, + 495, + 503.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 507, + 442, + 520 + ], + "lines": [ + { + "bbox": [ + 104, + 505, + 441, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 143, + 523 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 143, + 508, + 189, + 520 + ], + "score": 0.6, + "content": "m _ { 1 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 505, + 193, + 523 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 194, + 508, + 239, + 520 + ], + "score": 0.71, + "content": "m _ { 2 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 505, + 339, + 523 + ], + "score": 1.0, + "content": "is the unique solution in", + "type": "text" + }, + { + "bbox": [ + 340, + 507, + 428, + 520 + ], + "score": 0.92, + "content": "\\{ | m _ { 1 } | , | m _ { 2 } | < 1 / { \\mathfrak { T } } \\xi \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 505, + 441, + 523 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 505, + 441, + 523 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 524, + 478, + 585 + ], + "lines": [ + { + "bbox": [ + 119, + 524, + 478, + 585 + ], + "spans": [ + { + "bbox": [ + 119, + 524, + 478, + 585 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { m _ { 1 } ^ { - 1 } = - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - c _ { 1 } m _ { 2 } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - 2 \\tau c _ { 2 } m _ { 1 } m _ { 2 } + c _ { 2 } ^ { 2 } m _ { 1 } m _ { 2 } ^ { 2 } } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\\\ & { m _ { 2 } ^ { - 1 } = - \\xi - r \\gamma _ { 2 } m _ { 1 } + \\frac { \\gamma _ { 2 } c _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "b5a38e0b58ff533d613c40382db42c25a20f5c774b0213de7c35e0a32bfe7e6d.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 119, + 524, + 478, + 544.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 119, + 544.3333333333334, + 478, + 564.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 119, + 564.6666666666667, + 478, + 585.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 588, + 480, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 587, + 480, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 172, + 603 + ], + "score": 1.0, + "content": "where variables", + "type": "text" + }, + { + "bbox": [ + 173, + 590, + 198, + 601 + ], + "score": 0.91, + "content": "\\xi , \\rho , \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 587, + 233, + 603 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 234, + 589, + 265, + 601 + ], + "score": 0.9, + "content": "\\Im \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 587, + 277, + 603 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 277, + 589, + 302, + 601 + ], + "score": 0.82, + "content": "\\xi < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 587, + 305, + 603 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 305, + 590, + 349, + 601 + ], + "score": 0.78, + "content": "\\rho > \\tau > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 587, + 410, + 603 + ], + "score": 1.0, + "content": ", and constants", + "type": "text" + }, + { + "bbox": [ + 411, + 591, + 433, + 601 + ], + "score": 0.89, + "content": "c _ { 1 } , c _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 587, + 480, + 603 + ], + "score": 1.0, + "content": "defined as,", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 587, + 480, + 603 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 605, + 404, + 620 + ], + "lines": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "spans": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "score": 0.86, + "content": "c _ { 1 } = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , \\quad c _ { 2 } = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } ,", + "type": "interline_equation", + "image_path": "3b3fa5be4207d4f5ea8dff7f6542e69ad92e9a815e63fce4b8a7ab24e6234fba.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 206, + 605, + 404, + 620 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 625, + 342, + 637 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 342, + 641 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 120, + 641 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 625, + 174, + 637 + ], + "score": 0.93, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 622, + 194, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 626, + 207, + 637 + ], + "score": 0.86, + "content": "\\Im \\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 622, + 332, + 641 + ], + "score": 1.0, + "content": "denoting the imaginary part of", + "type": "text" + }, + { + "bbox": [ + 333, + 626, + 339, + 637 + ], + "score": 0.81, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 622, + 342, + 641 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 104, + 622, + 342, + 641 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 640, + 326, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 326, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 146, + 654 + ], + "score": 1.0, + "content": "Remark.", + "type": "text" + }, + { + "bbox": [ + 146, + 641, + 178, + 652 + ], + "score": 0.9, + "content": "c _ { 1 } \\geq c _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 639, + 281, + 654 + ], + "score": 1.0, + "content": "and the equality holds iff", + "type": "text" + }, + { + "bbox": [ + 282, + 641, + 289, + 652 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 639, + 326, + 654 + ], + "score": 1.0, + "content": "is linear.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 639, + 326, + 654 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 655, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 195, + 668 + ], + "score": 1.0, + "content": "Corollary 5. If we let", + "type": "text" + }, + { + "bbox": [ + 196, + 657, + 232, + 667 + ], + "score": 0.85, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 654, + 506, + 668 + ], + "score": 1.0, + "content": ", the variance is equal to the lowest value of the variance of the linear", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 282, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 133, + 681 + ], + "score": 1.0, + "content": "model", + "type": "text" + }, + { + "bbox": [ + 133, + 667, + 278, + 680 + ], + "score": 0.91, + "content": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\sigma ^ { 2 } \\mathrm { m i n } \\{ \\gamma _ { 2 } , 1 \\} / | 1 - \\gamma _ { 2 } |", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 666, + 282, + 681 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 654, + 506, + 681 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 104, + 685, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 104, + 685, + 507, + 702 + ], + "score": 1.0, + "content": "The proof of Theorem 4 largely follows from Hastie et al. (2019) with techniques similar to Cheng", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "and Singer (2013), but with modifications in otder to handle unnormalized and uncentered activation", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "functions. The above theorem holds irrespective of the underlying teacher model, and is consistent", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "score": 1.0, + "content": "with the double descent risk curve as it suggests that for all", + "type": "text" + }, + { + "bbox": [ + 341, + 722, + 351, + 732 + ], + "score": 0.84, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 720, + 506, + 733 + ], + "score": 1.0, + "content": ", variance of the random feature model", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 143, + 95 + ], + "score": 1.0, + "content": "peaks at", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 143, + 83, + 171, + 93 + ], + "score": 0.9, + "content": "h = n", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 172, + 82, + 230, + 95 + ], + "score": 1.0, + "content": "then drops as", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 230, + 84, + 241, + 94 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 241, + 82, + 370, + 95 + ], + "score": 1.0, + "content": "further increases. Note that as", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 370, + 84, + 408, + 94 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 409, + 82, + 505, + 95 + ], + "score": 1.0, + "content": ", a linear and nonlinear", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 312, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 312, + 106 + ], + "score": 1.0, + "content": "network would have the same asymptotic variance.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 685, + 507, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 143, + 95 + ], + "score": 1.0, + "content": "peaks at", + "type": "text" + }, + { + "bbox": [ + 143, + 83, + 171, + 93 + ], + "score": 0.9, + "content": "h = n", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 82, + 230, + 95 + ], + "score": 1.0, + "content": "then drops as", + "type": "text" + }, + { + "bbox": [ + 230, + 84, + 241, + 94 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 82, + 370, + 95 + ], + "score": 1.0, + "content": "further increases. Note that as", + "type": "text" + }, + { + "bbox": [ + 370, + 84, + 408, + 94 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 82, + 505, + 95 + ], + "score": 1.0, + "content": ", a linear and nonlinear", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 312, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 312, + 106 + ], + "score": 1.0, + "content": "network would have the same asymptotic variance.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 105, + 110, + 505, + 133 + ], + "lines": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 444, + 123 + ], + "score": 1.0, + "content": "Since double descent is observed in the variance term, we do not derive the bias for all", + "type": "text" + }, + { + "bbox": [ + 444, + 112, + 469, + 122 + ], + "score": 0.83, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 110, + 506, + 123 + ], + "score": 1.0, + "content": ". Instead,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 416, + 133 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 380, + 133 + ], + "score": 1.0, + "content": "we show that for linear teacher, the bias also becomes unbounded as", + "type": "text" + }, + { + "bbox": [ + 380, + 122, + 411, + 133 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 121, + 416, + 133 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 504, + 163 + ], + "lines": [ + { + "bbox": [ + 102, + 132, + 509, + 156 + ], + "spans": [ + { + "bbox": [ + 102, + 132, + 196, + 156 + ], + "score": 1.0, + "content": "Proposition 6. Given", + "type": "text" + }, + { + "bbox": [ + 197, + 140, + 234, + 151 + ], + "score": 0.27, + "content": "( A I ) ( A 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 132, + 252, + 156 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 252, + 141, + 266, + 151 + ], + "score": 0.48, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 132, + 285, + 156 + ], + "score": 1.0, + "content": "i.i.d. ∼", + "type": "text" + }, + { + "bbox": [ + 286, + 139, + 322, + 152 + ], + "score": 0.76, + "content": "\\mathcal { N } ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 132, + 346, + 156 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 346, + 140, + 381, + 150 + ], + "score": 0.85, + "content": "B \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 132, + 393, + 156 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 393, + 141, + 425, + 151 + ], + "score": 0.85, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 132, + 485, + 156 + ], + "score": 1.0, + "content": ". Furthermore,", + "type": "text" + }, + { + "bbox": [ + 486, + 141, + 495, + 150 + ], + "score": 0.75, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 132, + 509, + 156 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 185, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 152, + 163 + ], + "score": 1.0, + "content": "finite when", + "type": "text" + }, + { + "bbox": [ + 153, + 152, + 181, + 163 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 151, + 185, + 163 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 505, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 383, + 185 + ], + "score": 1.0, + "content": "Thus we have shown that a “cusp” in the population risk appears at", + "type": "text" + }, + { + "bbox": [ + 384, + 173, + 411, + 183 + ], + "score": 0.9, + "content": "h = n", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 172, + 505, + 185 + ], + "score": 1.0, + "content": ", which aligns with the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 293, + 196 + ], + "score": 1.0, + "content": "double descent phenomenon. Empirically as", + "type": "text" + }, + { + "bbox": [ + 293, + 185, + 331, + 195 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "the nonlinear model also shares the same", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "asymptotic bias with the linear model, as shown in Figure 2. We note that (Mei and Montanari, 2019,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "Thm. 1 & 3) analytically solved the risk of random feature model for a larger class of target functions", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 216, + 451, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 216, + 451, + 228 + ], + "score": 1.0, + "content": "than ours and confirmed that double descent appears in both the bias and the variance.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 244, + 397, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 244, + 398, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 398, + 259 + ], + "score": 1.0, + "content": "5 NONLINEAR MODEL: OPTIMIZING THE FIRST LAYER", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 270, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "Having observed the double descent phenomenon in optimizing the second layer, in the sequel we", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "consider a two-layer neural network with fixed second layer coefficients initialized from a Rademacher√ √", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 292, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 154, + 307 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 154, + 292, + 262, + 306 + ], + "score": 0.92, + "content": "a _ { i } \\sim \\operatorname { U n i f } \\{ - 1 / \\sqrt { h } , 1 \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 293, + 337, + 307 + ], + "score": 1.0, + "content": ", and the first layer", + "type": "text" + }, + { + "bbox": [ + 337, + 294, + 349, + 303 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 293, + 505, + 307 + ], + "score": 1.0, + "content": "is optimized with the following update", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 311, + 455, + 344 + ], + "lines": [ + { + "bbox": [ + 156, + 311, + 455, + 344 + ], + "spans": [ + { + "bbox": [ + 156, + 311, + 455, + 344 + ], + "score": 0.93, + "content": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { \\partial L ( X ; W ) } { \\partial W } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left[ y _ { i } - \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } _ { i } ) \\right] \\pmb { x } _ { i } [ \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } W ) \\circ \\pmb { a } ] ,", + "type": "interline_equation", + "image_path": "fb33ae8432099b540d82f7bb6276429643c466a862ef7cc060ce27adab37408b.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 156, + 311, + 455, + 322.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 156, + 322.0, + 455, + 333.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 156, + 333.0, + 455, + 344.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 506, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "which potentially has different stationary solutions with no explicit form, depending on the initializa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 358, + 507, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 300, + 374 + ], + "score": 1.0, + "content": "tion. We denote the solution of this flow at time", + "type": "text" + }, + { + "bbox": [ + 301, + 362, + 306, + 370 + ], + "score": 0.79, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 358, + 469, + 374 + ], + "score": 1.0, + "content": "started from designated initialization by", + "type": "text" + }, + { + "bbox": [ + 470, + 360, + 502, + 372 + ], + "score": 0.92, + "content": "W ^ { \\mathrm { i n i t } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 358, + 507, + 374 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 371, + 374, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 207, + 387 + ], + "score": 1.0, + "content": "its stationary solution by", + "type": "text" + }, + { + "bbox": [ + 207, + 371, + 219, + 384 + ], + "score": 0.87, + "content": "\\widehat { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 372, + 361, + 387 + ], + "score": 1.0, + "content": ", and the corresponding network by", + "type": "text" + }, + { + "bbox": [ + 362, + 372, + 369, + 385 + ], + "score": 0.84, + "content": "\\hat { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 372, + 374, + 387 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 108, + 387, + 503, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 386, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 209, + 400 + ], + "score": 1.0, + "content": "Remark. Although we let", + "type": "text" + }, + { + "bbox": [ + 209, + 388, + 242, + 397 + ], + "score": 0.88, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 386, + 505, + 400 + ], + "score": 1.0, + "content": ", this dynamics does not corresponds to the population gradient flow", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 288, + 411 + ], + "score": 1.0, + "content": "considered in Tian (2017). For instance when", + "type": "text" + }, + { + "bbox": [ + 288, + 398, + 351, + 410 + ], + "score": 0.93, + "content": "\\pmb { x } \\sim \\mathcal { N } ( 0 , I / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 397, + 506, + 411 + ], + "score": 1.0, + "content": ", the spectrum of the data covariance is", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 408, + 365, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 365, + 422 + ], + "score": 1.0, + "content": "Marcenko–Pastur, whereas the population covariance is identity. ˇ", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 504, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "As we cannot characterize the gradient flow solution from all possible initializations, we consider", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 439, + 250, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 250, + 451 + ], + "score": 1.0, + "content": "two specific scales of initialization:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 506, + 577 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 506, + 496 + ], + "score": 1.0, + "content": "Note that neither of the two initializations correspond to the “mean-field” regime (e.g. analyzed in√", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 493, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 223, + 508 + ], + "score": 1.0, + "content": "Mei et al. (2018)) due to the", + "type": "text" + }, + { + "bbox": [ + 224, + 493, + 249, + 507 + ], + "score": 0.92, + "content": "1 / { \\sqrt { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 493, + 439, + 508 + ], + "score": 1.0, + "content": "scaling of the second layer. In other words, as", + "type": "text" + }, + { + "bbox": [ + 440, + 495, + 447, + 505 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 493, + 506, + 508 + ], + "score": 1.0, + "content": "increases, we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "expect the distance traveled by each parameter to decrease under both initializations. The difference,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "however, is the “relative” amount the parameters traveled compared to their initialized magnitude,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 526, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 541 + ], + "score": 1.0, + "content": "which leads to solutions with contrasting properties. As we will see, under (A1)(A2) and vanishing", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 538, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 194, + 554 + ], + "score": 1.0, + "content": "initialization we have", + "type": "text" + }, + { + "bbox": [ + 194, + 538, + 322, + 553 + ], + "score": 0.93, + "content": "\\lVert W ( 0 ) - \\widehat { W } \\rVert _ { F } / \\lVert W ( 0 ) \\rVert _ { F } \\gg 1", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 540, + 505, + 554 + ], + "score": 1.0, + "content": ", i.e. the contribution of initialization vanishes", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "at the end of training, whereas for non-vanishing initialization the inequality is in the opposite", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 563, + 395, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 162, + 578 + ], + "score": 1.0, + "content": "direction, i.e.", + "type": "text" + }, + { + "bbox": [ + 162, + 563, + 174, + 575 + ], + "score": 0.86, + "content": "\\widehat { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 564, + 366, + 578 + ], + "score": 1.0, + "content": "“barely moves” and resembles the initialization", + "type": "text" + }, + { + "bbox": [ + 367, + 564, + 391, + 577 + ], + "score": 0.92, + "content": "W ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 564, + 395, + 578 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 252, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 253, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 253, + 604 + ], + "score": 1.0, + "content": "5.1 VANISHING INITIALIZATION", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 611, + 505, + 656 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 120, + 623 + ], + "score": 1.0, + "content": "As", + "type": "text" + }, + { + "bbox": [ + 120, + 612, + 162, + 623 + ], + "score": 0.91, + "content": "d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 612, + 505, + 623 + ], + "score": 1.0, + "content": ", the vanishing initialization becomes arbitrarily close to zero-initialization. We thus", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "expect the gradient flow under vanishing initialization to \"resemble\" that of starting from exactly", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "zero if the flow converges sufficiently fast and the gradient being Lipschitz. The Lipschitz condition", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 645, + 433, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 433, + 657 + ], + "score": 1.0, + "content": "(Lemma 18) can be established under the following assumption on the activation.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 114, + 662, + 493, + 676 + ], + "lines": [ + { + "bbox": [ + 117, + 661, + 493, + 677 + ], + "spans": [ + { + "bbox": [ + 117, + 661, + 143, + 677 + ], + "score": 1.0, + "content": "(A3):", + "type": "text" + }, + { + "bbox": [ + 143, + 664, + 150, + 675 + ], + "score": 0.8, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 661, + 313, + 677 + ], + "score": 1.0, + "content": "is smooth, Lipschitz and monotone with", + "type": "text" + }, + { + "bbox": [ + 313, + 661, + 493, + 676 + ], + "score": 0.78, + "content": "\\phi ^ { \\prime } ( 0 ) \\neq 0 ; | \\phi ^ { \\prime } ( \\pm x ) - \\phi ^ { \\prime } ( \\pm \\infty ) | = O ( e ^ { - x } ) .", + "type": "inline_equation" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 395, + 701 + ], + "score": 1.0, + "content": "The above assumption requires that the derivative of the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 395, + 688, + 403, + 699 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 686, + 506, + 701 + ], + "score": 1.0, + "content": "saturates beyond a O(1)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 507, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 507, + 711 + ], + "score": 1.0, + "content": "region, which holds true for the commonly-used smooth activations such as sigmoid and SoftPlus.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "In addition, the choice of scaling ensures that the gradient flow converges sufficiently fast. We thus", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "score": 1.0, + "content": "have the following characterization of the population risk:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 105, + 110, + 505, + 133 + ], + "lines": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 444, + 123 + ], + "score": 1.0, + "content": "Since double descent is observed in the variance term, we do not derive the bias for all", + "type": "text" + }, + { + "bbox": [ + 444, + 112, + 469, + 122 + ], + "score": 0.83, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 110, + 506, + 123 + ], + "score": 1.0, + "content": ". Instead,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 416, + 133 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 380, + 133 + ], + "score": 1.0, + "content": "we show that for linear teacher, the bias also becomes unbounded as", + "type": "text" + }, + { + "bbox": [ + 380, + 122, + 411, + 133 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 121, + 416, + 133 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 110, + 506, + 133 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 504, + 163 + ], + "lines": [ + { + "bbox": [ + 102, + 132, + 509, + 156 + ], + "spans": [ + { + "bbox": [ + 102, + 132, + 196, + 156 + ], + "score": 1.0, + "content": "Proposition 6. Given", + "type": "text" + }, + { + "bbox": [ + 197, + 140, + 234, + 151 + ], + "score": 0.27, + "content": "( A I ) ( A 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 132, + 252, + 156 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 252, + 141, + 266, + 151 + ], + "score": 0.48, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 132, + 285, + 156 + ], + "score": 1.0, + "content": "i.i.d. ∼", + "type": "text" + }, + { + "bbox": [ + 286, + 139, + 322, + 152 + ], + "score": 0.76, + "content": "\\mathcal { N } ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 132, + 346, + 156 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 346, + 140, + 381, + 150 + ], + "score": 0.85, + "content": "B \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 132, + 393, + 156 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 393, + 141, + 425, + 151 + ], + "score": 0.85, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 132, + 485, + 156 + ], + "score": 1.0, + "content": ". Furthermore,", + "type": "text" + }, + { + "bbox": [ + 486, + 141, + 495, + 150 + ], + "score": 0.75, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 132, + 509, + 156 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 185, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 152, + 163 + ], + "score": 1.0, + "content": "finite when", + "type": "text" + }, + { + "bbox": [ + 153, + 152, + 181, + 163 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 151, + 185, + 163 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 102, + 132, + 509, + 163 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 505, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 383, + 185 + ], + "score": 1.0, + "content": "Thus we have shown that a “cusp” in the population risk appears at", + "type": "text" + }, + { + "bbox": [ + 384, + 173, + 411, + 183 + ], + "score": 0.9, + "content": "h = n", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 172, + 505, + 185 + ], + "score": 1.0, + "content": ", which aligns with the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 293, + 196 + ], + "score": 1.0, + "content": "double descent phenomenon. Empirically as", + "type": "text" + }, + { + "bbox": [ + 293, + 185, + 331, + 195 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "the nonlinear model also shares the same", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "asymptotic bias with the linear model, as shown in Figure 2. We note that (Mei and Montanari, 2019,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "Thm. 1 & 3) analytically solved the risk of random feature model for a larger class of target functions", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 216, + 451, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 216, + 451, + 228 + ], + "score": 1.0, + "content": "than ours and confirmed that double descent appears in both the bias and the variance.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8, + "bbox_fs": [ + 104, + 172, + 506, + 228 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 244, + 397, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 244, + 398, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 398, + 259 + ], + "score": 1.0, + "content": "5 NONLINEAR MODEL: OPTIMIZING THE FIRST LAYER", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 270, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "Having observed the double descent phenomenon in optimizing the second layer, in the sequel we", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "consider a two-layer neural network with fixed second layer coefficients initialized from a Rademacher√ √", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 292, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 154, + 307 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 154, + 292, + 262, + 306 + ], + "score": 0.92, + "content": "a _ { i } \\sim \\operatorname { U n i f } \\{ - 1 / \\sqrt { h } , 1 \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 293, + 337, + 307 + ], + "score": 1.0, + "content": ", and the first layer", + "type": "text" + }, + { + "bbox": [ + 337, + 294, + 349, + 303 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 293, + 505, + 307 + ], + "score": 1.0, + "content": "is optimized with the following update", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 270, + 506, + 307 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 311, + 455, + 344 + ], + "lines": [ + { + "bbox": [ + 156, + 311, + 455, + 344 + ], + "spans": [ + { + "bbox": [ + 156, + 311, + 455, + 344 + ], + "score": 0.93, + "content": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { \\partial L ( X ; W ) } { \\partial W } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left[ y _ { i } - \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } _ { i } ) \\right] \\pmb { x } _ { i } [ \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } W ) \\circ \\pmb { a } ] ,", + "type": "interline_equation", + "image_path": "fb33ae8432099b540d82f7bb6276429643c466a862ef7cc060ce27adab37408b.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 156, + 311, + 455, + 322.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 156, + 322.0, + 455, + 333.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 156, + 333.0, + 455, + 344.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 506, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "which potentially has different stationary solutions with no explicit form, depending on the initializa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 358, + 507, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 300, + 374 + ], + "score": 1.0, + "content": "tion. We denote the solution of this flow at time", + "type": "text" + }, + { + "bbox": [ + 301, + 362, + 306, + 370 + ], + "score": 0.79, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 358, + 469, + 374 + ], + "score": 1.0, + "content": "started from designated initialization by", + "type": "text" + }, + { + "bbox": [ + 470, + 360, + 502, + 372 + ], + "score": 0.92, + "content": "W ^ { \\mathrm { i n i t } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 358, + 507, + 374 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 371, + 374, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 207, + 387 + ], + "score": 1.0, + "content": "its stationary solution by", + "type": "text" + }, + { + "bbox": [ + 207, + 371, + 219, + 384 + ], + "score": 0.87, + "content": "\\widehat { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 372, + 361, + 387 + ], + "score": 1.0, + "content": ", and the corresponding network by", + "type": "text" + }, + { + "bbox": [ + 362, + 372, + 369, + 385 + ], + "score": 0.84, + "content": "\\hat { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 372, + 374, + 387 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 349, + 507, + 387 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 387, + 503, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 386, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 209, + 400 + ], + "score": 1.0, + "content": "Remark. Although we let", + "type": "text" + }, + { + "bbox": [ + 209, + 388, + 242, + 397 + ], + "score": 0.88, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 386, + 505, + 400 + ], + "score": 1.0, + "content": ", this dynamics does not corresponds to the population gradient flow", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 288, + 411 + ], + "score": 1.0, + "content": "considered in Tian (2017). For instance when", + "type": "text" + }, + { + "bbox": [ + 288, + 398, + 351, + 410 + ], + "score": 0.93, + "content": "\\pmb { x } \\sim \\mathcal { N } ( 0 , I / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 397, + 506, + 411 + ], + "score": 1.0, + "content": ", the spectrum of the data covariance is", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 408, + 365, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 365, + 422 + ], + "score": 1.0, + "content": "Marcenko–Pastur, whereas the population covariance is identity. ˇ", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 386, + 506, + 422 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 504, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "As we cannot characterize the gradient flow solution from all possible initializations, we consider", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 439, + 250, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 250, + 451 + ], + "score": 1.0, + "content": "two specific scales of initialization:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 427, + 505, + 451 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 506, + 577 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 506, + 496 + ], + "score": 1.0, + "content": "Note that neither of the two initializations correspond to the “mean-field” regime (e.g. analyzed in√", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 493, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 223, + 508 + ], + "score": 1.0, + "content": "Mei et al. (2018)) due to the", + "type": "text" + }, + { + "bbox": [ + 224, + 493, + 249, + 507 + ], + "score": 0.92, + "content": "1 / { \\sqrt { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 493, + 439, + 508 + ], + "score": 1.0, + "content": "scaling of the second layer. In other words, as", + "type": "text" + }, + { + "bbox": [ + 440, + 495, + 447, + 505 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 493, + 506, + 508 + ], + "score": 1.0, + "content": "increases, we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "expect the distance traveled by each parameter to decrease under both initializations. The difference,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "however, is the “relative” amount the parameters traveled compared to their initialized magnitude,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 526, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 541 + ], + "score": 1.0, + "content": "which leads to solutions with contrasting properties. As we will see, under (A1)(A2) and vanishing", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 538, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 194, + 554 + ], + "score": 1.0, + "content": "initialization we have", + "type": "text" + }, + { + "bbox": [ + 194, + 538, + 322, + 553 + ], + "score": 0.93, + "content": "\\lVert W ( 0 ) - \\widehat { W } \\rVert _ { F } / \\lVert W ( 0 ) \\rVert _ { F } \\gg 1", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 540, + 505, + 554 + ], + "score": 1.0, + "content": ", i.e. the contribution of initialization vanishes", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "at the end of training, whereas for non-vanishing initialization the inequality is in the opposite", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 563, + 395, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 162, + 578 + ], + "score": 1.0, + "content": "direction, i.e.", + "type": "text" + }, + { + "bbox": [ + 162, + 563, + 174, + 575 + ], + "score": 0.86, + "content": "\\widehat { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 564, + 366, + 578 + ], + "score": 1.0, + "content": "“barely moves” and resembles the initialization", + "type": "text" + }, + { + "bbox": [ + 367, + 564, + 391, + 577 + ], + "score": 0.92, + "content": "W ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 564, + 395, + 578 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 481, + 506, + 578 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 252, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 253, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 253, + 604 + ], + "score": 1.0, + "content": "5.1 VANISHING INITIALIZATION", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 611, + 505, + 656 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 120, + 623 + ], + "score": 1.0, + "content": "As", + "type": "text" + }, + { + "bbox": [ + 120, + 612, + 162, + 623 + ], + "score": 0.91, + "content": "d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 612, + 505, + 623 + ], + "score": 1.0, + "content": ", the vanishing initialization becomes arbitrarily close to zero-initialization. We thus", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "expect the gradient flow under vanishing initialization to \"resemble\" that of starting from exactly", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "zero if the flow converges sufficiently fast and the gradient being Lipschitz. The Lipschitz condition", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 645, + 433, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 433, + 657 + ], + "score": 1.0, + "content": "(Lemma 18) can be established under the following assumption on the activation.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 612, + 505, + 657 + ] + }, + { + "type": "text", + "bbox": [ + 114, + 662, + 493, + 676 + ], + "lines": [ + { + "bbox": [ + 117, + 661, + 493, + 677 + ], + "spans": [ + { + "bbox": [ + 117, + 661, + 143, + 677 + ], + "score": 1.0, + "content": "(A3):", + "type": "text" + }, + { + "bbox": [ + 143, + 664, + 150, + 675 + ], + "score": 0.8, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 661, + 313, + 677 + ], + "score": 1.0, + "content": "is smooth, Lipschitz and monotone with", + "type": "text" + }, + { + "bbox": [ + 313, + 661, + 493, + 676 + ], + "score": 0.78, + "content": "\\phi ^ { \\prime } ( 0 ) \\neq 0 ; | \\phi ^ { \\prime } ( \\pm x ) - \\phi ^ { \\prime } ( \\pm \\infty ) | = O ( e ^ { - x } ) .", + "type": "inline_equation" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 117, + 661, + 493, + 677 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 395, + 701 + ], + "score": 1.0, + "content": "The above assumption requires that the derivative of the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 395, + 688, + 403, + 699 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 686, + 506, + 701 + ], + "score": 1.0, + "content": "saturates beyond a O(1)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 507, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 507, + 711 + ], + "score": 1.0, + "content": "region, which holds true for the commonly-used smooth activations such as sigmoid and SoftPlus.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "In addition, the choice of scaling ensures that the gradient flow converges sufficiently fast. We thus", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 340, + 733 + ], + "score": 1.0, + "content": "have the following characterization of the population risk:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 686, + 507, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 116 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 235, + 96 + ], + "score": 1.0, + "content": "Theorem 7. Given (A1-3). Let", + "type": "text" + }, + { + "bbox": [ + 235, + 82, + 307, + 95 + ], + "score": 0.92, + "content": "T = O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 79, + 326, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 81, + 415, + 95 + ], + "score": 0.92, + "content": "\\hat { f } ( \\cdot ) = f ^ { \\nu a n } ( \\cdot , W ( T ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 79, + 450, + 96 + ], + "score": 1.0, + "content": ", then as", + "type": "text" + }, + { + "bbox": [ + 451, + 83, + 503, + 94 + ], + "score": 0.87, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 79, + 507, + 96 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 92, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 92, + 216, + 106 + ], + "score": 1.0, + "content": "the gradient flow reaches a", + "type": "text" + }, + { + "bbox": [ + 216, + 94, + 234, + 105 + ], + "score": 0.87, + "content": "o ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 92, + 372, + 106 + ], + "score": 1.0, + "content": "first-order stationary point at time", + "type": "text" + }, + { + "bbox": [ + 373, + 95, + 380, + 104 + ], + "score": 0.63, + "content": "T ,", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 92, + 398, + 106 + ], + "score": 1.0, + "content": "i.e.", + "type": "text" + }, + { + "bbox": [ + 398, + 93, + 491, + 106 + ], + "score": 0.91, + "content": "\\| \\partial W ( T ) / \\partial t \\| _ { F } \\in o ( 1 ) ,", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 92, + 506, + 106 + ], + "score": 1.0, + "content": ", at", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 279, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 279, + 117 + ], + "score": 1.0, + "content": "which point the population risk is given as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 116, + 404, + 144 + ], + "lines": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "spans": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "score": 0.93, + "content": "R ( \\hat { f } ) \\operatorname* { m a x } \\{ 0 , \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } \\} r ^ { 2 } + \\frac { \\operatorname* { m i n } \\{ \\gamma _ { 1 } , 1 \\} } { | 1 - \\gamma _ { 1 } | } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "6149c6ff40f51527e17ac9a862dcf6f79d63aacf6bbd1f2ed3d0b379f0f723a6.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 156, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 423, + 169 + ], + "score": 1.0, + "content": "The expression above is the same as the risk of the least squares solution on input", + "type": "text" + }, + { + "bbox": [ + 423, + 155, + 464, + 168 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "; therefore", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 168, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 337, + 181 + ], + "score": 1.0, + "content": "the risk is independent to overparameterization (increasing", + "type": "text" + }, + { + "bbox": [ + 337, + 169, + 348, + 180 + ], + "score": 0.83, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 168, + 505, + 181 + ], + "score": 1.0, + "content": "). The intuition is that when the weights", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 179, + 506, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 191 + ], + "score": 1.0, + "content": "are initialized sufficiently small and travel infinitesimally, then the activation can be linearized around", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "score": 1.0, + "content": "0 and thus the model is equivalent to a two-layer linear network. Note that this result does not apply", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 201, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 345, + 212 + ], + "score": 1.0, + "content": "to the non-smooth ReLU activation. Instead, in Appendix", + "type": "text" + }, + { + "bbox": [ + 345, + 201, + 354, + 211 + ], + "score": 0.45, + "content": "\\mathrm { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 201, + 505, + 212 + ], + "score": 1.0, + "content": "we heuristically show that under the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 210, + 503, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 492, + 226 + ], + "score": 1.0, + "content": "additional assumption that the data is symmetric, the risk of ReLU network is also independent to", + "type": "text" + }, + { + "bbox": [ + 492, + 214, + 503, + 223 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 108, + 235, + 274, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 276, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 276, + 249 + ], + "score": 1.0, + "content": "5.2 NON-VANISHING INITIALIZATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "When initialization is sufficiently large, the amount each parameter travels to minimize the empirical", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 266, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 280 + ], + "score": 1.0, + "content": "risk becomes asymptotically negligible compared to the magnitude of initialization. In this case", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 277, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 291 + ], + "score": 1.0, + "content": "we establish under (A1-3) that (11) is asymptotically equivalent to the kernel gradient flow on the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 168, + 303 + ], + "score": 1.0, + "content": "tangent kernel:", + "type": "text" + }, + { + "bbox": [ + 168, + 289, + 314, + 302 + ], + "score": 0.89, + "content": "k ( { \\pmb x } , { \\pmb y } ) = \\langle \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb x } ) , \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb y } ) \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 289, + 505, + 303 + ], + "score": 1.0, + "content": ". The converged parameters under this linearized", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 301, + 272, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 272, + 313 + ], + "score": 1.0, + "content": "dynamics has the following closed-form:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 121, + 312, + 471, + 328 + ], + "lines": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "spans": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "score": 0.88, + "content": "\\mathrm { v e c } ( W ^ { * } ) \\approx \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) + \\Delta ; \\quad \\Delta = J ^ { \\dagger } ( { \\pmb y } - f ^ { \\mathrm { i n i t } } ( { \\pmb X } ) ) ; \\quad J _ { [ i , j ] } = \\nabla _ { \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) _ { j } } f ^ { \\mathrm { i n i t } } ( { \\pmb x } _ { i } ) ,", + "type": "interline_equation", + "image_path": "1c4303b859f171566dc09e6e10d5245a8e9ec21b71d979b638716501829868b7.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 132, + 344 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 330, + 192, + 342 + ], + "score": 0.93, + "content": "J \\in \\mathbb { R } ^ { n \\times ( d \\times h ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "is the Jacobian matrix w.r.t. to the model parameters. We remark that in contrast", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 342, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 355 + ], + "score": 1.0, + "content": "to most NTK-type global convergence results that require the width of the model to grow faster than", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 353, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 366 + ], + "score": 1.0, + "content": "the number of data points (e.g. Du et al. (2018)), our result is not built upon the overparameterization", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "on width, but instead an anti-concentration that relies on the scale of initialization. Consequently the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 374, + 416, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 416, + 388 + ], + "score": 1.0, + "content": "above initialization is larger than the scale that is commonly used in practice.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 495, + 404 + ], + "score": 1.0, + "content": "One may naturally expect the double descent phenomenon to appear in this kernel solution, as", + "type": "text" + }, + { + "bbox": [ + 495, + 392, + 505, + 402 + ], + "score": 0.79, + "content": "\\Delta", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 403, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 414 + ], + "score": 1.0, + "content": "exhibits the form of a least squares solution which contains a pseudo-inverse. However, we show that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 411, + 503, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 411, + 492, + 428 + ], + "score": 1.0, + "content": "this is not the case under the same assumptions in Section 4; in fact, the risk is also independent to", + "type": "text" + }, + { + "bbox": [ + 492, + 416, + 503, + 425 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 429, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 432, + 444 + ], + "score": 1.0, + "content": "An obstacle in computing the risk of the kernel model is the potentially non-zero", + "type": "text" + }, + { + "bbox": [ + 433, + 430, + 465, + 443 + ], + "score": 0.91, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 429, + 505, + 444 + ], + "score": 1.0, + "content": ". We thus", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 379, + 454 + ], + "score": 1.0, + "content": "adopt the \"doubling-trick\" from Chizat and Bach (2018b) to ensure", + "type": "text" + }, + { + "bbox": [ + 379, + 442, + 425, + 454 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 439, + 505, + 454 + ], + "score": 1.0, + "content": ", i.e. we assume the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 452, + 334, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 334, + 465 + ], + "score": 1.0, + "content": "following symmetric property on the initialized weights:", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 140, + 466, + 467, + 480 + ], + "lines": [ + { + "bbox": [ + 141, + 461, + 464, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 461, + 275, + 484 + ], + "score": 1.0, + "content": "(A4) Symmetric Initialization:", + "type": "text" + }, + { + "bbox": [ + 275, + 467, + 318, + 479 + ], + "score": 0.86, + "content": "\\forall i \\in [ 1 , h ]", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 461, + 330, + 484 + ], + "score": 1.0, + "content": ", ∃!", + "type": "text" + }, + { + "bbox": [ + 330, + 467, + 369, + 479 + ], + "score": 0.79, + "content": "j \\in [ 1 , h ]", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 461, + 387, + 484 + ], + "score": 1.0, + "content": "s.t.", + "type": "text" + }, + { + "bbox": [ + 387, + 466, + 464, + 480 + ], + "score": 0.86, + "content": "a _ { i } \\mathbf { w } _ { i } ^ { \\mathrm { i n i t } } = - a _ { j } \\mathbf { w } _ { j } ^ { \\mathrm { i n i t } }", + "type": "inline_equation" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 104, + 483, + 488, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 482, + 489, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 244, + 498 + ], + "score": 1.0, + "content": "Theorem 8. Given (A1-4) and let", + "type": "text" + }, + { + "bbox": [ + 245, + 484, + 297, + 495 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 482, + 392, + 498 + ], + "score": 1.0, + "content": ", the stationary solution", + "type": "text" + }, + { + "bbox": [ + 393, + 482, + 400, + 496 + ], + "score": 0.84, + "content": "\\hat { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 482, + 489, + 498 + ], + "score": 1.0, + "content": "has the following risk", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 497, + 448, + 564 + ], + "lines": [ + { + "bbox": [ + 163, + 497, + 448, + 564 + ], + "spans": [ + { + "bbox": [ + 163, + 497, + 448, + 564 + ], + "score": 0.87, + "content": "\\begin{array} { c } { { R ( \\hat { f } ) ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) r ^ { 2 } } } \\\\ { { + ( \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } - \\frac { 1 } { 4 } ) \\sigma ^ { 2 } , } } \\end{array}", + "type": "interline_equation", + "image_path": "524b0bbaa826651e5a48f9d5ad59b478de80ebfc70999f8e0225196119624bbe.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 163, + 497, + 448, + 519.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 163, + 519.3333333333334, + 448, + 541.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 163, + 541.6666666666667, + 448, + 564.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 564, + 413, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 562, + 414, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 133, + 579 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 564, + 178, + 577 + ], + "score": 0.64, + "content": "m = b _ { 1 } ^ { 2 } / b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 562, + 182, + 579 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 182, + 564, + 246, + 577 + ], + "score": 0.84, + "content": "b _ { 0 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 562, + 268, + 579 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 268, + 564, + 352, + 577 + ], + "score": 0.74, + "content": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 562, + 357, + 579 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 357, + 564, + 410, + 577 + ], + "score": 0.27, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 562, + 414, + 579 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 583, + 505, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 315, + 596 + ], + "score": 1.0, + "content": "Note that the population risk is again independent to", + "type": "text" + }, + { + "bbox": [ + 315, + 586, + 326, + 596 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 584, + 505, + 596 + ], + "score": 1.0, + "content": ", and thus double descent does not appear for", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "this initialization. Roughly speaking, the reason that the risk does not become unbounded at some", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 605, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 326, + 619 + ], + "score": 1.0, + "content": "point is that in the asymptotic limit the pseudo-inverse", + "type": "text" + }, + { + "bbox": [ + 327, + 605, + 359, + 618 + ], + "score": 0.92, + "content": "( J J ^ { \\top } ) ^ { \\dagger }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 605, + 505, + 619 + ], + "score": 1.0, + "content": "is stable due to the nonlinearity and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 617, + 299, + 629 + ], + "spans": [ + { + "bbox": [ + 107, + 617, + 140, + 628 + ], + "score": 0.89, + "content": "d h \\gg n", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 617, + 299, + 629 + ], + "score": 1.0, + "content": ". We make two additional observations:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 132, + 635, + 505, + 728 + ], + "lines": [ + { + "bbox": [ + 132, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 132, + 636, + 255, + 650 + ], + "score": 1.0, + "content": "• the stability of the inverse at", + "type": "text" + }, + { + "bbox": [ + 256, + 637, + 287, + 648 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "depends on the lowest eigenvalue of the tangent kernel", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 140, + 646, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 140, + 646, + 206, + 661 + ], + "score": 1.0, + "content": "matrix (smaller", + "type": "text" + }, + { + "bbox": [ + 206, + 647, + 244, + 659 + ], + "score": 0.93, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 646, + 505, + 661 + ], + "score": 1.0, + "content": "entails larger variance), which is determined by the nonlinearity;", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 137, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 137, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "While our result only holds for network with zero initial output, for non-symmetric (i.i.d.)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 142, + 672, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 142, + 672, + 384, + 685 + ], + "score": 1.0, + "content": "initialization we also observe that the risk is independent to", + "type": "text" + }, + { + "bbox": [ + 384, + 674, + 395, + 684 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 672, + 505, + 685 + ], + "score": 1.0, + "content": ", but the bias is higher than", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 681, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 141, + 681, + 471, + 696 + ], + "score": 1.0, + "content": "that in symmetric initialization as shown in Figure 8. We comment that the non-zero", + "type": "text" + }, + { + "bbox": [ + 471, + 683, + 505, + 695 + ], + "score": 0.92, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 141, + 694, + 505, + 706 + ], + "spans": [ + { + "bbox": [ + 141, + 694, + 505, + 706 + ], + "score": 1.0, + "content": "in (13) behaves as zero-mean Gaussian due to central limit theorem; therefore, in the kernel", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 705, + 506, + 717 + ], + "spans": [ + { + "bbox": [ + 141, + 705, + 477, + 717 + ], + "score": 1.0, + "content": "regime the function output at initialization is equivalent to additive noise to the labels", + "type": "text" + }, + { + "bbox": [ + 477, + 707, + 485, + 716 + ], + "score": 0.78, + "content": "\\textbf { { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 705, + 506, + 717 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 141, + 715, + 505, + 729 + ], + "spans": [ + { + "bbox": [ + 141, + 715, + 274, + 729 + ], + "score": 1.0, + "content": "we thus expect the magnitude of", + "type": "text" + }, + { + "bbox": [ + 274, + 716, + 307, + 728 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 715, + 505, + 729 + ], + "score": 1.0, + "content": "to negatively influence the model generalization.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 116 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 235, + 96 + ], + "score": 1.0, + "content": "Theorem 7. Given (A1-3). Let", + "type": "text" + }, + { + "bbox": [ + 235, + 82, + 307, + 95 + ], + "score": 0.92, + "content": "T = O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 79, + 326, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 81, + 415, + 95 + ], + "score": 0.92, + "content": "\\hat { f } ( \\cdot ) = f ^ { \\nu a n } ( \\cdot , W ( T ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 79, + 450, + 96 + ], + "score": 1.0, + "content": ", then as", + "type": "text" + }, + { + "bbox": [ + 451, + 83, + 503, + 94 + ], + "score": 0.87, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 79, + 507, + 96 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 92, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 92, + 216, + 106 + ], + "score": 1.0, + "content": "the gradient flow reaches a", + "type": "text" + }, + { + "bbox": [ + 216, + 94, + 234, + 105 + ], + "score": 0.87, + "content": "o ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 92, + 372, + 106 + ], + "score": 1.0, + "content": "first-order stationary point at time", + "type": "text" + }, + { + "bbox": [ + 373, + 95, + 380, + 104 + ], + "score": 0.63, + "content": "T ,", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 92, + 398, + 106 + ], + "score": 1.0, + "content": "i.e.", + "type": "text" + }, + { + "bbox": [ + 398, + 93, + 491, + 106 + ], + "score": 0.91, + "content": "\\| \\partial W ( T ) / \\partial t \\| _ { F } \\in o ( 1 ) ,", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 92, + 506, + 106 + ], + "score": 1.0, + "content": ", at", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 279, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 279, + 117 + ], + "score": 1.0, + "content": "which point the population risk is given as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 104, + 79, + 507, + 117 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 116, + 404, + 144 + ], + "lines": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "spans": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "score": 0.93, + "content": "R ( \\hat { f } ) \\operatorname* { m a x } \\{ 0 , \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } \\} r ^ { 2 } + \\frac { \\operatorname* { m i n } \\{ \\gamma _ { 1 } , 1 \\} } { | 1 - \\gamma _ { 1 } | } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "6149c6ff40f51527e17ac9a862dcf6f79d63aacf6bbd1f2ed3d0b379f0f723a6.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 207, + 116, + 404, + 144 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 156, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 423, + 169 + ], + "score": 1.0, + "content": "The expression above is the same as the risk of the least squares solution on input", + "type": "text" + }, + { + "bbox": [ + 423, + 155, + 464, + 168 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "; therefore", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 168, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 337, + 181 + ], + "score": 1.0, + "content": "the risk is independent to overparameterization (increasing", + "type": "text" + }, + { + "bbox": [ + 337, + 169, + 348, + 180 + ], + "score": 0.83, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 168, + 505, + 181 + ], + "score": 1.0, + "content": "). The intuition is that when the weights", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 179, + 506, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 191 + ], + "score": 1.0, + "content": "are initialized sufficiently small and travel infinitesimally, then the activation can be linearized around", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "score": 1.0, + "content": "0 and thus the model is equivalent to a two-layer linear network. Note that this result does not apply", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 201, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 345, + 212 + ], + "score": 1.0, + "content": "to the non-smooth ReLU activation. Instead, in Appendix", + "type": "text" + }, + { + "bbox": [ + 345, + 201, + 354, + 211 + ], + "score": 0.45, + "content": "\\mathrm { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 201, + 505, + 212 + ], + "score": 1.0, + "content": "we heuristically show that under the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 210, + 503, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 492, + 226 + ], + "score": 1.0, + "content": "additional assumption that the data is symmetric, the risk of ReLU network is also independent to", + "type": "text" + }, + { + "bbox": [ + 492, + 214, + 503, + 223 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 155, + 506, + 226 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 235, + 274, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 276, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 276, + 249 + ], + "score": 1.0, + "content": "5.2 NON-VANISHING INITIALIZATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "When initialization is sufficiently large, the amount each parameter travels to minimize the empirical", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 266, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 280 + ], + "score": 1.0, + "content": "risk becomes asymptotically negligible compared to the magnitude of initialization. In this case", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 277, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 291 + ], + "score": 1.0, + "content": "we establish under (A1-3) that (11) is asymptotically equivalent to the kernel gradient flow on the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 168, + 303 + ], + "score": 1.0, + "content": "tangent kernel:", + "type": "text" + }, + { + "bbox": [ + 168, + 289, + 314, + 302 + ], + "score": 0.89, + "content": "k ( { \\pmb x } , { \\pmb y } ) = \\langle \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb x } ) , \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb y } ) \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 289, + 505, + 303 + ], + "score": 1.0, + "content": ". The converged parameters under this linearized", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 301, + 272, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 272, + 313 + ], + "score": 1.0, + "content": "dynamics has the following closed-form:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 256, + 506, + 313 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 121, + 312, + 471, + 328 + ], + "lines": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "spans": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "score": 0.88, + "content": "\\mathrm { v e c } ( W ^ { * } ) \\approx \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) + \\Delta ; \\quad \\Delta = J ^ { \\dagger } ( { \\pmb y } - f ^ { \\mathrm { i n i t } } ( { \\pmb X } ) ) ; \\quad J _ { [ i , j ] } = \\nabla _ { \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) _ { j } } f ^ { \\mathrm { i n i t } } ( { \\pmb x } _ { i } ) ,", + "type": "interline_equation", + "image_path": "1c4303b859f171566dc09e6e10d5245a8e9ec21b71d979b638716501829868b7.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 121, + 312, + 471, + 328 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 132, + 344 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 330, + 192, + 342 + ], + "score": 0.93, + "content": "J \\in \\mathbb { R } ^ { n \\times ( d \\times h ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "is the Jacobian matrix w.r.t. to the model parameters. We remark that in contrast", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 342, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 355 + ], + "score": 1.0, + "content": "to most NTK-type global convergence results that require the width of the model to grow faster than", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 353, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 366 + ], + "score": 1.0, + "content": "the number of data points (e.g. Du et al. (2018)), our result is not built upon the overparameterization", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "on width, but instead an anti-concentration that relies on the scale of initialization. Consequently the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 374, + 416, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 416, + 388 + ], + "score": 1.0, + "content": "above initialization is larger than the scale that is commonly used in practice.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 329, + 506, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 495, + 404 + ], + "score": 1.0, + "content": "One may naturally expect the double descent phenomenon to appear in this kernel solution, as", + "type": "text" + }, + { + "bbox": [ + 495, + 392, + 505, + 402 + ], + "score": 0.79, + "content": "\\Delta", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 403, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 414 + ], + "score": 1.0, + "content": "exhibits the form of a least squares solution which contains a pseudo-inverse. However, we show that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 411, + 503, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 411, + 492, + 428 + ], + "score": 1.0, + "content": "this is not the case under the same assumptions in Section 4; in fact, the risk is also independent to", + "type": "text" + }, + { + "bbox": [ + 492, + 416, + 503, + 425 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 104, + 391, + 505, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 429, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 432, + 444 + ], + "score": 1.0, + "content": "An obstacle in computing the risk of the kernel model is the potentially non-zero", + "type": "text" + }, + { + "bbox": [ + 433, + 430, + 465, + 443 + ], + "score": 0.91, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 429, + 505, + 444 + ], + "score": 1.0, + "content": ". We thus", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 379, + 454 + ], + "score": 1.0, + "content": "adopt the \"doubling-trick\" from Chizat and Bach (2018b) to ensure", + "type": "text" + }, + { + "bbox": [ + 379, + 442, + 425, + 454 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 439, + 505, + 454 + ], + "score": 1.0, + "content": ", i.e. we assume the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 452, + 334, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 334, + 465 + ], + "score": 1.0, + "content": "following symmetric property on the initialized weights:", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 429, + 505, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 140, + 466, + 467, + 480 + ], + "lines": [ + { + "bbox": [ + 141, + 461, + 464, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 461, + 275, + 484 + ], + "score": 1.0, + "content": "(A4) Symmetric Initialization:", + "type": "text" + }, + { + "bbox": [ + 275, + 467, + 318, + 479 + ], + "score": 0.86, + "content": "\\forall i \\in [ 1 , h ]", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 461, + 330, + 484 + ], + "score": 1.0, + "content": ", ∃!", + "type": "text" + }, + { + "bbox": [ + 330, + 467, + 369, + 479 + ], + "score": 0.79, + "content": "j \\in [ 1 , h ]", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 461, + 387, + 484 + ], + "score": 1.0, + "content": "s.t.", + "type": "text" + }, + { + "bbox": [ + 387, + 466, + 464, + 480 + ], + "score": 0.86, + "content": "a _ { i } \\mathbf { w } _ { i } ^ { \\mathrm { i n i t } } = - a _ { j } \\mathbf { w } _ { j } ^ { \\mathrm { i n i t } }", + "type": "inline_equation" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 141, + 461, + 464, + 484 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 483, + 488, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 482, + 489, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 244, + 498 + ], + "score": 1.0, + "content": "Theorem 8. Given (A1-4) and let", + "type": "text" + }, + { + "bbox": [ + 245, + 484, + 297, + 495 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 482, + 392, + 498 + ], + "score": 1.0, + "content": ", the stationary solution", + "type": "text" + }, + { + "bbox": [ + 393, + 482, + 400, + 496 + ], + "score": 0.84, + "content": "\\hat { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 482, + 489, + 498 + ], + "score": 1.0, + "content": "has the following risk", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 482, + 489, + 498 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 497, + 448, + 564 + ], + "lines": [ + { + "bbox": [ + 163, + 497, + 448, + 564 + ], + "spans": [ + { + "bbox": [ + 163, + 497, + 448, + 564 + ], + "score": 0.87, + "content": "\\begin{array} { c } { { R ( \\hat { f } ) ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) r ^ { 2 } } } \\\\ { { + ( \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } - \\frac { 1 } { 4 } ) \\sigma ^ { 2 } , } } \\end{array}", + "type": "interline_equation", + "image_path": "524b0bbaa826651e5a48f9d5ad59b478de80ebfc70999f8e0225196119624bbe.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 163, + 497, + 448, + 519.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 163, + 519.3333333333334, + 448, + 541.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 163, + 541.6666666666667, + 448, + 564.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 564, + 413, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 562, + 414, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 133, + 579 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 564, + 178, + 577 + ], + "score": 0.64, + "content": "m = b _ { 1 } ^ { 2 } / b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 562, + 182, + 579 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 182, + 564, + 246, + 577 + ], + "score": 0.84, + "content": "b _ { 0 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 562, + 268, + 579 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 268, + 564, + 352, + 577 + ], + "score": 0.74, + "content": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 562, + 357, + 579 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 357, + 564, + 410, + 577 + ], + "score": 0.27, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 562, + 414, + 579 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 562, + 414, + 579 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 583, + 505, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 315, + 596 + ], + "score": 1.0, + "content": "Note that the population risk is again independent to", + "type": "text" + }, + { + "bbox": [ + 315, + 586, + 326, + 596 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 584, + 505, + 596 + ], + "score": 1.0, + "content": ", and thus double descent does not appear for", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "this initialization. Roughly speaking, the reason that the risk does not become unbounded at some", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 605, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 326, + 619 + ], + "score": 1.0, + "content": "point is that in the asymptotic limit the pseudo-inverse", + "type": "text" + }, + { + "bbox": [ + 327, + 605, + 359, + 618 + ], + "score": 0.92, + "content": "( J J ^ { \\top } ) ^ { \\dagger }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 605, + 505, + 619 + ], + "score": 1.0, + "content": "is stable due to the nonlinearity and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 617, + 299, + 629 + ], + "spans": [ + { + "bbox": [ + 107, + 617, + 140, + 628 + ], + "score": 0.89, + "content": "d h \\gg n", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 617, + 299, + 629 + ], + "score": 1.0, + "content": ". We make two additional observations:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 584, + 505, + 629 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 635, + 505, + 728 + ], + "lines": [ + { + "bbox": [ + 132, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 132, + 636, + 255, + 650 + ], + "score": 1.0, + "content": "• the stability of the inverse at", + "type": "text" + }, + { + "bbox": [ + 256, + 637, + 287, + 648 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "depends on the lowest eigenvalue of the tangent kernel", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 140, + 646, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 140, + 646, + 206, + 661 + ], + "score": 1.0, + "content": "matrix (smaller", + "type": "text" + }, + { + "bbox": [ + 206, + 647, + 244, + 659 + ], + "score": 0.93, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 646, + 505, + 661 + ], + "score": 1.0, + "content": "entails larger variance), which is determined by the nonlinearity;", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 137, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 137, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "While our result only holds for network with zero initial output, for non-symmetric (i.i.d.)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 142, + 672, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 142, + 672, + 384, + 685 + ], + "score": 1.0, + "content": "initialization we also observe that the risk is independent to", + "type": "text" + }, + { + "bbox": [ + 384, + 674, + 395, + 684 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 672, + 505, + 685 + ], + "score": 1.0, + "content": ", but the bias is higher than", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 681, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 141, + 681, + 471, + 696 + ], + "score": 1.0, + "content": "that in symmetric initialization as shown in Figure 8. We comment that the non-zero", + "type": "text" + }, + { + "bbox": [ + 471, + 683, + 505, + 695 + ], + "score": 0.92, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 141, + 694, + 505, + 706 + ], + "spans": [ + { + "bbox": [ + 141, + 694, + 505, + 706 + ], + "score": 1.0, + "content": "in (13) behaves as zero-mean Gaussian due to central limit theorem; therefore, in the kernel", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 705, + 506, + 717 + ], + "spans": [ + { + "bbox": [ + 141, + 705, + 477, + 717 + ], + "score": 1.0, + "content": "regime the function output at initialization is equivalent to additive noise to the labels", + "type": "text" + }, + { + "bbox": [ + 477, + 707, + 485, + 716 + ], + "score": 0.78, + "content": "\\textbf { { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 705, + 506, + 717 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 141, + 715, + 505, + 729 + ], + "spans": [ + { + "bbox": [ + 141, + 715, + 274, + 729 + ], + "score": 1.0, + "content": "we thus expect the magnitude of", + "type": "text" + }, + { + "bbox": [ + 274, + 716, + 307, + 728 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 715, + 505, + 729 + ], + "score": 1.0, + "content": "to negatively influence the model generalization.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5, + "bbox_fs": [ + 132, + 636, + 506, + 729 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 81, + 494, + 229 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 81, + 494, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 81, + 494, + 229 + ], + "spans": [ + { + "bbox": [ + 113, + 81, + 494, + 229 + ], + "score": 0.971, + "type": "image", + "image_path": "389923c17d029c0a38688561f57f18607a9ed4055a5409c2180d94b356e6ec61.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 81, + 494, + 130.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 130.33333333333334, + 494, + 179.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 179.66666666666669, + 494, + 229.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 237, + 505, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "score": 1.0, + "content": "Figure 3: Bias and variance of two-layer sigmoid network with optimized first layer under (A1)(A2). Individual", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 247, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 231, + 257 + ], + "score": 1.0, + "content": "dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 232, + 248, + 242, + 258 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 247, + 505, + 257 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. The bias and variance for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 257, + 349, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 349, + 268 + ], + "score": 1.0, + "content": "both initializations is well-aligned with Theorem 7 and Theorem 8.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 288, + 280, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 281, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 281, + 301 + ], + "score": 1.0, + "content": "5.3 COMPARING THE INITIALIZATIONS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "score": 1.0, + "content": "Figure 3 shows the agreement between theoretical prediction and experimental results. Although in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 319, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 336, + 333 + ], + "score": 1.0, + "content": "both cases the risk is independent to overparameterization", + "type": "text" + }, + { + "bbox": [ + 337, + 320, + 353, + 331 + ], + "score": 0.86, + "content": "( \\gamma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 319, + 506, + 333 + ], + "score": 1.0, + "content": ", the two initializations lead to models", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 331, + 450, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 450, + 343 + ], + "score": 1.0, + "content": "with contrasting properties, as demonstrated by the following comparison on the risk.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 504, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 196, + 361 + ], + "score": 1.0, + "content": "Corollary 9. For any", + "type": "text" + }, + { + "bbox": [ + 196, + 347, + 246, + 359 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 344, + 316, + 361 + ], + "score": 1.0, + "content": "and nonlinearity", + "type": "text" + }, + { + "bbox": [ + 317, + 347, + 324, + 359 + ], + "score": 0.6, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 344, + 327, + 361 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 327, + 345, + 425, + 359 + ], + "score": 0.94, + "content": "B ( \\hat { f } ^ { V a n } ) \\le B ( \\hat { f } ^ { N V } ) \\le 1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 344, + 506, + 361 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 357, + 450, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 133, + 374 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 134, + 360, + 162, + 371 + ], + "score": 0.87, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 357, + 165, + 374 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 165, + 359, + 231, + 372 + ], + "score": 0.91, + "content": "V ( \\hat { f } ^ { N V } ) = { \\cal { O } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 357, + 271, + 374 + ], + "score": 1.0, + "content": ", whereas", + "type": "text" + }, + { + "bbox": [ + 271, + 358, + 304, + 372 + ], + "score": 0.93, + "content": "V ( \\hat { f } ^ { V a n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 357, + 414, + 374 + ], + "score": 1.0, + "content": "can be arbitrarily large as", + "type": "text" + }, + { + "bbox": [ + 414, + 361, + 445, + 372 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 357, + 450, + 374 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 433, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 434, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 146, + 388 + ], + "score": 1.0, + "content": "Remark.", + "type": "text" + }, + { + "bbox": [ + 146, + 375, + 175, + 386 + ], + "score": 0.9, + "content": "m \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 373, + 279, + 388 + ], + "score": 1.0, + "content": "for all smooth activations", + "type": "text" + }, + { + "bbox": [ + 280, + 375, + 286, + 386 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 373, + 389, + 388 + ], + "score": 1.0, + "content": ", and the equality holds if", + "type": "text" + }, + { + "bbox": [ + 389, + 375, + 396, + 386 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 373, + 434, + 388 + ], + "score": 1.0, + "content": "is linear.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "Intuitively, small initialization enables the model \"evolve\" more during optimization and better align", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "with the data and target function. This potentially results in a lower bias, at the expense of overfitting", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "more to the noise (high variance). In contrast, with sufficiently large initialization the final model", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "score": 1.0, + "content": "becomes close to the initialized model, and thus we may expect it to be less “aligned” to the target", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 437, + 293, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 293, + 449 + ], + "score": 1.0, + "content": "(high bias) but is more stable (lower variance).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "In illustrate the different inductive bias of the two initializations, we plot the trajectory of neurons in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "Appendix A Figure 4. Observe that for vanishing initialization the neurons stay close to one another", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 476, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 506, + 488 + ], + "score": 1.0, + "content": "throughout the trajectory, which results in a low-rank weight matrix, as predicted by Theorem 7. In", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "score": 1.0, + "content": "contrast, for non-vanishing initialization the neurons stay close to initialization (therefore full-rank),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "which validates the kernel approximation. Last but not least, although the derived risk is only for", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "learning a linear target function, we empirically observe that when the teacher is also a two-layer", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "network, the population risk follows the same trend, i.e. double descent occurs when only the second", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 531, + 271, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 271, + 543 + ], + "score": 1.0, + "content": "layer is optimized, as shown in Figure 7.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 303, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 556, + 305, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 305, + 573 + ], + "score": 1.0, + "content": "6 DISCUSSION AND FUTURE WORKS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "score": 1.0, + "content": "We derived the exact population risk of high-dimensional two-layer neural networks in learning a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "linear target function over Gaussian data with additive label noise, and showed that optimizing the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "first or the second layer via gradient flow results in solutions with contrasting properties. Specifically,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 616, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 506, + 628 + ], + "score": 1.0, + "content": "double descent is present when the second layer coefficients are optimized, but not when the first", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "layer weights are optimized under certain initializations. Moreover, we highlight that the scale of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 638, + 399, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 399, + 650 + ], + "score": 1.0, + "content": "initialization leads to different inductive bias in optimizing the first layer.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "It should be noted that our analysis only applies to the unregularized objective: it has been shown", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 258, + 678 + ], + "score": 1.0, + "content": "that explicit regularization (such as", + "type": "text" + }, + { + "bbox": [ + 258, + 666, + 268, + 677 + ], + "score": 0.86, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 665, + 424, + 678 + ], + "score": 1.0, + "content": "penalty) stabilizes the singularity at", + "type": "text" + }, + { + "bbox": [ + 425, + 666, + 462, + 677 + ], + "score": 0.9, + "content": "\\gamma _ { 2 } ~ ~ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "(Mei and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "Montanari, 2019), and algorithmic regularization (Li et al., 2019a; Dong et al., 2019) also provides", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "robustness against noisy observations. We further remark that our findings do not directly contradict", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "the experimental double descent phenomenon, nor the practical benefit of overparameterization. In", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 291, + 721 + ], + "score": 1.0, + "content": "particular, the interpolation limit could occur at", + "type": "text" + }, + { + "bbox": [ + 292, + 711, + 323, + 721 + ], + "score": 0.89, + "content": "\\gamma _ { 2 } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 709, + 506, + 721 + ], + "score": 1.0, + "content": "which is beyond the regime we consider (such", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "as Figure 4 in (Belkin et al., 2018)). Thus what we conclude is that under the studied proportional", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 81, + 494, + 229 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 81, + 494, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 81, + 494, + 229 + ], + "spans": [ + { + "bbox": [ + 113, + 81, + 494, + 229 + ], + "score": 0.971, + "type": "image", + "image_path": "389923c17d029c0a38688561f57f18607a9ed4055a5409c2180d94b356e6ec61.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 81, + 494, + 130.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 130.33333333333334, + 494, + 179.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 179.66666666666669, + 494, + 229.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 237, + 505, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "score": 1.0, + "content": "Figure 3: Bias and variance of two-layer sigmoid network with optimized first layer under (A1)(A2). Individual", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 247, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 231, + 257 + ], + "score": 1.0, + "content": "dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 232, + 248, + 242, + 258 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 247, + 505, + 257 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. The bias and variance for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 257, + 349, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 349, + 268 + ], + "score": 1.0, + "content": "both initializations is well-aligned with Theorem 7 and Theorem 8.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 288, + 280, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 281, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 281, + 301 + ], + "score": 1.0, + "content": "5.3 COMPARING THE INITIALIZATIONS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "score": 1.0, + "content": "Figure 3 shows the agreement between theoretical prediction and experimental results. Although in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 319, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 336, + 333 + ], + "score": 1.0, + "content": "both cases the risk is independent to overparameterization", + "type": "text" + }, + { + "bbox": [ + 337, + 320, + 353, + 331 + ], + "score": 0.86, + "content": "( \\gamma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 319, + 506, + 333 + ], + "score": 1.0, + "content": ", the two initializations lead to models", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 331, + 450, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 450, + 343 + ], + "score": 1.0, + "content": "with contrasting properties, as demonstrated by the following comparison on the risk.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 308, + 506, + 343 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 504, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 196, + 361 + ], + "score": 1.0, + "content": "Corollary 9. For any", + "type": "text" + }, + { + "bbox": [ + 196, + 347, + 246, + 359 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 344, + 316, + 361 + ], + "score": 1.0, + "content": "and nonlinearity", + "type": "text" + }, + { + "bbox": [ + 317, + 347, + 324, + 359 + ], + "score": 0.6, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 344, + 327, + 361 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 327, + 345, + 425, + 359 + ], + "score": 0.94, + "content": "B ( \\hat { f } ^ { V a n } ) \\le B ( \\hat { f } ^ { N V } ) \\le 1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 344, + 506, + 361 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 357, + 450, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 133, + 374 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 134, + 360, + 162, + 371 + ], + "score": 0.87, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 357, + 165, + 374 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 165, + 359, + 231, + 372 + ], + "score": 0.91, + "content": "V ( \\hat { f } ^ { N V } ) = { \\cal { O } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 357, + 271, + 374 + ], + "score": 1.0, + "content": ", whereas", + "type": "text" + }, + { + "bbox": [ + 271, + 358, + 304, + 372 + ], + "score": 0.93, + "content": "V ( \\hat { f } ^ { V a n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 357, + 414, + 374 + ], + "score": 1.0, + "content": "can be arbitrarily large as", + "type": "text" + }, + { + "bbox": [ + 414, + 361, + 445, + 372 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 357, + 450, + 374 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 344, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 433, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 434, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 146, + 388 + ], + "score": 1.0, + "content": "Remark.", + "type": "text" + }, + { + "bbox": [ + 146, + 375, + 175, + 386 + ], + "score": 0.9, + "content": "m \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 373, + 279, + 388 + ], + "score": 1.0, + "content": "for all smooth activations", + "type": "text" + }, + { + "bbox": [ + 280, + 375, + 286, + 386 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 373, + 389, + 388 + ], + "score": 1.0, + "content": ", and the equality holds if", + "type": "text" + }, + { + "bbox": [ + 389, + 375, + 396, + 386 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 373, + 434, + 388 + ], + "score": 1.0, + "content": "is linear.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 373, + 434, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "Intuitively, small initialization enables the model \"evolve\" more during optimization and better align", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "with the data and target function. This potentially results in a lower bias, at the expense of overfitting", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "more to the noise (high variance). In contrast, with sufficiently large initialization the final model", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "score": 1.0, + "content": "becomes close to the initialized model, and thus we may expect it to be less “aligned” to the target", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 437, + 293, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 293, + 449 + ], + "score": 1.0, + "content": "(high bias) but is more stable (lower variance).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 393, + 505, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "In illustrate the different inductive bias of the two initializations, we plot the trajectory of neurons in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "Appendix A Figure 4. Observe that for vanishing initialization the neurons stay close to one another", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 476, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 506, + 488 + ], + "score": 1.0, + "content": "throughout the trajectory, which results in a low-rank weight matrix, as predicted by Theorem 7. In", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "score": 1.0, + "content": "contrast, for non-vanishing initialization the neurons stay close to initialization (therefore full-rank),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "which validates the kernel approximation. Last but not least, although the derived risk is only for", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "learning a linear target function, we empirically observe that when the teacher is also a two-layer", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "network, the population risk follows the same trend, i.e. double descent occurs when only the second", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 531, + 271, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 271, + 543 + ], + "score": 1.0, + "content": "layer is optimized, as shown in Figure 7.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 453, + 506, + 543 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 303, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 556, + 305, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 305, + 573 + ], + "score": 1.0, + "content": "6 DISCUSSION AND FUTURE WORKS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "score": 1.0, + "content": "We derived the exact population risk of high-dimensional two-layer neural networks in learning a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "linear target function over Gaussian data with additive label noise, and showed that optimizing the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "first or the second layer via gradient flow results in solutions with contrasting properties. Specifically,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 616, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 506, + 628 + ], + "score": 1.0, + "content": "double descent is present when the second layer coefficients are optimized, but not when the first", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "layer weights are optimized under certain initializations. Moreover, we highlight that the scale of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 638, + 399, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 399, + 650 + ], + "score": 1.0, + "content": "initialization leads to different inductive bias in optimizing the first layer.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 581, + 506, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "It should be noted that our analysis only applies to the unregularized objective: it has been shown", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 258, + 678 + ], + "score": 1.0, + "content": "that explicit regularization (such as", + "type": "text" + }, + { + "bbox": [ + 258, + 666, + 268, + 677 + ], + "score": 0.86, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 665, + 424, + 678 + ], + "score": 1.0, + "content": "penalty) stabilizes the singularity at", + "type": "text" + }, + { + "bbox": [ + 425, + 666, + 462, + 677 + ], + "score": 0.9, + "content": "\\gamma _ { 2 } ~ ~ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "(Mei and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "Montanari, 2019), and algorithmic regularization (Li et al., 2019a; Dong et al., 2019) also provides", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "robustness against noisy observations. We further remark that our findings do not directly contradict", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "the experimental double descent phenomenon, nor the practical benefit of overparameterization. In", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 291, + 721 + ], + "score": 1.0, + "content": "particular, the interpolation limit could occur at", + "type": "text" + }, + { + "bbox": [ + 292, + 711, + 323, + 721 + ], + "score": 0.89, + "content": "\\gamma _ { 2 } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 709, + 506, + 721 + ], + "score": 1.0, + "content": "which is beyond the regime we consider (such", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "as Figure 4 in (Belkin et al., 2018)). Thus what we conclude is that under the studied proportional", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "asymptotics, the mechanism that provably gives rise to double descent from previous works on least", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 456, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 456, + 106 + ], + "score": 1.0, + "content": "squares regression might not translate to neural networks trained with gradient descent.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 654, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "asymptotics, the mechanism that provably gives rise to double descent from previous works on least", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 456, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 456, + 106 + ], + "score": 1.0, + "content": "squares regression might not translate to neural networks trained with gradient descent.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "To simplify the computation, we rely on a set of strong assumptions similar to those in Hastie et al.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "score": 1.0, + "content": "(2019), some of which we believe can be relaxed in future works, such as isotropic Gaussian input", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "score": 1.0, + "content": "and linear target. Importantly, the two specific scales of initialization studied in Section 5 are by no", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 506, + 155 + ], + "score": 1.0, + "content": "means exhaustive, and thus one would expect that under a different initialization of the first layer, or", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 167, + 166 + ], + "score": 1.0, + "content": "the mean-field", + "type": "text" + }, + { + "bbox": [ + 168, + 154, + 185, + 166 + ], + "score": 0.86, + "content": "1 / h", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 155, + 506, + 166 + ], + "score": 1.0, + "content": "scaling of the second layer, the risk of the trained model can be very different.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 506, + 178 + ], + "score": 1.0, + "content": "Changing the loss function may also alter the generalization behavior of the network. Another", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "challenging problem is to extend the current analysis to beyond two layers. Last but not least, our", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 504, + 199 + ], + "score": 1.0, + "content": "result characterizes gradient flow which resembles gradient descent with small stepsize, and thus", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "it would be interesting to study the effect of learning rate schedule, which has a known impact on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 322, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 322, + 222 + ], + "score": 1.0, + "content": "generalization (Smith and Le, 2017; Li et al., 2019b).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 108, + 235, + 200, + 245 + ], + "lines": [ + { + "bbox": [ + 107, + 236, + 201, + 246 + ], + "spans": [ + { + "bbox": [ + 107, + 236, + 201, + 246 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "score": 1.0, + "content": "We thank Xiuyuan Cheng, Xuechen Li, Yiping Lu, Atsushi Nitanda, Shengyang Sun and anonymous", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 267, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 279 + ], + "score": 1.0, + "content": "reviewers for helpful comments and feedback. JB and DW were partially funded by LG Electronics", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "and NSERC. JB and MAE were supported by the CIFAR AI Chairs program. TS was partially", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 287, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 301 + ], + "score": 1.0, + "content": "supported by JSPS Kakenhi (26280009, 15H05707 and 18H03201), Japan Digital Design and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 299, + 160, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 160, + 311 + ], + "score": 1.0, + "content": "JST-CREST.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 108, + 340, + 175, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 176, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 176, + 354 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 105, + 355, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 358, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 374 + ], + "score": 1.0, + "content": "Madhu S Advani and Andrew M Saxe. High-dimensional dynamics of generalization error in neural", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 371, + 321, + 384 + ], + "spans": [ + { + "bbox": [ + 116, + 371, + 321, + 384 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1710.03667, 2017.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu and Yuanzhi Li. What can resnet learn efficiently, going beyond kernels? arXiv", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 114, + 402, + 254, + 414 + ], + "spans": [ + { + "bbox": [ + 114, + 402, + 254, + 414 + ], + "score": 1.0, + "content": "preprint arXiv:1905.10337, 2019.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 420, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 104, + 420, + 506, + 435 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 456, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 456, + 446 + ], + "score": 1.0, + "content": "neural networks, going beyond two layers. arXiv preprint arXiv:1811.04918, 2018a.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 450, + 507, + 467 + ], + "spans": [ + { + "bbox": [ + 104, + 450, + 507, + 467 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via over-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 464, + 357, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 464, + 357, + 476 + ], + "score": 1.0, + "content": "parameterization. arXiv preprint arXiv:1811.03962, 2018b.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "score": 1.0, + "content": "Sanjeev Arora, Rong Ge, Behnam Neyshabur, and Yi Zhang. Stronger generalization bounds for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 493, + 436, + 507 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 436, + 507 + ], + "score": 1.0, + "content": "deep nets via a compression approach. arXiv preprint arXiv:1802.05296, 2018.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Ruslan Salakhutdinov, and Ruosong Wang. On", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 525, + 497, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 525, + 497, + 538 + ], + "score": 1.0, + "content": "exact computation with an infinitely wide neural net. arXiv preprint arXiv:1904.11955, 2019a.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 543, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 559 + ], + "score": 1.0, + "content": "Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "score": 1.0, + "content": "optimization and generalization for overparameterized two-layer neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 567, + 223, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 223, + 580 + ], + "score": 1.0, + "content": "arXiv:1901.08584, 2019b.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 585, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 507, + 600 + ], + "score": 1.0, + "content": "Zhidong Bai and Jack W Silverstein. Spectral analysis of large dimensional random matrices,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 598, + 228, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 598, + 228, + 610 + ], + "score": 1.0, + "content": "volume 20. Springer, 2010.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 616, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 616, + 506, + 631 + ], + "score": 1.0, + "content": "Peter L Bartlett, Dylan J Foster, and Matus J Telgarsky. Spectrally-normalized margin bounds for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 628, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 115, + 628, + 507, + 641 + ], + "score": 1.0, + "content": "neural networks. In Advances in Neural Information Processing Systems, pages 6240–6249, 2017.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 648, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 648, + 506, + 663 + ], + "score": 1.0, + "content": "Peter L Bartlett, Philip M Long, Gábor Lugosi, and Alexander Tsigler. Benign overfitting in linear", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 659, + 326, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 326, + 673 + ], + "score": 1.0, + "content": "regression. arXiv preprint arXiv:1906.11300, 2019.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "Mikhail Belkin, Daniel Hsu, Siyuan Ma, and Soumik Mandal. Reconciling modern machine learning", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 690, + 406, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 406, + 703 + ], + "score": 1.0, + "content": "and the bias-variance trade-off. arXiv preprint arXiv:1812.11118, 2018.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "Mikhail Belkin, Daniel Hsu, and Ji Xu. Two models of double descent for weak features. arXiv", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 722, + 254, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 722, + 254, + 733 + ], + "score": 1.0, + "content": "preprint arXiv:1903.07571, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 31 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 506, + 106 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "To simplify the computation, we rely on a set of strong assumptions similar to those in Hastie et al.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "score": 1.0, + "content": "(2019), some of which we believe can be relaxed in future works, such as isotropic Gaussian input", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "score": 1.0, + "content": "and linear target. Importantly, the two specific scales of initialization studied in Section 5 are by no", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 506, + 155 + ], + "score": 1.0, + "content": "means exhaustive, and thus one would expect that under a different initialization of the first layer, or", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 167, + 166 + ], + "score": 1.0, + "content": "the mean-field", + "type": "text" + }, + { + "bbox": [ + 168, + 154, + 185, + 166 + ], + "score": 0.86, + "content": "1 / h", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 155, + 506, + 166 + ], + "score": 1.0, + "content": "scaling of the second layer, the risk of the trained model can be very different.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 506, + 178 + ], + "score": 1.0, + "content": "Changing the loss function may also alter the generalization behavior of the network. Another", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "challenging problem is to extend the current analysis to beyond two layers. Last but not least, our", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 504, + 199 + ], + "score": 1.0, + "content": "result characterizes gradient flow which resembles gradient descent with small stepsize, and thus", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "it would be interesting to study the effect of learning rate schedule, which has a known impact on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 322, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 322, + 222 + ], + "score": 1.0, + "content": "generalization (Smith and Le, 2017; Li et al., 2019b).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 110, + 506, + 222 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 235, + 200, + 245 + ], + "lines": [ + { + "bbox": [ + 107, + 236, + 201, + 246 + ], + "spans": [ + { + "bbox": [ + 107, + 236, + 201, + 246 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "score": 1.0, + "content": "We thank Xiuyuan Cheng, Xuechen Li, Yiping Lu, Atsushi Nitanda, Shengyang Sun and anonymous", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 267, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 279 + ], + "score": 1.0, + "content": "reviewers for helpful comments and feedback. JB and DW were partially funded by LG Electronics", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "and NSERC. JB and MAE were supported by the CIFAR AI Chairs program. TS was partially", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 287, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 301 + ], + "score": 1.0, + "content": "supported by JSPS Kakenhi (26280009, 15H05707 and 18H03201), Japan Digital Design and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 299, + 160, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 160, + 311 + ], + "score": 1.0, + "content": "JST-CREST.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 255, + 506, + 311 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 340, + 175, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 176, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 176, + 354 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "list", + "bbox": [ + 105, + 355, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 358, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 374 + ], + "score": 1.0, + "content": "Madhu S Advani and Andrew M Saxe. High-dimensional dynamics of generalization error in neural", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 371, + 321, + 384 + ], + "spans": [ + { + "bbox": [ + 116, + 371, + 321, + 384 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1710.03667, 2017.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu and Yuanzhi Li. What can resnet learn efficiently, going beyond kernels? arXiv", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 402, + 254, + 414 + ], + "spans": [ + { + "bbox": [ + 114, + 402, + 254, + 414 + ], + "score": 1.0, + "content": "preprint arXiv:1905.10337, 2019.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 420, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 104, + 420, + 506, + 435 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 432, + 456, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 456, + 446 + ], + "score": 1.0, + "content": "neural networks, going beyond two layers. arXiv preprint arXiv:1811.04918, 2018a.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 450, + 507, + 467 + ], + "spans": [ + { + "bbox": [ + 104, + 450, + 507, + 467 + ], + "score": 1.0, + "content": "Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via over-", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 464, + 357, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 464, + 357, + 476 + ], + "score": 1.0, + "content": "parameterization. arXiv preprint arXiv:1811.03962, 2018b.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "score": 1.0, + "content": "Sanjeev Arora, Rong Ge, Behnam Neyshabur, and Yi Zhang. Stronger generalization bounds for", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 493, + 436, + 507 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 436, + 507 + ], + "score": 1.0, + "content": "deep nets via a compression approach. arXiv preprint arXiv:1802.05296, 2018.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Ruslan Salakhutdinov, and Ruosong Wang. On", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 525, + 497, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 525, + 497, + 538 + ], + "score": 1.0, + "content": "exact computation with an infinitely wide neural net. arXiv preprint arXiv:1904.11955, 2019a.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 543, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 559 + ], + "score": 1.0, + "content": "Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "score": 1.0, + "content": "optimization and generalization for overparameterized two-layer neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 567, + 223, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 223, + 580 + ], + "score": 1.0, + "content": "arXiv:1901.08584, 2019b.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 585, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 507, + 600 + ], + "score": 1.0, + "content": "Zhidong Bai and Jack W Silverstein. Spectral analysis of large dimensional random matrices,", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 598, + 228, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 598, + 228, + 610 + ], + "score": 1.0, + "content": "volume 20. Springer, 2010.", + "type": "text" + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 616, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 616, + 506, + 631 + ], + "score": 1.0, + "content": "Peter L Bartlett, Dylan J Foster, and Matus J Telgarsky. Spectrally-normalized margin bounds for", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 628, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 115, + 628, + 507, + 641 + ], + "score": 1.0, + "content": "neural networks. In Advances in Neural Information Processing Systems, pages 6240–6249, 2017.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 648, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 648, + 506, + 663 + ], + "score": 1.0, + "content": "Peter L Bartlett, Philip M Long, Gábor Lugosi, and Alexander Tsigler. Benign overfitting in linear", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 659, + 326, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 326, + 673 + ], + "score": 1.0, + "content": "regression. arXiv preprint arXiv:1906.11300, 2019.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "Mikhail Belkin, Daniel Hsu, Siyuan Ma, and Soumik Mandal. Reconciling modern machine learning", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 690, + 406, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 406, + 703 + ], + "score": 1.0, + "content": "and the bias-variance trade-off. arXiv preprint arXiv:1812.11118, 2018.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "Mikhail Belkin, Daniel Hsu, and Ji Xu. Two models of double descent for weak features. arXiv", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 722, + 254, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 722, + 254, + 733 + ], + "score": 1.0, + "content": "preprint arXiv:1903.07571, 2019.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + } + ], + "index": 31, + "bbox_fs": [ + 104, + 358, + 507, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning over-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 420, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 420, + 106 + ], + "score": 1.0, + "content": "parameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 105, + 113, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Xiuyuan Cheng and Amit Singer. The spectrum of random inner-product kernel matrices. Random", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 351, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 351, + 137 + ], + "score": 1.0, + "content": "Matrices: Theory and Applications, 2(04):1350010, 2013.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 504, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "Lenaic Chizat and Francis Bach. On the global convergence of gradient descent for over-parameterized", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "models using optimal transport. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 117, + 165, + 195, + 177 + ], + "spans": [ + { + "bbox": [ + 117, + 165, + 195, + 177 + ], + "score": 1.0, + "content": "3036–3046, 2018a.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 504, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 505, + 199 + ], + "score": 1.0, + "content": "Lenaic Chizat and Francis Bach. A note on lazy training in supervised differentiable programming.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 196, + 283, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 196, + 283, + 208 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1812.07956, 2018b.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 506, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 366, + 228 + ], + "score": 1.0, + "content": "Bin Dong, Jikai Hou, Yiping Lu, and Zhihua Zhang. Distillation", + "type": "text" + }, + { + "bbox": [ + 366, + 217, + 376, + 226 + ], + "score": 0.69, + "content": "\\approx", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "early stopping? harvesting dark", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 227, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 116, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "knowledge utilizing anisotropic information retrieval for overparameterized neural network. arXiv", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 238, + 254, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 254, + 249 + ], + "score": 1.0, + "content": "preprint arXiv:1910.01255, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 504, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "Simon S Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 268, + 429, + 281 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 429, + 281 + ], + "score": 1.0, + "content": "over-parameterized neural networks. arXiv preprint arXiv:1810.02054, 2018.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 287, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "score": 1.0, + "content": "Gintare Karolina Dziugaite and Daniel M Roy. Computing nonvacuous generalization bounds for", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "deep (stochastic) neural networks with many more parameters than training data. arXiv preprint", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 220, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 220, + 321 + ], + "score": 1.0, + "content": "arXiv:1703.11008, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 105, + 329, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 343 + ], + "score": 1.0, + "content": "Noureddine El Karoui et al. The spectrum of kernel random matrices. The Annals of Statistics, 38(1):", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 339, + 165, + 352 + ], + "spans": [ + { + "bbox": [ + 114, + 339, + 165, + 352 + ], + "score": 1.0, + "content": "1–50, 2010.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 104, + 360, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 506, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 506, + 373 + ], + "score": 1.0, + "content": "Zhou Fan and Andrea Montanari. The spectral norm of random inner-product kernel matrices.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 371, + 366, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 371, + 366, + 383 + ], + "score": 1.0, + "content": "Probability Theory and Related Fields, 173(1-2):27–85, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "Mario Geiger, Stefano Spigler, Stéphane d’Ascoli, Levent Sagun, Marco Baity-Jesi, Giulio Biroli,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 402, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 402, + 506, + 414 + ], + "score": 1.0, + "content": "and Matthieu Wyart. The jamming transition as a paradigm to understand the loss landscape of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 412, + 370, + 425 + ], + "spans": [ + { + "bbox": [ + 116, + 412, + 370, + 425 + ], + "score": 1.0, + "content": "deep neural networks. arXiv preprint arXiv:1809.09349, 2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 503, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 446 + ], + "score": 1.0, + "content": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Limitations of lazy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 443, + 442, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 442, + 456 + ], + "score": 1.0, + "content": "training of two-layers neural networks. arXiv preprint arXiv:1906.08899, 2019a.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 503, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "score": 1.0, + "content": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Linearized two-layers", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 474, + 428, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 474, + 428, + 486 + ], + "score": 1.0, + "content": "neural networks in high dimension. arXiv preprint arXiv:1904.12191, 2019b.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 506, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "score": 1.0, + "content": "Sebastian Goldt, Madhu S Advani, Andrew M Saxe, Florent Krzakala, and Lenka Zdeborová.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "Generalisation dynamics of online learning in over-parameterised neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 515, + 219, + 526 + ], + "spans": [ + { + "bbox": [ + 115, + 515, + 219, + 526 + ], + "score": 1.0, + "content": "arXiv:1901.09085, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 506, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 547 + ], + "score": 1.0, + "content": "Suriya Gunasekar, Jason D Lee, Daniel Soudry, and Nati Srebro. Implicit bias of gradient descent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "on linear convolutional networks. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 556, + 191, + 568 + ], + "spans": [ + { + "bbox": [ + 116, + 556, + 191, + 568 + ], + "score": 1.0, + "content": "9461–9471, 2018.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 105, + 576, + 504, + 599 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "Trevor Hastie, Andrea Montanari, Saharon Rosset, and Ryan J Tibshirani. Surprises in high-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 587, + 478, + 600 + ], + "spans": [ + { + "bbox": [ + 116, + 587, + 478, + 600 + ], + "score": 1.0, + "content": "dimensional ridgeless least squares interpolation. arXiv preprint arXiv:1903.08560, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 620 + ], + "score": 1.0, + "content": "Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 616, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 114, + 616, + 506, + 632 + ], + "score": 1.0, + "content": "generalization in neural networks. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 629, + 191, + 640 + ], + "spans": [ + { + "bbox": [ + 116, + 629, + 191, + 640 + ], + "score": 1.0, + "content": "8571–8580, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 647, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 505, + 660 + ], + "score": 1.0, + "content": "Ziwei Ji and Matus Telgarsky. Gradient descent aligns the layers of deep linear networks. arXiv", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 659, + 254, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 254, + 672 + ], + "score": 1.0, + "content": "preprint arXiv:1810.02032, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "Ryo Karakida, Shotaro Akaho, and Shun-ichi Amari. Universal statistics of fisher information in", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 690, + 458, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 458, + 703 + ], + "score": 1.0, + "content": "deep neural networks: Mean field approach. arXiv preprint arXiv:1806.01316, 2018.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Anders Krogh and John A. Hertz. A simple weight decay can improve generalization. pages 950–957,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 114, + 719, + 142, + 733 + ], + "spans": [ + { + "bbox": [ + 114, + 719, + 142, + 733 + ], + "score": 1.0, + "content": "1992.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning over-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 420, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 420, + 106 + ], + "score": 1.0, + "content": "parameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 506, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 113, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Xiuyuan Cheng and Amit Singer. The spectrum of random inner-product kernel matrices. Random", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 351, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 351, + 137 + ], + "score": 1.0, + "content": "Matrices: Theory and Applications, 2(04):1350010, 2013.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 112, + 505, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 504, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "Lenaic Chizat and Francis Bach. On the global convergence of gradient descent for over-parameterized", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "models using optimal transport. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 117, + 165, + 195, + 177 + ], + "spans": [ + { + "bbox": [ + 117, + 165, + 195, + 177 + ], + "score": 1.0, + "content": "3036–3046, 2018a.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 142, + 506, + 177 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 504, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 505, + 199 + ], + "score": 1.0, + "content": "Lenaic Chizat and Francis Bach. A note on lazy training in supervised differentiable programming.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 196, + 283, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 196, + 283, + 208 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1812.07956, 2018b.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 182, + 505, + 208 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 506, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 366, + 228 + ], + "score": 1.0, + "content": "Bin Dong, Jikai Hou, Yiping Lu, and Zhihua Zhang. Distillation", + "type": "text" + }, + { + "bbox": [ + 366, + 217, + 376, + 226 + ], + "score": 0.69, + "content": "\\approx", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "early stopping? harvesting dark", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 227, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 116, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "knowledge utilizing anisotropic information retrieval for overparameterized neural network. arXiv", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 238, + 254, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 254, + 249 + ], + "score": 1.0, + "content": "preprint arXiv:1910.01255, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 215, + 505, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 504, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "Simon S Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 268, + 429, + 281 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 429, + 281 + ], + "score": 1.0, + "content": "over-parameterized neural networks. arXiv preprint arXiv:1810.02054, 2018.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 257, + 505, + 281 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 287, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "score": 1.0, + "content": "Gintare Karolina Dziugaite and Daniel M Roy. Computing nonvacuous generalization bounds for", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "deep (stochastic) neural networks with many more parameters than training data. arXiv preprint", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 220, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 220, + 321 + ], + "score": 1.0, + "content": "arXiv:1703.11008, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 287, + 505, + 321 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 329, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 343 + ], + "score": 1.0, + "content": "Noureddine El Karoui et al. The spectrum of kernel random matrices. The Annals of Statistics, 38(1):", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 339, + 165, + 352 + ], + "spans": [ + { + "bbox": [ + 114, + 339, + 165, + 352 + ], + "score": 1.0, + "content": "1–50, 2010.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 327, + 506, + 352 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 360, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 506, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 506, + 373 + ], + "score": 1.0, + "content": "Zhou Fan and Andrea Montanari. The spectral norm of random inner-product kernel matrices.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 371, + 366, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 371, + 366, + 383 + ], + "score": 1.0, + "content": "Probability Theory and Related Fields, 173(1-2):27–85, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 359, + 506, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "Mario Geiger, Stefano Spigler, Stéphane d’Ascoli, Levent Sagun, Marco Baity-Jesi, Giulio Biroli,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 402, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 402, + 506, + 414 + ], + "score": 1.0, + "content": "and Matthieu Wyart. The jamming transition as a paradigm to understand the loss landscape of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 412, + 370, + 425 + ], + "spans": [ + { + "bbox": [ + 116, + 412, + 370, + 425 + ], + "score": 1.0, + "content": "deep neural networks. arXiv preprint arXiv:1809.09349, 2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 390, + 506, + 425 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 503, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 446 + ], + "score": 1.0, + "content": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Limitations of lazy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 443, + 442, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 442, + 456 + ], + "score": 1.0, + "content": "training of two-layers neural networks. arXiv preprint arXiv:1906.08899, 2019a.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 430, + 505, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 503, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "score": 1.0, + "content": "Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Linearized two-layers", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 474, + 428, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 474, + 428, + 486 + ], + "score": 1.0, + "content": "neural networks in high dimension. arXiv preprint arXiv:1904.12191, 2019b.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 461, + 505, + 486 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 506, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "score": 1.0, + "content": "Sebastian Goldt, Madhu S Advani, Andrew M Saxe, Florent Krzakala, and Lenka Zdeborová.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "Generalisation dynamics of online learning in over-parameterised neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 515, + 219, + 526 + ], + "spans": [ + { + "bbox": [ + 115, + 515, + 219, + 526 + ], + "score": 1.0, + "content": "arXiv:1901.09085, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 493, + 506, + 526 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 506, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 547 + ], + "score": 1.0, + "content": "Suriya Gunasekar, Jason D Lee, Daniel Soudry, and Nati Srebro. Implicit bias of gradient descent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "on linear convolutional networks. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 556, + 191, + 568 + ], + "spans": [ + { + "bbox": [ + 116, + 556, + 191, + 568 + ], + "score": 1.0, + "content": "9461–9471, 2018.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 533, + 506, + 568 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 576, + 504, + 599 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "Trevor Hastie, Andrea Montanari, Saharon Rosset, and Ryan J Tibshirani. Surprises in high-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 587, + 478, + 600 + ], + "spans": [ + { + "bbox": [ + 116, + 587, + 478, + 600 + ], + "score": 1.0, + "content": "dimensional ridgeless least squares interpolation. arXiv preprint arXiv:1903.08560, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 575, + 506, + 600 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 620 + ], + "score": 1.0, + "content": "Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 616, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 114, + 616, + 506, + 632 + ], + "score": 1.0, + "content": "generalization in neural networks. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 629, + 191, + 640 + ], + "spans": [ + { + "bbox": [ + 116, + 629, + 191, + 640 + ], + "score": 1.0, + "content": "8571–8580, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 605, + 506, + 640 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 647, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 505, + 660 + ], + "score": 1.0, + "content": "Ziwei Ji and Matus Telgarsky. Gradient descent aligns the layers of deep linear networks. arXiv", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 659, + 254, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 254, + 672 + ], + "score": 1.0, + "content": "preprint arXiv:1810.02032, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 106, + 647, + 505, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "Ryo Karakida, Shotaro Akaho, and Shun-ichi Amari. Universal statistics of fisher information in", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 690, + 458, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 458, + 703 + ], + "score": 1.0, + "content": "deep neural networks: Mean field approach. arXiv preprint arXiv:1806.01316, 2018.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 106, + 678, + 505, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Anders Krogh and John A. Hertz. A simple weight decay can improve generalization. pages 950–957,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 114, + 719, + 142, + 733 + ], + "spans": [ + { + "bbox": [ + 114, + 719, + 142, + 733 + ], + "score": 1.0, + "content": "1992.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 708, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "Mingchen Li, Mahdi Soltanolkotabi, and Samet Oymak. Gradient descent with early stopping is prov-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "ably robust to label noise for overparameterized neural networks. arXiv preprint arXiv:1903.11680,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 147, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 147, + 117 + ], + "score": 1.0, + "content": "2019a.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "Yuanzhi Li and Yingyu Liang. Learning overparameterized neural networks via stochastic gradient", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 507, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 507, + 147 + ], + "score": 1.0, + "content": "descent on structured data. In Advances in Neural Information Processing Systems, pages 8157–", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 166, + 156 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 166, + 156 + ], + "score": 1.0, + "content": "8166, 2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 506, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 506, + 177 + ], + "score": 1.0, + "content": "Yuanzhi Li, Tengyu Ma, and Hongyang Zhang. Algorithmic regularization in over-parameterized", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 506, + 188 + ], + "score": 1.0, + "content": "matrix sensing and neural networks with quadratic activations. arXiv preprint arXiv:1712.09203,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 185, + 142, + 198 + ], + "spans": [ + { + "bbox": [ + 114, + 185, + 142, + 198 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 205, + 503, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 219 + ], + "score": 1.0, + "content": "Yuanzhi Li, Colin Wei, and Tengyu Ma. Towards explaining the regularization effect of initial large", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 450, + 228 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 450, + 228 + ], + "score": 1.0, + "content": "learning rate in training neural networks. arXiv preprint arXiv:1907.04595, 2019b.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 504, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "Tengyuan Liang and Alexander Rakhlin. Just interpolate: Kernel\" ridgeless\" regression can generalize.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 246, + 278, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 278, + 258 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1808.00387, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 265, + 504, + 288 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "score": 1.0, + "content": "Zhenyu Liao and Romain Couillet. On the spectrum of random features maps of high dimensional", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 276, + 301, + 289 + ], + "spans": [ + { + "bbox": [ + 116, + 276, + 301, + 289 + ], + "score": 1.0, + "content": "data. arXiv preprint arXiv:1805.11916, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 295, + 505, + 318 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Cosme Louart, Zhenyu Liao, Romain Couillet, et al. A random matrix approach to neural networks.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 306, + 356, + 319 + ], + "spans": [ + { + "bbox": [ + 115, + 306, + 356, + 319 + ], + "score": 1.0, + "content": "The Annals of Applied Probability, 28(2):1190–1248, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 108, + 324, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "V.A. Marcenko and Leonid Pastur. Distribution of eigenvalues for some sets of random matrices. ˇ", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 336, + 265, + 348 + ], + "spans": [ + { + "bbox": [ + 116, + 336, + 265, + 348 + ], + "score": 1.0, + "content": "Math USSR Sb, 1:457–483, 01 1967.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 505, + 368 + ], + "score": 1.0, + "content": "Song Mei and Andrea Montanari. The generalization error of random features regression: Precise", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 366, + 436, + 378 + ], + "spans": [ + { + "bbox": [ + 115, + 366, + 436, + 378 + ], + "score": 1.0, + "content": "asymptotics and double descent curve. arXiv preprint arXiv:1908.05355, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "Song Mei, Andrea Montanari, and Phan-Minh Nguyen. A mean field view of the landscape of two-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 396, + 507, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 507, + 408 + ], + "score": 1.0, + "content": "layer neural networks. Proceedings of the National Academy of Sciences, 115(33):E7665–E7671,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 406, + 142, + 418 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 142, + 418 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 425, + 504, + 449 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "score": 1.0, + "content": "Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Mean-field theory of two-layers neural", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 437, + 484, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 437, + 484, + 449 + ], + "score": 1.0, + "content": "networks: dimension-free bounds and kernel limit. arXiv preprint arXiv:1902.06015, 2019.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "Vaishnavh Nagarajan and J Zico Kolter. Generalization in deep networks: The role of distance from", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 466, + 335, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 466, + 335, + 479 + ], + "score": 1.0, + "content": "initialization. arXiv preprint arXiv:1901.01672, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 485, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 497, + 464, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 464, + 509 + ], + "score": 1.0, + "content": "role of implicit regularization in deep learning. arXiv preprint arXiv:1412.6614, 2014.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 504, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "Behnam Neyshabur, Zhiyuan Li, Srinadh Bhojanapalli, Yann LeCun, and Nathan Srebro. Towards", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 527, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 116, + 527, + 505, + 539 + ], + "score": 1.0, + "content": "understanding the role of over-parametrization in generalization of neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 537, + 219, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 219, + 550 + ], + "score": 1.0, + "content": "arXiv:1805.12076, 2018.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 103, + 556, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "Atsushi Nitanda and Taiji Suzuki. Stochastic particle gradient descent for infinite ensembles. arXiv", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 568, + 255, + 579 + ], + "spans": [ + { + "bbox": [ + 114, + 568, + 255, + 579 + ], + "score": 1.0, + "content": "preprint arXiv:1712.05438, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 586, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 600 + ], + "score": 1.0, + "content": "Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global con-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 598, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 598, + 506, + 609 + ], + "score": 1.0, + "content": "vergence guarantees for training shallow neural networks. arXiv preprint arXiv:1902.04674,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 608, + 142, + 620 + ], + "spans": [ + { + "bbox": [ + 115, + 608, + 142, + 620 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 104, + 627, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 642 + ], + "score": 1.0, + "content": "Jeffrey Pennington and Pratik Worah. Nonlinear random matrix theory for deep learning. In Advances", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 638, + 390, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 390, + 651 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems, pages 2637–2646, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "Jeffrey Pennington and Pratik Worah. The spectrum of the fisher information matrix of a single-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "hidden-layer neural network. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 680, + 191, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 191, + 691 + ], + "score": 1.0, + "content": "5410–5419, 2018.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "Grant M Rotskoff and Eric Vanden-Eijnden. Neural networks as interacting particle systems:", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Asymptotic convexity of the loss landscape and universal scaling of the approximation error. arXiv", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 721, + 254, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 254, + 732 + ], + "score": 1.0, + "content": "preprint arXiv:1805.00915, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "Mingchen Li, Mahdi Soltanolkotabi, and Samet Oymak. Gradient descent with early stopping is prov-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "ably robust to label noise for overparameterized neural networks. arXiv preprint arXiv:1903.11680,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 147, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 147, + 117 + ], + "score": 1.0, + "content": "2019a.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 506, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "Yuanzhi Li and Yingyu Liang. Learning overparameterized neural networks via stochastic gradient", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 507, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 507, + 147 + ], + "score": 1.0, + "content": "descent on structured data. In Advances in Neural Information Processing Systems, pages 8157–", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 166, + 156 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 166, + 156 + ], + "score": 1.0, + "content": "8166, 2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 123, + 507, + 156 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 506, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 506, + 177 + ], + "score": 1.0, + "content": "Yuanzhi Li, Tengyu Ma, and Hongyang Zhang. Algorithmic regularization in over-parameterized", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 506, + 188 + ], + "score": 1.0, + "content": "matrix sensing and neural networks with quadratic activations. arXiv preprint arXiv:1712.09203,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 185, + 142, + 198 + ], + "spans": [ + { + "bbox": [ + 114, + 185, + 142, + 198 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 164, + 506, + 198 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 205, + 503, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 219 + ], + "score": 1.0, + "content": "Yuanzhi Li, Colin Wei, and Tengyu Ma. Towards explaining the regularization effect of initial large", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 450, + 228 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 450, + 228 + ], + "score": 1.0, + "content": "learning rate in training neural networks. arXiv preprint arXiv:1907.04595, 2019b.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 203, + 505, + 228 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 504, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "Tengyuan Liang and Alexander Rakhlin. Just interpolate: Kernel\" ridgeless\" regression can generalize.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 246, + 278, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 278, + 258 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1808.00387, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 234, + 505, + 258 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 265, + 504, + 288 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "score": 1.0, + "content": "Zhenyu Liao and Romain Couillet. On the spectrum of random features maps of high dimensional", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 276, + 301, + 289 + ], + "spans": [ + { + "bbox": [ + 116, + 276, + 301, + 289 + ], + "score": 1.0, + "content": "data. arXiv preprint arXiv:1805.11916, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 264, + 505, + 289 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 295, + 505, + 318 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Cosme Louart, Zhenyu Liao, Romain Couillet, et al. A random matrix approach to neural networks.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 306, + 356, + 319 + ], + "spans": [ + { + "bbox": [ + 115, + 306, + 356, + 319 + ], + "score": 1.0, + "content": "The Annals of Applied Probability, 28(2):1190–1248, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 295, + 505, + 319 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 324, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "V.A. Marcenko and Leonid Pastur. Distribution of eigenvalues for some sets of random matrices. ˇ", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 336, + 265, + 348 + ], + "spans": [ + { + "bbox": [ + 116, + 336, + 265, + 348 + ], + "score": 1.0, + "content": "Math USSR Sb, 1:457–483, 01 1967.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 106, + 325, + 506, + 348 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 505, + 368 + ], + "score": 1.0, + "content": "Song Mei and Andrea Montanari. The generalization error of random features regression: Precise", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 366, + 436, + 378 + ], + "spans": [ + { + "bbox": [ + 115, + 366, + 436, + 378 + ], + "score": 1.0, + "content": "asymptotics and double descent curve. arXiv preprint arXiv:1908.05355, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 354, + 505, + 378 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "Song Mei, Andrea Montanari, and Phan-Minh Nguyen. A mean field view of the landscape of two-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 396, + 507, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 507, + 408 + ], + "score": 1.0, + "content": "layer neural networks. Proceedings of the National Academy of Sciences, 115(33):E7665–E7671,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 406, + 142, + 418 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 142, + 418 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 385, + 507, + 418 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 425, + 504, + 449 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "score": 1.0, + "content": "Song Mei, Theodor Misiakiewicz, and Andrea Montanari. Mean-field theory of two-layers neural", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 437, + 484, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 437, + 484, + 449 + ], + "score": 1.0, + "content": "networks: dimension-free bounds and kernel limit. arXiv preprint arXiv:1902.06015, 2019.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 425, + 505, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "Vaishnavh Nagarajan and J Zico Kolter. Generalization in deep networks: The role of distance from", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 466, + 335, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 466, + 335, + 479 + ], + "score": 1.0, + "content": "initialization. arXiv preprint arXiv:1901.01672, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 106, + 455, + 505, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 485, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 497, + 464, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 464, + 509 + ], + "score": 1.0, + "content": "role of implicit regularization in deep learning. arXiv preprint arXiv:1412.6614, 2014.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 106, + 486, + 505, + 509 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 504, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "Behnam Neyshabur, Zhiyuan Li, Srinadh Bhojanapalli, Yann LeCun, and Nathan Srebro. Towards", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 527, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 116, + 527, + 505, + 539 + ], + "score": 1.0, + "content": "understanding the role of over-parametrization in generalization of neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 537, + 219, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 219, + 550 + ], + "score": 1.0, + "content": "arXiv:1805.12076, 2018.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 515, + 505, + 550 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 556, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "Atsushi Nitanda and Taiji Suzuki. Stochastic particle gradient descent for infinite ensembles. arXiv", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 568, + 255, + 579 + ], + "spans": [ + { + "bbox": [ + 114, + 568, + 255, + 579 + ], + "score": 1.0, + "content": "preprint arXiv:1712.05438, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 556, + 505, + 579 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 586, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 600 + ], + "score": 1.0, + "content": "Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global con-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 598, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 598, + 506, + 609 + ], + "score": 1.0, + "content": "vergence guarantees for training shallow neural networks. arXiv preprint arXiv:1902.04674,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 608, + 142, + 620 + ], + "spans": [ + { + "bbox": [ + 115, + 608, + 142, + 620 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 586, + 506, + 620 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 627, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 642 + ], + "score": 1.0, + "content": "Jeffrey Pennington and Pratik Worah. Nonlinear random matrix theory for deep learning. In Advances", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 638, + 390, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 390, + 651 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems, pages 2637–2646, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 626, + 505, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "Jeffrey Pennington and Pratik Worah. The spectrum of the fisher information matrix of a single-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "hidden-layer neural network. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 680, + 191, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 191, + 691 + ], + "score": 1.0, + "content": "5410–5419, 2018.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 657, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "Grant M Rotskoff and Eric Vanden-Eijnden. Neural networks as interacting particle systems:", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Asymptotic convexity of the loss landscape and universal scaling of the approximation error. arXiv", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 721, + 254, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 254, + 732 + ], + "score": 1.0, + "content": "preprint arXiv:1805.00915, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 697, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "score": 1.0, + "content": "Justin Sirignano and Konstantinos Spiliopoulos. Mean field analysis of neural networks: A central", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 338, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 338, + 106 + ], + "score": 1.0, + "content": "limit theorem. arXiv preprint arXiv:1808.09372, 2018.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 105, + 112, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Samuel L Smith and Quoc V Le. A bayesian perspective on generalization and stochastic gradient", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 315, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 315, + 135 + ], + "score": 1.0, + "content": "descent. arXiv preprint arXiv:1710.06451, 2017.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 105, + 141, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 506, + 154 + ], + "score": 1.0, + "content": "Zhao Song and Xin Yang. Quadratic suffices for over-parametrization via matrix chernoff bound.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 279, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 279, + 165 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1906.03593, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 108, + 171, + 505, + 205 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 506, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 506, + 185 + ], + "score": 1.0, + "content": "Stefano Spigler, Mario Geiger, Stéphane d’Ascoli, Levent Sagun, Giulio Biroli, and Matthieu Wyart.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "A jamming transition from under-to over-parametrization affects loss landscape and generalization.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 279, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 279, + 206 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1810.09665, 2018.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 105, + 213, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "score": 1.0, + "content": "Taiji Suzuki. Adaptivity of deep relu network for learning in besov and mixed smooth besov spaces:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 224, + 445, + 236 + ], + "spans": [ + { + "bbox": [ + 116, + 224, + 445, + 236 + ], + "score": 1.0, + "content": "optimal rate and curse of dimensionality. arXiv preprint arXiv:1810.08033, 2018.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 105, + 242, + 492, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 492, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 492, + 256 + ], + "score": 1.0, + "content": "Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 109, + 261, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 107, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 107, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "Yuandong Tian. An analytical formula of population gradient for two-layered relu network and its", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 116, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "applications in convergence and critical point analysis. In Proceedings of the 34th International", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 283, + 449, + 297 + ], + "spans": [ + { + "bbox": [ + 116, + 283, + 449, + 297 + ], + "score": 1.0, + "content": "Conference on Machine Learning-Volume 70, pages 3404–3413. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 108, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "Colin Wei, Jason D Lee, Qiang Liu, and Tengyu Ma. On the margin theory of feedforward neural", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 314, + 321, + 326 + ], + "spans": [ + { + "bbox": [ + 115, + 314, + 321, + 326 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1810.05369, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 504, + 355 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Francis Williams, Matthew Trager, Claudio Silva, Daniele Panozzo, Denis Zorin, and Joan Bruna.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 343, + 504, + 356 + ], + "spans": [ + { + "bbox": [ + 116, + 343, + 504, + 356 + ], + "score": 1.0, + "content": "Gradient dynamics of shallow univariate relu networks. arXiv preprint arXiv:1906.07842, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 108, + 362, + 503, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Blake Woodworth, Suriya Gunasekar, Jason Lee, Daniel Soudry, and Nathan Srebro. Kernel and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 374, + 451, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 451, + 385 + ], + "score": 1.0, + "content": "deep regimes in overparametrized models. arXiv preprint arXiv:1906.05827, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 502, + 405 + ], + "lines": [ + { + "bbox": [ + 104, + 390, + 502, + 406 + ], + "spans": [ + { + "bbox": [ + 104, + 390, + 502, + 406 + ], + "score": 1.0, + "content": "Ji Xu and Daniel Hsu. On the number of variables to use in principal component regression. 2019.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 105, + 410, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 506, + 425 + ], + "score": 1.0, + "content": "Gilad Yehudai and Ohad Shamir. On the power and limitations of random features for understanding", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 422, + 349, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 422, + 349, + 434 + ], + "score": 1.0, + "content": "neural networks. arXiv preprint arXiv:1904.00687, 2019.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 440, + 504, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 455 + ], + "score": 1.0, + "content": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 452, + 476, + 464 + ], + "spans": [ + { + "bbox": [ + 116, + 452, + 476, + 464 + ], + "score": 1.0, + "content": "deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "Kai Zhong, Zhao Song, Prateek Jain, Peter L Bartlett, and Inderjit S Dhillon. Recovery guarantees", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 481, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 115, + 481, + 505, + 496 + ], + "score": 1.0, + "content": "for one-hidden-layer neural networks. In Proceedings of the 34th International Conference on", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 493, + 388, + 505 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 388, + 505 + ], + "score": 1.0, + "content": "Machine Learning-Volume 70, pages 4140–4149. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "score": 1.0, + "content": "Justin Sirignano and Konstantinos Spiliopoulos. Mean field analysis of neural networks: A central", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 338, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 338, + 106 + ], + "score": 1.0, + "content": "limit theorem. arXiv preprint arXiv:1808.09372, 2018.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 83, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 112, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Samuel L Smith and Quoc V Le. A bayesian perspective on generalization and stochastic gradient", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 315, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 315, + 135 + ], + "score": 1.0, + "content": "descent. arXiv preprint arXiv:1710.06451, 2017.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 112, + 505, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 141, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 506, + 154 + ], + "score": 1.0, + "content": "Zhao Song and Xin Yang. Quadratic suffices for over-parametrization via matrix chernoff bound.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 279, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 279, + 165 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1906.03593, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 141, + 506, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 171, + 505, + 205 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 506, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 506, + 185 + ], + "score": 1.0, + "content": "Stefano Spigler, Mario Geiger, Stéphane d’Ascoli, Levent Sagun, Giulio Biroli, and Matthieu Wyart.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "A jamming transition from under-to over-parametrization affects loss landscape and generalization.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 279, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 279, + 206 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1810.09665, 2018.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 171, + 506, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 213, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "score": 1.0, + "content": "Taiji Suzuki. Adaptivity of deep relu network for learning in besov and mixed smooth besov spaces:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 224, + 445, + 236 + ], + "spans": [ + { + "bbox": [ + 116, + 224, + 445, + 236 + ], + "score": 1.0, + "content": "optimal rate and curse of dimensionality. arXiv preprint arXiv:1810.08033, 2018.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 212, + 506, + 236 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 242, + 492, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 492, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 492, + 256 + ], + "score": 1.0, + "content": "Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 241, + 492, + 256 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 261, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 107, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 107, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "Yuandong Tian. An analytical formula of population gradient for two-layered relu network and its", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 116, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "applications in convergence and critical point analysis. In Proceedings of the 34th International", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 283, + 449, + 297 + ], + "spans": [ + { + "bbox": [ + 116, + 283, + 449, + 297 + ], + "score": 1.0, + "content": "Conference on Machine Learning-Volume 70, pages 3404–3413. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 107, + 261, + 505, + 297 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "Colin Wei, Jason D Lee, Qiang Liu, and Tengyu Ma. On the margin theory of feedforward neural", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 314, + 321, + 326 + ], + "spans": [ + { + "bbox": [ + 115, + 314, + 321, + 326 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1810.05369, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 303, + 505, + 326 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 504, + 355 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Francis Williams, Matthew Trager, Claudio Silva, Daniele Panozzo, Denis Zorin, and Joan Bruna.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 343, + 504, + 356 + ], + "spans": [ + { + "bbox": [ + 116, + 343, + 504, + 356 + ], + "score": 1.0, + "content": "Gradient dynamics of shallow univariate relu networks. arXiv preprint arXiv:1906.07842, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 332, + 505, + 356 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 362, + 503, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Blake Woodworth, Suriya Gunasekar, Jason Lee, Daniel Soudry, and Nathan Srebro. Kernel and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 374, + 451, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 451, + 385 + ], + "score": 1.0, + "content": "deep regimes in overparametrized models. arXiv preprint arXiv:1906.05827, 2019.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 362, + 505, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 502, + 405 + ], + "lines": [ + { + "bbox": [ + 104, + 390, + 502, + 406 + ], + "spans": [ + { + "bbox": [ + 104, + 390, + 502, + 406 + ], + "score": 1.0, + "content": "Ji Xu and Daniel Hsu. On the number of variables to use in principal component regression. 2019.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 390, + 502, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 410, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 506, + 425 + ], + "score": 1.0, + "content": "Gilad Yehudai and Ohad Shamir. On the power and limitations of random features for understanding", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 422, + 349, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 422, + 349, + 434 + ], + "score": 1.0, + "content": "neural networks. arXiv preprint arXiv:1904.00687, 2019.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 409, + 506, + 434 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 440, + 504, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 455 + ], + "score": 1.0, + "content": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 452, + 476, + 464 + ], + "spans": [ + { + "bbox": [ + 116, + 452, + 476, + 464 + ], + "score": 1.0, + "content": "deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 439, + 506, + 464 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "Kai Zhong, Zhao Song, Prateek Jain, Peter L Bartlett, and Inderjit S Dhillon. Recovery guarantees", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 481, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 115, + 481, + 505, + 496 + ], + "score": 1.0, + "content": "for one-hidden-layer neural networks. In Proceedings of the 34th International Conference on", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 493, + 388, + 505 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 388, + 505 + ], + "score": 1.0, + "content": "Machine Learning-Volume 70, pages 4140–4149. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 470, + 505, + 505 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 111, + 494, + 248 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 111, + 494, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 111, + 494, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 494, + 248 + ], + "score": 0.965, + "type": "image", + "image_path": "8be263e99069d2ff2756929d4f59b8c08885627a89e302d107583385553bcd06.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 111, + 494, + 156.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 156.66666666666666, + 494, + 202.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 202.33333333333331, + 494, + 247.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 256, + 506, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 256, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 505, + 268 + ], + "score": 1.0, + "content": "Figure 4: trajectory of neurons from initialization (dark blue) to optimum (orange) on the first two dimensions", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 277, + 277 + ], + "score": 1.0, + "content": "(two-layer SoftPlus student and linear teacher;", + "type": "text" + }, + { + "bbox": [ + 277, + 267, + 320, + 277 + ], + "score": 0.89, + "content": "\\mathrm { S N R } { = } 1 / 4 \\mathrm { \\Omega }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 266, + 504, + 277 + ], + "score": 1.0, + "content": ". For vanishing initialization the neurons stay close", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "to one another throughout the trajectory, whereas for non-vanishing initialization the neurons stay close to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 285, + 156, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 156, + 298 + ], + "score": 1.0, + "content": "initialization.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 119, + 315, + 488, + 465 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 315, + 488, + 465 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 315, + 488, + 465 + ], + "spans": [ + { + "bbox": [ + 119, + 315, + 488, + 465 + ], + "score": 0.971, + "type": "image", + "image_path": "67d5f63ec5bc3ba3b56af6a95e487287d5f924b0f1e30f617bcb3ec06520d044.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 119, + 315, + 488, + 365.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 119, + 365.0, + 488, + 415.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 119, + 415.0, + 488, + 465.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 472, + 506, + 496 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 473, + 504, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 461, + 485 + ], + "score": 1.0, + "content": "Figure 5: Population risk of two-layer linear network with fixed random 1st layer with", + "type": "text" + }, + { + "bbox": [ + 462, + 473, + 504, + 483 + ], + "score": 0.72, + "content": "S _ { \\mathrm { { N R = } } 2 5 / 1 6 }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 482, + 406, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 391, + 498 + ], + "score": 1.0, + "content": "under Gaussian input and linear teacher. Brighter color indicates larger", + "type": "text" + }, + { + "bbox": [ + 391, + 486, + 402, + 496 + ], + "score": 0.84, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 482, + 406, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "table", + "bbox": [ + 106, + 536, + 450, + 573 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 518, + 380, + 529 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 518, + 380, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 380, + 530 + ], + "score": 1.0, + "content": "SUMMARY OF THE PRESENCE / ABSENCE OF DOUBLE DESCENT", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 106, + 536, + 450, + 573 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 536, + 450, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 450, + 573 + ], + "score": 0.972, + "html": "
Singularity in2nd Layer Trained (RF)Vanishing Init.Non-vanishing Init.
Bias1: No; Y2: Yesγ1: No; γ2: Noγ1: No; 2: No
VarianceY1: No; 2: Yes1: Yes; Y2: NoY1: No; 2: No
", + "type": "table", + "image_path": "daf90c88b28403b003b2e689cb5162fcef4c8474104f827875680e451e9cedd4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 536, + 450, + 548.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 548.3333333333334, + 450, + 560.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 560.6666666666667, + 450, + 573.0000000000001 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 82, + 306, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 307, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 307, + 95 + ], + "score": 1.0, + "content": "A ADDITIONAL FIGURES AND PLOTS", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 111, + 494, + 248 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 111, + 494, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 111, + 494, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 494, + 248 + ], + "score": 0.965, + "type": "image", + "image_path": "8be263e99069d2ff2756929d4f59b8c08885627a89e302d107583385553bcd06.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 111, + 494, + 156.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 156.66666666666666, + 494, + 202.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 202.33333333333331, + 494, + 247.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 256, + 506, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 256, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 505, + 268 + ], + "score": 1.0, + "content": "Figure 4: trajectory of neurons from initialization (dark blue) to optimum (orange) on the first two dimensions", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 277, + 277 + ], + "score": 1.0, + "content": "(two-layer SoftPlus student and linear teacher;", + "type": "text" + }, + { + "bbox": [ + 277, + 267, + 320, + 277 + ], + "score": 0.89, + "content": "\\mathrm { S N R } { = } 1 / 4 \\mathrm { \\Omega }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 266, + 504, + 277 + ], + "score": 1.0, + "content": ". For vanishing initialization the neurons stay close", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "to one another throughout the trajectory, whereas for non-vanishing initialization the neurons stay close to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 285, + 156, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 156, + 298 + ], + "score": 1.0, + "content": "initialization.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 119, + 315, + 488, + 465 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 315, + 488, + 465 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 315, + 488, + 465 + ], + "spans": [ + { + "bbox": [ + 119, + 315, + 488, + 465 + ], + "score": 0.971, + "type": "image", + "image_path": "67d5f63ec5bc3ba3b56af6a95e487287d5f924b0f1e30f617bcb3ec06520d044.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 119, + 315, + 488, + 365.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 119, + 365.0, + 488, + 415.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 119, + 415.0, + 488, + 465.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 472, + 506, + 496 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 473, + 504, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 461, + 485 + ], + "score": 1.0, + "content": "Figure 5: Population risk of two-layer linear network with fixed random 1st layer with", + "type": "text" + }, + { + "bbox": [ + 462, + 473, + 504, + 483 + ], + "score": 0.72, + "content": "S _ { \\mathrm { { N R = } } 2 5 / 1 6 }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 482, + 406, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 391, + 498 + ], + "score": 1.0, + "content": "under Gaussian input and linear teacher. Brighter color indicates larger", + "type": "text" + }, + { + "bbox": [ + 391, + 486, + 402, + 496 + ], + "score": 0.84, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 482, + 406, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "table", + "bbox": [ + 106, + 536, + 450, + 573 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 518, + 380, + 529 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 518, + 380, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 380, + 530 + ], + "score": 1.0, + "content": "SUMMARY OF THE PRESENCE / ABSENCE OF DOUBLE DESCENT", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 106, + 536, + 450, + 573 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 536, + 450, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 450, + 573 + ], + "score": 0.972, + "html": "
Singularity in2nd Layer Trained (RF)Vanishing Init.Non-vanishing Init.
Bias1: No; Y2: Yesγ1: No; γ2: Noγ1: No; 2: No
VarianceY1: No; 2: Yes1: Yes; Y2: NoY1: No; 2: No
", + "type": "table", + "image_path": "daf90c88b28403b003b2e689cb5162fcef4c8474104f827875680e451e9cedd4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 536, + 450, + 548.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 548.3333333333334, + 450, + 560.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 560.6666666666667, + 450, + 573.0000000000001 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 80, + 486, + 217 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 80, + 486, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 80, + 486, + 217 + ], + "spans": [ + { + "bbox": [ + 123, + 80, + 486, + 217 + ], + "score": 0.971, + "type": "image", + "image_path": "487235692f241523abf1600424158205f71f19e5f44d0aed9c5fb66d2d8f7cb9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 80, + 486, + 125.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 125.66666666666666, + 486, + 171.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 171.33333333333331, + 486, + 216.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 219, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 218, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 229 + ], + "score": 1.0, + "content": "Figure 6: Bias and variance of two-layer SoftPlus network with optimized first layer under (A1)(A2). Individual", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 231, + 240 + ], + "score": 1.0, + "content": "dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 232, + 230, + 242, + 239 + ], + "score": 0.83, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. The bias and variance for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 237, + 396, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 396, + 250 + ], + "score": 1.0, + "content": "both initializations is well-aligned with Theorem 7 and Theorem 8, respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 122, + 265, + 486, + 403 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 122, + 265, + 486, + 403 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 122, + 265, + 486, + 403 + ], + "spans": [ + { + "bbox": [ + 122, + 265, + 486, + 403 + ], + "score": 0.968, + "type": "image", + "image_path": "0a41e6cc5c4bd20c793bbe18e10c8b870f315c2459db916ab633ef96972f668a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 122, + 265, + 486, + 311.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 122, + 311.0, + 486, + 357.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 122, + 357.0, + 486, + 403.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 404, + 505, + 435 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 240, + 416 + ], + "score": 1.0, + "content": "Figure 7: Population risk (scaled by", + "type": "text" + }, + { + "bbox": [ + 241, + 405, + 258, + 415 + ], + "score": 0.84, + "content": "1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "of two-layer ReLU network trained to fit a two-layer ReLU teacher", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 150, + 425 + ], + "score": 1.0, + "content": "model with", + "type": "text" + }, + { + "bbox": [ + 150, + 415, + 174, + 424 + ], + "score": 0.9, + "content": "h = d", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 414, + 340, + 425 + ], + "score": 1.0, + "content": "neurons. Brighter color corresponds to larger", + "type": "text" + }, + { + "bbox": [ + 340, + 416, + 350, + 425 + ], + "score": 0.85, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 414, + 505, + 425 + ], + "score": 1.0, + "content": ". Similar to the linear teacher case, double", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 423, + 479, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 479, + 436 + ], + "score": 1.0, + "content": "descent is observed when the second layer is optimized (a) but not when the first layer is optimized (b).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 122, + 452, + 486, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 122, + 452, + 486, + 588 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 122, + 452, + 486, + 588 + ], + "spans": [ + { + "bbox": [ + 122, + 452, + 486, + 588 + ], + "score": 0.97, + "type": "image", + "image_path": "c366b578fe991ae26f542e54255e90708f233ee5480d69cce84eed842879b63d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 122, + 452, + 486, + 497.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 122, + 497.3333333333333, + 486, + 542.6666666666666 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 122, + 542.6666666666666, + 486, + 588.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 590, + 505, + 621 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Figure 8: Bias of (a) SoftPlus and (b) sigmoid two-layer network with optimized first layer under (A1)(A2).", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 600, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 407, + 611 + ], + "score": 1.0, + "content": "Note that bias under i.i.d. initialization is also independent to overparameterization", + "type": "text" + }, + { + "bbox": [ + 408, + 600, + 423, + 610 + ], + "score": 0.82, + "content": "\\left( \\gamma _ { 2 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 600, + 505, + 611 + ], + "score": 1.0, + "content": ", but is higher than the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 609, + 483, + 621 + ], + "spans": [ + { + "bbox": [ + 104, + 609, + 468, + 621 + ], + "score": 1.0, + "content": "bias under symmetric initialization (“doubling trick”) and not always upper-bounded by the null risk", + "type": "text" + }, + { + "bbox": [ + 469, + 609, + 478, + 619 + ], + "score": 0.84, + "content": "r ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 609, + 483, + 621 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 646, + 202, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 645, + 204, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 204, + 661 + ], + "score": 1.0, + "content": "B BACKGROUND", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 108, + 674, + 245, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 673, + 246, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 246, + 687 + ], + "score": 1.0, + "content": "B.1 ROTATIONAL INVARIANCE", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 697, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 507, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 507, + 711 + ], + "score": 1.0, + "content": "The rotational invariance of Gaussian distribution is crucial in our analysis throughout this paper.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 345, + 721 + ], + "score": 1.0, + "content": "A basic observation is that for a random Gaussian matrix", + "type": "text" + }, + { + "bbox": [ + 345, + 710, + 355, + 719 + ], + "score": 0.85, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 709, + 477, + 721 + ], + "score": 1.0, + "content": "and any fixed unitary matrix", + "type": "text" + }, + { + "bbox": [ + 477, + 710, + 486, + 719 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 709, + 505, + 721 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 719, + 268, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 165, + 732 + ], + "score": 1.0, + "content": "distribution of", + "type": "text" + }, + { + "bbox": [ + 166, + 720, + 176, + 730 + ], + "score": 0.84, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 719, + 194, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 720, + 212, + 730 + ], + "score": 0.78, + "content": "U X", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 719, + 268, + 732 + ], + "score": 1.0, + "content": "are the same.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 80, + 486, + 217 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 80, + 486, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 80, + 486, + 217 + ], + "spans": [ + { + "bbox": [ + 123, + 80, + 486, + 217 + ], + "score": 0.971, + "type": "image", + "image_path": "487235692f241523abf1600424158205f71f19e5f44d0aed9c5fb66d2d8f7cb9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 80, + 486, + 125.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 125.66666666666666, + 486, + 171.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 171.33333333333331, + 486, + 216.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 219, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 218, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 229 + ], + "score": 1.0, + "content": "Figure 6: Bias and variance of two-layer SoftPlus network with optimized first layer under (A1)(A2). Individual", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 231, + 240 + ], + "score": 1.0, + "content": "dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 232, + 230, + 242, + 239 + ], + "score": 0.83, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. The bias and variance for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 237, + 396, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 396, + 250 + ], + "score": 1.0, + "content": "both initializations is well-aligned with Theorem 7 and Theorem 8, respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 122, + 265, + 486, + 403 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 122, + 265, + 486, + 403 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 122, + 265, + 486, + 403 + ], + "spans": [ + { + "bbox": [ + 122, + 265, + 486, + 403 + ], + "score": 0.968, + "type": "image", + "image_path": "0a41e6cc5c4bd20c793bbe18e10c8b870f315c2459db916ab633ef96972f668a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 122, + 265, + 486, + 311.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 122, + 311.0, + 486, + 357.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 122, + 357.0, + 486, + 403.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 404, + 505, + 435 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 240, + 416 + ], + "score": 1.0, + "content": "Figure 7: Population risk (scaled by", + "type": "text" + }, + { + "bbox": [ + 241, + 405, + 258, + 415 + ], + "score": 0.84, + "content": "1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "of two-layer ReLU network trained to fit a two-layer ReLU teacher", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 150, + 425 + ], + "score": 1.0, + "content": "model with", + "type": "text" + }, + { + "bbox": [ + 150, + 415, + 174, + 424 + ], + "score": 0.9, + "content": "h = d", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 414, + 340, + 425 + ], + "score": 1.0, + "content": "neurons. Brighter color corresponds to larger", + "type": "text" + }, + { + "bbox": [ + 340, + 416, + 350, + 425 + ], + "score": 0.85, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 414, + 505, + 425 + ], + "score": 1.0, + "content": ". Similar to the linear teacher case, double", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 423, + 479, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 479, + 436 + ], + "score": 1.0, + "content": "descent is observed when the second layer is optimized (a) but not when the first layer is optimized (b).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 122, + 452, + 486, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 122, + 452, + 486, + 588 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 122, + 452, + 486, + 588 + ], + "spans": [ + { + "bbox": [ + 122, + 452, + 486, + 588 + ], + "score": 0.97, + "type": "image", + "image_path": "c366b578fe991ae26f542e54255e90708f233ee5480d69cce84eed842879b63d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 122, + 452, + 486, + 497.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 122, + 497.3333333333333, + 486, + 542.6666666666666 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 122, + 542.6666666666666, + 486, + 588.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 590, + 505, + 621 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Figure 8: Bias of (a) SoftPlus and (b) sigmoid two-layer network with optimized first layer under (A1)(A2).", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 600, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 407, + 611 + ], + "score": 1.0, + "content": "Note that bias under i.i.d. initialization is also independent to overparameterization", + "type": "text" + }, + { + "bbox": [ + 408, + 600, + 423, + 610 + ], + "score": 0.82, + "content": "\\left( \\gamma _ { 2 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 600, + 505, + 611 + ], + "score": 1.0, + "content": ", but is higher than the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 609, + 483, + 621 + ], + "spans": [ + { + "bbox": [ + 104, + 609, + 468, + 621 + ], + "score": 1.0, + "content": "bias under symmetric initialization (“doubling trick”) and not always upper-bounded by the null risk", + "type": "text" + }, + { + "bbox": [ + 469, + 609, + 478, + 619 + ], + "score": 0.84, + "content": "r ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 609, + 483, + 621 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 646, + 202, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 645, + 204, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 204, + 661 + ], + "score": 1.0, + "content": "B BACKGROUND", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 108, + 674, + 245, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 673, + 246, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 246, + 687 + ], + "score": 1.0, + "content": "B.1 ROTATIONAL INVARIANCE", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 697, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 507, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 507, + 711 + ], + "score": 1.0, + "content": "The rotational invariance of Gaussian distribution is crucial in our analysis throughout this paper.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 345, + 721 + ], + "score": 1.0, + "content": "A basic observation is that for a random Gaussian matrix", + "type": "text" + }, + { + "bbox": [ + 345, + 710, + 355, + 719 + ], + "score": 0.85, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 709, + 477, + 721 + ], + "score": 1.0, + "content": "and any fixed unitary matrix", + "type": "text" + }, + { + "bbox": [ + 477, + 710, + 486, + 719 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 709, + 505, + 721 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 719, + 268, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 165, + 732 + ], + "score": 1.0, + "content": "distribution of", + "type": "text" + }, + { + "bbox": [ + 166, + 720, + 176, + 730 + ], + "score": 0.84, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 719, + 194, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 720, + 212, + 730 + ], + "score": 0.78, + "content": "U X", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 719, + 268, + 732 + ], + "score": 1.0, + "content": "are the same.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 696, + 507, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 104, + 78, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 281, + 97 + ], + "score": 1.0, + "content": "Lemma 10 (Rotational Invariance). Denote", + "type": "text" + }, + { + "bbox": [ + 282, + 81, + 342, + 94 + ], + "score": 0.91, + "content": "A ( X ) \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 78, + 421, + 97 + ], + "score": 1.0, + "content": "a matrix function of", + "type": "text" + }, + { + "bbox": [ + 422, + 82, + 467, + 93 + ], + "score": 0.91, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 78, + 479, + 97 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 480, + 82, + 505, + 95 + ], + "score": 0.9, + "content": "A ( X )", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 344, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 159, + 106 + ], + "score": 1.0, + "content": "satisfies that", + "type": "text" + }, + { + "bbox": [ + 159, + 93, + 251, + 106 + ], + "score": 0.93, + "content": "A ( U X ) = U A ( X ) U ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 93, + 311, + 106 + ], + "score": 1.0, + "content": "for all unitary", + "type": "text" + }, + { + "bbox": [ + 311, + 95, + 320, + 104 + ], + "score": 0.8, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 93, + 344, + 106 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 107, + 391, + 131 + ], + "lines": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "spans": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "score": 0.94, + "content": "\\operatorname { \\mathbb { E } } _ { X } [ \\beta ^ { T } A ( X ) \\beta ] = { \\frac { 1 } { d } } \\beta ^ { T } \\beta \\operatorname { \\mathbb { E } } _ { X } [ \\operatorname { t r } \\left( A ( X ) \\right) ] .", + "type": "interline_equation", + "image_path": "833071b965fe2b8b3823963911062089ca34d7e99f1a2b40fc6e6d4f65e91370.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 133, + 470, + 146 + ], + "lines": [ + { + "bbox": [ + 103, + 132, + 469, + 148 + ], + "spans": [ + { + "bbox": [ + 103, + 132, + 192, + 148 + ], + "score": 1.0, + "content": "for any fixed nonzero", + "type": "text" + }, + { + "bbox": [ + 193, + 133, + 225, + 145 + ], + "score": 0.93, + "content": "\\beta \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 132, + 306, + 148 + ], + "score": 1.0, + "content": "and random matrix", + "type": "text" + }, + { + "bbox": [ + 306, + 134, + 316, + 144 + ], + "score": 0.84, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 132, + 402, + 148 + ], + "score": 1.0, + "content": "with each entry i.i.d.", + "type": "text" + }, + { + "bbox": [ + 402, + 133, + 469, + 146 + ], + "score": 0.92, + "content": "X _ { i j } \\sim \\mathcal { N } ( 0 , \\sigma ^ { 2 } )", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 506, + 189 + ], + "lines": [ + { + "bbox": [ + 104, + 152, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 104, + 152, + 277, + 169 + ], + "score": 1.0, + "content": "Proof. Choose a set of Unitary matrices", + "type": "text" + }, + { + "bbox": [ + 277, + 154, + 311, + 167 + ], + "score": 0.93, + "content": "\\{ U _ { i } \\} _ { i = 1 } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 152, + 353, + 169 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 353, + 154, + 418, + 167 + ], + "score": 0.91, + "content": "U _ { i } ^ { \\top } \\beta = \\| \\beta \\| e _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 152, + 450, + 169 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 451, + 156, + 461, + 166 + ], + "score": 0.84, + "content": "e _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 152, + 488, + 169 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 488, + 156, + 492, + 165 + ], + "score": 0.76, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 152, + 506, + 169 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 166, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 186, + 180 + ], + "score": 1.0, + "content": "canonical vector in", + "type": "text" + }, + { + "bbox": [ + 187, + 166, + 199, + 177 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 166, + 230, + 180 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 230, + 167, + 273, + 178 + ], + "score": 0.91, + "content": "U _ { i } X \\sim X", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 166, + 313, + 180 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 313, + 167, + 486, + 179 + ], + "score": 0.87, + "content": "\\mathbb { E } [ A ( X ) ] = \\mathbb { E } [ A ( U _ { i } X ) ] = U _ { i } \\mathbb { E } [ A ( X ) ] U _ { i } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 166, + 506, + 180 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 177, + 134, + 190 + ], + "spans": [ + { + "bbox": [ + 104, + 177, + 134, + 190 + ], + "score": 1.0, + "content": "hence", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 189, + 481, + 224 + ], + "lines": [ + { + "bbox": [ + 114, + 189, + 481, + 224 + ], + "spans": [ + { + "bbox": [ + 114, + 189, + 481, + 224 + ], + "score": 0.93, + "content": "\\mathbb { E } [ \\beta ^ { T } A ( X ) \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } \\mathbb { E } [ \\beta ^ { T } U _ { i } A ( X ) U _ { i } ^ { \\top } \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } e _ { i } ^ { T } \\mathbb { E } [ A ( X ) ] e _ { i } = \\frac { \\beta ^ { T } \\beta } { d } \\mathbb { E } [ \\mathrm { t r } ( A ( X ) ) ] .", + "type": "interline_equation", + "image_path": "0f0a8bdb9f31ba01e5ce82b1e4ed0b7f59808e7937c7a2f8da7c0eba8f7617dc.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 114, + 189, + 481, + 200.66666666666666 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 114, + 200.66666666666666, + 481, + 212.33333333333331 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 114, + 212.33333333333331, + 481, + 223.99999999999997 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 507, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 314, + 240 + ], + "score": 1.0, + "content": "Note that the property also holds for matrix function", + "type": "text" + }, + { + "bbox": [ + 315, + 227, + 339, + 239 + ], + "score": 0.93, + "content": "A ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 225, + 392, + 240 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 392, + 226, + 484, + 239 + ], + "score": 0.93, + "content": "A ( X U ) = U A ( X ) U ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 225, + 506, + 240 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 506, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 248, + 251 + ], + "score": 1.0, + "content": "can be extended to matrix function", + "type": "text" + }, + { + "bbox": [ + 248, + 239, + 256, + 248 + ], + "score": 0.82, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 237, + 406, + 251 + ], + "score": 1.0, + "content": "that takes multiple matrices as input.", + "type": "text" + }, + { + "bbox": [ + 494, + 238, + 506, + 250 + ], + "score": 0.996, + "content": "□", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 400, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 260, + 401, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 401, + 275 + ], + "score": 1.0, + "content": "For brevity we will refer to rotational invariance instead of equation (15).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 285, + 246, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 247, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 247, + 298 + ], + "score": 1.0, + "content": "B.2 MARCENKO ˇ –PASTUR LAW", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 305, + 453, + 318 + ], + "lines": [ + { + "bbox": [ + 104, + 304, + 456, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 253, + 321 + ], + "score": 1.0, + "content": "For a real symmetric random matrix", + "type": "text" + }, + { + "bbox": [ + 253, + 306, + 295, + 317 + ], + "score": 0.91, + "content": "A \\in \\mathbb { R } ^ { p \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 304, + 456, + 321 + ], + "score": 1.0, + "content": ", define its empirical spectral density as", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 320, + 367, + 353 + ], + "lines": [ + { + "bbox": [ + 243, + 320, + 367, + 353 + ], + "spans": [ + { + "bbox": [ + 243, + 320, + 367, + 353 + ], + "score": 0.95, + "content": "\\mu _ { A } ( d \\lambda ) = \\frac { 1 } { p } \\sum _ { i = 1 } ^ { p } \\delta _ { \\lambda _ { i } ( A ) } ( \\lambda ) d \\lambda ,", + "type": "interline_equation", + "image_path": "bd285da91081ea0ebf82d1825ac45c0adacc1c9e692c65d7e2093dcd9bafee6d.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 320, + 367, + 336.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 243, + 336.5, + 367, + 353.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 506, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 134, + 368 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 355, + 209, + 367 + ], + "score": 0.92, + "content": "\\delta _ { a } ( x ) = \\delta ( x - a )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 354, + 361, + 368 + ], + "score": 1.0, + "content": "is the Dirac delta function. Assume", + "type": "text" + }, + { + "bbox": [ + 362, + 355, + 452, + 368 + ], + "score": 0.92, + "content": "A = W _ { p } \\sim W _ { p } ( I , n )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "is a Wishart", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 157, + 380 + ], + "score": 1.0, + "content": "matrix, i.e.", + "type": "text" + }, + { + "bbox": [ + 158, + 367, + 223, + 380 + ], + "score": 0.91, + "content": "W _ { p } = X ^ { \\top } X / n", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 366, + 244, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 244, + 367, + 292, + 378 + ], + "score": 0.91, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "is random Gaussian matrix with each column i.i.d.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 378, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 378, + 165, + 390 + ], + "score": 0.92, + "content": "X _ { i } \\sim \\mathcal { N } ( \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 378, + 350, + 391 + ], + "score": 1.0, + "content": ". Marcenko and Pastur ˇ (1967) showed that as", + "type": "text" + }, + { + "bbox": [ + 350, + 380, + 392, + 390 + ], + "score": 0.89, + "content": "n , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 378, + 411, + 391 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 411, + 378, + 486, + 390 + ], + "score": 0.92, + "content": "p / n = \\gamma \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 378, + 506, + 391 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 389, + 454, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 211, + 403 + ], + "score": 1.0, + "content": "empirical spectral density", + "type": "text" + }, + { + "bbox": [ + 212, + 390, + 249, + 403 + ], + "score": 0.93, + "content": "\\mu _ { W _ { p } } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 389, + 407, + 403 + ], + "score": 1.0, + "content": "converges weakly to a limiting density", + "type": "text" + }, + { + "bbox": [ + 407, + 390, + 448, + 403 + ], + "score": 0.92, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 389, + 454, + 403 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 404, + 468, + 430 + ], + "lines": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "spans": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "score": 0.91, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( d \\lambda ) = [ 1 - \\gamma ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\frac { 1 } { 2 \\pi \\gamma \\lambda } \\sqrt { ( ( 1 + \\sqrt \\gamma ) ^ { 2 } - \\lambda ) ( \\lambda - ( 1 - \\sqrt \\gamma ) ^ { 2 } ) } d \\lambda .", + "type": "interline_equation", + "image_path": "982a95475ba530700c71603522d670cc909fbc0f9a160aa61cf0f58c7ff4736f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 506, + 468 + ], + "lines": [ + { + "bbox": [ + 104, + 430, + 507, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 136, + 446 + ], + "score": 1.0, + "content": "We say", + "type": "text" + }, + { + "bbox": [ + 137, + 434, + 165, + 446 + ], + "score": 0.89, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 430, + 416, + 446 + ], + "score": 1.0, + "content": "is the density of the Marˇcenko–Pastur distribution with support", + "type": "text" + }, + { + "bbox": [ + 416, + 432, + 507, + 446 + ], + "score": 0.86, + "content": "S = [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 +", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 444, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 444, + 133, + 458 + ], + "score": 0.9, + "content": "\\sqrt { \\gamma } ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 444, + 151, + 458 + ], + "score": 1.0, + "content": "(for", + "type": "text" + }, + { + "bbox": [ + 151, + 446, + 195, + 457 + ], + "score": 0.89, + "content": "0 < \\gamma < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 444, + 210, + 458 + ], + "score": 1.0, + "content": ") or", + "type": "text" + }, + { + "bbox": [ + 210, + 444, + 351, + 458 + ], + "score": 0.91, + "content": "S = \\{ 0 \\} \\cup [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 + \\sqrt { \\gamma } ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 444, + 369, + 458 + ], + "score": 1.0, + "content": "(for", + "type": "text" + }, + { + "bbox": [ + 369, + 445, + 394, + 456 + ], + "score": 0.88, + "content": "\\gamma \\geq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 444, + 506, + 458 + ], + "score": 1.0, + "content": "). Note that this implies that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 456, + 419, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 250, + 469 + ], + "score": 1.0, + "content": "the smallest non-zero eigenvalue of", + "type": "text" + }, + { + "bbox": [ + 250, + 457, + 259, + 466 + ], + "score": 0.83, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 456, + 389, + 469 + ], + "score": 1.0, + "content": "is bounded away from 0 a.s. for", + "type": "text" + }, + { + "bbox": [ + 389, + 457, + 414, + 468 + ], + "score": 0.9, + "content": "\\gamma \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 456, + 419, + 469 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 506, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "The explicit form of Marcenko–Pastur distribution allows us to investigate the asymptotic properties ˇ", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "of random matrices. Generally speaking, by Pormanteau theorem one can translate any bounded", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 496, + 507, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 507, + 507 + ], + "score": 1.0, + "content": "continuous function on the empirical spectral density to the one on Marcenko–Pastur distribution, i.e. ˇ", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 505, + 403, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 174, + 519 + ], + "score": 1.0, + "content": "for any bounded", + "type": "text" + }, + { + "bbox": [ + 174, + 506, + 229, + 518 + ], + "score": 0.94, + "content": "f ( \\lambda ) \\in C ( S )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 505, + 244, + 519 + ], + "score": 1.0, + "content": ", as", + "type": "text" + }, + { + "bbox": [ + 244, + 507, + 286, + 518 + ], + "score": 0.89, + "content": "n , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 505, + 307, + 519 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 308, + 506, + 344, + 518 + ], + "score": 0.92, + "content": "p / n = \\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 505, + 403, + 519 + ], + "score": 1.0, + "content": ", almost surely", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 520, + 394, + 546 + ], + "lines": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "spans": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "score": 0.95, + "content": "\\int _ { S } f ( \\lambda ) \\cdot \\mu _ { W } ( d \\lambda ) \\to \\int f _ { S } ( \\lambda ) \\cdot \\mu _ { M P } ( d \\lambda ) .", + "type": "interline_equation", + "image_path": "b5b5c50caa17b9bae2674f02aa659aac7984332e27c225a9b1c289dc84c7a551.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 479, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 480, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 451, + 567 + ], + "score": 1.0, + "content": "One implication is the following trace concentration on the inverse Wishart matrix for", + "type": "text" + }, + { + "bbox": [ + 451, + 554, + 476, + 565 + ], + "score": 0.91, + "content": "\\gamma < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 552, + 480, + 567 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 567, + 494, + 600 + ], + "lines": [ + { + "bbox": [ + 115, + 567, + 494, + 600 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 494, + 600 + ], + "score": 0.94, + "content": "\\operatorname { t r } ( X ^ { \\top } X ) = \\operatorname { t r } ( { \\frac { 1 } { p } } W _ { p } ^ { - 1 } ) = { \\frac { 1 } { p } } \\sum _ { i = 1 } ^ { p } { \\frac { 1 } { \\lambda _ { i } ( W _ { p } ) } } = \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { W _ { p } } ( d \\lambda ) \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { \\operatorname { M P } ( \\gamma ) } ( d \\lambda ) = { \\frac { 1 } { 1 - \\gamma } } .", + "type": "interline_equation", + "image_path": "2d0f780ecfa10baaa2208e09863b4e1e023a7dda5466a5cd80525455e3c92adb.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 115, + 567, + 494, + 578.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 115, + 578.0, + 494, + 589.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 115, + 589.0, + 494, + 600.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 506, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 381, + 626 + ], + "score": 1.0, + "content": "We remark that instability of the trace of the invert Wishart matrix as", + "type": "text" + }, + { + "bbox": [ + 382, + 614, + 410, + 625 + ], + "score": 0.91, + "content": "\\gamma 1", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 613, + 505, + 626 + ], + "score": 1.0, + "content": "plays an important role", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 624, + 249, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 249, + 636 + ], + "score": 1.0, + "content": "in the double descent phenomenon.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "title", + "bbox": [ + 108, + 648, + 260, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 261, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 261, + 662 + ], + "score": 1.0, + "content": "B.3 ORTHOGONAL POLYNOMIALS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 506, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 479, + 683 + ], + "score": 1.0, + "content": "Orthogonal polynomials are useful in the analysis of nonlinear random matrices. Suppose", + "type": "text" + }, + { + "bbox": [ + 479, + 670, + 487, + 681 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 667, + 506, + 683 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 679, + 508, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 679, + 151, + 696 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 152, + 682, + 196, + 694 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 679, + 226, + 696 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 227, + 682, + 280, + 694 + ], + "score": 0.77, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 679, + 284, + 696 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 284, + 680, + 406, + 694 + ], + "score": 0.89, + "content": "\\mu _ { G } ( d x ) = ( \\sqrt { 2 \\pi } ) ^ { - 1 } e ^ { - x ^ { 2 } / 2 } d x", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 679, + 508, + 696 + ], + "score": 1.0, + "content": "is the Gaussian measure.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 693, + 280, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 122, + 705 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 694, + 148, + 704 + ], + "score": 0.88, + "content": "n \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 693, + 280, + 705 + ], + "score": 1.0, + "content": ", define the Hermite polynomials", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 707, + 378, + 731 + ], + "lines": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "spans": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "score": 0.94, + "content": "H _ { n } ( x ) = ( - 1 ) ^ { n } e ^ { - x ^ { 2 } / 2 } \\frac { \\partial ^ { n } } { \\partial x ^ { n } } e ^ { - x ^ { 2 } / 2 } ,", + "type": "interline_equation", + "image_path": "b7dfb5d7f01e6357a4b55bb87789baf72f245abf1cc48ff117c370fea81ed43c.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "spans": [], + "index": 40 + } + ] + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 104, + 78, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 281, + 97 + ], + "score": 1.0, + "content": "Lemma 10 (Rotational Invariance). Denote", + "type": "text" + }, + { + "bbox": [ + 282, + 81, + 342, + 94 + ], + "score": 0.91, + "content": "A ( X ) \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 78, + 421, + 97 + ], + "score": 1.0, + "content": "a matrix function of", + "type": "text" + }, + { + "bbox": [ + 422, + 82, + 467, + 93 + ], + "score": 0.91, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 78, + 479, + 97 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 480, + 82, + 505, + 95 + ], + "score": 0.9, + "content": "A ( X )", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 344, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 159, + 106 + ], + "score": 1.0, + "content": "satisfies that", + "type": "text" + }, + { + "bbox": [ + 159, + 93, + 251, + 106 + ], + "score": 0.93, + "content": "A ( U X ) = U A ( X ) U ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 93, + 311, + 106 + ], + "score": 1.0, + "content": "for all unitary", + "type": "text" + }, + { + "bbox": [ + 311, + 95, + 320, + 104 + ], + "score": 0.8, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 93, + 344, + 106 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 104, + 78, + 505, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 107, + 391, + 131 + ], + "lines": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "spans": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "score": 0.94, + "content": "\\operatorname { \\mathbb { E } } _ { X } [ \\beta ^ { T } A ( X ) \\beta ] = { \\frac { 1 } { d } } \\beta ^ { T } \\beta \\operatorname { \\mathbb { E } } _ { X } [ \\operatorname { t r } \\left( A ( X ) \\right) ] .", + "type": "interline_equation", + "image_path": "833071b965fe2b8b3823963911062089ca34d7e99f1a2b40fc6e6d4f65e91370.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 219, + 107, + 391, + 131 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 133, + 470, + 146 + ], + "lines": [ + { + "bbox": [ + 103, + 132, + 469, + 148 + ], + "spans": [ + { + "bbox": [ + 103, + 132, + 192, + 148 + ], + "score": 1.0, + "content": "for any fixed nonzero", + "type": "text" + }, + { + "bbox": [ + 193, + 133, + 225, + 145 + ], + "score": 0.93, + "content": "\\beta \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 132, + 306, + 148 + ], + "score": 1.0, + "content": "and random matrix", + "type": "text" + }, + { + "bbox": [ + 306, + 134, + 316, + 144 + ], + "score": 0.84, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 132, + 402, + 148 + ], + "score": 1.0, + "content": "with each entry i.i.d.", + "type": "text" + }, + { + "bbox": [ + 402, + 133, + 469, + 146 + ], + "score": 0.92, + "content": "X _ { i j } \\sim \\mathcal { N } ( 0 , \\sigma ^ { 2 } )", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 103, + 132, + 469, + 148 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 506, + 189 + ], + "lines": [ + { + "bbox": [ + 104, + 152, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 104, + 152, + 277, + 169 + ], + "score": 1.0, + "content": "Proof. Choose a set of Unitary matrices", + "type": "text" + }, + { + "bbox": [ + 277, + 154, + 311, + 167 + ], + "score": 0.93, + "content": "\\{ U _ { i } \\} _ { i = 1 } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 152, + 353, + 169 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 353, + 154, + 418, + 167 + ], + "score": 0.91, + "content": "U _ { i } ^ { \\top } \\beta = \\| \\beta \\| e _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 152, + 450, + 169 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 451, + 156, + 461, + 166 + ], + "score": 0.84, + "content": "e _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 152, + 488, + 169 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 488, + 156, + 492, + 165 + ], + "score": 0.76, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 152, + 506, + 169 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 166, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 186, + 180 + ], + "score": 1.0, + "content": "canonical vector in", + "type": "text" + }, + { + "bbox": [ + 187, + 166, + 199, + 177 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 166, + 230, + 180 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 230, + 167, + 273, + 178 + ], + "score": 0.91, + "content": "U _ { i } X \\sim X", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 166, + 313, + 180 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 313, + 167, + 486, + 179 + ], + "score": 0.87, + "content": "\\mathbb { E } [ A ( X ) ] = \\mathbb { E } [ A ( U _ { i } X ) ] = U _ { i } \\mathbb { E } [ A ( X ) ] U _ { i } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 166, + 506, + 180 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 177, + 134, + 190 + ], + "spans": [ + { + "bbox": [ + 104, + 177, + 134, + 190 + ], + "score": 1.0, + "content": "hence", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 104, + 152, + 506, + 190 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 189, + 481, + 224 + ], + "lines": [ + { + "bbox": [ + 114, + 189, + 481, + 224 + ], + "spans": [ + { + "bbox": [ + 114, + 189, + 481, + 224 + ], + "score": 0.93, + "content": "\\mathbb { E } [ \\beta ^ { T } A ( X ) \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } \\mathbb { E } [ \\beta ^ { T } U _ { i } A ( X ) U _ { i } ^ { \\top } \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } e _ { i } ^ { T } \\mathbb { E } [ A ( X ) ] e _ { i } = \\frac { \\beta ^ { T } \\beta } { d } \\mathbb { E } [ \\mathrm { t r } ( A ( X ) ) ] .", + "type": "interline_equation", + "image_path": "0f0a8bdb9f31ba01e5ce82b1e4ed0b7f59808e7937c7a2f8da7c0eba8f7617dc.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 114, + 189, + 481, + 200.66666666666666 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 114, + 200.66666666666666, + 481, + 212.33333333333331 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 114, + 212.33333333333331, + 481, + 223.99999999999997 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 507, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 314, + 240 + ], + "score": 1.0, + "content": "Note that the property also holds for matrix function", + "type": "text" + }, + { + "bbox": [ + 315, + 227, + 339, + 239 + ], + "score": 0.93, + "content": "A ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 225, + 392, + 240 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 392, + 226, + 484, + 239 + ], + "score": 0.93, + "content": "A ( X U ) = U A ( X ) U ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 225, + 506, + 240 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 506, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 248, + 251 + ], + "score": 1.0, + "content": "can be extended to matrix function", + "type": "text" + }, + { + "bbox": [ + 248, + 239, + 256, + 248 + ], + "score": 0.82, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 237, + 406, + 251 + ], + "score": 1.0, + "content": "that takes multiple matrices as input.", + "type": "text" + }, + { + "bbox": [ + 494, + 238, + 506, + 250 + ], + "score": 0.996, + "content": "□", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 225, + 506, + 251 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 400, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 260, + 401, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 401, + 275 + ], + "score": 1.0, + "content": "For brevity we will refer to rotational invariance instead of equation (15).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 260, + 401, + 275 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 285, + 246, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 247, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 247, + 298 + ], + "score": 1.0, + "content": "B.2 MARCENKO ˇ –PASTUR LAW", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 305, + 453, + 318 + ], + "lines": [ + { + "bbox": [ + 104, + 304, + 456, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 253, + 321 + ], + "score": 1.0, + "content": "For a real symmetric random matrix", + "type": "text" + }, + { + "bbox": [ + 253, + 306, + 295, + 317 + ], + "score": 0.91, + "content": "A \\in \\mathbb { R } ^ { p \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 304, + 456, + 321 + ], + "score": 1.0, + "content": ", define its empirical spectral density as", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 304, + 456, + 321 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 320, + 367, + 353 + ], + "lines": [ + { + "bbox": [ + 243, + 320, + 367, + 353 + ], + "spans": [ + { + "bbox": [ + 243, + 320, + 367, + 353 + ], + "score": 0.95, + "content": "\\mu _ { A } ( d \\lambda ) = \\frac { 1 } { p } \\sum _ { i = 1 } ^ { p } \\delta _ { \\lambda _ { i } ( A ) } ( \\lambda ) d \\lambda ,", + "type": "interline_equation", + "image_path": "bd285da91081ea0ebf82d1825ac45c0adacc1c9e692c65d7e2093dcd9bafee6d.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 320, + 367, + 336.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 243, + 336.5, + 367, + 353.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 506, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 134, + 368 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 355, + 209, + 367 + ], + "score": 0.92, + "content": "\\delta _ { a } ( x ) = \\delta ( x - a )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 354, + 361, + 368 + ], + "score": 1.0, + "content": "is the Dirac delta function. Assume", + "type": "text" + }, + { + "bbox": [ + 362, + 355, + 452, + 368 + ], + "score": 0.92, + "content": "A = W _ { p } \\sim W _ { p } ( I , n )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "is a Wishart", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 157, + 380 + ], + "score": 1.0, + "content": "matrix, i.e.", + "type": "text" + }, + { + "bbox": [ + 158, + 367, + 223, + 380 + ], + "score": 0.91, + "content": "W _ { p } = X ^ { \\top } X / n", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 366, + 244, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 244, + 367, + 292, + 378 + ], + "score": 0.91, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "is random Gaussian matrix with each column i.i.d.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 378, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 378, + 165, + 390 + ], + "score": 0.92, + "content": "X _ { i } \\sim \\mathcal { N } ( \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 378, + 350, + 391 + ], + "score": 1.0, + "content": ". Marcenko and Pastur ˇ (1967) showed that as", + "type": "text" + }, + { + "bbox": [ + 350, + 380, + 392, + 390 + ], + "score": 0.89, + "content": "n , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 378, + 411, + 391 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 411, + 378, + 486, + 390 + ], + "score": 0.92, + "content": "p / n = \\gamma \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 378, + 506, + 391 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 389, + 454, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 211, + 403 + ], + "score": 1.0, + "content": "empirical spectral density", + "type": "text" + }, + { + "bbox": [ + 212, + 390, + 249, + 403 + ], + "score": 0.93, + "content": "\\mu _ { W _ { p } } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 389, + 407, + 403 + ], + "score": 1.0, + "content": "converges weakly to a limiting density", + "type": "text" + }, + { + "bbox": [ + 407, + 390, + 448, + 403 + ], + "score": 0.92, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 389, + 454, + 403 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 354, + 506, + 403 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 404, + 468, + 430 + ], + "lines": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "spans": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "score": 0.91, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( d \\lambda ) = [ 1 - \\gamma ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\frac { 1 } { 2 \\pi \\gamma \\lambda } \\sqrt { ( ( 1 + \\sqrt \\gamma ) ^ { 2 } - \\lambda ) ( \\lambda - ( 1 - \\sqrt \\gamma ) ^ { 2 } ) } d \\lambda .", + "type": "interline_equation", + "image_path": "982a95475ba530700c71603522d670cc909fbc0f9a160aa61cf0f58c7ff4736f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 127, + 404, + 468, + 430 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 506, + 468 + ], + "lines": [ + { + "bbox": [ + 104, + 430, + 507, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 136, + 446 + ], + "score": 1.0, + "content": "We say", + "type": "text" + }, + { + "bbox": [ + 137, + 434, + 165, + 446 + ], + "score": 0.89, + "content": "\\mu _ { \\mathrm { M P } ( \\gamma ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 430, + 416, + 446 + ], + "score": 1.0, + "content": "is the density of the Marˇcenko–Pastur distribution with support", + "type": "text" + }, + { + "bbox": [ + 416, + 432, + 507, + 446 + ], + "score": 0.86, + "content": "S = [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 +", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 444, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 444, + 133, + 458 + ], + "score": 0.9, + "content": "\\sqrt { \\gamma } ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 444, + 151, + 458 + ], + "score": 1.0, + "content": "(for", + "type": "text" + }, + { + "bbox": [ + 151, + 446, + 195, + 457 + ], + "score": 0.89, + "content": "0 < \\gamma < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 444, + 210, + 458 + ], + "score": 1.0, + "content": ") or", + "type": "text" + }, + { + "bbox": [ + 210, + 444, + 351, + 458 + ], + "score": 0.91, + "content": "S = \\{ 0 \\} \\cup [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 + \\sqrt { \\gamma } ) ^ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 444, + 369, + 458 + ], + "score": 1.0, + "content": "(for", + "type": "text" + }, + { + "bbox": [ + 369, + 445, + 394, + 456 + ], + "score": 0.88, + "content": "\\gamma \\geq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 444, + 506, + 458 + ], + "score": 1.0, + "content": "). Note that this implies that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 456, + 419, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 250, + 469 + ], + "score": 1.0, + "content": "the smallest non-zero eigenvalue of", + "type": "text" + }, + { + "bbox": [ + 250, + 457, + 259, + 466 + ], + "score": 0.83, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 456, + 389, + 469 + ], + "score": 1.0, + "content": "is bounded away from 0 a.s. for", + "type": "text" + }, + { + "bbox": [ + 389, + 457, + 414, + 468 + ], + "score": 0.9, + "content": "\\gamma \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 456, + 419, + 469 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 104, + 430, + 507, + 469 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 506, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "The explicit form of Marcenko–Pastur distribution allows us to investigate the asymptotic properties ˇ", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "of random matrices. Generally speaking, by Pormanteau theorem one can translate any bounded", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 496, + 507, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 507, + 507 + ], + "score": 1.0, + "content": "continuous function on the empirical spectral density to the one on Marcenko–Pastur distribution, i.e. ˇ", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 505, + 403, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 174, + 519 + ], + "score": 1.0, + "content": "for any bounded", + "type": "text" + }, + { + "bbox": [ + 174, + 506, + 229, + 518 + ], + "score": 0.94, + "content": "f ( \\lambda ) \\in C ( S )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 505, + 244, + 519 + ], + "score": 1.0, + "content": ", as", + "type": "text" + }, + { + "bbox": [ + 244, + 507, + 286, + 518 + ], + "score": 0.89, + "content": "n , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 505, + 307, + 519 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 308, + 506, + 344, + 518 + ], + "score": 0.92, + "content": "p / n = \\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 505, + 403, + 519 + ], + "score": 1.0, + "content": ", almost surely", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 473, + 507, + 519 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 520, + 394, + 546 + ], + "lines": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "spans": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "score": 0.95, + "content": "\\int _ { S } f ( \\lambda ) \\cdot \\mu _ { W } ( d \\lambda ) \\to \\int f _ { S } ( \\lambda ) \\cdot \\mu _ { M P } ( d \\lambda ) .", + "type": "interline_equation", + "image_path": "b5b5c50caa17b9bae2674f02aa659aac7984332e27c225a9b1c289dc84c7a551.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 216, + 520, + 394, + 546 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 479, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 480, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 451, + 567 + ], + "score": 1.0, + "content": "One implication is the following trace concentration on the inverse Wishart matrix for", + "type": "text" + }, + { + "bbox": [ + 451, + 554, + 476, + 565 + ], + "score": 0.91, + "content": "\\gamma < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 552, + 480, + 567 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 552, + 480, + 567 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 567, + 494, + 600 + ], + "lines": [ + { + "bbox": [ + 115, + 567, + 494, + 600 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 494, + 600 + ], + "score": 0.94, + "content": "\\operatorname { t r } ( X ^ { \\top } X ) = \\operatorname { t r } ( { \\frac { 1 } { p } } W _ { p } ^ { - 1 } ) = { \\frac { 1 } { p } } \\sum _ { i = 1 } ^ { p } { \\frac { 1 } { \\lambda _ { i } ( W _ { p } ) } } = \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { W _ { p } } ( d \\lambda ) \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { \\operatorname { M P } ( \\gamma ) } ( d \\lambda ) = { \\frac { 1 } { 1 - \\gamma } } .", + "type": "interline_equation", + "image_path": "2d0f780ecfa10baaa2208e09863b4e1e023a7dda5466a5cd80525455e3c92adb.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 115, + 567, + 494, + 578.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 115, + 578.0, + 494, + 589.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 115, + 589.0, + 494, + 600.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 506, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 381, + 626 + ], + "score": 1.0, + "content": "We remark that instability of the trace of the invert Wishart matrix as", + "type": "text" + }, + { + "bbox": [ + 382, + 614, + 410, + 625 + ], + "score": 0.91, + "content": "\\gamma 1", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 613, + 505, + 626 + ], + "score": 1.0, + "content": "plays an important role", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 624, + 249, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 249, + 636 + ], + "score": 1.0, + "content": "in the double descent phenomenon.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 613, + 505, + 636 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 648, + 260, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 261, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 261, + 662 + ], + "score": 1.0, + "content": "B.3 ORTHOGONAL POLYNOMIALS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 506, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 479, + 683 + ], + "score": 1.0, + "content": "Orthogonal polynomials are useful in the analysis of nonlinear random matrices. Suppose", + "type": "text" + }, + { + "bbox": [ + 479, + 670, + 487, + 681 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 667, + 506, + 683 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 679, + 508, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 679, + 151, + 696 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 152, + 682, + 196, + 694 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 679, + 226, + 696 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 227, + 682, + 280, + 694 + ], + "score": 0.77, + "content": "G \\sim \\mathcal { N } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 679, + 284, + 696 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 284, + 680, + 406, + 694 + ], + "score": 0.89, + "content": "\\mu _ { G } ( d x ) = ( \\sqrt { 2 \\pi } ) ^ { - 1 } e ^ { - x ^ { 2 } / 2 } d x", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 679, + 508, + 696 + ], + "score": 1.0, + "content": "is the Gaussian measure.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 693, + 280, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 122, + 705 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 694, + 148, + 704 + ], + "score": 0.88, + "content": "n \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 693, + 280, + 705 + ], + "score": 1.0, + "content": ", define the Hermite polynomials", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 104, + 667, + 508, + 705 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 707, + 378, + 731 + ], + "lines": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "spans": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "score": 0.94, + "content": "H _ { n } ( x ) = ( - 1 ) ^ { n } e ^ { - x ^ { 2 } / 2 } \\frac { \\partial ^ { n } } { \\partial x ^ { n } } e ^ { - x ^ { 2 } / 2 } ,", + "type": "interline_equation", + "image_path": "b7dfb5d7f01e6357a4b55bb87789baf72f245abf1cc48ff117c370fea81ed43c.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 232, + 707, + 378, + 731 + ], + "spans": [], + "index": 40 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 291, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 291, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 291, + 96 + ], + "score": 1.0, + "content": "Note that orthogonality can be easily verified:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 100, + 418, + 127 + ], + "lines": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "spans": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "score": 0.95, + "content": "\\mathbb { E } [ H _ { j } ( G ) H _ { k } ( G ) ] = \\int _ { \\mathbb { R } } H _ { j } ( x ) H _ { k } ( x ) \\mu _ { G } ( d x ) = j ! \\cdot \\delta _ { j k } .", + "type": "interline_equation", + "image_path": "2e9d9c0c84ec4d7edfddbeca264979db59bb9c3273641cfe0a87f5b12ec5abcf.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 134, + 504, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 131, + 148 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 135, + 180, + 147 + ], + "score": 0.92, + "content": "\\{ H _ { i } ( x ) \\} _ { i = 0 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 133, + 316, + 148 + ], + "score": 1.0, + "content": "forms a set of orthogonal basis in", + "type": "text" + }, + { + "bbox": [ + 317, + 134, + 361, + 147 + ], + "score": 0.93, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 133, + 415, + 148 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 415, + 135, + 435, + 147 + ], + "score": 0.92, + "content": "\\phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 133, + 506, + 148 + ], + "score": 1.0, + "content": "can be expanded", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 145, + 215, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 215, + 159 + ], + "score": 1.0, + "content": "under the Hermite basis as", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 162, + 434, + 196 + ], + "lines": [ + { + "bbox": [ + 177, + 162, + 434, + 196 + ], + "spans": [ + { + "bbox": [ + 177, + 162, + 434, + 196 + ], + "score": 0.95, + "content": "\\phi ( x ) = \\sum _ { i = 0 } ^ { \\infty } c _ { i } H _ { i } ( x ) = \\sum _ { i = 0 } ^ { \\infty } \\left( \\frac { 1 } { k ! } \\int _ { \\mathbb { R } } \\phi ( a ) H _ { i } ( a ) \\mu _ { G } ( d a ) \\right) H _ { i } ( x ) .", + "type": "interline_equation", + "image_path": "cf3e2ec794a62ef6df6913e0a506cd1108f960db43f48819f91342e73f19bdbb.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 177, + 162, + 434, + 173.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 173.33333333333334, + 434, + 184.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 184.66666666666669, + 434, + 196.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 207, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "Following Cheng and Singer (2013), we mainly focus on the first few terms of the Hermite expansion.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 218, + 404, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 185, + 231 + ], + "score": 1.0, + "content": "One can check that", + "type": "text" + }, + { + "bbox": [ + 185, + 219, + 231, + 231 + ], + "score": 0.89, + "content": "H _ { 0 } ( x ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 218, + 234, + 231 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 235, + 219, + 281, + 231 + ], + "score": 0.88, + "content": "H _ { 1 } ( x ) = x", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 218, + 302, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 302, + 219, + 322, + 231 + ], + "score": 0.92, + "content": "\\phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 218, + 404, + 231 + ], + "score": 1.0, + "content": "can be expanded as", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 237, + 361, + 252 + ], + "lines": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "spans": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "score": 0.92, + "content": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x + \\phi _ { \\perp } ( x ) ,", + "type": "interline_equation", + "image_path": "226531a5330bd604d496dcddb6c029eb33b8f83d7b3f79483fb402113d0504ba.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 506, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 134, + 272 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 258, + 195, + 271 + ], + "score": 0.84, + "content": "c _ { 0 } = \\mathbb { E } [ \\phi ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 257, + 200, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 200, + 258, + 268, + 271 + ], + "score": 0.9, + "content": "c _ { 1 } = \\mathbb { E } [ G \\phi ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 257, + 506, + 272 + ], + "score": 1.0, + "content": ", and terms in the RHS are orthogonal to one another in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 267, + 369, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 269, + 151, + 281 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 267, + 369, + 283 + ], + "score": 1.0, + "content": ". Taking square and expectation over both sides yields", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 198, + 288, + 409, + 303 + ], + "lines": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "spans": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] = \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] , } \\end{array}", + "type": "interline_equation", + "image_path": "12684a4b97f9202c539740eb08f6762f801dfc54acbb2ad6395521a01bc8ca6b.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 506, + 345 + ], + "lines": [ + { + "bbox": [ + 104, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 104, + 309, + 169, + 325 + ], + "score": 1.0, + "content": "which indicates", + "type": "text" + }, + { + "bbox": [ + 169, + 311, + 391, + 324 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } - \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } = \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] \\ge 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 309, + 506, + 325 + ], + "score": 1.0, + "content": ". Note that the equality holds", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 321, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 159, + 335 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 159, + 323, + 223, + 335 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\phi _ { \\bot } ( G ) ^ { 2 } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 321, + 241, + 335 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 242, + 323, + 308, + 335 + ], + "score": 0.91, + "content": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 321, + 505, + 335 + ], + "score": 1.0, + "content": "is linear. One application of this decomposition is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "score": 1.0, + "content": "to “linearize” a non-linear matrix in high dimensions, which will be useful for the following sections.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 362, + 262, + 376 + ], + "lines": [ + { + "bbox": [ + 106, + 361, + 263, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 263, + 378 + ], + "score": 1.0, + "content": "C PROOF OF MAIN RESULTS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 219, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 221, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 221, + 402 + ], + "score": 1.0, + "content": "C.1 PROOF OF LEMMA 1", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 108, + 410, + 504, + 437 + ], + "lines": [ + { + "bbox": [ + 106, + 408, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 169, + 424 + ], + "score": 1.0, + "content": "Given features", + "type": "text" + }, + { + "bbox": [ + 169, + 410, + 215, + 421 + ], + "score": 0.91, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 408, + 246, + 424 + ], + "score": 1.0, + "content": ", labels", + "type": "text" + }, + { + "bbox": [ + 246, + 410, + 279, + 423 + ], + "score": 0.92, + "content": "\\pmb { y } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 408, + 374, + 424 + ], + "score": 1.0, + "content": "and model parameters", + "type": "text" + }, + { + "bbox": [ + 374, + 411, + 381, + 421 + ], + "score": 0.79, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 408, + 469, + 424 + ], + "score": 1.0, + "content": ", the gradient flow of", + "type": "text" + }, + { + "bbox": [ + 469, + 411, + 475, + 421 + ], + "score": 0.79, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 408, + 506, + 424 + ], + "score": 1.0, + "content": "on the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 416, + 289, + 441 + ], + "spans": [ + { + "bbox": [ + 103, + 416, + 158, + 441 + ], + "score": 1.0, + "content": "squared loss", + "type": "text" + }, + { + "bbox": [ + 159, + 422, + 215, + 438 + ], + "score": 0.93, + "content": "\\left\\| \\pmb { y } - X ^ { \\top } \\pmb { \\theta } \\right\\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 416, + 289, + 441 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 444, + 366, + 470 + ], + "lines": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "spans": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "score": 0.94, + "content": "\\frac { \\partial \\pmb { \\theta } ( t ) } { \\partial t } = \\frac { 1 } { n } X ( y - X ^ { \\top } \\pmb { \\theta } ( t ) ) .", + "type": "interline_equation", + "image_path": "dbdad43c1e0beba1bc12f91a6bd87045cd6f5dcdf6c7b08528a7ae7dcfaaf568.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 476, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 477, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 201, + 489 + ], + "score": 1.0, + "content": "Thus with initialization", + "type": "text" + }, + { + "bbox": [ + 202, + 477, + 213, + 487 + ], + "score": 0.89, + "content": "\\pmb { \\theta } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 474, + 347, + 489 + ], + "score": 1.0, + "content": ", the solution of this ODE at time", + "type": "text" + }, + { + "bbox": [ + 347, + 477, + 352, + 486 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 474, + 477, + 489 + ], + "score": 1.0, + "content": "can be written in explicit form", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 195, + 494, + 415, + 516 + ], + "lines": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "spans": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "score": 0.93, + "content": "\\pmb { \\theta } ( t ) = e ^ { - \\frac { t } { n } X X ^ { \\top } } \\pmb { \\theta } _ { 0 } + ( X X ^ { \\top } ) ^ { \\dagger } \\left( I - e ^ { - \\frac { t } { n } X X ^ { \\top } } \\right) X \\pmb { y } .", + "type": "interline_equation", + "image_path": "2d757828b13e9155af54df8b08c0c8971ae620a6a40afc9f48108c39ecdde395.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 322, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 522, + 322, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 131, + 535 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 523, + 161, + 534 + ], + "score": 0.91, + "content": "\\pmb { \\theta } _ { 0 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 522, + 192, + 535 + ], + "score": 1.0, + "content": ", taking", + "type": "text" + }, + { + "bbox": [ + 192, + 524, + 222, + 533 + ], + "score": 0.9, + "content": "t \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 522, + 322, + 535 + ], + "score": 1.0, + "content": "yields the desired result.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 555, + 230, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 554, + 231, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 231, + 568 + ], + "score": 1.0, + "content": "C.2 PROOF OF THEOREM 2", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 576, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 333, + 589 + ], + "score": 1.0, + "content": "We compute the bias and variance for different cases of", + "type": "text" + }, + { + "bbox": [ + 333, + 579, + 358, + 588 + ], + "score": 0.78, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 577, + 505, + 589 + ], + "score": 1.0, + "content": ". We first discuss the case where the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 588, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 169, + 599 + ], + "score": 1.0, + "content": "random feature", + "type": "text" + }, + { + "bbox": [ + 170, + 588, + 185, + 599 + ], + "score": 0.92, + "content": "\\Phi _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 588, + 352, + 599 + ], + "score": 1.0, + "content": "is not full rank (Case I). Otherwise when", + "type": "text" + }, + { + "bbox": [ + 352, + 589, + 367, + 599 + ], + "score": 0.92, + "content": "\\Phi _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 588, + 505, + 599 + ], + "score": 1.0, + "content": "is full rank, we discuss whether it", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 599, + 330, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 330, + 610 + ], + "score": 1.0, + "content": "is full column rank (Case II) or full row rank (Case III).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 624, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 139, + 638 + ], + "score": 1.0, + "content": "Case I:", + "type": "text" + }, + { + "bbox": [ + 140, + 624, + 167, + 635 + ], + "score": 0.92, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 623, + 261, + 638 + ], + "score": 1.0, + "content": "is not full rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 261, + 626, + 329, + 636 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } < 1 , \\gamma _ { 2 } > \\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 623, + 405, + 638 + ], + "score": 1.0, + "content": ". In this case rank", + "type": "text" + }, + { + "bbox": [ + 405, + 624, + 502, + 637 + ], + "score": 0.87, + "content": "( \\Phi _ { X } ) = d < \\operatorname* { m i n } ( n , h )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 623, + 506, + 638 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 635, + 334, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 334, + 648 + ], + "score": 1.0, + "content": "and thus by taking the Moore-Penrose inverse we obtain", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 654, + 383, + 670 + ], + "lines": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "spans": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = W ( X ^ { \\top } W ) ^ { \\dagger } { \\boldsymbol { y } } = ( X X ^ { \\top } ) ^ { - 1 } X { \\boldsymbol { y } } .", + "type": "interline_equation", + "image_path": "5ef80b31e9945ee11acf8d4a05a6697c17d69130393268c46722d4b3c1e888d6.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 682, + 506, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "It is clear that the mean and variance is identical to the underparameterized regime in Hastie et al.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 692, + 234, + 707 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 177, + 707 + ], + "score": 1.0, + "content": "(2019), i.e. when", + "type": "text" + }, + { + "bbox": [ + 177, + 694, + 229, + 705 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 692, + 234, + 707 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 712, + 361, + 736 + ], + "lines": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "spans": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "score": 0.94, + "content": "B 0 ; \\quad V \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "474990e4ccfc71621442e1ac6495e4e3d2d2543a9b05f97006a93d40244500cd.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "spans": [], + "index": 33 + } + ] + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 523, + 504, + 533 + ], + "lines": [ + { + "bbox": [ + 496, + 524, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 496, + 524, + 504, + 534 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 291, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 291, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 291, + 96 + ], + "score": 1.0, + "content": "Note that orthogonality can be easily verified:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 81, + 291, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 100, + 418, + 127 + ], + "lines": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "spans": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "score": 0.95, + "content": "\\mathbb { E } [ H _ { j } ( G ) H _ { k } ( G ) ] = \\int _ { \\mathbb { R } } H _ { j } ( x ) H _ { k } ( x ) \\mu _ { G } ( d x ) = j ! \\cdot \\delta _ { j k } .", + "type": "interline_equation", + "image_path": "2e9d9c0c84ec4d7edfddbeca264979db59bb9c3273641cfe0a87f5b12ec5abcf.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 192, + 100, + 418, + 127 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 134, + 504, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 131, + 148 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 135, + 180, + 147 + ], + "score": 0.92, + "content": "\\{ H _ { i } ( x ) \\} _ { i = 0 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 133, + 316, + 148 + ], + "score": 1.0, + "content": "forms a set of orthogonal basis in", + "type": "text" + }, + { + "bbox": [ + 317, + 134, + 361, + 147 + ], + "score": 0.93, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 133, + 415, + 148 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 415, + 135, + 435, + 147 + ], + "score": 0.92, + "content": "\\phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 133, + 506, + 148 + ], + "score": 1.0, + "content": "can be expanded", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 145, + 215, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 215, + 159 + ], + "score": 1.0, + "content": "under the Hermite basis as", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 133, + 506, + 159 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 162, + 434, + 196 + ], + "lines": [ + { + "bbox": [ + 177, + 162, + 434, + 196 + ], + "spans": [ + { + "bbox": [ + 177, + 162, + 434, + 196 + ], + "score": 0.95, + "content": "\\phi ( x ) = \\sum _ { i = 0 } ^ { \\infty } c _ { i } H _ { i } ( x ) = \\sum _ { i = 0 } ^ { \\infty } \\left( \\frac { 1 } { k ! } \\int _ { \\mathbb { R } } \\phi ( a ) H _ { i } ( a ) \\mu _ { G } ( d a ) \\right) H _ { i } ( x ) .", + "type": "interline_equation", + "image_path": "cf3e2ec794a62ef6df6913e0a506cd1108f960db43f48819f91342e73f19bdbb.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 177, + 162, + 434, + 173.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 173.33333333333334, + 434, + 184.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 184.66666666666669, + 434, + 196.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 207, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "Following Cheng and Singer (2013), we mainly focus on the first few terms of the Hermite expansion.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 218, + 404, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 185, + 231 + ], + "score": 1.0, + "content": "One can check that", + "type": "text" + }, + { + "bbox": [ + 185, + 219, + 231, + 231 + ], + "score": 0.89, + "content": "H _ { 0 } ( x ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 218, + 234, + 231 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 235, + 219, + 281, + 231 + ], + "score": 0.88, + "content": "H _ { 1 } ( x ) = x", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 218, + 302, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 302, + 219, + 322, + 231 + ], + "score": 0.92, + "content": "\\phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 218, + 404, + 231 + ], + "score": 1.0, + "content": "can be expanded as", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 206, + 506, + 231 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 237, + 361, + 252 + ], + "lines": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "spans": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "score": 0.92, + "content": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x + \\phi _ { \\perp } ( x ) ,", + "type": "interline_equation", + "image_path": "226531a5330bd604d496dcddb6c029eb33b8f83d7b3f79483fb402113d0504ba.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 249, + 237, + 361, + 252 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 506, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 134, + 272 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 258, + 195, + 271 + ], + "score": 0.84, + "content": "c _ { 0 } = \\mathbb { E } [ \\phi ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 257, + 200, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 200, + 258, + 268, + 271 + ], + "score": 0.9, + "content": "c _ { 1 } = \\mathbb { E } [ G \\phi ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 257, + 506, + 272 + ], + "score": 1.0, + "content": ", and terms in the RHS are orthogonal to one another in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 267, + 369, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 269, + 151, + 281 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 267, + 369, + 283 + ], + "score": 1.0, + "content": ". Taking square and expectation over both sides yields", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 257, + 506, + 283 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 198, + 288, + 409, + 303 + ], + "lines": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "spans": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] = \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] , } \\end{array}", + "type": "interline_equation", + "image_path": "12684a4b97f9202c539740eb08f6762f801dfc54acbb2ad6395521a01bc8ca6b.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 198, + 288, + 409, + 303 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 506, + 345 + ], + "lines": [ + { + "bbox": [ + 104, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 104, + 309, + 169, + 325 + ], + "score": 1.0, + "content": "which indicates", + "type": "text" + }, + { + "bbox": [ + 169, + 311, + 391, + 324 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } - \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } = \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] \\ge 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 309, + 506, + 325 + ], + "score": 1.0, + "content": ". Note that the equality holds", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 321, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 159, + 335 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 159, + 323, + 223, + 335 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\phi _ { \\bot } ( G ) ^ { 2 } ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 321, + 241, + 335 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 242, + 323, + 308, + 335 + ], + "score": 0.91, + "content": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 321, + 505, + 335 + ], + "score": 1.0, + "content": "is linear. One application of this decomposition is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "score": 1.0, + "content": "to “linearize” a non-linear matrix in high dimensions, which will be useful for the following sections.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 309, + 506, + 345 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 362, + 262, + 376 + ], + "lines": [ + { + "bbox": [ + 106, + 361, + 263, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 263, + 378 + ], + "score": 1.0, + "content": "C PROOF OF MAIN RESULTS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 219, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 221, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 221, + 402 + ], + "score": 1.0, + "content": "C.1 PROOF OF LEMMA 1", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 108, + 410, + 504, + 437 + ], + "lines": [ + { + "bbox": [ + 106, + 408, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 169, + 424 + ], + "score": 1.0, + "content": "Given features", + "type": "text" + }, + { + "bbox": [ + 169, + 410, + 215, + 421 + ], + "score": 0.91, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 408, + 246, + 424 + ], + "score": 1.0, + "content": ", labels", + "type": "text" + }, + { + "bbox": [ + 246, + 410, + 279, + 423 + ], + "score": 0.92, + "content": "\\pmb { y } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 408, + 374, + 424 + ], + "score": 1.0, + "content": "and model parameters", + "type": "text" + }, + { + "bbox": [ + 374, + 411, + 381, + 421 + ], + "score": 0.79, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 408, + 469, + 424 + ], + "score": 1.0, + "content": ", the gradient flow of", + "type": "text" + }, + { + "bbox": [ + 469, + 411, + 475, + 421 + ], + "score": 0.79, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 408, + 506, + 424 + ], + "score": 1.0, + "content": "on the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 416, + 289, + 441 + ], + "spans": [ + { + "bbox": [ + 103, + 416, + 158, + 441 + ], + "score": 1.0, + "content": "squared loss", + "type": "text" + }, + { + "bbox": [ + 159, + 422, + 215, + 438 + ], + "score": 0.93, + "content": "\\left\\| \\pmb { y } - X ^ { \\top } \\pmb { \\theta } \\right\\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 416, + 289, + 441 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 103, + 408, + 506, + 441 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 444, + 366, + 470 + ], + "lines": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "spans": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "score": 0.94, + "content": "\\frac { \\partial \\pmb { \\theta } ( t ) } { \\partial t } = \\frac { 1 } { n } X ( y - X ^ { \\top } \\pmb { \\theta } ( t ) ) .", + "type": "interline_equation", + "image_path": "dbdad43c1e0beba1bc12f91a6bd87045cd6f5dcdf6c7b08528a7ae7dcfaaf568.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 246, + 444, + 366, + 470 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 476, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 477, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 201, + 489 + ], + "score": 1.0, + "content": "Thus with initialization", + "type": "text" + }, + { + "bbox": [ + 202, + 477, + 213, + 487 + ], + "score": 0.89, + "content": "\\pmb { \\theta } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 474, + 347, + 489 + ], + "score": 1.0, + "content": ", the solution of this ODE at time", + "type": "text" + }, + { + "bbox": [ + 347, + 477, + 352, + 486 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 474, + 477, + 489 + ], + "score": 1.0, + "content": "can be written in explicit form", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 474, + 477, + 489 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 195, + 494, + 415, + 516 + ], + "lines": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "spans": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "score": 0.93, + "content": "\\pmb { \\theta } ( t ) = e ^ { - \\frac { t } { n } X X ^ { \\top } } \\pmb { \\theta } _ { 0 } + ( X X ^ { \\top } ) ^ { \\dagger } \\left( I - e ^ { - \\frac { t } { n } X X ^ { \\top } } \\right) X \\pmb { y } .", + "type": "interline_equation", + "image_path": "2d757828b13e9155af54df8b08c0c8971ae620a6a40afc9f48108c39ecdde395.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 195, + 494, + 415, + 516 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 322, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 522, + 322, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 131, + 535 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 523, + 161, + 534 + ], + "score": 0.91, + "content": "\\pmb { \\theta } _ { 0 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 522, + 192, + 535 + ], + "score": 1.0, + "content": ", taking", + "type": "text" + }, + { + "bbox": [ + 192, + 524, + 222, + 533 + ], + "score": 0.9, + "content": "t \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 522, + 322, + 535 + ], + "score": 1.0, + "content": "yields the desired result.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 522, + 322, + 535 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 555, + 230, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 554, + 231, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 231, + 568 + ], + "score": 1.0, + "content": "C.2 PROOF OF THEOREM 2", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 576, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 333, + 589 + ], + "score": 1.0, + "content": "We compute the bias and variance for different cases of", + "type": "text" + }, + { + "bbox": [ + 333, + 579, + 358, + 588 + ], + "score": 0.78, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 577, + 505, + 589 + ], + "score": 1.0, + "content": ". We first discuss the case where the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 588, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 169, + 599 + ], + "score": 1.0, + "content": "random feature", + "type": "text" + }, + { + "bbox": [ + 170, + 588, + 185, + 599 + ], + "score": 0.92, + "content": "\\Phi _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 588, + 352, + 599 + ], + "score": 1.0, + "content": "is not full rank (Case I). Otherwise when", + "type": "text" + }, + { + "bbox": [ + 352, + 589, + 367, + 599 + ], + "score": 0.92, + "content": "\\Phi _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 588, + 505, + 599 + ], + "score": 1.0, + "content": "is full rank, we discuss whether it", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 599, + 330, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 330, + 610 + ], + "score": 1.0, + "content": "is full column rank (Case II) or full row rank (Case III).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 577, + 505, + 610 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 624, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 139, + 638 + ], + "score": 1.0, + "content": "Case I:", + "type": "text" + }, + { + "bbox": [ + 140, + 624, + 167, + 635 + ], + "score": 0.92, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 623, + 261, + 638 + ], + "score": 1.0, + "content": "is not full rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 261, + 626, + 329, + 636 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } < 1 , \\gamma _ { 2 } > \\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 623, + 405, + 638 + ], + "score": 1.0, + "content": ". In this case rank", + "type": "text" + }, + { + "bbox": [ + 405, + 624, + 502, + 637 + ], + "score": 0.87, + "content": "( \\Phi _ { X } ) = d < \\operatorname* { m i n } ( n , h )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 623, + 506, + 638 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 635, + 334, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 334, + 648 + ], + "score": 1.0, + "content": "and thus by taking the Moore-Penrose inverse we obtain", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 623, + 506, + 648 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 654, + 383, + 670 + ], + "lines": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "spans": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "score": 0.92, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = W ( X ^ { \\top } W ) ^ { \\dagger } { \\boldsymbol { y } } = ( X X ^ { \\top } ) ^ { - 1 } X { \\boldsymbol { y } } .", + "type": "interline_equation", + "image_path": "5ef80b31e9945ee11acf8d4a05a6697c17d69130393268c46722d4b3c1e888d6.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 228, + 654, + 383, + 670 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 682, + 506, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "It is clear that the mean and variance is identical to the underparameterized regime in Hastie et al.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 692, + 234, + 707 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 177, + 707 + ], + "score": 1.0, + "content": "(2019), i.e. when", + "type": "text" + }, + { + "bbox": [ + 177, + 694, + 229, + 705 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 692, + 234, + 707 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 682, + 506, + 707 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 712, + 361, + 736 + ], + "lines": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "spans": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "score": 0.94, + "content": "B 0 ; \\quad V \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "474990e4ccfc71621442e1ac6495e4e3d2d2543a9b05f97006a93d40244500cd.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 250, + 712, + 361, + 736 + ], + "spans": [], + "index": 33 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 144, + 95 + ], + "score": 1.0, + "content": "Case II:", + "type": "text" + }, + { + "bbox": [ + 144, + 81, + 172, + 93 + ], + "score": 0.85, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 80, + 292, + 95 + ], + "score": 1.0, + "content": "has full column rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 293, + 83, + 362, + 95 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } < 1 , \\gamma _ { 1 } > \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 80, + 506, + 95 + ], + "score": 1.0, + "content": ". By Lemma 1, the solution of the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 217, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 217, + 106 + ], + "score": 1.0, + "content": "second layer coefficients is", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 110, + 376, + 126 + ], + "lines": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "spans": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\hat { \\beta } = W ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X y . } \\end{array}", + "type": "interline_equation", + "image_path": "db6a7ab2e5d37354a1a7e2bfefe5b814e4b18a174d7d8cf315f09918a89cd80c.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 486, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 131, + 486, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 138, + 146 + ], + "score": 1.0, + "content": "Denote", + "type": "text" + }, + { + "bbox": [ + 138, + 132, + 192, + 143 + ], + "score": 0.92, + "content": "W = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 131, + 486, + 146 + ], + "score": 1.0, + "content": "the singular value decomposition. We perform the block decomposition:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 149, + 356, + 178 + ], + "lines": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "spans": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "score": 0.95, + "content": "\\Sigma = \\left[ \\Sigma _ { 0 } \\right] , X = \\left[ \\Sigma _ { 1 } \\right] ,", + "type": "interline_equation", + "image_path": "8e724e50ca9424395b9283a4a5889c12e652bf61e7a42bfd5db110cf4a12fed3.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 218 + ], + "lines": [ + { + "bbox": [ + 104, + 180, + 507, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 180, + 133, + 198 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 182, + 310, + 196 + ], + "score": 0.9, + "content": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { h \\times h } , X _ { 0 } \\in \\mathbb { R } ^ { h \\times n } , X _ { 1 } \\in \\mathbb { R } ^ { ( d - h ) \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 180, + 378, + 198 + ], + "score": 1.0, + "content": ", and notice that", + "type": "text" + }, + { + "bbox": [ + 378, + 184, + 409, + 195 + ], + "score": 0.91, + "content": "X _ { 0 } , X _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 180, + 507, + 198 + ], + "score": 1.0, + "content": "are independent. By a", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "concentration of measure argument (e.g. Tao (2012); Hastie et al. (2019)) one can show that the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 206, + 425, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 425, + 218 + ], + "score": 1.0, + "content": "quantity below tightly concentrates at its expectation. For the variance we have", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 222, + 464, + 291 + ], + "lines": [ + { + "bbox": [ + 147, + 222, + 464, + 291 + ], + "spans": [ + { + "bbox": [ + 147, + 222, + 464, + 291 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } X ^ { \\top } \\sigma ^ { 2 } X W \\left( \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } \\right) \\right. } \\\\ & { \\left. = \\sigma ^ { 2 } \\mathrm { t r } \\left( W ^ { \\top } W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } U ^ { \\top } X X ^ { \\top } U \\Sigma \\right) ^ { - 1 } \\right) \\right. } \\\\ & { \\left. \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\left( X _ { 0 } X _ { 0 } ^ { \\top } \\right) ^ { - 1 } \\right) \\right. \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "a478871fcdaef4bc974ffb67e8b80c1135afc95829ea44be74511ef48dcc3d72.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 147, + 222, + 464, + 245.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 147, + 245.0, + 464, + 268.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 147, + 268.0, + 464, + 291.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 294, + 457, + 307 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 457, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 457, + 308 + ], + "score": 1.0, + "content": "where the last equality follows from Appendix B.2. Similarly for the bias term we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 313, + 497, + 474 + ], + "lines": [ + { + "bbox": [ + 114, + 313, + 497, + 474 + ], + "spans": [ + { + "bbox": [ + 114, + 313, + 497, + 474 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { B = \\| { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } \\beta - \\beta \\| _ { 2 } ^ { 2 } } \\\\ & { \\quad = \\beta ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) \\beta } \\\\ & { \\stackrel { ( \\psi ) } { = } \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma X ^ { \\top } ( V \\Sigma X ^ { \\top } X X ^ { \\top } U \\Sigma \\Sigma { \\cal { W } } ^ { \\top } ) ^ { - 1 } V \\Sigma \\Sigma ^ { \\top } { \\cal { U } } ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( \\cdots ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma ^ { \\top } \\Sigma \\sigma \\Sigma ^ { \\top } \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( \\Sigma \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\Sigma \\Big ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( ( \\Sigma ^ { \\top } X ^ { \\top } \\Sigma ) ^ { - 1 } \\end{array}", + "type": "interline_equation", + "image_path": "515891f2ef048c678fbfca2d4a5db9c09d60df581a642c374b48b8a5384a2777.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 114, + 313, + 497, + 366.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 114, + 366.6666666666667, + 497, + 420.33333333333337 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 114, + 420.33333333333337, + 497, + 474.00000000000006 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 478, + 504, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 288, + 491 + ], + "score": 1.0, + "content": "where symmetric arguments are omitted as", + "type": "text" + }, + { + "bbox": [ + 289, + 479, + 309, + 489 + ], + "score": 0.81, + "content": "( \\cdot \\cdot \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 477, + 505, + 491 + ], + "score": 1.0, + "content": ", and (i) follows from the rotational invariance", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 489, + 466, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 281, + 504 + ], + "score": 1.0, + "content": "argument introduced in Lemma 10 and that", + "type": "text" + }, + { + "bbox": [ + 282, + 489, + 325, + 502 + ], + "score": 0.91, + "content": "\\beta ^ { \\top } \\beta = r ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 489, + 466, + 504 + ], + "score": 1.0, + "content": ". By the block decomposition (30),", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 507, + 439, + 535 + ], + "lines": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "spans": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "score": 0.91, + "content": "\\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } = \\left[ \\begin{array} { l l } { 0 } & { ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } } \\\\ { 0 } & { - I _ { d - h } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "75b68ce49c070f2e797f82c3695a9dbb8c5ffcaddcd04477af0fd67b3b103535.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 539, + 253, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 253, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 253, + 552 + ], + "score": 1.0, + "content": "Therefore the bias term simplifies to", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 128, + 556, + 482, + 639 + ], + "lines": [ + { + "bbox": [ + 128, + 556, + 482, + 639 + ], + "spans": [ + { + "bbox": [ + 128, + 556, + 482, + 639 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) ^ { \\top } \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) \\right) } } \\\\ { { \\mathrm { } = \\frac { r ^ { 2 } } { d } \\left( \\mathrm { t r } \\left( ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } X _ { 1 } X _ { 0 } ^ { \\top } ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } \\right) + ( d - h ) \\right) } } \\\\ { { \\mathrm { } \\to \\frac { r ^ { 2 } } { d } \\left( \\frac { ( d - h ) h } { n - h - 1 } + d - h \\right) \\to \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } \\left( 1 - \\gamma _ { 2 } \\right) } r ^ { 2 } . } } \\end{array}", + "type": "interline_equation", + "image_path": "d5f2f2ce220c66d82124355105f5bd55076362bad3c56e92df6e32ab92705ff6.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 128, + 556, + 482, + 583.6666666666666 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 128, + 583.6666666666666, + 482, + 611.3333333333333 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 128, + 611.3333333333333, + 482, + 638.9999999999999 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 282, + 654 + ], + "lines": [ + { + "bbox": [ + 106, + 641, + 281, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 228, + 656 + ], + "score": 1.0, + "content": "Thus we have obtained that as", + "type": "text" + }, + { + "bbox": [ + 229, + 643, + 281, + 654 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 660, + 348, + 685 + ], + "lines": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "spans": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "score": 0.95, + "content": "B \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } ( 1 - \\gamma _ { 2 } ) } r ^ { 2 } .", + "type": "interline_equation", + "image_path": "b44679d17b6fe7ed1fadb754db3d0f4639acf3b59a5c7f8f6e646d0ca2791004.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 696, + 495, + 710 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 496, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 146, + 712 + ], + "score": 1.0, + "content": "Case III:", + "type": "text" + }, + { + "bbox": [ + 147, + 696, + 174, + 708 + ], + "score": 0.88, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 694, + 276, + 712 + ], + "score": 1.0, + "content": "has full row rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 277, + 698, + 338, + 709 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } > 1 , \\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 694, + 496, + 712 + ], + "score": 1.0, + "content": ". Similarly, the least squares solution is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 714, + 375, + 730 + ], + "lines": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "spans": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "score": 0.91, + "content": "\\hat { \\beta } = W W ^ { \\top } X ( X ^ { \\top } W W ^ { \\top } X ) ^ { - 1 } \\pmb { y } ,", + "type": "interline_equation", + "image_path": "06daaa58c150186f04888890add9047d6e277956a9d567db4bf4a049dbeaf530.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "spans": [], + "index": 25 + } + ] + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 144, + 95 + ], + "score": 1.0, + "content": "Case II:", + "type": "text" + }, + { + "bbox": [ + 144, + 81, + 172, + 93 + ], + "score": 0.85, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 80, + 292, + 95 + ], + "score": 1.0, + "content": "has full column rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 293, + 83, + 362, + 95 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } < 1 , \\gamma _ { 1 } > \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 80, + 506, + 95 + ], + "score": 1.0, + "content": ". By Lemma 1, the solution of the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 217, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 217, + 106 + ], + "score": 1.0, + "content": "second layer coefficients is", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 80, + 506, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 110, + 376, + 126 + ], + "lines": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "spans": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\hat { \\beta } = W ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X y . } \\end{array}", + "type": "interline_equation", + "image_path": "db6a7ab2e5d37354a1a7e2bfefe5b814e4b18a174d7d8cf315f09918a89cd80c.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 234, + 110, + 376, + 126 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 486, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 131, + 486, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 138, + 146 + ], + "score": 1.0, + "content": "Denote", + "type": "text" + }, + { + "bbox": [ + 138, + 132, + 192, + 143 + ], + "score": 0.92, + "content": "W = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 131, + 486, + 146 + ], + "score": 1.0, + "content": "the singular value decomposition. We perform the block decomposition:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 131, + 486, + 146 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 149, + 356, + 178 + ], + "lines": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "spans": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "score": 0.95, + "content": "\\Sigma = \\left[ \\Sigma _ { 0 } \\right] , X = \\left[ \\Sigma _ { 1 } \\right] ,", + "type": "interline_equation", + "image_path": "8e724e50ca9424395b9283a4a5889c12e652bf61e7a42bfd5db110cf4a12fed3.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 253, + 149, + 356, + 178 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 218 + ], + "lines": [ + { + "bbox": [ + 104, + 180, + 507, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 180, + 133, + 198 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 182, + 310, + 196 + ], + "score": 0.9, + "content": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { h \\times h } , X _ { 0 } \\in \\mathbb { R } ^ { h \\times n } , X _ { 1 } \\in \\mathbb { R } ^ { ( d - h ) \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 180, + 378, + 198 + ], + "score": 1.0, + "content": ", and notice that", + "type": "text" + }, + { + "bbox": [ + 378, + 184, + 409, + 195 + ], + "score": 0.91, + "content": "X _ { 0 } , X _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 180, + 507, + 198 + ], + "score": 1.0, + "content": "are independent. By a", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "concentration of measure argument (e.g. Tao (2012); Hastie et al. (2019)) one can show that the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 206, + 425, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 425, + 218 + ], + "score": 1.0, + "content": "quantity below tightly concentrates at its expectation. For the variance we have", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 104, + 180, + 507, + 218 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 222, + 464, + 291 + ], + "lines": [ + { + "bbox": [ + 147, + 222, + 464, + 291 + ], + "spans": [ + { + "bbox": [ + 147, + 222, + 464, + 291 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } X ^ { \\top } \\sigma ^ { 2 } X W \\left( \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } \\right) \\right. } \\\\ & { \\left. = \\sigma ^ { 2 } \\mathrm { t r } \\left( W ^ { \\top } W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } U ^ { \\top } X X ^ { \\top } U \\Sigma \\right) ^ { - 1 } \\right) \\right. } \\\\ & { \\left. \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\left( X _ { 0 } X _ { 0 } ^ { \\top } \\right) ^ { - 1 } \\right) \\right. \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "a478871fcdaef4bc974ffb67e8b80c1135afc95829ea44be74511ef48dcc3d72.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 147, + 222, + 464, + 245.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 147, + 245.0, + 464, + 268.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 147, + 268.0, + 464, + 291.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 294, + 457, + 307 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 457, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 457, + 308 + ], + "score": 1.0, + "content": "where the last equality follows from Appendix B.2. Similarly for the bias term we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 294, + 457, + 308 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 313, + 497, + 474 + ], + "lines": [ + { + "bbox": [ + 114, + 313, + 497, + 474 + ], + "spans": [ + { + "bbox": [ + 114, + 313, + 497, + 474 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { B = \\| { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } \\beta - \\beta \\| _ { 2 } ^ { 2 } } \\\\ & { \\quad = \\beta ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) \\beta } \\\\ & { \\stackrel { ( \\psi ) } { = } \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma X ^ { \\top } ( V \\Sigma X ^ { \\top } X X ^ { \\top } U \\Sigma \\Sigma { \\cal { W } } ^ { \\top } ) ^ { - 1 } V \\Sigma \\Sigma ^ { \\top } { \\cal { U } } ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( \\cdots ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma ^ { \\top } \\Sigma \\sigma \\Sigma ^ { \\top } \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( \\Sigma \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\Sigma \\Big ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( ( \\Sigma ^ { \\top } X ^ { \\top } \\Sigma ) ^ { - 1 } \\end{array}", + "type": "interline_equation", + "image_path": "515891f2ef048c678fbfca2d4a5db9c09d60df581a642c374b48b8a5384a2777.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 114, + 313, + 497, + 366.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 114, + 366.6666666666667, + 497, + 420.33333333333337 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 114, + 420.33333333333337, + 497, + 474.00000000000006 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 478, + 504, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 288, + 491 + ], + "score": 1.0, + "content": "where symmetric arguments are omitted as", + "type": "text" + }, + { + "bbox": [ + 289, + 479, + 309, + 489 + ], + "score": 0.81, + "content": "( \\cdot \\cdot \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 477, + 505, + 491 + ], + "score": 1.0, + "content": ", and (i) follows from the rotational invariance", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 489, + 466, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 281, + 504 + ], + "score": 1.0, + "content": "argument introduced in Lemma 10 and that", + "type": "text" + }, + { + "bbox": [ + 282, + 489, + 325, + 502 + ], + "score": 0.91, + "content": "\\beta ^ { \\top } \\beta = r ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 489, + 466, + 504 + ], + "score": 1.0, + "content": ". By the block decomposition (30),", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 477, + 505, + 504 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 507, + 439, + 535 + ], + "lines": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "spans": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "score": 0.91, + "content": "\\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } = \\left[ \\begin{array} { l l } { 0 } & { ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } } \\\\ { 0 } & { - I _ { d - h } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "75b68ce49c070f2e797f82c3695a9dbb8c5ffcaddcd04477af0fd67b3b103535.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 171, + 507, + 439, + 535 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 539, + 253, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 253, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 253, + 552 + ], + "score": 1.0, + "content": "Therefore the bias term simplifies to", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 538, + 253, + 552 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 128, + 556, + 482, + 639 + ], + "lines": [ + { + "bbox": [ + 128, + 556, + 482, + 639 + ], + "spans": [ + { + "bbox": [ + 128, + 556, + 482, + 639 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) ^ { \\top } \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) \\right) } } \\\\ { { \\mathrm { } = \\frac { r ^ { 2 } } { d } \\left( \\mathrm { t r } \\left( ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } X _ { 1 } X _ { 0 } ^ { \\top } ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } \\right) + ( d - h ) \\right) } } \\\\ { { \\mathrm { } \\to \\frac { r ^ { 2 } } { d } \\left( \\frac { ( d - h ) h } { n - h - 1 } + d - h \\right) \\to \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } \\left( 1 - \\gamma _ { 2 } \\right) } r ^ { 2 } . } } \\end{array}", + "type": "interline_equation", + "image_path": "d5f2f2ce220c66d82124355105f5bd55076362bad3c56e92df6e32ab92705ff6.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 128, + 556, + 482, + 583.6666666666666 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 128, + 583.6666666666666, + 482, + 611.3333333333333 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 128, + 611.3333333333333, + 482, + 638.9999999999999 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 282, + 654 + ], + "lines": [ + { + "bbox": [ + 106, + 641, + 281, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 228, + 656 + ], + "score": 1.0, + "content": "Thus we have obtained that as", + "type": "text" + }, + { + "bbox": [ + 229, + 643, + 281, + 654 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 641, + 281, + 656 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 660, + 348, + 685 + ], + "lines": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "spans": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "score": 0.95, + "content": "B \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } ( 1 - \\gamma _ { 2 } ) } r ^ { 2 } .", + "type": "interline_equation", + "image_path": "b44679d17b6fe7ed1fadb754db3d0f4639acf3b59a5c7f8f6e646d0ca2791004.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 263, + 660, + 348, + 685 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 696, + 495, + 710 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 496, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 146, + 712 + ], + "score": 1.0, + "content": "Case III:", + "type": "text" + }, + { + "bbox": [ + 147, + 696, + 174, + 708 + ], + "score": 0.88, + "content": "W ^ { \\top } X", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 694, + 276, + 712 + ], + "score": 1.0, + "content": "has full row rank, i.e.", + "type": "text" + }, + { + "bbox": [ + 277, + 698, + 338, + 709 + ], + "score": 0.92, + "content": "\\gamma _ { 1 } > 1 , \\gamma _ { 2 } > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 694, + 496, + 712 + ], + "score": 1.0, + "content": ". Similarly, the least squares solution is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 694, + 496, + 712 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 714, + 375, + 730 + ], + "lines": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "spans": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "score": 0.91, + "content": "\\hat { \\beta } = W W ^ { \\top } X ( X ^ { \\top } W W ^ { \\top } X ) ^ { - 1 } \\pmb { y } ,", + "type": "interline_equation", + "image_path": "06daaa58c150186f04888890add9047d6e277956a9d567db4bf4a049dbeaf530.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 235, + 714, + 375, + 730 + ], + "spans": [], + "index": 25 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 209, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 210, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 210, + 95 + ], + "score": 1.0, + "content": "Simplifying the variance:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 101, + 455, + 168 + ], + "lines": [ + { + "bbox": [ + 156, + 101, + 455, + 168 + ], + "spans": [ + { + "bbox": [ + 156, + 101, + 455, + 168 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W W ^ { \\top } X \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } \\sigma ^ { 2 } \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } X ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } U \\Sigma V ^ { \\top } \\left( V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } U \\Sigma V ^ { \\top } \\right) ^ { - 2 } V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma \\right) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "fb55650dbf8ef0fa3382b662b0fa85e7de89f0fd1ffb2fb001ed04409ef08846.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 156, + 101, + 455, + 123.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 123.33333333333333, + 455, + 145.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 156, + 145.66666666666666, + 455, + 168.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 506, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 227, + 189 + ], + "score": 1.0, + "content": "where we applied the SVD of", + "type": "text" + }, + { + "bbox": [ + 228, + 175, + 280, + 186 + ], + "score": 0.91, + "content": "X = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "and the rotational invariance argument. Using a similar", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 221, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 205, + 199 + ], + "score": 1.0, + "content": "block decomposition on", + "type": "text" + }, + { + "bbox": [ + 205, + 187, + 217, + 197 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 186, + 221, + 199 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 205, + 358, + 234 + ], + "lines": [ + { + "bbox": [ + 252, + 205, + 358, + 234 + ], + "spans": [ + { + "bbox": [ + 252, + 205, + 358, + 234 + ], + "score": 0.93, + "content": "\\Sigma = \\left[ \\stackrel { \\Sigma _ { 0 } } { 0 } \\right] , W = \\left[ \\stackrel { W _ { 0 } } { W _ { 1 } } \\right] ,", + "type": "interline_equation", + "image_path": "23adeaa39b60c9788b5a183b8ad08c7b80dab595e2cb883c50a3b0aef6a65dd6.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 205, + 358, + 219.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 252, + 219.5, + 358, + 234.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 241, + 505, + 256 + ], + "lines": [ + { + "bbox": [ + 104, + 239, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 239, + 132, + 258 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 241, + 306, + 255 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { n \\times n } , W _ { 0 } \\in \\mathbb { R } ^ { n \\times h } , W _ { 1 } \\in \\mathbb { R } ^ { ( d - n ) \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 239, + 326, + 258 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 326, + 243, + 360, + 255 + ], + "score": 0.9, + "content": "W _ { 0 } , W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 239, + 484, + 258 + ], + "score": 1.0, + "content": "independent. We thus simplify", + "type": "text" + }, + { + "bbox": [ + 484, + 244, + 493, + 253 + ], + "score": 0.8, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 239, + 506, + 258 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 261, + 514, + 352 + ], + "lines": [ + { + "bbox": [ + 113, + 261, + 514, + 352 + ], + "spans": [ + { + "bbox": [ + 113, + 261, + 514, + 352 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { V = \\sigma ^ { 2 } \\mathrm { t r } ( W W ^ { \\top } \\Sigma ( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma ) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } ) } \\\\ & \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( [ \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { \\cdots } & { \\cdots } \\\\ { ( \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { W _ { 1 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 1 } ^ { \\top } } \\end{array} ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( \\mathrm { t r } ( \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ) + \\mathrm { t r } ( W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } W _ { 1 } ^ { \\top } ) ) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( ( X ^ { \\top } X ) ^ { - 1 } ) + \\sigma ^ { 2 } \\mathrm { t r } ( W _ { 1 } ^ { \\top } W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } ) \\cdot \\ ( 3 9 ) } \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "af984b7aada2fb32194ea1fbcbe8852714b6279464135d08e84fcc1015befb4a.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 113, + 261, + 514, + 291.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 113, + 291.3333333333333, + 514, + 321.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 113, + 321.66666666666663, + 514, + 351.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 357, + 340, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 340, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 340, + 370 + ], + "score": 1.0, + "content": "Hence we obtain the following expression on the variance", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 376, + 499, + 404 + ], + "lines": [ + { + "bbox": [ + 110, + 376, + 499, + 404 + ], + "spans": [ + { + "bbox": [ + 110, + 376, + 499, + 404 + ], + "score": 0.91, + "content": "V \\sigma ^ { 2 } \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\sigma ^ { 2 } ( d - n ) \\mathbb { E } _ { W , X } V \\mathrm { t r } ( ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ) \\sigma ^ { 2 } ( \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\frac { 1 } { \\gamma _ { 2 } - 1 } ) .", + "type": "interline_equation", + "image_path": "874afd9111c23f1cff8c896a34da4f7616adad78aecef68652ac8bc445969cd3.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 110, + 376, + 499, + 385.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 385.3333333333333, + 499, + 394.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 394.66666666666663, + 499, + 403.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 423, + 372, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 372, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 372, + 436 + ], + "score": 1.0, + "content": "We omit the derivation of bias, which follows a similar derivation:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 442, + 348, + 469 + ], + "lines": [ + { + "bbox": [ + 263, + 442, + 348, + 469 + ], + "spans": [ + { + "bbox": [ + 263, + 442, + 348, + 469 + ], + "score": 0.95, + "content": "B \\frac { \\gamma _ { 2 } ( \\gamma _ { 1 } - 1 ) } { \\gamma _ { 1 } ( \\gamma _ { 2 } - 1 ) } r ^ { 2 } .", + "type": "interline_equation", + "image_path": "2288a3ebd351e85f1503cb89a8d9d6caf04b57eb29d22f1ff4b5a50629a20461.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 442, + 348, + 455.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 263, + 455.5, + 348, + 469.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 278, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 280, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 280, + 496 + ], + "score": 1.0, + "content": "Combining Case I, II, III yields theorem 2.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "title", + "bbox": [ + 107, + 515, + 245, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 515, + 246, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 246, + 528 + ], + "score": 1.0, + "content": "C.3 PROOF OF PROPOSITION 3", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 105, + 537, + 499, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 500, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 328, + 550 + ], + "score": 1.0, + "content": "Given the squared loss, one can derive the dynamics of", + "type": "text" + }, + { + "bbox": [ + 328, + 538, + 340, + 548 + ], + "score": 0.77, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 536, + 436, + 550 + ], + "score": 1.0, + "content": "with fixed second layer", + "type": "text" + }, + { + "bbox": [ + 437, + 540, + 444, + 548 + ], + "score": 0.73, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 536, + 500, + 550 + ], + "score": 1.0, + "content": "w.r.t the loss:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 556, + 384, + 582 + ], + "lines": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "spans": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "score": 0.94, + "content": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - X ^ { \\top } W ( t ) \\pmb { a } ) \\pmb { a } ^ { \\top } .", + "type": "interline_equation", + "image_path": "0ca5d134c0e3bb46bde6687e000ff933f332606ecfe415aa7dd013398706531c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 203, + 609 + ], + "score": 1.0, + "content": "Note that the update of", + "type": "text" + }, + { + "bbox": [ + 204, + 596, + 216, + 605 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 594, + 387, + 609 + ], + "score": 1.0, + "content": "can be written as a linear combination of", + "type": "text" + }, + { + "bbox": [ + 388, + 598, + 395, + 605 + ], + "score": 0.76, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 594, + 425, + 609 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 426, + 595, + 470, + 608 + ], + "score": 0.93, + "content": "W ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 594, + 505, + 609 + ], + "score": 1.0, + "content": ", we can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 606, + 383, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 129, + 619 + ], + "score": 1.0, + "content": "write", + "type": "text" + }, + { + "bbox": [ + 129, + 606, + 198, + 619 + ], + "score": 0.95, + "content": "W ( t ) = \\bar { \\hat { w } } ( t ) \\mathbf { { a } } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 606, + 237, + 619 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 238, + 607, + 247, + 617 + ], + "score": 0.83, + "content": "\\hat { \\textbf { \\textit { w } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 606, + 362, + 619 + ], + "score": 1.0, + "content": ". The corresponding flow on", + "type": "text" + }, + { + "bbox": [ + 362, + 607, + 371, + 617 + ], + "score": 0.84, + "content": "\\hat { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 606, + 383, + 619 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 626, + 383, + 651 + ], + "lines": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "spans": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "score": 0.94, + "content": "\\frac { \\partial \\pmb { \\hat { w } } ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - \\boldsymbol { X } ^ { \\top } \\pmb { \\hat { w } } ( t ) \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } ) ,", + "type": "interline_equation", + "image_path": "b992fc90e218018b684b44a84a932474a897f282e9308707633ba3c4fb4c0fab.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 246, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 246, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 246, + 670 + ], + "score": 1.0, + "content": "which gives the following solution", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 677, + 389, + 705 + ], + "lines": [ + { + "bbox": [ + 222, + 677, + 389, + 705 + ], + "spans": [ + { + "bbox": [ + 222, + 677, + 389, + 705 + ], + "score": 0.94, + "content": "{ \\hat { \\pmb { w } } } ^ { * } = { \\frac { 1 } { \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } } } { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } \\Rightarrow { \\hat { \\beta } } = { \\boldsymbol { W } } ^ { * } \\pmb { a } = { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } .", + "type": "interline_equation", + "image_path": "81713cb939b19c512a2893faf164a0bb87be96797d7b1b1e201e30a5a003185b.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 677, + 389, + 691.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 222, + 691.0, + 389, + 705.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 712, + 486, + 725 + ], + "lines": [ + { + "bbox": [ + 105, + 711, + 486, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 486, + 726 + ], + "score": 1.0, + "content": "Thus gradient flow on the first layer leads to the minimum-norm solution on the input features.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 482, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 495, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 495, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 713, + 504, + 723 + ], + "lines": [ + { + "bbox": [ + 496, + 714, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 496, + 714, + 505, + 724 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 209, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 210, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 210, + 95 + ], + "score": 1.0, + "content": "Simplifying the variance:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 82, + 210, + 95 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 101, + 455, + 168 + ], + "lines": [ + { + "bbox": [ + 156, + 101, + 455, + 168 + ], + "spans": [ + { + "bbox": [ + 156, + 101, + 455, + 168 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W W ^ { \\top } X \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } \\sigma ^ { 2 } \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } X ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } U \\Sigma V ^ { \\top } \\left( V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } U \\Sigma V ^ { \\top } \\right) ^ { - 2 } V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma \\right) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "fb55650dbf8ef0fa3382b662b0fa85e7de89f0fd1ffb2fb001ed04409ef08846.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 156, + 101, + 455, + 123.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 123.33333333333333, + 455, + 145.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 156, + 145.66666666666666, + 455, + 168.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 506, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 227, + 189 + ], + "score": 1.0, + "content": "where we applied the SVD of", + "type": "text" + }, + { + "bbox": [ + 228, + 175, + 280, + 186 + ], + "score": 0.91, + "content": "X = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "and the rotational invariance argument. Using a similar", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 221, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 205, + 199 + ], + "score": 1.0, + "content": "block decomposition on", + "type": "text" + }, + { + "bbox": [ + 205, + 187, + 217, + 197 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 186, + 221, + 199 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 174, + 506, + 199 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 205, + 358, + 234 + ], + "lines": [ + { + "bbox": [ + 252, + 205, + 358, + 234 + ], + "spans": [ + { + "bbox": [ + 252, + 205, + 358, + 234 + ], + "score": 0.93, + "content": "\\Sigma = \\left[ \\stackrel { \\Sigma _ { 0 } } { 0 } \\right] , W = \\left[ \\stackrel { W _ { 0 } } { W _ { 1 } } \\right] ,", + "type": "interline_equation", + "image_path": "23adeaa39b60c9788b5a183b8ad08c7b80dab595e2cb883c50a3b0aef6a65dd6.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 205, + 358, + 219.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 252, + 219.5, + 358, + 234.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 241, + 505, + 256 + ], + "lines": [ + { + "bbox": [ + 104, + 239, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 239, + 132, + 258 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 241, + 306, + 255 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { n \\times n } , W _ { 0 } \\in \\mathbb { R } ^ { n \\times h } , W _ { 1 } \\in \\mathbb { R } ^ { ( d - n ) \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 239, + 326, + 258 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 326, + 243, + 360, + 255 + ], + "score": 0.9, + "content": "W _ { 0 } , W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 239, + 484, + 258 + ], + "score": 1.0, + "content": "independent. We thus simplify", + "type": "text" + }, + { + "bbox": [ + 484, + 244, + 493, + 253 + ], + "score": 0.8, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 239, + 506, + 258 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 104, + 239, + 506, + 258 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 261, + 514, + 352 + ], + "lines": [ + { + "bbox": [ + 113, + 261, + 514, + 352 + ], + "spans": [ + { + "bbox": [ + 113, + 261, + 514, + 352 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { V = \\sigma ^ { 2 } \\mathrm { t r } ( W W ^ { \\top } \\Sigma ( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma ) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } ) } \\\\ & \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( [ \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { \\cdots } & { \\cdots } \\\\ { ( \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { W _ { 1 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 1 } ^ { \\top } } \\end{array} ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( \\mathrm { t r } ( \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ) + \\mathrm { t r } ( W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } W _ { 1 } ^ { \\top } ) ) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( ( X ^ { \\top } X ) ^ { - 1 } ) + \\sigma ^ { 2 } \\mathrm { t r } ( W _ { 1 } ^ { \\top } W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } ) \\cdot \\ ( 3 9 ) } \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "af984b7aada2fb32194ea1fbcbe8852714b6279464135d08e84fcc1015befb4a.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 113, + 261, + 514, + 291.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 113, + 291.3333333333333, + 514, + 321.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 113, + 321.66666666666663, + 514, + 351.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 357, + 340, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 340, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 340, + 370 + ], + "score": 1.0, + "content": "Hence we obtain the following expression on the variance", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 357, + 340, + 370 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 376, + 499, + 404 + ], + "lines": [ + { + "bbox": [ + 110, + 376, + 499, + 404 + ], + "spans": [ + { + "bbox": [ + 110, + 376, + 499, + 404 + ], + "score": 0.91, + "content": "V \\sigma ^ { 2 } \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\sigma ^ { 2 } ( d - n ) \\mathbb { E } _ { W , X } V \\mathrm { t r } ( ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ) \\sigma ^ { 2 } ( \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\frac { 1 } { \\gamma _ { 2 } - 1 } ) .", + "type": "interline_equation", + "image_path": "874afd9111c23f1cff8c896a34da4f7616adad78aecef68652ac8bc445969cd3.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 110, + 376, + 499, + 385.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 385.3333333333333, + 499, + 394.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 394.66666666666663, + 499, + 403.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 423, + 372, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 372, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 372, + 436 + ], + "score": 1.0, + "content": "We omit the derivation of bias, which follows a similar derivation:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 423, + 372, + 436 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 442, + 348, + 469 + ], + "lines": [ + { + "bbox": [ + 263, + 442, + 348, + 469 + ], + "spans": [ + { + "bbox": [ + 263, + 442, + 348, + 469 + ], + "score": 0.95, + "content": "B \\frac { \\gamma _ { 2 } ( \\gamma _ { 1 } - 1 ) } { \\gamma _ { 1 } ( \\gamma _ { 2 } - 1 ) } r ^ { 2 } .", + "type": "interline_equation", + "image_path": "2288a3ebd351e85f1503cb89a8d9d6caf04b57eb29d22f1ff4b5a50629a20461.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 442, + 348, + 455.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 263, + 455.5, + 348, + 469.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 278, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 280, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 280, + 496 + ], + "score": 1.0, + "content": "Combining Case I, II, III yields theorem 2.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 480, + 280, + 496 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 515, + 245, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 515, + 246, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 246, + 528 + ], + "score": 1.0, + "content": "C.3 PROOF OF PROPOSITION 3", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 105, + 537, + 499, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 500, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 328, + 550 + ], + "score": 1.0, + "content": "Given the squared loss, one can derive the dynamics of", + "type": "text" + }, + { + "bbox": [ + 328, + 538, + 340, + 548 + ], + "score": 0.77, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 536, + 436, + 550 + ], + "score": 1.0, + "content": "with fixed second layer", + "type": "text" + }, + { + "bbox": [ + 437, + 540, + 444, + 548 + ], + "score": 0.73, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 536, + 500, + 550 + ], + "score": 1.0, + "content": "w.r.t the loss:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 536, + 500, + 550 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 556, + 384, + 582 + ], + "lines": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "spans": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "score": 0.94, + "content": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - X ^ { \\top } W ( t ) \\pmb { a } ) \\pmb { a } ^ { \\top } .", + "type": "interline_equation", + "image_path": "0ca5d134c0e3bb46bde6687e000ff933f332606ecfe415aa7dd013398706531c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 227, + 556, + 384, + 582 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 203, + 609 + ], + "score": 1.0, + "content": "Note that the update of", + "type": "text" + }, + { + "bbox": [ + 204, + 596, + 216, + 605 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 594, + 387, + 609 + ], + "score": 1.0, + "content": "can be written as a linear combination of", + "type": "text" + }, + { + "bbox": [ + 388, + 598, + 395, + 605 + ], + "score": 0.76, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 594, + 425, + 609 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 426, + 595, + 470, + 608 + ], + "score": 0.93, + "content": "W ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 594, + 505, + 609 + ], + "score": 1.0, + "content": ", we can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 606, + 383, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 129, + 619 + ], + "score": 1.0, + "content": "write", + "type": "text" + }, + { + "bbox": [ + 129, + 606, + 198, + 619 + ], + "score": 0.95, + "content": "W ( t ) = \\bar { \\hat { w } } ( t ) \\mathbf { { a } } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 606, + 237, + 619 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 238, + 607, + 247, + 617 + ], + "score": 0.83, + "content": "\\hat { \\textbf { \\textit { w } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 606, + 362, + 619 + ], + "score": 1.0, + "content": ". The corresponding flow on", + "type": "text" + }, + { + "bbox": [ + 362, + 607, + 371, + 617 + ], + "score": 0.84, + "content": "\\hat { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 606, + 383, + 619 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 594, + 505, + 619 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 626, + 383, + 651 + ], + "lines": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "spans": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "score": 0.94, + "content": "\\frac { \\partial \\pmb { \\hat { w } } ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - \\boldsymbol { X } ^ { \\top } \\pmb { \\hat { w } } ( t ) \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } ) ,", + "type": "interline_equation", + "image_path": "b992fc90e218018b684b44a84a932474a897f282e9308707633ba3c4fb4c0fab.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 227, + 626, + 383, + 651 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 246, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 246, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 246, + 670 + ], + "score": 1.0, + "content": "which gives the following solution", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 658, + 246, + 670 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 677, + 389, + 705 + ], + "lines": [ + { + "bbox": [ + 222, + 677, + 389, + 705 + ], + "spans": [ + { + "bbox": [ + 222, + 677, + 389, + 705 + ], + "score": 0.94, + "content": "{ \\hat { \\pmb { w } } } ^ { * } = { \\frac { 1 } { \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } } } { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } \\Rightarrow { \\hat { \\beta } } = { \\boldsymbol { W } } ^ { * } \\pmb { a } = { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } .", + "type": "interline_equation", + "image_path": "81713cb939b19c512a2893faf164a0bb87be96797d7b1b1e201e30a5a003185b.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 677, + 389, + 691.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 222, + 691.0, + 389, + 705.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 712, + 486, + 725 + ], + "lines": [ + { + "bbox": [ + 105, + 711, + 486, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 486, + 726 + ], + "score": 1.0, + "content": "Thus gradient flow on the first layer leads to the minimum-norm solution on the input features.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 711, + 486, + 726 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 231, + 94 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 231, + 94 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 231, + 94 + ], + "score": 1.0, + "content": "C.4 PROOF OF THEOREM 4", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 102, + 492, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 492, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 442, + 117 + ], + "score": 1.0, + "content": "Following the bias-variance decomposition (7), the variance term can be written as (", + "type": "text" + }, + { + "bbox": [ + 442, + 102, + 453, + 113 + ], + "score": 0.84, + "content": "\\cdot \\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 101, + 492, + 117 + ], + "score": 1.0, + "content": "omitted)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 116, + 432, + 207 + ], + "lines": [ + { + "bbox": [ + 177, + 116, + 432, + 207 + ], + "spans": [ + { + "bbox": [ + 177, + 116, + 432, + 207 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { V = \\mathbb { E } _ { \\pmb { x } , \\pmb { \\varepsilon } } \\Big [ \\| \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) - \\mathbb { E } _ { \\pmb { \\varepsilon } } \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\Big \\| \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\phi ( W ^ { \\top } \\pmb { x } ) \\Big \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] \\right) } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } K _ { W } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "2cbbd84b03cbf4922e3d6e17f5f9ef4c7ad64a61dccd27f8a5bbfea7a801866a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 177, + 116, + 432, + 146.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 177, + 146.33333333333334, + 432, + 176.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 177, + 176.66666666666669, + 432, + 207.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 390, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 390, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 324, + 221 + ], + "score": 1.0, + "content": "where we define the expected non-linear Gram matrix", + "type": "text" + }, + { + "bbox": [ + 324, + 208, + 378, + 220 + ], + "score": 0.92, + "content": "K _ { W } \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 207, + 390, + 221 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 222, + 375, + 243 + ], + "lines": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "spans": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "score": 0.88, + "content": "K _ { W } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] .", + "type": "interline_equation", + "image_path": "4150c3e50a9017d8f65e32db158b3b719961d1c95653b8261f1a875cfcf0d895.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 366, + 265 + ], + "lines": [ + { + "bbox": [ + 102, + 242, + 363, + 268 + ], + "spans": [ + { + "bbox": [ + 102, + 242, + 217, + 268 + ], + "score": 1.0, + "content": "and for each entry we have", + "type": "text" + }, + { + "bbox": [ + 217, + 245, + 363, + 266 + ], + "score": 0.84, + "content": "( K _ { W } ) _ { [ i , j ] } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { j } ^ { \\top } \\pmb { x } ) \\Big ] .", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 269, + 506, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 267, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 282 + ], + "score": 1.0, + "content": "Random matrix in the form of covariance matrix of nonlinear features has been studied in many", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "works Hastie et al. (2019); Mei and Montanari (2019); Liao and Couillet (2018); Louart et al. (2018);", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 291, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 506, + 303 + ], + "score": 1.0, + "content": "Pennington and Worah (2017). We note that our setup for the variance term is very similar to that for", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 302, + 420, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 420, + 315 + ], + "score": 1.0, + "content": "nonlinear features in Hastie et al. (2019) with modifications mentioned below.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 105, + 319, + 504, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "In contrast to the linear network in Section C.2, the Gram matrix of a nonlinear activation is almost", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 329, + 465, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 465, + 342 + ], + "score": 1.0, + "content": "surely full-rank, as specified in the following lemma from Pennington and Worah (2017):", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 344, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 104, + 342, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 342, + 196, + 357 + ], + "score": 1.0, + "content": "Lemma 11. Suppose √", + "type": "text" + }, + { + "bbox": [ + 196, + 344, + 203, + 355 + ], + "score": 0.8, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 342, + 405, + 357 + ], + "score": 1.0, + "content": "is not linear. Then the smallest singular value of", + "type": "text" + }, + { + "bbox": [ + 405, + 343, + 504, + 356 + ], + "score": 0.91, + "content": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 155, + 369 + ], + "score": 1.0, + "content": "is of order", + "type": "text" + }, + { + "bbox": [ + 155, + 355, + 185, + 367 + ], + "score": 0.92, + "content": "O ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 353, + 423, + 369 + ], + "score": 1.0, + "content": ". To be precise, consider the empirical spectral density", + "type": "text" + }, + { + "bbox": [ + 423, + 356, + 475, + 368 + ], + "score": 0.91, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 353, + 506, + 369 + ], + "score": 1.0, + "content": ", when", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 366, + 425, + 380 + ], + "spans": [ + { + "bbox": [ + 107, + 367, + 159, + 379 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 366, + 180, + 380 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 180, + 367, + 222, + 379 + ], + "score": 0.92, + "content": "d / n \\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 366, + 241, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 241, + 367, + 284, + 379 + ], + "score": 0.7, + "content": "h / n \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 366, + 288, + 380 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 288, + 367, + 340, + 380 + ], + "score": 0.87, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 366, + 425, + 380 + ], + "score": 1.0, + "content": "converges weakly to", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 382, + 401, + 398 + ], + "lines": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "spans": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "score": 0.86, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda ) [ 1 - \\gamma _ { 2 } ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\mu ^ { + } ( d \\lambda ) ,", + "type": "interline_equation", + "image_path": "30105f33e88d205694fab25035f259f3b7cc7cada4bf9e48291fa6222d7b70c1.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 348, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 347, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 133, + 414 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 400, + 165, + 412 + ], + "score": 0.92, + "content": "\\mu ^ { + } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 398, + 270, + 414 + ], + "score": 1.0, + "content": "has non-negative support", + "type": "text" + }, + { + "bbox": [ + 271, + 400, + 297, + 412 + ], + "score": 0.87, + "content": "\\lbrack \\rho , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 398, + 318, + 414 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 318, + 401, + 343, + 412 + ], + "score": 0.88, + "content": "\\rho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 398, + 347, + 414 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 105, + 419, + 483, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 484, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 256, + 433 + ], + "score": 1.0, + "content": "We therefore consider two scenarios:", + "type": "text" + }, + { + "bbox": [ + 257, + 420, + 265, + 430 + ], + "score": 0.85, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 418, + 484, + 433 + ], + "score": 1.0, + "content": "is full column rank (Case I) or full row rank (Case II).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 442, + 307, + 454 + ], + "lines": [ + { + "bbox": [ + 105, + 440, + 307, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 140, + 456 + ], + "score": 1.0, + "content": "Case 1.", + "type": "text" + }, + { + "bbox": [ + 140, + 443, + 166, + 453 + ], + "score": 0.87, + "content": "h < n", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 440, + 307, + 456 + ], + "score": 1.0, + "content": ". In this case (45) simplifies into", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 126, + 456, + 483, + 483 + ], + "lines": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "spans": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "score": 0.91, + "content": "V = \\operatorname { t r } \\left( \\left( \\phi ( W ^ { \\top } X ) \\phi ( X ^ { \\top } W ) \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } \\operatorname { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } V _ { \\xi } ,", + "type": "interline_equation", + "image_path": "3668294f9bd1d547e0fbc6005c68eed1228bfc1f53d35acd53c1b9fcb29b594f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 486, + 506, + 522 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 258, + 498 + ], + "score": 1.0, + "content": "where the continuity and boundness of", + "type": "text" + }, + { + "bbox": [ + 258, + 486, + 269, + 498 + ], + "score": 0.89, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 484, + 279, + 498 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 279, + 486, + 310, + 497 + ], + "score": 0.92, + "content": "\\xi = 0 ^ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 484, + 444, + 498 + ], + "score": 1.0, + "content": "is guaranteed by Lemma 11 when", + "type": "text" + }, + { + "bbox": [ + 444, + 486, + 503, + 498 + ], + "score": 0.93, + "content": "\\gamma _ { 2 } = h / n \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 504, + 484, + 507, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 140, + 509 + ], + "score": 1.0, + "content": "A ridge", + "type": "text" + }, + { + "bbox": [ + 141, + 497, + 147, + 508 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 495, + 506, + 509 + ], + "score": 1.0, + "content": "is added make use of (Louart et al., 2018, Theorem 1), which derived the asymptotic", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 507, + 410, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 215, + 524 + ], + "score": 1.0, + "content": "equivalent of the resolvent", + "type": "text" + }, + { + "bbox": [ + 216, + 507, + 279, + 524 + ], + "score": 0.94, + "content": "\\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 508, + 352, + 524 + ], + "score": 1.0, + "content": ". It follows that as", + "type": "text" + }, + { + "bbox": [ + 352, + 510, + 405, + 522 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 508, + 410, + 524 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 525, + 465, + 675 + ], + "lines": [ + { + "bbox": [ + 141, + 525, + 465, + 675 + ], + "spans": [ + { + "bbox": [ + 141, + 525, + 465, + 675 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\mathrm { t r } \\left( h ^ { - 1 } \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) - h ^ { - 1 } I \\right) } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } \\left\\| \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } - \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) ^ { - 1 } \\right\\| _ { F } } \\\\ & { \\quad \\cdot \\left\\| \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right\\| _ { 2 } } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } O ( n ^ { - 1 / 2 + \\varepsilon } ) O ( n ^ { 1 / 2 } ) \\to 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "526904641963f8ba69ef39d63c4d674173d9230b06c35a9dd75ff16689448313.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 141, + 525, + 465, + 575.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 141, + 575.0, + 465, + 625.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 141, + 625.0, + 465, + 675.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 243, + 694 + ], + "score": 1.0, + "content": "where we have used the inequality", + "type": "text" + }, + { + "bbox": [ + 244, + 679, + 339, + 692 + ], + "score": 0.92, + "content": "\\operatorname { t r } \\left( A B \\right) \\leq \\left\\| A \\right\\| _ { F } \\left\\| B \\right\\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 677, + 449, + 694 + ], + "score": 1.0, + "content": ", and equivalently by taking", + "type": "text" + }, + { + "bbox": [ + 449, + 680, + 475, + 691 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 677, + 506, + 694 + ], + "score": 1.0, + "content": "we get", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 694, + 409, + 736 + ], + "lines": [ + { + "bbox": [ + 201, + 694, + 409, + 736 + ], + "spans": [ + { + "bbox": [ + 201, + 694, + 409, + 736 + ], + "score": 0.94, + "content": "\\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } \\frac { n } { h } \\frac { \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } { 1 + \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } = 1 .", + "type": "interline_equation", + "image_path": "47918eab7f209ed0d7ec24e3972688136f5931475e7d0360095f3688a840a5b9.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 694, + 409, + 715.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 201, + 715.0, + 409, + 736.0 + ], + "spans": [], + "index": 30 + } + ] + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 231, + 94 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 231, + 94 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 231, + 94 + ], + "score": 1.0, + "content": "C.4 PROOF OF THEOREM 4", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 102, + 492, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 492, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 442, + 117 + ], + "score": 1.0, + "content": "Following the bias-variance decomposition (7), the variance term can be written as (", + "type": "text" + }, + { + "bbox": [ + 442, + 102, + 453, + 113 + ], + "score": 0.84, + "content": "\\cdot \\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 101, + 492, + 117 + ], + "score": 1.0, + "content": "omitted)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 101, + 492, + 117 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 116, + 432, + 207 + ], + "lines": [ + { + "bbox": [ + 177, + 116, + 432, + 207 + ], + "spans": [ + { + "bbox": [ + 177, + 116, + 432, + 207 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { V = \\mathbb { E } _ { \\pmb { x } , \\pmb { \\varepsilon } } \\Big [ \\| \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) - \\mathbb { E } _ { \\pmb { \\varepsilon } } \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\Big \\| \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\phi ( W ^ { \\top } \\pmb { x } ) \\Big \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] \\right) } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } K _ { W } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "2cbbd84b03cbf4922e3d6e17f5f9ef4c7ad64a61dccd27f8a5bbfea7a801866a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 177, + 116, + 432, + 146.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 177, + 146.33333333333334, + 432, + 176.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 177, + 176.66666666666669, + 432, + 207.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 390, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 390, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 324, + 221 + ], + "score": 1.0, + "content": "where we define the expected non-linear Gram matrix", + "type": "text" + }, + { + "bbox": [ + 324, + 208, + 378, + 220 + ], + "score": 0.92, + "content": "K _ { W } \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 207, + 390, + 221 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 207, + 390, + 221 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 222, + 375, + 243 + ], + "lines": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "spans": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "score": 0.88, + "content": "K _ { W } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] .", + "type": "interline_equation", + "image_path": "4150c3e50a9017d8f65e32db158b3b719961d1c95653b8261f1a875cfcf0d895.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 235, + 222, + 375, + 243 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 366, + 265 + ], + "lines": [ + { + "bbox": [ + 102, + 242, + 363, + 268 + ], + "spans": [ + { + "bbox": [ + 102, + 242, + 217, + 268 + ], + "score": 1.0, + "content": "and for each entry we have", + "type": "text" + }, + { + "bbox": [ + 217, + 245, + 363, + 266 + ], + "score": 0.84, + "content": "( K _ { W } ) _ { [ i , j ] } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { j } ^ { \\top } \\pmb { x } ) \\Big ] .", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 102, + 242, + 363, + 268 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 269, + 506, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 267, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 282 + ], + "score": 1.0, + "content": "Random matrix in the form of covariance matrix of nonlinear features has been studied in many", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "works Hastie et al. (2019); Mei and Montanari (2019); Liao and Couillet (2018); Louart et al. (2018);", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 291, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 506, + 303 + ], + "score": 1.0, + "content": "Pennington and Worah (2017). We note that our setup for the variance term is very similar to that for", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 302, + 420, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 420, + 315 + ], + "score": 1.0, + "content": "nonlinear features in Hastie et al. (2019) with modifications mentioned below.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 267, + 506, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 319, + 504, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "In contrast to the linear network in Section C.2, the Gram matrix of a nonlinear activation is almost", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 329, + 465, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 465, + 342 + ], + "score": 1.0, + "content": "surely full-rank, as specified in the following lemma from Pennington and Worah (2017):", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 318, + 506, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 344, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 104, + 342, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 342, + 196, + 357 + ], + "score": 1.0, + "content": "Lemma 11. Suppose √", + "type": "text" + }, + { + "bbox": [ + 196, + 344, + 203, + 355 + ], + "score": 0.8, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 342, + 405, + 357 + ], + "score": 1.0, + "content": "is not linear. Then the smallest singular value of", + "type": "text" + }, + { + "bbox": [ + 405, + 343, + 504, + 356 + ], + "score": 0.91, + "content": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 155, + 369 + ], + "score": 1.0, + "content": "is of order", + "type": "text" + }, + { + "bbox": [ + 155, + 355, + 185, + 367 + ], + "score": 0.92, + "content": "O ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 353, + 423, + 369 + ], + "score": 1.0, + "content": ". To be precise, consider the empirical spectral density", + "type": "text" + }, + { + "bbox": [ + 423, + 356, + 475, + 368 + ], + "score": 0.91, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 353, + 506, + 369 + ], + "score": 1.0, + "content": ", when", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 366, + 425, + 380 + ], + "spans": [ + { + "bbox": [ + 107, + 367, + 159, + 379 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 366, + 180, + 380 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 180, + 367, + 222, + 379 + ], + "score": 0.92, + "content": "d / n \\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 366, + 241, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 241, + 367, + 284, + 379 + ], + "score": 0.7, + "content": "h / n \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 366, + 288, + 380 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 288, + 367, + 340, + 380 + ], + "score": 0.87, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 366, + 425, + 380 + ], + "score": 1.0, + "content": "converges weakly to", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 104, + 342, + 506, + 380 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 382, + 401, + 398 + ], + "lines": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "spans": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "score": 0.86, + "content": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda ) [ 1 - \\gamma _ { 2 } ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\mu ^ { + } ( d \\lambda ) ,", + "type": "interline_equation", + "image_path": "30105f33e88d205694fab25035f259f3b7cc7cada4bf9e48291fa6222d7b70c1.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 208, + 382, + 401, + 398 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 348, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 347, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 133, + 414 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 400, + 165, + 412 + ], + "score": 0.92, + "content": "\\mu ^ { + } ( d \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 398, + 270, + 414 + ], + "score": 1.0, + "content": "has non-negative support", + "type": "text" + }, + { + "bbox": [ + 271, + 400, + 297, + 412 + ], + "score": 0.87, + "content": "\\lbrack \\rho , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 398, + 318, + 414 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 318, + 401, + 343, + 412 + ], + "score": 0.88, + "content": "\\rho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 398, + 347, + 414 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 398, + 347, + 414 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 419, + 483, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 484, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 256, + 433 + ], + "score": 1.0, + "content": "We therefore consider two scenarios:", + "type": "text" + }, + { + "bbox": [ + 257, + 420, + 265, + 430 + ], + "score": 0.85, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 418, + 484, + 433 + ], + "score": 1.0, + "content": "is full column rank (Case I) or full row rank (Case II).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 418, + 484, + 433 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 442, + 307, + 454 + ], + "lines": [ + { + "bbox": [ + 105, + 440, + 307, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 140, + 456 + ], + "score": 1.0, + "content": "Case 1.", + "type": "text" + }, + { + "bbox": [ + 140, + 443, + 166, + 453 + ], + "score": 0.87, + "content": "h < n", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 440, + 307, + 456 + ], + "score": 1.0, + "content": ". In this case (45) simplifies into", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 440, + 307, + 456 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 126, + 456, + 483, + 483 + ], + "lines": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "spans": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "score": 0.91, + "content": "V = \\operatorname { t r } \\left( \\left( \\phi ( W ^ { \\top } X ) \\phi ( X ^ { \\top } W ) \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } \\operatorname { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } V _ { \\xi } ,", + "type": "interline_equation", + "image_path": "3668294f9bd1d547e0fbc6005c68eed1228bfc1f53d35acd53c1b9fcb29b594f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 126, + 456, + 483, + 483 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 486, + 506, + 522 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 258, + 498 + ], + "score": 1.0, + "content": "where the continuity and boundness of", + "type": "text" + }, + { + "bbox": [ + 258, + 486, + 269, + 498 + ], + "score": 0.89, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 484, + 279, + 498 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 279, + 486, + 310, + 497 + ], + "score": 0.92, + "content": "\\xi = 0 ^ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 484, + 444, + 498 + ], + "score": 1.0, + "content": "is guaranteed by Lemma 11 when", + "type": "text" + }, + { + "bbox": [ + 444, + 486, + 503, + 498 + ], + "score": 0.93, + "content": "\\gamma _ { 2 } = h / n \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 504, + 484, + 507, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 140, + 509 + ], + "score": 1.0, + "content": "A ridge", + "type": "text" + }, + { + "bbox": [ + 141, + 497, + 147, + 508 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 495, + 506, + 509 + ], + "score": 1.0, + "content": "is added make use of (Louart et al., 2018, Theorem 1), which derived the asymptotic", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 507, + 410, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 215, + 524 + ], + "score": 1.0, + "content": "equivalent of the resolvent", + "type": "text" + }, + { + "bbox": [ + 216, + 507, + 279, + 524 + ], + "score": 0.94, + "content": "\\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 508, + 352, + 524 + ], + "score": 1.0, + "content": ". It follows that as", + "type": "text" + }, + { + "bbox": [ + 352, + 510, + 405, + 522 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 508, + 410, + 524 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 484, + 507, + 524 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 525, + 465, + 675 + ], + "lines": [ + { + "bbox": [ + 141, + 525, + 465, + 675 + ], + "spans": [ + { + "bbox": [ + 141, + 525, + 465, + 675 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\mathrm { t r } \\left( h ^ { - 1 } \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) - h ^ { - 1 } I \\right) } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } \\left\\| \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } - \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) ^ { - 1 } \\right\\| _ { F } } \\\\ & { \\quad \\cdot \\left\\| \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right\\| _ { 2 } } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } O ( n ^ { - 1 / 2 + \\varepsilon } ) O ( n ^ { 1 / 2 } ) \\to 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "526904641963f8ba69ef39d63c4d674173d9230b06c35a9dd75ff16689448313.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 141, + 525, + 465, + 575.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 141, + 575.0, + 465, + 625.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 141, + 625.0, + 465, + 675.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 243, + 694 + ], + "score": 1.0, + "content": "where we have used the inequality", + "type": "text" + }, + { + "bbox": [ + 244, + 679, + 339, + 692 + ], + "score": 0.92, + "content": "\\operatorname { t r } \\left( A B \\right) \\leq \\left\\| A \\right\\| _ { F } \\left\\| B \\right\\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 677, + 449, + 694 + ], + "score": 1.0, + "content": ", and equivalently by taking", + "type": "text" + }, + { + "bbox": [ + 449, + 680, + 475, + 691 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 677, + 506, + 694 + ], + "score": 1.0, + "content": "we get", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 677, + 506, + 694 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 694, + 409, + 736 + ], + "lines": [ + { + "bbox": [ + 201, + 694, + 409, + 736 + ], + "spans": [ + { + "bbox": [ + 201, + 694, + 409, + 736 + ], + "score": 0.94, + "content": "\\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } \\frac { n } { h } \\frac { \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } { 1 + \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } = 1 .", + "type": "interline_equation", + "image_path": "47918eab7f209ed0d7ec24e3972688136f5931475e7d0360095f3688a840a5b9.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 694, + 409, + 715.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 201, + 715.0, + 409, + 736.0 + ], + "spans": [], + "index": 30 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 101 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 147, + 99 + ], + "score": 1.0, + "content": "Therefore", + "type": "text" + }, + { + "bbox": [ + 147, + 80, + 418, + 101 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } n / h \\cdot \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\gamma _ { 2 } / ( 1 - \\gamma _ { 2 } ) . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 80, + 460, + 99 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 461, + 83, + 506, + 96 + ], + "score": 0.9, + "content": "\\partial V _ { \\xi } / \\partial \\xi =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 97, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 210, + 113 + ], + "score": 0.9, + "content": "\\mathrm { t r } \\left( \\xi ( \\Phi \\Phi ^ { \\top } - \\xi I ) ^ { - 2 } K _ { W } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 97, + 374, + 113 + ], + "score": 1.0, + "content": "is bounded around the neighbourhood of", + "type": "text" + }, + { + "bbox": [ + 374, + 100, + 398, + 111 + ], + "score": 0.89, + "content": "\\xi = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 97, + 506, + 113 + ], + "score": 1.0, + "content": ", hence following the same", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 110, + 487, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 390, + 124 + ], + "score": 1.0, + "content": "argument as (Hastie et al., 2019, Theorem 4), we exchange the limit of", + "type": "text" + }, + { + "bbox": [ + 390, + 111, + 417, + 123 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 110, + 435, + 124 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 435, + 111, + 483, + 123 + ], + "score": 0.93, + "content": "n , d , h 0", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 110, + 487, + 124 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 127, + 334, + 151 + ], + "lines": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "spans": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "score": 0.93, + "content": "V \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } .", + "type": "interline_equation", + "image_path": "c92aa2600d8044011547634f0b384701c1f6c8ff7d58733fb78e4194f2952399.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 160, + 506, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 143, + 173 + ], + "score": 1.0, + "content": "Case 2.", + "type": "text" + }, + { + "bbox": [ + 144, + 162, + 173, + 172 + ], + "score": 0.88, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". Techniques used in the current proof are largely borrowed from Hastie et al.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 172, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 505, + 184 + ], + "score": 1.0, + "content": "(2019); Cheng and Singer (2013), and we include the full proof for completeness. It should be noted", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "that compared to Hastie et al. (2019) we handle the non-zero expectation of the nonlinearity under", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Gaussian distribution, i.e. the off-diagonal entries of the kernel matrix is no longer zero-centered. For", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 205, + 377, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 377, + 217 + ], + "score": 1.0, + "content": "simplicity we mainly adhere to the notations in Hastie et al. (2019).", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 313, + 235 + ], + "score": 1.0, + "content": "We briefly summarizes the procedure for deriving", + "type": "text" + }, + { + "bbox": [ + 313, + 222, + 322, + 232 + ], + "score": 0.74, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 221, + 506, + 235 + ], + "score": 1.0, + "content": ". Instead of calculating the variance directly,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 233, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 234, + 245 + ], + "score": 1.0, + "content": "we analyze a modified quantity", + "type": "text" + }, + { + "bbox": [ + 234, + 233, + 246, + 245 + ], + "score": 0.88, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 233, + 303, + 245 + ], + "score": 1.0, + "content": "and then take", + "type": "text" + }, + { + "bbox": [ + 303, + 234, + 329, + 245 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 233, + 505, + 245 + ], + "score": 1.0, + "content": ", which can be connected to the trace of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 185, + 259 + ], + "score": 1.0, + "content": "resolvent of matrix", + "type": "text" + }, + { + "bbox": [ + 185, + 245, + 194, + 256 + ], + "score": 0.85, + "content": "\\tilde { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 245, + 388, + 259 + ], + "score": 1.0, + "content": "defined in (56); this translates the calculation of", + "type": "text" + }, + { + "bbox": [ + 388, + 246, + 399, + 259 + ], + "score": 0.88, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "into the calculation of the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 258, + 266, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 193, + 271 + ], + "score": 1.0, + "content": "Stieltjes transform of", + "type": "text" + }, + { + "bbox": [ + 194, + 258, + 202, + 269 + ], + "score": 0.8, + "content": "\\tilde { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 258, + 266, + 271 + ], + "score": 1.0, + "content": "(57), (61), (62).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 283, + 302, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 283, + 301, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 244, + 297 + ], + "score": 1.0, + "content": "C.5 DERIVING THE VARIANCE", + "type": "text" + }, + { + "bbox": [ + 244, + 284, + 254, + 294 + ], + "score": 0.67, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 283, + 274, + 297 + ], + "score": 1.0, + "content": "FOR", + "type": "text" + }, + { + "bbox": [ + 274, + 284, + 301, + 294 + ], + "score": 0.81, + "content": "h > n", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 104, + 304, + 463, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 464, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 452, + 318 + ], + "score": 1.0, + "content": "Step 1. An equivalent expression. For notational simplicity we omit the magnitude", + "type": "text" + }, + { + "bbox": [ + 452, + 307, + 459, + 315 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 303, + 464, + 318 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 321, + 433, + 349 + ], + "lines": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "spans": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "score": 0.92, + "content": "\\begin{array} { r } { V = \\mathrm { t r } \\bigg ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } \\bigg ) . } \\end{array}", + "type": "interline_equation", + "image_path": "deb1415ea32e2554feb854247738d950e9a7cc3b2c9c5d080c17a325eb6e09ad.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 356, + 365 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 357, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 357, + 366 + ], + "score": 1.0, + "content": "and due to the same continuity argument as in Case 1 we have", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 369, + 418, + 394 + ], + "lines": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "spans": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "score": 0.93, + "content": "V = \\operatorname * { l i m } _ { \\xi 0 } \\frac { 1 } { n } \\Big [ \\mathrm { t r } ( S ( S ^ { \\top } S - \\xi I _ { n } ) ^ { - 2 } S ^ { \\top } K _ { W } ) \\Big ] = \\operatorname * { l i m } _ { \\xi 0 } V _ { \\xi } .", + "type": "interline_equation", + "image_path": "8b2d551025e06da7434a2b9f5271a75e04ab44cce52bdc7d9e4404774b445e77.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 372, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 372, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 133, + 415 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 400, + 282, + 414 + ], + "score": 0.9, + "content": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n } = \\Phi / \\sqrt { n } , \\xi \\in \\mathbb { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 399, + 300, + 415 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 301, + 401, + 332, + 413 + ], + "score": 0.89, + "content": "\\Im \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 399, + 344, + 415 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 345, + 402, + 369, + 413 + ], + "score": 0.88, + "content": "\\xi < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 399, + 372, + 415 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 506, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 298, + 430 + ], + "score": 1.0, + "content": "We decompose the normalized feature matrix", + "type": "text" + }, + { + "bbox": [ + 299, + 417, + 382, + 430 + ], + "score": 0.91, + "content": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 416, + 396, + 430 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 396, + 417, + 450, + 428 + ], + "score": 0.88, + "content": "S = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 416, + 483, + 430 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 483, + 418, + 505, + 429 + ], + "score": 0.85, + "content": "\\Sigma =", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 426, + 507, + 442 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 231, + 442 + ], + "score": 0.9, + "content": "\\mathrm { d i a g } _ { h \\times n } \\big ( \\mathring { \\phi _ { 1 } } , \\cdot \\cdot \\cdot , \\phi _ { n } \\big ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 426, + 346, + 442 + ], + "score": 1.0, + "content": "is a tall diagonal matrix, and", + "type": "text" + }, + { + "bbox": [ + 347, + 429, + 456, + 441 + ], + "score": 0.89, + "content": "U \\overset { ^ { \\prime } } { = } [ \\pmb { u } _ { 1 } , \\cdots , \\pmb { u } _ { h } ] \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 426, + 507, + 442 + ], + "score": 1.0, + "content": "is the set of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 440, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 217, + 455 + ], + "score": 1.0, + "content": "orthogonal eigenvectors of", + "type": "text" + }, + { + "bbox": [ + 217, + 441, + 343, + 454 + ], + "score": 0.89, + "content": "S \\underline { { S } } ^ { \\top } = \\phi ( \\underline { { W } } ^ { \\top } X ) \\phi ( X ^ { \\top } W ) / n", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 440, + 363, + 455 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 364, + 442, + 408, + 452 + ], + "score": 0.91, + "content": "V \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 440, + 506, + 455 + ], + "score": 1.0, + "content": "is the set of orthogonal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 452, + 443, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 169, + 465 + ], + "score": 1.0, + "content": "eigenvectors of", + "type": "text" + }, + { + "bbox": [ + 170, + 452, + 295, + 465 + ], + "score": 0.9, + "content": "S ^ { \\top } S = \\phi ( X ^ { \\top } W ) \\phi ( \\dot { W } ^ { \\top } X ) / n", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 452, + 443, + 465 + ], + "score": 1.0, + "content": ". Now the variance can be written as", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 470, + 395, + 493 + ], + "lines": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "spans": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "score": 0.92, + "content": "V _ { \\xi } = \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) .", + "type": "interline_equation", + "image_path": "1e601b2a761065abacfb294a61d5e371d9d05fe2d3cc49ea946ce1d251bd3f17.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 503, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 496, + 504, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 451, + 511 + ], + "score": 1.0, + "content": "By the same argument as in Lemma 13 of Hastie et al. (2019) one can show that when", + "type": "text" + }, + { + "bbox": [ + 452, + 498, + 504, + 510 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 514, + 479, + 538 + ], + "lines": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "score": 0.92, + "content": "V _ { \\xi } = \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) \\Big ] \\to \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) \\Big ] ,", + "type": "interline_equation", + "image_path": "1d70b869882ea158d5128f5c765fadd3fbb0a6758b7069413ff1fe35e856aec0.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 507, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 143, + 559 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 144, + 543, + 162, + 556 + ], + "score": 0.9, + "content": "\\tilde { K } _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 543, + 255, + 559 + ], + "score": 1.0, + "content": "is the approxmation of", + "type": "text" + }, + { + "bbox": [ + 255, + 546, + 273, + 556 + ], + "score": 0.88, + "content": "K _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 543, + 505, + 559 + ], + "score": 1.0, + "content": "defined in Lemma 16. Writing the trace explicitly (denote", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 555, + 343, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 155, + 569 + ], + "score": 1.0, + "content": "eigenvalues", + "type": "text" + }, + { + "bbox": [ + 156, + 556, + 189, + 568 + ], + "score": 0.95, + "content": "\\lambda _ { i } = \\phi _ { i } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 555, + 210, + 569 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 210, + 556, + 302, + 568 + ], + "score": 0.9, + "content": "\\phi _ { n + 1 } = \\cdot \\cdot \\cdot = \\phi _ { h } = 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 555, + 343, + 569 + ], + "score": 1.0, + "content": "), we have", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 573, + 462, + 608 + ], + "lines": [ + { + "bbox": [ + 149, + 573, + 462, + 608 + ], + "spans": [ + { + "bbox": [ + 149, + 573, + 462, + 608 + ], + "score": 0.94, + "content": "V _ { \\xi } \\to \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) = \\gamma _ { 2 } \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } { \\pmb { u } } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\pmb { u } } _ { i } .", + "type": "interline_equation", + "image_path": "5f7703358b5aaabc04bb952542ba6f16e46959915b54dd57af799b2688729815.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 149, + 573, + 462, + 584.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 149, + 584.6666666666666, + 462, + 596.3333333333333 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 149, + 596.3333333333333, + 462, + 607.9999999999999 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 503, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 504, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 262, + 629 + ], + "score": 1.0, + "content": "Since the positive support of spectrum", + "type": "text" + }, + { + "bbox": [ + 262, + 615, + 268, + 624 + ], + "score": 0.83, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 612, + 430, + 629 + ], + "score": 1.0, + "content": "is lower bounded and the density at 0 is", + "type": "text" + }, + { + "bbox": [ + 430, + 613, + 464, + 626 + ], + "score": 0.92, + "content": "1 - \\gamma _ { 2 } ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 612, + 504, + 629 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 631, + 513, + 666 + ], + "lines": [ + { + "bbox": [ + 111, + 631, + 513, + 666 + ], + "spans": [ + { + "bbox": [ + 111, + 631, + 513, + 666 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\gamma _ { \\xi } \\to \\gamma _ { 2 } \\displaystyle \\frac { 1 } { h } \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } = \\gamma _ { 2 } \\displaystyle \\int \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\gamma _ { 2 } \\displaystyle \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) , } \\end{array}", + "type": "interline_equation", + "image_path": "98350499ada2aed4c3ced6b99bb30d22702b96ffa89753c2282abce0bdfe94ba.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 111, + 631, + 513, + 642.6666666666666 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 111, + 642.6666666666666, + 513, + 654.3333333333333 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 111, + 654.3333333333333, + 513, + 665.9999999999999 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 682, + 492, + 699 + ], + "lines": [ + { + "bbox": [ + 104, + 681, + 493, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 681, + 173, + 700 + ], + "score": 1.0, + "content": "where we define", + "type": "text" + }, + { + "bbox": [ + 174, + 683, + 315, + 699 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { n } ( x ) = \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\delta _ { \\lambda _ { i } } ( x ) \\pmb { u } _ { i } ^ { \\top } \\tilde { K } _ { W } \\pmb { u } _ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 681, + 397, + 700 + ], + "score": 1.0, + "content": "and its positive part", + "type": "text" + }, + { + "bbox": [ + 397, + 685, + 424, + 697 + ], + "score": 0.92, + "content": "\\mu _ { n } ^ { + } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 681, + 493, + 700 + ], + "score": 1.0, + "content": ". Hence we have", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 165, + 703, + 445, + 731 + ], + "lines": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "spans": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "score": 0.94, + "content": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } V _ { \\xi } = \\operatorname* { l i m } _ { \\xi \\to 0 } \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda } \\mu _ { \\infty } ^ { + } ( d \\lambda ) .", + "type": "interline_equation", + "image_path": "d2b089387b970da85b422f5e2b23f06cc9985d07a89aa42c5bb61dbfb5b0b019.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 101 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 147, + 99 + ], + "score": 1.0, + "content": "Therefore", + "type": "text" + }, + { + "bbox": [ + 147, + 80, + 418, + 101 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } n / h \\cdot \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\gamma _ { 2 } / ( 1 - \\gamma _ { 2 } ) . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 80, + 460, + 99 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 461, + 83, + 506, + 96 + ], + "score": 0.9, + "content": "\\partial V _ { \\xi } / \\partial \\xi =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 97, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 210, + 113 + ], + "score": 0.9, + "content": "\\mathrm { t r } \\left( \\xi ( \\Phi \\Phi ^ { \\top } - \\xi I ) ^ { - 2 } K _ { W } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 97, + 374, + 113 + ], + "score": 1.0, + "content": "is bounded around the neighbourhood of", + "type": "text" + }, + { + "bbox": [ + 374, + 100, + 398, + 111 + ], + "score": 0.89, + "content": "\\xi = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 97, + 506, + 113 + ], + "score": 1.0, + "content": ", hence following the same", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 110, + 487, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 390, + 124 + ], + "score": 1.0, + "content": "argument as (Hastie et al., 2019, Theorem 4), we exchange the limit of", + "type": "text" + }, + { + "bbox": [ + 390, + 111, + 417, + 123 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 110, + 435, + 124 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 435, + 111, + 483, + 123 + ], + "score": 0.93, + "content": "n , d , h 0", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 110, + 487, + 124 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 80, + 506, + 124 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 127, + 334, + 151 + ], + "lines": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "spans": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "score": 0.93, + "content": "V \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } .", + "type": "interline_equation", + "image_path": "c92aa2600d8044011547634f0b384701c1f6c8ff7d58733fb78e4194f2952399.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 276, + 127, + 334, + 151 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 160, + 506, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 143, + 173 + ], + "score": 1.0, + "content": "Case 2.", + "type": "text" + }, + { + "bbox": [ + 144, + 162, + 173, + 172 + ], + "score": 0.88, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". Techniques used in the current proof are largely borrowed from Hastie et al.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 172, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 505, + 184 + ], + "score": 1.0, + "content": "(2019); Cheng and Singer (2013), and we include the full proof for completeness. It should be noted", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "that compared to Hastie et al. (2019) we handle the non-zero expectation of the nonlinearity under", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Gaussian distribution, i.e. the off-diagonal entries of the kernel matrix is no longer zero-centered. For", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 205, + 377, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 377, + 217 + ], + "score": 1.0, + "content": "simplicity we mainly adhere to the notations in Hastie et al. (2019).", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 159, + 506, + 217 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 313, + 235 + ], + "score": 1.0, + "content": "We briefly summarizes the procedure for deriving", + "type": "text" + }, + { + "bbox": [ + 313, + 222, + 322, + 232 + ], + "score": 0.74, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 221, + 506, + 235 + ], + "score": 1.0, + "content": ". Instead of calculating the variance directly,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 233, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 234, + 245 + ], + "score": 1.0, + "content": "we analyze a modified quantity", + "type": "text" + }, + { + "bbox": [ + 234, + 233, + 246, + 245 + ], + "score": 0.88, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 233, + 303, + 245 + ], + "score": 1.0, + "content": "and then take", + "type": "text" + }, + { + "bbox": [ + 303, + 234, + 329, + 245 + ], + "score": 0.92, + "content": "\\xi 0", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 233, + 505, + 245 + ], + "score": 1.0, + "content": ", which can be connected to the trace of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 185, + 259 + ], + "score": 1.0, + "content": "resolvent of matrix", + "type": "text" + }, + { + "bbox": [ + 185, + 245, + 194, + 256 + ], + "score": 0.85, + "content": "\\tilde { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 245, + 388, + 259 + ], + "score": 1.0, + "content": "defined in (56); this translates the calculation of", + "type": "text" + }, + { + "bbox": [ + 388, + 246, + 399, + 259 + ], + "score": 0.88, + "content": "V _ { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "into the calculation of the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 258, + 266, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 193, + 271 + ], + "score": 1.0, + "content": "Stieltjes transform of", + "type": "text" + }, + { + "bbox": [ + 194, + 258, + 202, + 269 + ], + "score": 0.8, + "content": "\\tilde { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 258, + 266, + 271 + ], + "score": 1.0, + "content": "(57), (61), (62).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 221, + 506, + 271 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 283, + 302, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 283, + 301, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 244, + 297 + ], + "score": 1.0, + "content": "C.5 DERIVING THE VARIANCE", + "type": "text" + }, + { + "bbox": [ + 244, + 284, + 254, + 294 + ], + "score": 0.67, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 283, + 274, + 297 + ], + "score": 1.0, + "content": "FOR", + "type": "text" + }, + { + "bbox": [ + 274, + 284, + 301, + 294 + ], + "score": 0.81, + "content": "h > n", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 283, + 301, + 297 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 304, + 463, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 464, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 452, + 318 + ], + "score": 1.0, + "content": "Step 1. An equivalent expression. For notational simplicity we omit the magnitude", + "type": "text" + }, + { + "bbox": [ + 452, + 307, + 459, + 315 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 303, + 464, + 318 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 303, + 464, + 318 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 321, + 433, + 349 + ], + "lines": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "spans": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "score": 0.92, + "content": "\\begin{array} { r } { V = \\mathrm { t r } \\bigg ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } \\bigg ) . } \\end{array}", + "type": "interline_equation", + "image_path": "deb1415ea32e2554feb854247738d950e9a7cc3b2c9c5d080c17a325eb6e09ad.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 178, + 321, + 433, + 349 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 356, + 365 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 357, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 357, + 366 + ], + "score": 1.0, + "content": "and due to the same continuity argument as in Case 1 we have", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 352, + 357, + 366 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 369, + 418, + 394 + ], + "lines": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "spans": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "score": 0.93, + "content": "V = \\operatorname * { l i m } _ { \\xi 0 } \\frac { 1 } { n } \\Big [ \\mathrm { t r } ( S ( S ^ { \\top } S - \\xi I _ { n } ) ^ { - 2 } S ^ { \\top } K _ { W } ) \\Big ] = \\operatorname * { l i m } _ { \\xi 0 } V _ { \\xi } .", + "type": "interline_equation", + "image_path": "8b2d551025e06da7434a2b9f5271a75e04ab44cce52bdc7d9e4404774b445e77.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 192, + 369, + 418, + 394 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 372, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 372, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 133, + 415 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 400, + 282, + 414 + ], + "score": 0.9, + "content": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n } = \\Phi / \\sqrt { n } , \\xi \\in \\mathbb { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 399, + 300, + 415 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 301, + 401, + 332, + 413 + ], + "score": 0.89, + "content": "\\Im \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 399, + 344, + 415 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 345, + 402, + 369, + 413 + ], + "score": 0.88, + "content": "\\xi < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 399, + 372, + 415 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 399, + 372, + 415 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 506, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 298, + 430 + ], + "score": 1.0, + "content": "We decompose the normalized feature matrix", + "type": "text" + }, + { + "bbox": [ + 299, + 417, + 382, + 430 + ], + "score": 0.91, + "content": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 416, + 396, + 430 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 396, + 417, + 450, + 428 + ], + "score": 0.88, + "content": "S = U \\Sigma V ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 416, + 483, + 430 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 483, + 418, + 505, + 429 + ], + "score": 0.85, + "content": "\\Sigma =", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 426, + 507, + 442 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 231, + 442 + ], + "score": 0.9, + "content": "\\mathrm { d i a g } _ { h \\times n } \\big ( \\mathring { \\phi _ { 1 } } , \\cdot \\cdot \\cdot , \\phi _ { n } \\big ) \\in \\mathbb { R } ^ { h \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 426, + 346, + 442 + ], + "score": 1.0, + "content": "is a tall diagonal matrix, and", + "type": "text" + }, + { + "bbox": [ + 347, + 429, + 456, + 441 + ], + "score": 0.89, + "content": "U \\overset { ^ { \\prime } } { = } [ \\pmb { u } _ { 1 } , \\cdots , \\pmb { u } _ { h } ] \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 426, + 507, + 442 + ], + "score": 1.0, + "content": "is the set of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 440, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 217, + 455 + ], + "score": 1.0, + "content": "orthogonal eigenvectors of", + "type": "text" + }, + { + "bbox": [ + 217, + 441, + 343, + 454 + ], + "score": 0.89, + "content": "S \\underline { { S } } ^ { \\top } = \\phi ( \\underline { { W } } ^ { \\top } X ) \\phi ( X ^ { \\top } W ) / n", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 440, + 363, + 455 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 364, + 442, + 408, + 452 + ], + "score": 0.91, + "content": "V \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 440, + 506, + 455 + ], + "score": 1.0, + "content": "is the set of orthogonal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 452, + 443, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 169, + 465 + ], + "score": 1.0, + "content": "eigenvectors of", + "type": "text" + }, + { + "bbox": [ + 170, + 452, + 295, + 465 + ], + "score": 0.9, + "content": "S ^ { \\top } S = \\phi ( X ^ { \\top } W ) \\phi ( \\dot { W } ^ { \\top } X ) / n", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 452, + 443, + 465 + ], + "score": 1.0, + "content": ". Now the variance can be written as", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 416, + 507, + 465 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 470, + 395, + 493 + ], + "lines": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "spans": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "score": 0.92, + "content": "V _ { \\xi } = \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) .", + "type": "interline_equation", + "image_path": "1e601b2a761065abacfb294a61d5e371d9d05fe2d3cc49ea946ce1d251bd3f17.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 216, + 470, + 395, + 493 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 503, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 496, + 504, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 451, + 511 + ], + "score": 1.0, + "content": "By the same argument as in Lemma 13 of Hastie et al. (2019) one can show that when", + "type": "text" + }, + { + "bbox": [ + 452, + 498, + 504, + 510 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 496, + 504, + 511 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 514, + 479, + 538 + ], + "lines": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "score": 0.92, + "content": "V _ { \\xi } = \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) \\Big ] \\to \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) \\Big ] ,", + "type": "interline_equation", + "image_path": "1d70b869882ea158d5128f5c765fadd3fbb0a6758b7069413ff1fe35e856aec0.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 115, + 514, + 479, + 538 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 507, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 143, + 559 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 144, + 543, + 162, + 556 + ], + "score": 0.9, + "content": "\\tilde { K } _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 543, + 255, + 559 + ], + "score": 1.0, + "content": "is the approxmation of", + "type": "text" + }, + { + "bbox": [ + 255, + 546, + 273, + 556 + ], + "score": 0.88, + "content": "K _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 543, + 505, + 559 + ], + "score": 1.0, + "content": "defined in Lemma 16. Writing the trace explicitly (denote", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 555, + 343, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 155, + 569 + ], + "score": 1.0, + "content": "eigenvalues", + "type": "text" + }, + { + "bbox": [ + 156, + 556, + 189, + 568 + ], + "score": 0.95, + "content": "\\lambda _ { i } = \\phi _ { i } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 555, + 210, + 569 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 210, + 556, + 302, + 568 + ], + "score": 0.9, + "content": "\\phi _ { n + 1 } = \\cdot \\cdot \\cdot = \\phi _ { h } = 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 555, + 343, + 569 + ], + "score": 1.0, + "content": "), we have", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 543, + 505, + 569 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 573, + 462, + 608 + ], + "lines": [ + { + "bbox": [ + 149, + 573, + 462, + 608 + ], + "spans": [ + { + "bbox": [ + 149, + 573, + 462, + 608 + ], + "score": 0.94, + "content": "V _ { \\xi } \\to \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) = \\gamma _ { 2 } \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } { \\pmb { u } } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\pmb { u } } _ { i } .", + "type": "interline_equation", + "image_path": "5f7703358b5aaabc04bb952542ba6f16e46959915b54dd57af799b2688729815.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 149, + 573, + 462, + 584.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 149, + 584.6666666666666, + 462, + 596.3333333333333 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 149, + 596.3333333333333, + 462, + 607.9999999999999 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 503, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 504, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 262, + 629 + ], + "score": 1.0, + "content": "Since the positive support of spectrum", + "type": "text" + }, + { + "bbox": [ + 262, + 615, + 268, + 624 + ], + "score": 0.83, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 612, + 430, + 629 + ], + "score": 1.0, + "content": "is lower bounded and the density at 0 is", + "type": "text" + }, + { + "bbox": [ + 430, + 613, + 464, + 626 + ], + "score": 0.92, + "content": "1 - \\gamma _ { 2 } ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 612, + 504, + 629 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 612, + 504, + 629 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 631, + 513, + 666 + ], + "lines": [ + { + "bbox": [ + 111, + 631, + 513, + 666 + ], + "spans": [ + { + "bbox": [ + 111, + 631, + 513, + 666 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\gamma _ { \\xi } \\to \\gamma _ { 2 } \\displaystyle \\frac { 1 } { h } \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } = \\gamma _ { 2 } \\displaystyle \\int \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\gamma _ { 2 } \\displaystyle \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) , } \\end{array}", + "type": "interline_equation", + "image_path": "98350499ada2aed4c3ced6b99bb30d22702b96ffa89753c2282abce0bdfe94ba.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 111, + 631, + 513, + 642.6666666666666 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 111, + 642.6666666666666, + 513, + 654.3333333333333 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 111, + 654.3333333333333, + 513, + 665.9999999999999 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 682, + 492, + 699 + ], + "lines": [ + { + "bbox": [ + 104, + 681, + 493, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 681, + 173, + 700 + ], + "score": 1.0, + "content": "where we define", + "type": "text" + }, + { + "bbox": [ + 174, + 683, + 315, + 699 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { n } ( x ) = \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\delta _ { \\lambda _ { i } } ( x ) \\pmb { u } _ { i } ^ { \\top } \\tilde { K } _ { W } \\pmb { u } _ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 681, + 397, + 700 + ], + "score": 1.0, + "content": "and its positive part", + "type": "text" + }, + { + "bbox": [ + 397, + 685, + 424, + 697 + ], + "score": 0.92, + "content": "\\mu _ { n } ^ { + } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 681, + 493, + 700 + ], + "score": 1.0, + "content": ". Hence we have", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 104, + 681, + 493, + 700 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 165, + 703, + 445, + 731 + ], + "lines": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "spans": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "score": 0.94, + "content": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } V _ { \\xi } = \\operatorname* { l i m } _ { \\xi \\to 0 } \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda } \\mu _ { \\infty } ^ { + } ( d \\lambda ) .", + "type": "interline_equation", + "image_path": "d2b089387b970da85b422f5e2b23f06cc9985d07a89aa42c5bb61dbfb5b0b019.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 165, + 703, + 445, + 731 + ], + "spans": [], + "index": 36 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 398, + 95 + ], + "lines": [ + { + "bbox": [ + 104, + 78, + 399, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 232, + 97 + ], + "score": 1.0, + "content": "We define the following matrix", + "type": "text" + }, + { + "bbox": [ + 233, + 81, + 318, + 95 + ], + "score": 0.92, + "content": "\\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) \\in \\mathbb { R } ^ { N \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 78, + 347, + 97 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 347, + 83, + 394, + 93 + ], + "score": 0.93, + "content": "N = n + h", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 78, + 399, + 97 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 98, + 398, + 126 + ], + "lines": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "spans": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "score": 0.93, + "content": "\\boldsymbol { \\tilde { A } _ { n } } ( \\rho , \\varsigma , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\varsigma \\mathbf { 1 } _ { h } \\mathbf { 1 } _ { h } ^ { \\top } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "7098dd044cf0ce45d2d339dc153c061a34188a1721556182854930ef68fe8937.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 132, + 282, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 283, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 256, + 146 + ], + "score": 1.0, + "content": "And denote the Stieltjes transform of", + "type": "text" + }, + { + "bbox": [ + 256, + 131, + 270, + 144 + ], + "score": 0.91, + "content": "{ \\tilde { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 130, + 283, + 146 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 148, + 404, + 173 + ], + "lines": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "spans": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "score": 0.91, + "content": "\\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) .", + "type": "interline_equation", + "image_path": "ca8963aa61947f8a7a7e07c2f564b1831096eb0b4cdefd30ae14d19cde68b56a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 330, + 190 + ], + "lines": [ + { + "bbox": [ + 105, + 176, + 331, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 236, + 192 + ], + "score": 1.0, + "content": "Then following the definition of", + "type": "text" + }, + { + "bbox": [ + 237, + 177, + 255, + 190 + ], + "score": 0.92, + "content": "\\tilde { K } _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 176, + 331, + 192 + ], + "score": 1.0, + "content": "one can show that", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 194, + 422, + 228 + ], + "lines": [ + { + "bbox": [ + 189, + 194, + 422, + 228 + ], + "spans": [ + { + "bbox": [ + 189, + 194, + 422, + 228 + ], + "score": 0.94, + "content": "\\tilde { m } _ { n } ( \\xi , r x , s x , t x ) = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } x - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) ,", + "type": "interline_equation", + "image_path": "d7cfa9ab41dcac8f1118b47b56f863d3a6d347394000145d246e81194df0a232.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 194, + 422, + 211.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 189, + 211.0, + 422, + 228.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 244, + 244 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 244, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 244, + 245 + ], + "score": 1.0, + "content": "and taking matrix derivative gives", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 248, + 495, + 352 + ], + "lines": [ + { + "bbox": [ + 114, + 248, + 495, + 352 + ], + "spans": [ + { + "bbox": [ + 114, + 248, + 495, + 352 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { - \\displaystyle \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\xi ^ { 2 } I _ { h } + S S ^ { \\top } } & { 0 } \\\\ { 0 } & { I _ { n } + S ^ { \\top } S } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( U ( \\Sigma \\Sigma ^ { \\top } + \\xi ^ { 2 } I _ { h } ) ^ { - 1 } U ^ { \\top } \\tilde { K } _ { W } \\right) = \\frac { 1 } { n } \\displaystyle \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } . } \\end{array}", + "type": "interline_equation", + "image_path": "af41f24faadcea66d2bba2d1806e83d830fc6bd694112deb886e27e03a9b1ced.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 114, + 248, + 495, + 282.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 114, + 282.6666666666667, + 495, + 317.33333333333337 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 114, + 317.33333333333337, + 495, + 352.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 464, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 365, + 465, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 173, + 381 + ], + "score": 1.0, + "content": "Denote the limit", + "type": "text" + }, + { + "bbox": [ + 174, + 367, + 343, + 380 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau ) = \\operatorname* { l i m } _ { n , h , d \\infty } \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 365, + 465, + 381 + ], + "score": 1.0, + "content": ", and the derivative is given as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 384, + 483, + 448 + ], + "lines": [ + { + "bbox": [ + 109, + 384, + 483, + 448 + ], + "spans": [ + { + "bbox": [ + 109, + 384, + 483, + 448 + ], + "score": 0.95, + "content": "\\begin{array} { l } { \\displaystyle \\left. - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\right. _ { x = 0 } = \\displaystyle \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\frac { 1 } { n } \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } { \\boldsymbol u } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\boldsymbol u } _ { i } } \\\\ { \\displaystyle = \\gamma _ { 2 } \\int _ { \\lambda \\geq 0 } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } + \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) . } \\end{array}", + "type": "interline_equation", + "image_path": "e0099db1440f89576a9471d07dd9d9a25693c783b2187c1c40d990dbfa28b03f.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 109, + 384, + 483, + 405.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 405.3333333333333, + 483, + 426.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 109, + 426.66666666666663, + 483, + 447.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 320, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 321, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 309, + 465 + ], + "score": 1.0, + "content": "For simplicity we define the following function on", + "type": "text" + }, + { + "bbox": [ + 310, + 451, + 316, + 462 + ], + "score": 0.82, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 448, + 321, + 465 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 466, + 375, + 491 + ], + "lines": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "spans": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "score": 0.94, + "content": "q ( \\xi ) = - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } ,", + "type": "interline_equation", + "image_path": "8378bc5c34eda2f679b92343c138dedffe13146e09d609d8147748dbb5a9ff2a.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 462, + 511 + ], + "lines": [ + { + "bbox": [ + 154, + 495, + 344, + 512 + ], + "spans": [ + { + "bbox": [ + 154, + 495, + 344, + 512 + ], + "score": 0.92, + "content": "\\begin{array} { r } { q _ { + } ( \\xi ) = q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac 1 { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) } \\end{array}", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 271, + 515, + 339, + 535 + ], + "lines": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "spans": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "score": 0.9, + "content": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) .", + "type": "interline_equation", + "image_path": "96522d192b8d179a6fe57aff2ebd3476d5d0a88738dfd0126e5d7acd9cedef32.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 546, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 192, + 559 + ], + "score": 1.0, + "content": "Step 2. Calculating", + "type": "text" + }, + { + "bbox": [ + 192, + 547, + 210, + 558 + ], + "score": 0.89, + "content": "q ( \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 546, + 231, + 559 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 232, + 546, + 288, + 559 + ], + "score": 0.92, + "content": "m _ { n } ( \\xi , \\rho , \\varsigma , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 546, + 435, + 559 + ], + "score": 1.0, + "content": "This subsection aims to calculate", + "type": "text" + }, + { + "bbox": [ + 435, + 546, + 486, + 559 + ], + "score": 0.92, + "content": "\\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 556, + 484, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 234, + 570 + ], + "score": 0.9, + "content": "q ( \\xi ) = - \\tilde { m } _ { x } ^ { \\prime } ( \\xi , r x , s x , t x ) | _ { x = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 556, + 484, + 570 + ], + "score": 1.0, + "content": ", from which the variance can be computed from (57)(61)(62).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 439, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 438, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 148, + 590 + ], + "score": 1.0, + "content": "We define", + "type": "text" + }, + { + "bbox": [ + 149, + 577, + 162, + 587 + ], + "score": 0.92, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 574, + 421, + 590 + ], + "score": 1.0, + "content": "by subtracting the off-diagonal entries of the upper-left block of", + "type": "text" + }, + { + "bbox": [ + 421, + 574, + 434, + 587 + ], + "score": 0.91, + "content": "{ \\tilde { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 574, + 438, + 590 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 236, + 592, + 374, + 620 + ], + "lines": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "spans": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "score": 0.94, + "content": "A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] ,", + "type": "interline_equation", + "image_path": "ccbb6c63072a3e4058671b0e8a6e601a81aac9ee807b7a8860cc04cdf5762e7c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 624, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 104, + 623, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 104, + 623, + 134, + 640 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 624, + 208, + 639 + ], + "score": 0.93, + "content": "S = \\tilde { S } - a _ { 0 } I _ { p \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 623, + 230, + 640 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 230, + 624, + 381, + 638 + ], + "score": 0.87, + "content": "S _ { i k } = \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } ) - a _ { 0 } = \\varphi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 623, + 386, + 640 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 386, + 626, + 444, + 638 + ], + "score": 0.91, + "content": "a _ { 0 } = \\mathbb { E } [ \\phi ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 623, + 506, + 640 + ], + "score": 1.0, + "content": ". The Stieltjes", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 159, + 651 + ], + "score": 1.0, + "content": "transform of", + "type": "text" + }, + { + "bbox": [ + 159, + 639, + 173, + 650 + ], + "score": 0.89, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 637, + 212, + 651 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 212, + 638, + 381, + 651 + ], + "score": 0.87, + "content": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 637, + 505, + 651 + ], + "score": 1.0, + "content": ". The following Lemma shows", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 648, + 257, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 124, + 662 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 650, + 139, + 661 + ], + "score": 0.89, + "content": "\\tilde { m } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 648, + 157, + 662 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 158, + 651, + 173, + 660 + ], + "score": 0.87, + "content": "m _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 648, + 257, + 662 + ], + "score": 1.0, + "content": "have the same limit:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 663, + 475, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 663, + 473, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 182, + 678 + ], + "score": 1.0, + "content": "Lemma 12. when", + "type": "text" + }, + { + "bbox": [ + 183, + 666, + 216, + 675 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 663, + 248, + 678 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 249, + 664, + 317, + 676 + ], + "score": 0.38, + "content": "\\Im \\xi > 0 o r \\xi < 0 ", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 663, + 355, + 678 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 355, + 664, + 473, + 676 + ], + "score": 0.9, + "content": "m _ { n } ( \\xi , \\rho , \\tau ) \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) .", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 684, + 253, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 254, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 203, + 699 + ], + "score": 1.0, + "content": "Proof. By definition of", + "type": "text" + }, + { + "bbox": [ + 204, + 684, + 217, + 696 + ], + "score": 0.9, + "content": "{ \\bar { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 682, + 236, + 699 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 236, + 685, + 249, + 696 + ], + "score": 0.89, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 682, + 254, + 699 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 700, + 440, + 731 + ], + "lines": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "spans": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] \\left[ \\begin{array} { c c } { \\xi } & { a _ { 0 } } \\\\ { a _ { 0 } } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "8af219b2bfb34b777549f260c1474405eea56646dceb8fd9b6e82824732dfa15.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 25, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "21", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 398, + 95 + ], + "lines": [ + { + "bbox": [ + 104, + 78, + 399, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 232, + 97 + ], + "score": 1.0, + "content": "We define the following matrix", + "type": "text" + }, + { + "bbox": [ + 233, + 81, + 318, + 95 + ], + "score": 0.92, + "content": "\\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) \\in \\mathbb { R } ^ { N \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 78, + 347, + 97 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 347, + 83, + 394, + 93 + ], + "score": 0.93, + "content": "N = n + h", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 78, + 399, + 97 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 104, + 78, + 399, + 97 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 98, + 398, + 126 + ], + "lines": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "spans": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "score": 0.93, + "content": "\\boldsymbol { \\tilde { A } _ { n } } ( \\rho , \\varsigma , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\varsigma \\mathbf { 1 } _ { h } \\mathbf { 1 } _ { h } ^ { \\top } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "7098dd044cf0ce45d2d339dc153c061a34188a1721556182854930ef68fe8937.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 212, + 98, + 398, + 126 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 132, + 282, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 283, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 256, + 146 + ], + "score": 1.0, + "content": "And denote the Stieltjes transform of", + "type": "text" + }, + { + "bbox": [ + 256, + 131, + 270, + 144 + ], + "score": 0.91, + "content": "{ \\tilde { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 130, + 283, + 146 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 130, + 283, + 146 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 148, + 404, + 173 + ], + "lines": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "spans": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "score": 0.91, + "content": "\\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) .", + "type": "interline_equation", + "image_path": "ca8963aa61947f8a7a7e07c2f564b1831096eb0b4cdefd30ae14d19cde68b56a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 207, + 148, + 404, + 173 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 330, + 190 + ], + "lines": [ + { + "bbox": [ + 105, + 176, + 331, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 236, + 192 + ], + "score": 1.0, + "content": "Then following the definition of", + "type": "text" + }, + { + "bbox": [ + 237, + 177, + 255, + 190 + ], + "score": 0.92, + "content": "\\tilde { K } _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 176, + 331, + 192 + ], + "score": 1.0, + "content": "one can show that", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 176, + 331, + 192 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 194, + 422, + 228 + ], + "lines": [ + { + "bbox": [ + 189, + 194, + 422, + 228 + ], + "spans": [ + { + "bbox": [ + 189, + 194, + 422, + 228 + ], + "score": 0.94, + "content": "\\tilde { m } _ { n } ( \\xi , r x , s x , t x ) = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } x - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) ,", + "type": "interline_equation", + "image_path": "d7cfa9ab41dcac8f1118b47b56f863d3a6d347394000145d246e81194df0a232.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 194, + 422, + 211.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 189, + 211.0, + 422, + 228.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 244, + 244 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 244, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 244, + 245 + ], + "score": 1.0, + "content": "and taking matrix derivative gives", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 231, + 244, + 245 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 248, + 495, + 352 + ], + "lines": [ + { + "bbox": [ + 114, + 248, + 495, + 352 + ], + "spans": [ + { + "bbox": [ + 114, + 248, + 495, + 352 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { - \\displaystyle \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\xi ^ { 2 } I _ { h } + S S ^ { \\top } } & { 0 } \\\\ { 0 } & { I _ { n } + S ^ { \\top } S } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( U ( \\Sigma \\Sigma ^ { \\top } + \\xi ^ { 2 } I _ { h } ) ^ { - 1 } U ^ { \\top } \\tilde { K } _ { W } \\right) = \\frac { 1 } { n } \\displaystyle \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } . } \\end{array}", + "type": "interline_equation", + "image_path": "af41f24faadcea66d2bba2d1806e83d830fc6bd694112deb886e27e03a9b1ced.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 114, + 248, + 495, + 282.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 114, + 282.6666666666667, + 495, + 317.33333333333337 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 114, + 317.33333333333337, + 495, + 352.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 464, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 365, + 465, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 173, + 381 + ], + "score": 1.0, + "content": "Denote the limit", + "type": "text" + }, + { + "bbox": [ + 174, + 367, + 343, + 380 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau ) = \\operatorname* { l i m } _ { n , h , d \\infty } \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 365, + 465, + 381 + ], + "score": 1.0, + "content": ", and the derivative is given as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 365, + 465, + 381 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 384, + 483, + 448 + ], + "lines": [ + { + "bbox": [ + 109, + 384, + 483, + 448 + ], + "spans": [ + { + "bbox": [ + 109, + 384, + 483, + 448 + ], + "score": 0.95, + "content": "\\begin{array} { l } { \\displaystyle \\left. - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\right. _ { x = 0 } = \\displaystyle \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\frac { 1 } { n } \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } { \\boldsymbol u } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\boldsymbol u } _ { i } } \\\\ { \\displaystyle = \\gamma _ { 2 } \\int _ { \\lambda \\geq 0 } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } + \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) . } \\end{array}", + "type": "interline_equation", + "image_path": "e0099db1440f89576a9471d07dd9d9a25693c783b2187c1c40d990dbfa28b03f.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 109, + 384, + 483, + 405.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 405.3333333333333, + 483, + 426.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 109, + 426.66666666666663, + 483, + 447.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 320, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 321, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 309, + 465 + ], + "score": 1.0, + "content": "For simplicity we define the following function on", + "type": "text" + }, + { + "bbox": [ + 310, + 451, + 316, + 462 + ], + "score": 0.82, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 448, + 321, + 465 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 448, + 321, + 465 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 466, + 375, + 491 + ], + "lines": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "spans": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "score": 0.94, + "content": "q ( \\xi ) = - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } ,", + "type": "interline_equation", + "image_path": "8378bc5c34eda2f679b92343c138dedffe13146e09d609d8147748dbb5a9ff2a.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 235, + 466, + 375, + 491 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 462, + 511 + ], + "lines": [ + { + "bbox": [ + 154, + 495, + 344, + 512 + ], + "spans": [ + { + "bbox": [ + 154, + 495, + 344, + 512 + ], + "score": 0.92, + "content": "\\begin{array} { r } { q _ { + } ( \\xi ) = q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac 1 { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) } \\end{array}", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 154, + 495, + 344, + 512 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 271, + 515, + 339, + 535 + ], + "lines": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "spans": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "score": 0.9, + "content": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) .", + "type": "interline_equation", + "image_path": "96522d192b8d179a6fe57aff2ebd3476d5d0a88738dfd0126e5d7acd9cedef32.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 271, + 515, + 339, + 535 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 546, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 192, + 559 + ], + "score": 1.0, + "content": "Step 2. Calculating", + "type": "text" + }, + { + "bbox": [ + 192, + 547, + 210, + 558 + ], + "score": 0.89, + "content": "q ( \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 546, + 231, + 559 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 232, + 546, + 288, + 559 + ], + "score": 0.92, + "content": "m _ { n } ( \\xi , \\rho , \\varsigma , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 546, + 435, + 559 + ], + "score": 1.0, + "content": "This subsection aims to calculate", + "type": "text" + }, + { + "bbox": [ + 435, + 546, + 486, + 559 + ], + "score": 0.92, + "content": "\\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 556, + 484, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 234, + 570 + ], + "score": 0.9, + "content": "q ( \\xi ) = - \\tilde { m } _ { x } ^ { \\prime } ( \\xi , r x , s x , t x ) | _ { x = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 556, + 484, + 570 + ], + "score": 1.0, + "content": ", from which the variance can be computed from (57)(61)(62).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 546, + 505, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 439, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 438, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 148, + 590 + ], + "score": 1.0, + "content": "We define", + "type": "text" + }, + { + "bbox": [ + 149, + 577, + 162, + 587 + ], + "score": 0.92, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 574, + 421, + 590 + ], + "score": 1.0, + "content": "by subtracting the off-diagonal entries of the upper-left block of", + "type": "text" + }, + { + "bbox": [ + 421, + 574, + 434, + 587 + ], + "score": 0.91, + "content": "{ \\tilde { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 574, + 438, + 590 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 574, + 438, + 590 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 236, + 592, + 374, + 620 + ], + "lines": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "spans": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "score": 0.94, + "content": "A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] ,", + "type": "interline_equation", + "image_path": "ccbb6c63072a3e4058671b0e8a6e601a81aac9ee807b7a8860cc04cdf5762e7c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 236, + 592, + 374, + 620 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 624, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 104, + 623, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 104, + 623, + 134, + 640 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 624, + 208, + 639 + ], + "score": 0.93, + "content": "S = \\tilde { S } - a _ { 0 } I _ { p \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 623, + 230, + 640 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 230, + 624, + 381, + 638 + ], + "score": 0.87, + "content": "S _ { i k } = \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } ) - a _ { 0 } = \\varphi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 623, + 386, + 640 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 386, + 626, + 444, + 638 + ], + "score": 0.91, + "content": "a _ { 0 } = \\mathbb { E } [ \\phi ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 623, + 506, + 640 + ], + "score": 1.0, + "content": ". The Stieltjes", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 159, + 651 + ], + "score": 1.0, + "content": "transform of", + "type": "text" + }, + { + "bbox": [ + 159, + 639, + 173, + 650 + ], + "score": 0.89, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 637, + 212, + 651 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 212, + 638, + 381, + 651 + ], + "score": 0.87, + "content": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 637, + 505, + 651 + ], + "score": 1.0, + "content": ". The following Lemma shows", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 648, + 257, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 124, + 662 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 650, + 139, + 661 + ], + "score": 0.89, + "content": "\\tilde { m } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 648, + 157, + 662 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 158, + 651, + 173, + 660 + ], + "score": 0.87, + "content": "m _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 648, + 257, + 662 + ], + "score": 1.0, + "content": "have the same limit:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 104, + 623, + 506, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 663, + 475, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 663, + 473, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 182, + 678 + ], + "score": 1.0, + "content": "Lemma 12. when", + "type": "text" + }, + { + "bbox": [ + 183, + 666, + 216, + 675 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 663, + 248, + 678 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 249, + 664, + 317, + 676 + ], + "score": 0.38, + "content": "\\Im \\xi > 0 o r \\xi < 0 ", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 663, + 355, + 678 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 355, + 664, + 473, + 676 + ], + "score": 0.9, + "content": "m _ { n } ( \\xi , \\rho , \\tau ) \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) .", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 663, + 473, + 678 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 684, + 253, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 254, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 203, + 699 + ], + "score": 1.0, + "content": "Proof. By definition of", + "type": "text" + }, + { + "bbox": [ + 204, + 684, + 217, + 696 + ], + "score": 0.9, + "content": "{ \\bar { A } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 682, + 236, + 699 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 236, + 685, + 249, + 696 + ], + "score": 0.89, + "content": "A _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 682, + 254, + 699 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 682, + 254, + 699 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 700, + 440, + 731 + ], + "lines": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "spans": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] \\left[ \\begin{array} { c c } { \\xi } & { a _ { 0 } } \\\\ { a _ { 0 } } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "8af219b2bfb34b777549f260c1474405eea56646dceb8fd9b6e82824732dfa15.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 170, + 700, + 440, + 731 + ], + "spans": [], + "index": 28 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 93 + ], + "score": 1.0, + "content": "which is a rank-2 matrix. By theorem A.43 from Bai and Silverstein (2010), which characterizes the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 371, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 371, + 106 + ], + "score": 1.0, + "content": "effect of finite-rank perturbation on the e.s.d. of random matrices:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 110, + 382, + 134 + ], + "lines": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "spans": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "score": 0.94, + "content": "\\operatorname* { s u p } _ { x } | F ^ { \\tilde { A } _ { n } } ( x ) - F ^ { A _ { n } } ( x ) | \\leq O \\left( n ^ { - 1 } \\right) ,", + "type": "interline_equation", + "image_path": "3aeaf525edaa1ae902f9fdb47acff8d4dad67e9db825334c4a19d53080e11e4c.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 140, + 505, + 164 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 133, + 153 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 140, + 150, + 150 + ], + "score": 0.9, + "content": "F ^ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 137, + 308, + 153 + ], + "score": 1.0, + "content": "is the empirical spectral distribution of", + "type": "text" + }, + { + "bbox": [ + 308, + 140, + 355, + 151 + ], + "score": 0.92, + "content": "M \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 137, + 506, + 153 + ], + "score": 1.0, + "content": ". The claim follows from the Stieltjes", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 151, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 315, + 164 + ], + "score": 1.0, + "content": "continuity theorem (e.g. Section 2.4 in Tao (2012)).", + "type": "text" + }, + { + "bbox": [ + 496, + 153, + 504, + 161 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 457, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 454, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 247, + 188 + ], + "score": 1.0, + "content": "To calculate the Stieltjes transform", + "type": "text" + }, + { + "bbox": [ + 248, + 177, + 262, + 186 + ], + "score": 0.87, + "content": "m _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 174, + 441, + 188 + ], + "score": 1.0, + "content": ", we take advantage of the block structure of", + "type": "text" + }, + { + "bbox": [ + 441, + 176, + 454, + 186 + ], + "score": 0.9, + "content": "A _ { n }", + "type": "inline_equation" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 192, + 431, + 243 + ], + "lines": [ + { + "bbox": [ + 180, + 192, + 431, + 243 + ], + "spans": [ + { + "bbox": [ + 180, + 192, + 431, + 243 + ], + "score": 0.95, + "content": "\\begin{array} { l } { { m _ { 1 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { p } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ 1 . . p , 1 . . p ] } ^ { - 1 } \\right) , } } \\\\ { { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ p + 1 . . p + n , p + 1 . . p + n ] } ^ { - 1 } \\right) . } } \\end{array}", + "type": "interline_equation", + "image_path": "9ebf10c37453fa053d2e979503fda8f10419487e6f42fd71f42c6c7d2f799d00.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 180, + 192, + 431, + 209.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 180, + 209.0, + 431, + 226.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 180, + 226.0, + 431, + 243.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 247, + 384, + 261 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 382, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 192, + 262 + ], + "score": 1.0, + "content": "One can observe that", + "type": "text" + }, + { + "bbox": [ + 193, + 247, + 382, + 261 + ], + "score": 0.88, + "content": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) + m _ { 2 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 264, + 506, + 311 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 309, + 278 + ], + "score": 1.0, + "content": "In the following equations we omit the subscript", + "type": "text" + }, + { + "bbox": [ + 310, + 267, + 317, + 275 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 263, + 430, + 278 + ], + "score": 1.0, + "content": ", as well as dependency on", + "type": "text" + }, + { + "bbox": [ + 430, + 267, + 455, + 276 + ], + "score": 0.87, + "content": "\\rho , \\varsigma , \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 263, + 505, + 278 + ], + "score": 1.0, + "content": ". Following", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 271, + 294 + ], + "score": 1.0, + "content": "Hastie et al. (2019), we rewrite the matrix", + "type": "text" + }, + { + "bbox": [ + 271, + 276, + 370, + 302 + ], + "score": 0.9, + "content": "A _ { n } = A = { \\left[ \\begin{array} { l l } { A _ { * } } & { a } \\\\ { \\mathbf { 1 } } & { 0 } \\end{array} \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 282, + 400, + 295 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 400, + 282, + 413, + 292 + ], + "score": 0.84, + "content": "A ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 282, + 430, + 295 + ], + "score": 1.0, + "content": "is a", + "type": "text" + }, + { + "bbox": [ + 430, + 282, + 505, + 294 + ], + "score": 0.91, + "content": "\\left( N - 1 \\right) \\times \\left( N - 1 \\right)", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 299, + 407, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 249, + 312 + ], + "score": 1.0, + "content": "matrix with last column and row of", + "type": "text" + }, + { + "bbox": [ + 250, + 300, + 258, + 309 + ], + "score": 0.83, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 299, + 313, + 312 + ], + "score": 1.0, + "content": "removed and", + "type": "text" + }, + { + "bbox": [ + 313, + 302, + 320, + 309 + ], + "score": 0.77, + "content": "\\pmb { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 299, + 407, + 312 + ], + "score": 1.0, + "content": "the activation vector:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 314, + 463, + 342 + ], + "lines": [ + { + "bbox": [ + 149, + 314, + 463, + 342 + ], + "spans": [ + { + "bbox": [ + 149, + 314, + 463, + 342 + ], + "score": 0.95, + "content": "\\begin{array} { r } { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S _ { * } } \\\\ { S _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] ; \\quad \\pmb { a } ^ { \\top } = [ \\phi ( \\boldsymbol { W } ^ { \\top } \\mathbf { x } _ { n } ) ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] = [ \\boldsymbol { s } ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "b34f7a5369954cf11554bfe26ab8230e571a2fe0fc7b1912585e7c823e2d5555.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 149, + 314, + 463, + 323.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 149, + 323.3333333333333, + 463, + 332.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 149, + 332.66666666666663, + 463, + 341.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 347, + 280, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "score": 1.0, + "content": "Hence by the block matrix inverse formula", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 364, + 426, + 392 + ], + "lines": [ + { + "bbox": [ + 185, + 364, + 426, + 392 + ], + "spans": [ + { + "bbox": [ + 185, + 364, + 426, + 392 + ], + "score": 0.92, + "content": "\\begin{array} { r } { ( A - \\xi I _ { N } ) ^ { - 1 } = \\left[ \\begin{array} { l l } { * } & { * } \\\\ { * } & { [ - \\xi - { \\pmb a } ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } { \\pmb a } ] ^ { - 1 } } \\end{array} \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "a7bf8105f5cfca014049958264580ebadc742124d6f5b90b1327ed4d973b3844.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 185, + 364, + 426, + 378.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 185, + 378.0, + 426, + 392.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 318, + 409 + ], + "lines": [ + { + "bbox": [ + 106, + 397, + 318, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 318, + 410 + ], + "score": 1.0, + "content": "Plugging this back in the Stieltjes transform we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 134, + 414, + 477, + 464 + ], + "lines": [ + { + "bbox": [ + 134, + 414, + 477, + 464 + ], + "spans": [ + { + "bbox": [ + 134, + 414, + 477, + 464 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) _ { [ h + 1 , N ] } ^ { - 1 } \\right) = \\mathbb { E } _ { a } \\left[ ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right] _ { N N } } \\\\ { \\displaystyle = \\mathbb { E } _ { a } \\bigg [ \\Big ( - \\xi - a ^ { \\top } ( A ^ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ) ^ { - 1 } \\bigg ] . } \\end{array}", + "type": "interline_equation", + "image_path": "f176ad0980439202a213c14bd6a4d34054a238f6980d3ef3c754459f486fb731.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 134, + 414, + 477, + 430.6666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 134, + 430.6666666666667, + 477, + 447.33333333333337 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 134, + 447.33333333333337, + 477, + 464.00000000000006 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 506, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 276, + 486 + ], + "score": 1.0, + "content": "To obtain an asymptotic description of", + "type": "text" + }, + { + "bbox": [ + 276, + 475, + 297, + 486 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 472, + 506, + 486 + ], + "score": 1.0, + "content": ", we perform the orthonormal decomposition on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 177, + 498 + ], + "score": 1.0, + "content": "the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 178, + 486, + 186, + 496 + ], + "score": 0.79, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 483, + 370, + 498 + ], + "score": 1.0, + "content": "introduced in Cheng and Singer (2013):", + "type": "text" + }, + { + "bbox": [ + 370, + 484, + 471, + 496 + ], + "score": 0.91, + "content": "\\varphi ( x ) \\ = \\ a _ { 1 } x + \\varphi _ { \\perp } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 483, + 507, + 498 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 495, + 507, + 508 + ], + "spans": [ + { + "bbox": [ + 107, + 495, + 207, + 508 + ], + "score": 0.92, + "content": "a _ { 1 } = \\mathbb { E } _ { x \\sim \\mathcal { N } ( 0 , 1 ) } [ x \\varphi ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 495, + 254, + 508 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 254, + 495, + 322, + 507 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ( \\bar { 1 } \\leq i \\leq \\bar { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 495, + 507, + 508 + ], + "score": 1.0, + "content": ", we perform the following orthonormal de-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 505, + 447, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 249, + 519 + ], + "score": 1.0, + "content": "composition (along the direction of", + "type": "text" + }, + { + "bbox": [ + 250, + 508, + 263, + 517 + ], + "score": 0.88, + "content": "{ \\bf { x } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 505, + 344, + 519 + ], + "score": 1.0, + "content": "and the direction of", + "type": "text" + }, + { + "bbox": [ + 344, + 506, + 357, + 517 + ], + "score": 0.89, + "content": "\\tilde { \\mathbf { \\pmb { w } } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 505, + 426, + 519 + ], + "score": 1.0, + "content": "perpendicular to", + "type": "text" + }, + { + "bbox": [ + 426, + 507, + 439, + 517 + ], + "score": 0.85, + "content": "{ \\mathbf { \\mathcal { x } } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 505, + 447, + 519 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 523, + 394, + 555 + ], + "lines": [ + { + "bbox": [ + 216, + 523, + 394, + 555 + ], + "spans": [ + { + "bbox": [ + 216, + 523, + 394, + 555 + ], + "score": 0.93, + "content": "\\pmb { w } _ { i } = \\underbrace { \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { n } } _ { \\eta _ { i } } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } = \\eta _ { i } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } .", + "type": "interline_equation", + "image_path": "cee414213d3b40bac825a65641fde07e4ad883d5b7cb0bba8c495de72d91b4ea.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 523, + 394, + 539.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 216, + 539.0, + 394, + 555.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 294, + 572 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 295, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 276, + 573 + ], + "score": 1.0, + "content": "We can thus simplify the activation vector", + "type": "text" + }, + { + "bbox": [ + 276, + 563, + 283, + 570 + ], + "score": 0.8, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 560, + 295, + 573 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 577, + 449, + 733 + ], + "lines": [ + { + "bbox": [ + 162, + 577, + 449, + 733 + ], + "spans": [ + { + "bbox": [ + 162, + 577, + 449, + 733 + ], + "score": 0.5, + "content": "\\begin{array} { r l } { a ^ { \\top } = \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { 1 } ) } & { \\cdots \\cdot \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { = \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { 1 } } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { h } \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { + \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { 1 } ) } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdot \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] . } \\end{array}", + "type": "interline_equation", + "image_path": "dfd9392f1a095abc0522233c533911ce1fef4318c4702efd92bf8bde56cf40af.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 162, + 577, + 449, + 629.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 162, + 629.0, + 449, + 681.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 162, + 681.0, + 449, + 733.0 + ], + "spans": [], + "index": 32 + } + ] + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 93 + ], + "score": 1.0, + "content": "which is a rank-2 matrix. By theorem A.43 from Bai and Silverstein (2010), which characterizes the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 371, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 371, + 106 + ], + "score": 1.0, + "content": "effect of finite-rank perturbation on the e.s.d. of random matrices:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 504, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 110, + 382, + 134 + ], + "lines": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "spans": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "score": 0.94, + "content": "\\operatorname* { s u p } _ { x } | F ^ { \\tilde { A } _ { n } } ( x ) - F ^ { A _ { n } } ( x ) | \\leq O \\left( n ^ { - 1 } \\right) ,", + "type": "interline_equation", + "image_path": "3aeaf525edaa1ae902f9fdb47acff8d4dad67e9db825334c4a19d53080e11e4c.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 227, + 110, + 382, + 134 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 140, + 505, + 164 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 133, + 153 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 140, + 150, + 150 + ], + "score": 0.9, + "content": "F ^ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 137, + 308, + 153 + ], + "score": 1.0, + "content": "is the empirical spectral distribution of", + "type": "text" + }, + { + "bbox": [ + 308, + 140, + 355, + 151 + ], + "score": 0.92, + "content": "M \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 137, + 506, + 153 + ], + "score": 1.0, + "content": ". The claim follows from the Stieltjes", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 151, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 315, + 164 + ], + "score": 1.0, + "content": "continuity theorem (e.g. Section 2.4 in Tao (2012)).", + "type": "text" + }, + { + "bbox": [ + 496, + 153, + 504, + 161 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 137, + 506, + 164 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 457, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 454, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 247, + 188 + ], + "score": 1.0, + "content": "To calculate the Stieltjes transform", + "type": "text" + }, + { + "bbox": [ + 248, + 177, + 262, + 186 + ], + "score": 0.87, + "content": "m _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 174, + 441, + 188 + ], + "score": 1.0, + "content": ", we take advantage of the block structure of", + "type": "text" + }, + { + "bbox": [ + 441, + 176, + 454, + 186 + ], + "score": 0.9, + "content": "A _ { n }", + "type": "inline_equation" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 174, + 454, + 188 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 192, + 431, + 243 + ], + "lines": [ + { + "bbox": [ + 180, + 192, + 431, + 243 + ], + "spans": [ + { + "bbox": [ + 180, + 192, + 431, + 243 + ], + "score": 0.95, + "content": "\\begin{array} { l } { { m _ { 1 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { p } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ 1 . . p , 1 . . p ] } ^ { - 1 } \\right) , } } \\\\ { { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ p + 1 . . p + n , p + 1 . . p + n ] } ^ { - 1 } \\right) . } } \\end{array}", + "type": "interline_equation", + "image_path": "9ebf10c37453fa053d2e979503fda8f10419487e6f42fd71f42c6c7d2f799d00.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 180, + 192, + 431, + 209.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 180, + 209.0, + 431, + 226.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 180, + 226.0, + 431, + 243.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 247, + 384, + 261 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 382, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 192, + 262 + ], + "score": 1.0, + "content": "One can observe that", + "type": "text" + }, + { + "bbox": [ + 193, + 247, + 382, + 261 + ], + "score": 0.88, + "content": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) + m _ { 2 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 246, + 382, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 264, + 506, + 311 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 309, + 278 + ], + "score": 1.0, + "content": "In the following equations we omit the subscript", + "type": "text" + }, + { + "bbox": [ + 310, + 267, + 317, + 275 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 263, + 430, + 278 + ], + "score": 1.0, + "content": ", as well as dependency on", + "type": "text" + }, + { + "bbox": [ + 430, + 267, + 455, + 276 + ], + "score": 0.87, + "content": "\\rho , \\varsigma , \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 263, + 505, + 278 + ], + "score": 1.0, + "content": ". Following", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 271, + 294 + ], + "score": 1.0, + "content": "Hastie et al. (2019), we rewrite the matrix", + "type": "text" + }, + { + "bbox": [ + 271, + 276, + 370, + 302 + ], + "score": 0.9, + "content": "A _ { n } = A = { \\left[ \\begin{array} { l l } { A _ { * } } & { a } \\\\ { \\mathbf { 1 } } & { 0 } \\end{array} \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 282, + 400, + 295 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 400, + 282, + 413, + 292 + ], + "score": 0.84, + "content": "A ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 282, + 430, + 295 + ], + "score": 1.0, + "content": "is a", + "type": "text" + }, + { + "bbox": [ + 430, + 282, + 505, + 294 + ], + "score": 0.91, + "content": "\\left( N - 1 \\right) \\times \\left( N - 1 \\right)", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 299, + 407, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 249, + 312 + ], + "score": 1.0, + "content": "matrix with last column and row of", + "type": "text" + }, + { + "bbox": [ + 250, + 300, + 258, + 309 + ], + "score": 0.83, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 299, + 313, + 312 + ], + "score": 1.0, + "content": "removed and", + "type": "text" + }, + { + "bbox": [ + 313, + 302, + 320, + 309 + ], + "score": 0.77, + "content": "\\pmb { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 299, + 407, + 312 + ], + "score": 1.0, + "content": "the activation vector:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 263, + 505, + 312 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 314, + 463, + 342 + ], + "lines": [ + { + "bbox": [ + 149, + 314, + 463, + 342 + ], + "spans": [ + { + "bbox": [ + 149, + 314, + 463, + 342 + ], + "score": 0.95, + "content": "\\begin{array} { r } { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S _ { * } } \\\\ { S _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] ; \\quad \\pmb { a } ^ { \\top } = [ \\phi ( \\boldsymbol { W } ^ { \\top } \\mathbf { x } _ { n } ) ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] = [ \\boldsymbol { s } ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "b34f7a5369954cf11554bfe26ab8230e571a2fe0fc7b1912585e7c823e2d5555.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 149, + 314, + 463, + 323.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 149, + 323.3333333333333, + 463, + 332.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 149, + 332.66666666666663, + 463, + 341.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 347, + 280, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "score": 1.0, + "content": "Hence by the block matrix inverse formula", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 347, + 280, + 360 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 364, + 426, + 392 + ], + "lines": [ + { + "bbox": [ + 185, + 364, + 426, + 392 + ], + "spans": [ + { + "bbox": [ + 185, + 364, + 426, + 392 + ], + "score": 0.92, + "content": "\\begin{array} { r } { ( A - \\xi I _ { N } ) ^ { - 1 } = \\left[ \\begin{array} { l l } { * } & { * } \\\\ { * } & { [ - \\xi - { \\pmb a } ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } { \\pmb a } ] ^ { - 1 } } \\end{array} \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "a7bf8105f5cfca014049958264580ebadc742124d6f5b90b1327ed4d973b3844.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 185, + 364, + 426, + 378.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 185, + 378.0, + 426, + 392.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 318, + 409 + ], + "lines": [ + { + "bbox": [ + 106, + 397, + 318, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 318, + 410 + ], + "score": 1.0, + "content": "Plugging this back in the Stieltjes transform we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 397, + 318, + 410 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 134, + 414, + 477, + 464 + ], + "lines": [ + { + "bbox": [ + 134, + 414, + 477, + 464 + ], + "spans": [ + { + "bbox": [ + 134, + 414, + 477, + 464 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) _ { [ h + 1 , N ] } ^ { - 1 } \\right) = \\mathbb { E } _ { a } \\left[ ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right] _ { N N } } \\\\ { \\displaystyle = \\mathbb { E } _ { a } \\bigg [ \\Big ( - \\xi - a ^ { \\top } ( A ^ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ) ^ { - 1 } \\bigg ] . } \\end{array}", + "type": "interline_equation", + "image_path": "f176ad0980439202a213c14bd6a4d34054a238f6980d3ef3c754459f486fb731.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 134, + 414, + 477, + 430.6666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 134, + 430.6666666666667, + 477, + 447.33333333333337 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 134, + 447.33333333333337, + 477, + 464.00000000000006 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 506, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 276, + 486 + ], + "score": 1.0, + "content": "To obtain an asymptotic description of", + "type": "text" + }, + { + "bbox": [ + 276, + 475, + 297, + 486 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 472, + 506, + 486 + ], + "score": 1.0, + "content": ", we perform the orthonormal decomposition on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 177, + 498 + ], + "score": 1.0, + "content": "the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 178, + 486, + 186, + 496 + ], + "score": 0.79, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 483, + 370, + 498 + ], + "score": 1.0, + "content": "introduced in Cheng and Singer (2013):", + "type": "text" + }, + { + "bbox": [ + 370, + 484, + 471, + 496 + ], + "score": 0.91, + "content": "\\varphi ( x ) \\ = \\ a _ { 1 } x + \\varphi _ { \\perp } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 483, + 507, + 498 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 495, + 507, + 508 + ], + "spans": [ + { + "bbox": [ + 107, + 495, + 207, + 508 + ], + "score": 0.92, + "content": "a _ { 1 } = \\mathbb { E } _ { x \\sim \\mathcal { N } ( 0 , 1 ) } [ x \\varphi ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 495, + 254, + 508 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 254, + 495, + 322, + 507 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ( \\bar { 1 } \\leq i \\leq \\bar { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 495, + 507, + 508 + ], + "score": 1.0, + "content": ", we perform the following orthonormal de-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 505, + 447, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 249, + 519 + ], + "score": 1.0, + "content": "composition (along the direction of", + "type": "text" + }, + { + "bbox": [ + 250, + 508, + 263, + 517 + ], + "score": 0.88, + "content": "{ \\bf { x } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 505, + 344, + 519 + ], + "score": 1.0, + "content": "and the direction of", + "type": "text" + }, + { + "bbox": [ + 344, + 506, + 357, + 517 + ], + "score": 0.89, + "content": "\\tilde { \\mathbf { \\pmb { w } } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 505, + 426, + 519 + ], + "score": 1.0, + "content": "perpendicular to", + "type": "text" + }, + { + "bbox": [ + 426, + 507, + 439, + 517 + ], + "score": 0.85, + "content": "{ \\mathbf { \\mathcal { x } } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 505, + 447, + 519 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 472, + 507, + 519 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 523, + 394, + 555 + ], + "lines": [ + { + "bbox": [ + 216, + 523, + 394, + 555 + ], + "spans": [ + { + "bbox": [ + 216, + 523, + 394, + 555 + ], + "score": 0.93, + "content": "\\pmb { w } _ { i } = \\underbrace { \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { n } } _ { \\eta _ { i } } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } = \\eta _ { i } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } .", + "type": "interline_equation", + "image_path": "cee414213d3b40bac825a65641fde07e4ad883d5b7cb0bba8c495de72d91b4ea.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 523, + 394, + 539.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 216, + 539.0, + 394, + 555.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 294, + 572 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 295, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 276, + 573 + ], + "score": 1.0, + "content": "We can thus simplify the activation vector", + "type": "text" + }, + { + "bbox": [ + 276, + 563, + 283, + 570 + ], + "score": 0.8, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 560, + 295, + 573 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 560, + 295, + 573 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 577, + 449, + 733 + ], + "lines": [ + { + "bbox": [ + 162, + 577, + 449, + 733 + ], + "spans": [ + { + "bbox": [ + 162, + 577, + 449, + 733 + ], + "score": 0.5, + "content": "\\begin{array} { r l } { a ^ { \\top } = \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { 1 } ) } & { \\cdots \\cdot \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { = \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { 1 } } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { h } \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { + \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { 1 } ) } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdot \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] . } \\end{array}", + "type": "interline_equation", + "image_path": "dfd9392f1a095abc0522233c533911ce1fef4318c4702efd92bf8bde56cf40af.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 162, + 577, + 449, + 629.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 162, + 629.0, + 449, + 681.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 162, + 681.0, + 449, + 733.0 + ], + "spans": [], + "index": 32 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 266, + 95 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 266, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 137, + 95 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 138, + 82, + 263, + 94 + ], + "score": 0.89, + "content": "1 \\leq i \\neq j \\leq h , 1 \\leq k \\leq n - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 82, + 266, + 95 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 175, + 96, + 435, + 137 + ], + "lines": [ + { + "bbox": [ + 175, + 96, + 435, + 137 + ], + "spans": [ + { + "bbox": [ + 175, + 96, + 435, + 137 + ], + "score": 0.92, + "content": "Q _ { i j } = \\left( \\eta _ { i } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { i } \\right) ^ { \\top } \\left( \\eta _ { j } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { j } \\right) = \\eta _ { i } \\eta _ { j } + \\underbrace { \\tilde { \\pmb { w } } _ { i } ^ { \\top } \\tilde { \\pmb { w } } _ { j } } _ { \\tilde { Q } _ { i j } } .", + "type": "interline_equation", + "image_path": "181ef276eb865e4cd29ca48f27514054f218948bd7346c449c4145ffd515e926.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 175, + 96, + 435, + 109.66666666666667 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 175, + 109.66666666666667, + 435, + 123.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 175, + 123.33333333333334, + 435, + 137.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 321, + 152 + ], + "lines": [ + { + "bbox": [ + 106, + 139, + 322, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 309, + 153 + ], + "score": 1.0, + "content": "Similarly we decompose the activation function in", + "type": "text" + }, + { + "bbox": [ + 309, + 141, + 317, + 150 + ], + "score": 0.81, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 139, + 322, + 153 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 154, + 514, + 227 + ], + "lines": [ + { + "bbox": [ + 112, + 154, + 514, + 227 + ], + "spans": [ + { + "bbox": [ + 112, + 154, + 514, + 227 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle { S _ { i k } = \\frac { 1 } { \\sqrt n } \\varphi \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) = \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } a _ { 1 } \\tilde { w } _ { i } ^ { \\top } x _ { k } + \\frac { 1 } { \\sqrt n } \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) } } \\\\ { { \\displaystyle { \\quad = \\frac { 1 } { \\sqrt n } \\varphi ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) + \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } \\left[ \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) - \\varphi _ { \\bot } ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) \\right] } } \\cdot { \\displaystyle ( 7 4 ) } } } \\end{array}", + "type": "interline_equation", + "image_path": "618584fd5f5e9c29878a555c52c0ac3adb13e0962d66ea03df40eb535d077628.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 112, + 154, + 514, + 178.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 112, + 178.33333333333334, + 514, + 202.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 202.66666666666669, + 514, + 227.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 231, + 313, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 314, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 301, + 244 + ], + "score": 1.0, + "content": "We thus have an equivalent expression of matrix", + "type": "text" + }, + { + "bbox": [ + 302, + 231, + 314, + 242 + ], + "score": 0.87, + "content": "A _ { * }", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 244, + 500, + 336 + ], + "lines": [ + { + "bbox": [ + 108, + 244, + 500, + 336 + ], + "spans": [ + { + "bbox": [ + 108, + 244, + 500, + 336 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { t \\eta \\eta ^ { \\top } } & { a _ { 1 } \\eta u ^ { \\top } } \\\\ { a _ { 1 } u \\eta ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } \\\\ & { \\quad = \\underbrace { \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } _ { \\tilde { A } _ { * } } + \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] } _ { U } \\underbrace { \\left[ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} \\right] } _ { C } \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] ^ { \\top } } _ { U ^ { \\top } } + \\underbrace { \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { \\mathbf { 0 } _ { n - 1 } } \\end{array} \\right] } _ { E } } \\\\ & { \\quad = \\tilde { A } _ { * } + U C U ^ { \\top } + E . } \\end{array}", + "type": "interline_equation", + "image_path": "7f605ae8ef8e552482a0d07a320a468883d23f6f7287b64cee9511c5716f210a.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 108, + 244, + 500, + 274.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 274.6666666666667, + 500, + 305.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 108, + 305.33333333333337, + 500, + 336.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 339, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 311, + 352 + ], + "score": 1.0, + "content": "By argument similar to (Hastie et al., 2019, B.1.2),", + "type": "text" + }, + { + "bbox": [ + 311, + 340, + 320, + 350 + ], + "score": 0.81, + "content": "E", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 339, + 395, + 352 + ], + "score": 1.0, + "content": "diminishes to 0 as", + "type": "text" + }, + { + "bbox": [ + 395, + 342, + 428, + 350 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "with respect to the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 464, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 411, + 364 + ], + "score": 1.0, + "content": "Frobenius norm, therefore by the Woodbury’s identity and the expression of", + "type": "text" + }, + { + "bbox": [ + 412, + 353, + 433, + 363 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 351, + 464, + 364 + ], + "score": 1.0, + "content": "in (70)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 365, + 514, + 484 + ], + "lines": [ + { + "bbox": [ + 111, + 365, + 514, + 484 + ], + "spans": [ + { + "bbox": [ + 111, + 365, + 514, + 484 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { n _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } + U C U ^ { \\top } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } } _ { u } + } \\\\ & { \\qquad \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U } _ { v ^ { \\top } } \\underbrace { ( C ^ { - 1 } + U ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U ) ^ { - 1 } } _ { S } \\underbrace { U ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a } _ { v } \\bigg ) ^ { - 1 } \\bigg ] } \\end{array}", + "type": "interline_equation", + "image_path": "4bb68b057174ebc232ea9b5b5c2a800e4f5a9317e52bb2d552e4465104d95399.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 111, + 365, + 514, + 404.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 404.6666666666667, + 514, + 444.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 111, + 444.33333333333337, + 514, + 484.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 318, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 494, + 317, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 190, + 512 + ], + "score": 1.0, + "content": "We bound each term", + "type": "text" + }, + { + "bbox": [ + 191, + 498, + 219, + 509 + ], + "score": 0.92, + "content": "u , v , S", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 494, + 267, + 512 + ], + "score": 1.0, + "content": "to compute", + "type": "text" + }, + { + "bbox": [ + 267, + 499, + 289, + 510 + ], + "score": 0.88, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 494, + 309, + 512 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 310, + 500, + 317, + 507 + ], + "score": 0.78, + "content": "u", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 513, + 498, + 534 + ], + "lines": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "spans": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\mathbb { E } _ { a } u = \\mathbb { E } _ { a } \\Big [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ] = \\mathrm { t r } \\left( \\mathbb { E } _ { s } \\big [ s s ^ { \\top } \\big ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . { h } , 1 . { h } ] } ^ { - 1 } \\right) = b \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) , } \\end{array}", + "type": "interline_equation", + "image_path": "6465813f446dd9de9e8b3f200a891ab0a68ab06f69e90248d875f54672e8fcf2.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 548, + 505, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 548, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 133, + 564 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 549, + 375, + 563 + ], + "score": 0.88, + "content": "b = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ \\varphi ( \\boldsymbol { x } ) ^ { 2 } ] = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ ( \\phi ( \\boldsymbol { x } ) - \\mathbb { E } \\phi ( \\boldsymbol { x } ) ) ^ { 2 } ] = r _ { ! }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 548, + 506, + 564 + ], + "score": 1.0, + "content": ". From a standard concentration", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 561, + 309, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 256, + 573 + ], + "score": 1.0, + "content": "of measure argument we have that as", + "type": "text" + }, + { + "bbox": [ + 257, + 562, + 309, + 572 + ], + "score": 0.89, + "content": "n , h , d \\infty", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 576, + 368, + 590 + ], + "lines": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "spans": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "score": 0.89, + "content": "u \\mathbb { E } _ { \\pmb { a } } u = r \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) .", + "type": "interline_equation", + "image_path": "3de955f14284b3bc42d4fd78c3cbe1cbe6c04cd4be13063cb696def49ea7fdd1.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 276, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 277, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 140, + 607 + ], + "score": 1.0, + "content": "And for", + "type": "text" + }, + { + "bbox": [ + 140, + 595, + 147, + 603 + ], + "score": 0.76, + "content": "\\textbf { { v } }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 590, + 189, + 607 + ], + "score": 1.0, + "content": "(note that", + "type": "text" + }, + { + "bbox": [ + 189, + 594, + 198, + 603 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 590, + 264, + 607 + ], + "score": 1.0, + "content": "is dependent on", + "type": "text" + }, + { + "bbox": [ + 264, + 595, + 272, + 603 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 590, + 277, + 607 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 608, + 498, + 671 + ], + "lines": [ + { + "bbox": [ + 110, + 608, + 498, + 671 + ], + "spans": [ + { + "bbox": [ + 110, + 608, + 498, + 671 + ], + "score": 0.91, + "content": "\\begin{array} { r l } { \\mathbb { E } _ { a } \\boldsymbol { v } ^ { \\top } = \\mathbb { E } _ { a } [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } \\boldsymbol { U } ] = \\mathbb { E } _ { a } [ [ \\boldsymbol { s } ^ { \\top } , \\boldsymbol { 0 } _ { n - 1 } ^ { \\top } ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\boldsymbol { \\eta } } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { \\boldsymbol { u } } \\end{array} ] ] } & { } \\\\ { = [ \\mathbb { E } _ { s } [ \\boldsymbol { s } ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\boldsymbol { \\eta } ] ] } & { 0 ] = [ \\underbrace { \\mathrm { t r } ( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\boldsymbol { \\eta } \\boldsymbol { s } ^ { \\top } ] ) } _ { \\boldsymbol { v } } } & { 0 ] . } \\end{array}", + "type": "interline_equation", + "image_path": "c83a48b452e463c3496ecb37ceb21e3ebf147dda5286c7d6effeb35823e84ee6.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 110, + 608, + 498, + 629.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 110, + 629.0, + 498, + 650.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 110, + 650.0, + 498, + 671.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 685, + 472, + 705 + ], + "lines": [ + { + "bbox": [ + 103, + 682, + 476, + 708 + ], + "spans": [ + { + "bbox": [ + 103, + 682, + 132, + 708 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 686, + 414, + 705 + ], + "score": 0.83, + "content": "\\begin{array} { r } { v = \\mathrm { t r } \\left( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . . h , 1 . . h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\eta s ^ { \\top } ] \\right) = a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 682, + 476, + 708 + ], + "score": 1.0, + "content": "We thus have", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 709, + 396, + 736 + ], + "lines": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "spans": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "score": 0.92, + "content": "v \\mathbb { E } _ { a } v = [ a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\quad 0 ] .", + "type": "interline_equation", + "image_path": "be198bd9a9c589f5ffdfb83ba80d6f26c337394810f47059d0839259a99f3a4f.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "spans": [], + "index": 27 + } + ] + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 266, + 95 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 266, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 137, + 95 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 138, + 82, + 263, + 94 + ], + "score": 0.89, + "content": "1 \\leq i \\neq j \\leq h , 1 \\leq k \\leq n - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 82, + 266, + 95 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 82, + 266, + 95 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 175, + 96, + 435, + 137 + ], + "lines": [ + { + "bbox": [ + 175, + 96, + 435, + 137 + ], + "spans": [ + { + "bbox": [ + 175, + 96, + 435, + 137 + ], + "score": 0.92, + "content": "Q _ { i j } = \\left( \\eta _ { i } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { i } \\right) ^ { \\top } \\left( \\eta _ { j } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { j } \\right) = \\eta _ { i } \\eta _ { j } + \\underbrace { \\tilde { \\pmb { w } } _ { i } ^ { \\top } \\tilde { \\pmb { w } } _ { j } } _ { \\tilde { Q } _ { i j } } .", + "type": "interline_equation", + "image_path": "181ef276eb865e4cd29ca48f27514054f218948bd7346c449c4145ffd515e926.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 175, + 96, + 435, + 109.66666666666667 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 175, + 109.66666666666667, + 435, + 123.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 175, + 123.33333333333334, + 435, + 137.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 321, + 152 + ], + "lines": [ + { + "bbox": [ + 106, + 139, + 322, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 309, + 153 + ], + "score": 1.0, + "content": "Similarly we decompose the activation function in", + "type": "text" + }, + { + "bbox": [ + 309, + 141, + 317, + 150 + ], + "score": 0.81, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 139, + 322, + 153 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 139, + 322, + 153 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 154, + 514, + 227 + ], + "lines": [ + { + "bbox": [ + 112, + 154, + 514, + 227 + ], + "spans": [ + { + "bbox": [ + 112, + 154, + 514, + 227 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle { S _ { i k } = \\frac { 1 } { \\sqrt n } \\varphi \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) = \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } a _ { 1 } \\tilde { w } _ { i } ^ { \\top } x _ { k } + \\frac { 1 } { \\sqrt n } \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) } } \\\\ { { \\displaystyle { \\quad = \\frac { 1 } { \\sqrt n } \\varphi ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) + \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } \\left[ \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) - \\varphi _ { \\bot } ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) \\right] } } \\cdot { \\displaystyle ( 7 4 ) } } } \\end{array}", + "type": "interline_equation", + "image_path": "618584fd5f5e9c29878a555c52c0ac3adb13e0962d66ea03df40eb535d077628.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 112, + 154, + 514, + 178.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 112, + 178.33333333333334, + 514, + 202.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 202.66666666666669, + 514, + 227.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 231, + 313, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 314, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 301, + 244 + ], + "score": 1.0, + "content": "We thus have an equivalent expression of matrix", + "type": "text" + }, + { + "bbox": [ + 302, + 231, + 314, + 242 + ], + "score": 0.87, + "content": "A _ { * }", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 230, + 314, + 244 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 244, + 500, + 336 + ], + "lines": [ + { + "bbox": [ + 108, + 244, + 500, + 336 + ], + "spans": [ + { + "bbox": [ + 108, + 244, + 500, + 336 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { t \\eta \\eta ^ { \\top } } & { a _ { 1 } \\eta u ^ { \\top } } \\\\ { a _ { 1 } u \\eta ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } \\\\ & { \\quad = \\underbrace { \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } _ { \\tilde { A } _ { * } } + \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] } _ { U } \\underbrace { \\left[ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} \\right] } _ { C } \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] ^ { \\top } } _ { U ^ { \\top } } + \\underbrace { \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { \\mathbf { 0 } _ { n - 1 } } \\end{array} \\right] } _ { E } } \\\\ & { \\quad = \\tilde { A } _ { * } + U C U ^ { \\top } + E . } \\end{array}", + "type": "interline_equation", + "image_path": "7f605ae8ef8e552482a0d07a320a468883d23f6f7287b64cee9511c5716f210a.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 108, + 244, + 500, + 274.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 274.6666666666667, + 500, + 305.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 108, + 305.33333333333337, + 500, + 336.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 339, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 311, + 352 + ], + "score": 1.0, + "content": "By argument similar to (Hastie et al., 2019, B.1.2),", + "type": "text" + }, + { + "bbox": [ + 311, + 340, + 320, + 350 + ], + "score": 0.81, + "content": "E", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 339, + 395, + 352 + ], + "score": 1.0, + "content": "diminishes to 0 as", + "type": "text" + }, + { + "bbox": [ + 395, + 342, + 428, + 350 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "with respect to the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 464, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 411, + 364 + ], + "score": 1.0, + "content": "Frobenius norm, therefore by the Woodbury’s identity and the expression of", + "type": "text" + }, + { + "bbox": [ + 412, + 353, + 433, + 363 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 351, + 464, + 364 + ], + "score": 1.0, + "content": "in (70)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 339, + 505, + 364 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 365, + 514, + 484 + ], + "lines": [ + { + "bbox": [ + 111, + 365, + 514, + 484 + ], + "spans": [ + { + "bbox": [ + 111, + 365, + 514, + 484 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { n _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } + U C U ^ { \\top } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } } _ { u } + } \\\\ & { \\qquad \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U } _ { v ^ { \\top } } \\underbrace { ( C ^ { - 1 } + U ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U ) ^ { - 1 } } _ { S } \\underbrace { U ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a } _ { v } \\bigg ) ^ { - 1 } \\bigg ] } \\end{array}", + "type": "interline_equation", + "image_path": "4bb68b057174ebc232ea9b5b5c2a800e4f5a9317e52bb2d552e4465104d95399.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 111, + 365, + 514, + 404.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 404.6666666666667, + 514, + 444.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 111, + 444.33333333333337, + 514, + 484.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 318, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 494, + 317, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 190, + 512 + ], + "score": 1.0, + "content": "We bound each term", + "type": "text" + }, + { + "bbox": [ + 191, + 498, + 219, + 509 + ], + "score": 0.92, + "content": "u , v , S", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 494, + 267, + 512 + ], + "score": 1.0, + "content": "to compute", + "type": "text" + }, + { + "bbox": [ + 267, + 499, + 289, + 510 + ], + "score": 0.88, + "content": "m _ { 2 , n }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 494, + 309, + 512 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 310, + 500, + 317, + 507 + ], + "score": 0.78, + "content": "u", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 494, + 317, + 512 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 513, + 498, + 534 + ], + "lines": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "spans": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\mathbb { E } _ { a } u = \\mathbb { E } _ { a } \\Big [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ] = \\mathrm { t r } \\left( \\mathbb { E } _ { s } \\big [ s s ^ { \\top } \\big ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . { h } , 1 . { h } ] } ^ { - 1 } \\right) = b \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) , } \\end{array}", + "type": "interline_equation", + "image_path": "6465813f446dd9de9e8b3f200a891ab0a68ab06f69e90248d875f54672e8fcf2.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 109, + 513, + 498, + 534 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 548, + 505, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 548, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 133, + 564 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 549, + 375, + 563 + ], + "score": 0.88, + "content": "b = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ \\varphi ( \\boldsymbol { x } ) ^ { 2 } ] = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ ( \\phi ( \\boldsymbol { x } ) - \\mathbb { E } \\phi ( \\boldsymbol { x } ) ) ^ { 2 } ] = r _ { ! }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 548, + 506, + 564 + ], + "score": 1.0, + "content": ". From a standard concentration", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 561, + 309, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 256, + 573 + ], + "score": 1.0, + "content": "of measure argument we have that as", + "type": "text" + }, + { + "bbox": [ + 257, + 562, + 309, + 572 + ], + "score": 0.89, + "content": "n , h , d \\infty", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 548, + 506, + 573 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 576, + 368, + 590 + ], + "lines": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "spans": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "score": 0.89, + "content": "u \\mathbb { E } _ { \\pmb { a } } u = r \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) .", + "type": "interline_equation", + "image_path": "3de955f14284b3bc42d4fd78c3cbe1cbe6c04cd4be13063cb696def49ea7fdd1.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 243, + 576, + 368, + 590 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 276, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 277, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 140, + 607 + ], + "score": 1.0, + "content": "And for", + "type": "text" + }, + { + "bbox": [ + 140, + 595, + 147, + 603 + ], + "score": 0.76, + "content": "\\textbf { { v } }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 590, + 189, + 607 + ], + "score": 1.0, + "content": "(note that", + "type": "text" + }, + { + "bbox": [ + 189, + 594, + 198, + 603 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 590, + 264, + 607 + ], + "score": 1.0, + "content": "is dependent on", + "type": "text" + }, + { + "bbox": [ + 264, + 595, + 272, + 603 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 590, + 277, + 607 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 590, + 277, + 607 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 608, + 498, + 671 + ], + "lines": [ + { + "bbox": [ + 110, + 608, + 498, + 671 + ], + "spans": [ + { + "bbox": [ + 110, + 608, + 498, + 671 + ], + "score": 0.91, + "content": "\\begin{array} { r l } { \\mathbb { E } _ { a } \\boldsymbol { v } ^ { \\top } = \\mathbb { E } _ { a } [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } \\boldsymbol { U } ] = \\mathbb { E } _ { a } [ [ \\boldsymbol { s } ^ { \\top } , \\boldsymbol { 0 } _ { n - 1 } ^ { \\top } ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\boldsymbol { \\eta } } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { \\boldsymbol { u } } \\end{array} ] ] } & { } \\\\ { = [ \\mathbb { E } _ { s } [ \\boldsymbol { s } ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\boldsymbol { \\eta } ] ] } & { 0 ] = [ \\underbrace { \\mathrm { t r } ( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\boldsymbol { \\eta } \\boldsymbol { s } ^ { \\top } ] ) } _ { \\boldsymbol { v } } } & { 0 ] . } \\end{array}", + "type": "interline_equation", + "image_path": "c83a48b452e463c3496ecb37ceb21e3ebf147dda5286c7d6effeb35823e84ee6.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 110, + 608, + 498, + 629.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 110, + 629.0, + 498, + 650.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 110, + 650.0, + 498, + 671.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 685, + 472, + 705 + ], + "lines": [ + { + "bbox": [ + 103, + 682, + 476, + 708 + ], + "spans": [ + { + "bbox": [ + 103, + 682, + 132, + 708 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 686, + 414, + 705 + ], + "score": 0.83, + "content": "\\begin{array} { r } { v = \\mathrm { t r } \\left( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . . h , 1 . . h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\eta s ^ { \\top } ] \\right) = a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 682, + 476, + 708 + ], + "score": 1.0, + "content": "We thus have", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 103, + 682, + 476, + 708 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 709, + 396, + 736 + ], + "lines": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "spans": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "score": 0.92, + "content": "v \\mathbb { E } _ { a } v = [ a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\quad 0 ] .", + "type": "interline_equation", + "image_path": "be198bd9a9c589f5ffdfb83ba80d6f26c337394810f47059d0839259a99f3a4f.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 215, + 709, + 396, + 736 + ], + "spans": [], + "index": 27 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 247, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 247, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 117, + 96 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 118, + 83, + 169, + 94 + ], + "score": 0.93, + "content": "n , d , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 81, + 235, + 96 + ], + "score": 1.0, + "content": ". And finally for", + "type": "text" + }, + { + "bbox": [ + 235, + 83, + 243, + 92 + ], + "score": 0.78, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 81, + 247, + 96 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 99, + 514, + 287 + ], + "lines": [ + { + "bbox": [ + 111, + 99, + 514, + 287 + ], + "spans": [ + { + "bbox": [ + 111, + 99, + 514, + 287 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\iota _ { \\alpha } S ^ { - 1 } = \\mathbb { E } _ { \\alpha } [ C ^ { - 1 } + U ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } U ] } \\\\ { = } & { [ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} ] ^ { - 1 } + \\mathbb { E } _ { \\alpha } [ [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ] } \\\\ { = } & { [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + \\mathbb { E } _ { \\alpha } [ \\begin{array} { c c } { \\eta ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\eta } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } ] } \\\\ { 0 } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } } \\end{array} ] } \\\\ { = } & [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + [ \\begin{array} { c c } { \\mathbb { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\mathbb { E } _ { \\alpha } | \\eta \\eta ^ { \\top } ) } & \\mathrm { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } | \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "e0317d45bcbd655a5a3ab8fc2a92a602f4d095bf3f002fe76c8b9ee801e910ac.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 111, + 99, + 514, + 161.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 161.66666666666666, + 514, + 224.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 224.33333333333331, + 514, + 287.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 290, + 218, + 302 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 219, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 162, + 303 + ], + "score": 1.0, + "content": "And hence as", + "type": "text" + }, + { + "bbox": [ + 163, + 290, + 215, + 302 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 289, + 219, + 303 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 167, + 307, + 444, + 335 + ], + "lines": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "spans": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "score": 0.91, + "content": "S ^ { - 1 } \\to \\mathbb { E } _ { a } S ^ { - 1 } = \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "77d228e451b8149b74f1076da3e7f58b07ed4bf3f12d8722b505970c17cd28d9.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 458, + 352 + ], + "lines": [ + { + "bbox": [ + 104, + 336, + 454, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 336, + 433, + 357 + ], + "score": 1.0, + "content": "Therefore by combining (78), (80), (82), we arrive at the following expression on", + "type": "text" + }, + { + "bbox": [ + 433, + 343, + 454, + 353 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 358, + 511, + 451 + ], + "lines": [ + { + "bbox": [ + 111, + 358, + 511, + 451 + ], + "spans": [ + { + "bbox": [ + 111, + 358, + 511, + 451 + ], + "score": 0.92, + "content": "\\begin{array} { r l r } & { } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) \\to \\mathbb { E } _ { a } \\Big [ \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } \\Big ] \\to \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } } \\\\ & { \\to \\Big ( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\Big ( a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } \\Big ) ^ { 2 } \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] _ { [ 1 , 1 ] } ^ { - 1 } \\Big ) ^ { - 1 } } \\\\ & { } & { = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\frac { \\gamma _ { 2 } a _ { 1 } ^ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) } { m _ { 1 , n } \\left( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } . } \\end{array}", + "type": "interline_equation", + "image_path": "2e463b72ecdc08c758158ed164610061d27ce21a56c00e98a0c5e918bcb14259.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 111, + 358, + 511, + 389.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 389.0, + 511, + 420.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 420.0, + 511, + 451.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 280, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 281, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 214, + 470 + ], + "score": 1.0, + "content": "Similarly we can calculate", + "type": "text" + }, + { + "bbox": [ + 215, + 456, + 268, + 469 + ], + "score": 0.93, + "content": "m _ { 1 , n } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 453, + 281, + 470 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 475, + 500, + 523 + ], + "lines": [ + { + "bbox": [ + 104, + 475, + 500, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 475, + 500, + 523 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\to } \\\\ & { \\left( - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 , n } - r m _ { 2 , n } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 } - \\tau ) - 2 \\tau a _ { 1 } ^ { 2 } m _ { 1 , n } m _ { 2 , n } + a _ { 1 } ^ { 4 } m _ { 1 , n } m _ { 2 , n } ^ { 2 } } { m _ { 1 , n } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } } \\end{array}", + "type": "interline_equation", + "image_path": "7168d04b9d261062afc1d4d9ed42de994590f68307a13b5ac077e17e14cd5241.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 104, + 475, + 500, + 491.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 104, + 491.0, + 500, + 507.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 104, + 507.0, + 500, + 523.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 433, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 434, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 434, + 554 + ], + "score": 1.0, + "content": "Uniqueness in (84)(83) follows from (Hastie et al., 2019, Sec B.1) and is omitted.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 107, + 572, + 241, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 242, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 242, + 586 + ], + "score": 1.0, + "content": "C.6 PROOF OF COROLLARY 5", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 399, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 399, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 234, + 608 + ], + "score": 1.0, + "content": "In this section we take the limit", + "type": "text" + }, + { + "bbox": [ + 234, + 595, + 270, + 605 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 591, + 399, + 608 + ], + "score": 1.0, + "content": ". In this case (8), (9) simplify to", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 610, + 385, + 646 + ], + "lines": [ + { + "bbox": [ + 226, + 610, + 385, + 646 + ], + "spans": [ + { + "bbox": [ + 226, + 610, + 385, + 646 + ], + "score": 0.91, + "content": "\\begin{array} { l } { { m _ { 2 } = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 } \\right) ^ { - 1 } , } } \\\\ { { \\nonumber } } \\\\ { { m _ { 1 } = \\left( - \\xi - \\rho - \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - r m _ { 2 } \\right) ^ { - 1 } . } } \\end{array}", + "type": "interline_equation", + "image_path": "9cfbc26107344339a87cccee67503863e23a242db4b518e43119b45a91cd3d13.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 226, + 610, + 385, + 628.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 226, + 628.0, + 385, + 646.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 651, + 457, + 665 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 457, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 152, + 666 + ], + "score": 1.0, + "content": "Recall that", + "type": "text" + }, + { + "bbox": [ + 152, + 651, + 320, + 664 + ], + "score": 0.92, + "content": "m ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 } ( \\xi , \\rho , \\tau ) + m _ { 2 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 650, + 457, + 666 + ], + "score": 1.0, + "content": ". By taking the derivative we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 669, + 514, + 725 + ], + "lines": [ + { + "bbox": [ + 111, + 669, + 514, + 725 + ], + "spans": [ + { + "bbox": [ + 111, + 669, + 514, + 725 + ], + "score": 0.91, + "content": "\\begin{array} { l } { \\displaystyle - q ( \\xi ) = \\frac { \\partial } { \\partial x } m ( \\xi , r x , t x ) \\Big | _ { x = 0 } = \\left. r \\frac { \\partial } { \\partial \\rho } m ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } } \\\\ { \\displaystyle \\quad = \\left. r \\gamma _ { 2 } \\frac { \\partial } { \\partial \\rho } m _ { 1 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. r \\frac { \\partial } { \\partial \\rho } m _ { 2 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\gamma _ { 2 } \\frac { \\partial } { \\partial \\tau } m _ { 1 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m _ { 2 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "48ede76d93106e807ad8bb633e222d0ffb6ee52649e86b9606a60815bb7ef8d4.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 111, + 669, + 514, + 687.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 687.6666666666666, + 514, + 706.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 111, + 706.3333333333333, + 514, + 724.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 16, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 541, + 505, + 552 + ], + "lines": [ + { + "bbox": [ + 496, + 543, + 504, + 552 + ], + "spans": [ + { + "bbox": [ + 496, + 543, + 504, + 552 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 247, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 247, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 117, + 96 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 118, + 83, + 169, + 94 + ], + "score": 0.93, + "content": "n , d , p \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 81, + 235, + 96 + ], + "score": 1.0, + "content": ". And finally for", + "type": "text" + }, + { + "bbox": [ + 235, + 83, + 243, + 92 + ], + "score": 0.78, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 81, + 247, + 96 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 81, + 247, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 99, + 514, + 287 + ], + "lines": [ + { + "bbox": [ + 111, + 99, + 514, + 287 + ], + "spans": [ + { + "bbox": [ + 111, + 99, + 514, + 287 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\iota _ { \\alpha } S ^ { - 1 } = \\mathbb { E } _ { \\alpha } [ C ^ { - 1 } + U ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } U ] } \\\\ { = } & { [ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} ] ^ { - 1 } + \\mathbb { E } _ { \\alpha } [ [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ] } \\\\ { = } & { [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + \\mathbb { E } _ { \\alpha } [ \\begin{array} { c c } { \\eta ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\eta } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } ] } \\\\ { 0 } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } } \\end{array} ] } \\\\ { = } & [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + [ \\begin{array} { c c } { \\mathbb { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\mathbb { E } _ { \\alpha } | \\eta \\eta ^ { \\top } ) } & \\mathrm { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } | \\end{array} \\end{array}", + "type": "interline_equation", + "image_path": "e0317d45bcbd655a5a3ab8fc2a92a602f4d095bf3f002fe76c8b9ee801e910ac.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 111, + 99, + 514, + 161.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 161.66666666666666, + 514, + 224.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 224.33333333333331, + 514, + 287.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 290, + 218, + 302 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 219, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 162, + 303 + ], + "score": 1.0, + "content": "And hence as", + "type": "text" + }, + { + "bbox": [ + 163, + 290, + 215, + 302 + ], + "score": 0.92, + "content": "n , d , h \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 289, + 219, + 303 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 289, + 219, + 303 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 167, + 307, + 444, + 335 + ], + "lines": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "spans": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "score": 0.91, + "content": "S ^ { - 1 } \\to \\mathbb { E } _ { a } S ^ { - 1 } = \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] .", + "type": "interline_equation", + "image_path": "77d228e451b8149b74f1076da3e7f58b07ed4bf3f12d8722b505970c17cd28d9.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 167, + 307, + 444, + 335 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 458, + 352 + ], + "lines": [ + { + "bbox": [ + 104, + 336, + 454, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 336, + 433, + 357 + ], + "score": 1.0, + "content": "Therefore by combining (78), (80), (82), we arrive at the following expression on", + "type": "text" + }, + { + "bbox": [ + 433, + 343, + 454, + 353 + ], + "score": 0.89, + "content": "m _ { 2 , n }", + "type": "inline_equation" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 104, + 336, + 454, + 357 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 358, + 511, + 451 + ], + "lines": [ + { + "bbox": [ + 111, + 358, + 511, + 451 + ], + "spans": [ + { + "bbox": [ + 111, + 358, + 511, + 451 + ], + "score": 0.92, + "content": "\\begin{array} { r l r } & { } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) \\to \\mathbb { E } _ { a } \\Big [ \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } \\Big ] \\to \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } } \\\\ & { \\to \\Big ( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\Big ( a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } \\Big ) ^ { 2 } \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] _ { [ 1 , 1 ] } ^ { - 1 } \\Big ) ^ { - 1 } } \\\\ & { } & { = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\frac { \\gamma _ { 2 } a _ { 1 } ^ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) } { m _ { 1 , n } \\left( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } . } \\end{array}", + "type": "interline_equation", + "image_path": "2e463b72ecdc08c758158ed164610061d27ce21a56c00e98a0c5e918bcb14259.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 111, + 358, + 511, + 389.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 389.0, + 511, + 420.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 420.0, + 511, + 451.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 280, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 281, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 214, + 470 + ], + "score": 1.0, + "content": "Similarly we can calculate", + "type": "text" + }, + { + "bbox": [ + 215, + 456, + 268, + 469 + ], + "score": 0.93, + "content": "m _ { 1 , n } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 453, + 281, + 470 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 453, + 281, + 470 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 475, + 500, + 523 + ], + "lines": [ + { + "bbox": [ + 104, + 475, + 500, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 475, + 500, + 523 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\to } \\\\ & { \\left( - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 , n } - r m _ { 2 , n } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 } - \\tau ) - 2 \\tau a _ { 1 } ^ { 2 } m _ { 1 , n } m _ { 2 , n } + a _ { 1 } ^ { 4 } m _ { 1 , n } m _ { 2 , n } ^ { 2 } } { m _ { 1 , n } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } } \\end{array}", + "type": "interline_equation", + "image_path": "7168d04b9d261062afc1d4d9ed42de994590f68307a13b5ac077e17e14cd5241.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 104, + 475, + 500, + 491.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 104, + 491.0, + 500, + 507.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 104, + 507.0, + 500, + 523.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 433, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 434, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 434, + 554 + ], + "score": 1.0, + "content": "Uniqueness in (84)(83) follows from (Hastie et al., 2019, Sec B.1) and is omitted.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 541, + 434, + 554 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 572, + 241, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 242, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 242, + 586 + ], + "score": 1.0, + "content": "C.6 PROOF OF COROLLARY 5", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 399, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 399, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 234, + 608 + ], + "score": 1.0, + "content": "In this section we take the limit", + "type": "text" + }, + { + "bbox": [ + 234, + 595, + 270, + 605 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 591, + 399, + 608 + ], + "score": 1.0, + "content": ". In this case (8), (9) simplify to", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 591, + 399, + 608 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 610, + 385, + 646 + ], + "lines": [ + { + "bbox": [ + 226, + 610, + 385, + 646 + ], + "spans": [ + { + "bbox": [ + 226, + 610, + 385, + 646 + ], + "score": 0.91, + "content": "\\begin{array} { l } { { m _ { 2 } = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 } \\right) ^ { - 1 } , } } \\\\ { { \\nonumber } } \\\\ { { m _ { 1 } = \\left( - \\xi - \\rho - \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - r m _ { 2 } \\right) ^ { - 1 } . } } \\end{array}", + "type": "interline_equation", + "image_path": "9cfbc26107344339a87cccee67503863e23a242db4b518e43119b45a91cd3d13.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 226, + 610, + 385, + 628.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 226, + 628.0, + 385, + 646.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 651, + 457, + 665 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 457, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 152, + 666 + ], + "score": 1.0, + "content": "Recall that", + "type": "text" + }, + { + "bbox": [ + 152, + 651, + 320, + 664 + ], + "score": 0.92, + "content": "m ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 } ( \\xi , \\rho , \\tau ) + m _ { 2 } ( \\xi , \\rho , \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 650, + 457, + 666 + ], + "score": 1.0, + "content": ". By taking the derivative we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 650, + 457, + 666 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 669, + 514, + 725 + ], + "lines": [ + { + "bbox": [ + 111, + 669, + 514, + 725 + ], + "spans": [ + { + "bbox": [ + 111, + 669, + 514, + 725 + ], + "score": 0.91, + "content": "\\begin{array} { l } { \\displaystyle - q ( \\xi ) = \\frac { \\partial } { \\partial x } m ( \\xi , r x , t x ) \\Big | _ { x = 0 } = \\left. r \\frac { \\partial } { \\partial \\rho } m ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } } \\\\ { \\displaystyle \\quad = \\left. r \\gamma _ { 2 } \\frac { \\partial } { \\partial \\rho } m _ { 1 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. r \\frac { \\partial } { \\partial \\rho } m _ { 2 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\gamma _ { 2 } \\frac { \\partial } { \\partial \\tau } m _ { 1 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m _ { 2 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "48ede76d93106e807ad8bb633e222d0ffb6ee52649e86b9606a60815bb7ef8d4.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 111, + 669, + 514, + 687.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 687.6666666666666, + 514, + 706.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 111, + 706.3333333333333, + 514, + 724.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Observe that (85), (86) constitutes a set of implicit functions. Thus differentiating the two functions", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 342, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 167, + 107 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 167, + 96, + 183, + 105 + ], + "score": 0.86, + "content": "\\tau , \\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 93, + 274, + 107 + ], + "score": 1.0, + "content": "and then substitute by", + "type": "text" + }, + { + "bbox": [ + 274, + 94, + 317, + 105 + ], + "score": 0.91, + "content": "\\rho = \\tau = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 93, + 342, + 107 + ], + "score": 1.0, + "content": "gives", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 110, + 467, + 159 + ], + "lines": [ + { + "bbox": [ + 144, + 110, + 467, + 159 + ], + "spans": [ + { + "bbox": [ + 144, + 110, + 467, + 159 + ], + "score": 0.94, + "content": "q ( \\xi ) = \\frac { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) - \\xi ^ { 2 } \\right) \\left( \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } + r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) } { 2 \\xi ^ { 2 } \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } } .", + "type": "interline_equation", + "image_path": "f3a30b56497382899bcdbcb36652db9ead678eaf172d9441fad1f710afe7a1c4.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 144, + 110, + 467, + 126.33333333333333 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 144, + 126.33333333333333, + 467, + 142.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 144, + 142.66666666666666, + 467, + 159.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 306, + 175 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 307, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 307, + 177 + ], + "score": 1.0, + "content": "Hence by (62) we obtain the asymptotic variance:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 180, + 426, + 208 + ], + "lines": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "spans": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "score": 0.94, + "content": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) = \\operatorname* { l i m } _ { \\xi \\to 0 } \\left( q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } \\right) = \\frac { 1 } { \\gamma _ { 2 } - 1 } .", + "type": "interline_equation", + "image_path": "86eeef0b3abd7ad987a9ea07a9ae6f290701e7389f424b75e6dfa7996e256ba1.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 385, + 224 + ], + "lines": [ + { + "bbox": [ + 107, + 211, + 385, + 226 + ], + "spans": [ + { + "bbox": [ + 107, + 211, + 214, + 226 + ], + "score": 1.0, + "content": "Combining the case where", + "type": "text" + }, + { + "bbox": [ + 215, + 213, + 244, + 223 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 211, + 385, + 226 + ], + "score": 1.0, + "content": "in Theorem 4 completes the proof.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 230, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 229, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 165, + 244 + ], + "score": 1.0, + "content": "Remark. For", + "type": "text" + }, + { + "bbox": [ + 165, + 230, + 242, + 243 + ], + "score": 0.88, + "content": "\\phi ( x ) = \\mathrm { R e L U } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 229, + 247, + 244 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 248, + 231, + 333, + 243 + ], + "score": 0.77, + "content": "c _ { 1 } = 1 / 2 - 1 / ( 2 \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 229, + 337, + 244 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 337, + 231, + 380, + 243 + ], + "score": 0.79, + "content": "c _ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 229, + 403, + 244 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 403, + 230, + 505, + 243 + ], + "score": 0.87, + "content": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 240, + 372, + 254 + ], + "spans": [ + { + "bbox": [ + 107, + 242, + 155, + 254 + ], + "score": 0.91, + "content": "\\log ( 1 + e ^ { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 240, + 273, + 254 + ], + "score": 1.0, + "content": ", numerical integration yields", + "type": "text" + }, + { + "bbox": [ + 273, + 243, + 324, + 253 + ], + "score": 0.83, + "content": "c _ { 1 } \\approx 0 . 2 7 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 240, + 329, + 254 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 329, + 242, + 367, + 254 + ], + "score": 0.89, + "content": "c _ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 240, + 372, + 254 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 241, + 278 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 243, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 243, + 279 + ], + "score": 1.0, + "content": "C.7 PROOF OF COROLLARY 6", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 108, + 286, + 279, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 278, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 251, + 300 + ], + "score": 1.0, + "content": "C.7.1 UNBOUNDED BIAS WHEN", + "type": "text" + }, + { + "bbox": [ + 252, + 288, + 278, + 298 + ], + "score": 0.73, + "content": "h = n", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 304, + 504, + 319 + ], + "lines": [ + { + "bbox": [ + 104, + 302, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 302, + 307, + 321 + ], + "score": 1.0, + "content": "From the bias-variance decomposition (7), the bias", + "type": "text" + }, + { + "bbox": [ + 308, + 307, + 316, + 316 + ], + "score": 0.86, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 302, + 367, + 321 + ], + "score": 1.0, + "content": "is written as", + "type": "text" + }, + { + "bbox": [ + 367, + 304, + 475, + 319 + ], + "score": 0.93, + "content": "\\begin{array} { r } { B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( Q _ { 1 } + Q _ { 2 } + I _ { d } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 302, + 506, + 321 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 160, + 324, + 451, + 339 + ], + "lines": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "spans": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "score": 0.87, + "content": "Q _ { 1 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } ; \\quad Q _ { 2 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } W ^ { \\top } .", + "type": "interline_equation", + "image_path": "d7c0e84d892b7cc2080851eaa9255a8cffe4c12a695558b64413bdd223c190de.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 507, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 135, + 360 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 135, + 347, + 168, + 356 + ], + "score": 0.89, + "content": "\\textit { h } = \\textit { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 344, + 289, + 360 + ], + "score": 1.0, + "content": ", due to the nonlinearity of", + "type": "text" + }, + { + "bbox": [ + 289, + 347, + 296, + 358 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 344, + 342, + 360 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 342, + 345, + 383, + 358 + ], + "score": 0.92, + "content": "\\phi ( W ^ { \\top } X )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 344, + 506, + 360 + ], + "score": 1.0, + "content": "is full rank a.s., and hence", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 355, + 379, + 371 + ], + "spans": [ + { + "bbox": [ + 107, + 357, + 227, + 370 + ], + "score": 0.9, + "content": "[ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } = [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 355, + 365, + 371 + ], + "score": 1.0, + "content": ". We have the following bound for", + "type": "text" + }, + { + "bbox": [ + 366, + 358, + 379, + 369 + ], + "score": 0.87, + "content": "Q _ { 1 }", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 373, + 514, + 426 + ], + "lines": [ + { + "bbox": [ + 111, + 373, + 514, + 426 + ], + "spans": [ + { + "bbox": [ + 111, + 373, + 514, + 426 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\overset { ! } { \\operatorname { t r } } ( Q _ { 1 } ) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } \\right) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\\\ { \\geq \\displaystyle \\frac { 1 } { d } \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) \\mathrm { t r } \\left( [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) = \\frac { \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } { n } \\mathrm { t r } \\left( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "5766ed0a2fd875e0e35d4aa77091b0681c5f63f7e856db1527c5fff9a055611d.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 373, + 514, + 390.6666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 111, + 390.6666666666667, + 514, + 408.33333333333337 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 408.33333333333337, + 514, + 426.00000000000006 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 506, + 456 + ], + "lines": [ + { + "bbox": [ + 104, + 429, + 507, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 133, + 446 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 133, + 433, + 145, + 443 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 429, + 167, + 446 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 167, + 431, + 195, + 445 + ], + "score": 0.93, + "content": "X / { \\sqrt { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 429, + 214, + 446 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 214, + 432, + 280, + 443 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { d \\times n } ~ = ~ \\mathbb { R } ^ { d \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 429, + 507, + 446 + ], + "score": 1.0, + "content": "follows the same distribution where each entry i.i.d.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 441, + 155, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 444, + 149, + 456 + ], + "score": 0.93, + "content": "\\mathcal { N } ( 0 , 1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 441, + 155, + 458 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 460, + 469, + 515 + ], + "lines": [ + { + "bbox": [ + 125, + 460, + 469, + 515 + ], + "spans": [ + { + "bbox": [ + 125, + 460, + 469, + 515 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { d \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } \\mathrm { t r } ( Q _ { 1 } ) \\geq \\frac { 1 } { n } \\mathrm { t r } ( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X ) \\sim \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot W ^ { \\top } W ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot ( I + Q ) ) = \\operatorname* { l i m } _ { \\xi 0 } - \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , x , 0 , x ) \\infty , } \\end{array}", + "type": "interline_equation", + "image_path": "9800d639b0ce67d163d79772a44df6e0175a04082a36777a81e4a888953acb7e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 125, + 460, + 469, + 478.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 125, + 478.3333333333333, + 469, + 496.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 125, + 496.66666666666663, + 469, + 515.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 519, + 506, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 366, + 532 + ], + "score": 1.0, + "content": "where in Section C.5 we have showed (91) is unbounded when", + "type": "text" + }, + { + "bbox": [ + 366, + 522, + 400, + 529 + ], + "score": 0.88, + "content": "n \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 519, + 505, + 532 + ], + "score": 1.0, + "content": ". Moreover, by (154) and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 529, + 500, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 206, + 543 + ], + "score": 1.0, + "content": "Weyl’s theorem we have", + "type": "text" + }, + { + "bbox": [ + 206, + 530, + 286, + 543 + ], + "score": 0.92, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 529, + 326, + 543 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + }, + { + "bbox": [ + 326, + 530, + 372, + 542 + ], + "score": 0.93, + "content": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 1 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 529, + 449, + 543 + ], + "score": 1.0, + "content": "is unbounded. For", + "type": "text" + }, + { + "bbox": [ + 449, + 531, + 495, + 542 + ], + "score": 0.92, + "content": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 2 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 529, + 500, + 543 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 547, + 469, + 571 + ], + "lines": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "spans": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "score": 0.91, + "content": "\\frac { 1 } { d } \\mathrm { t r } ( Q _ { 2 } ) = \\frac { 1 } { d } \\mathrm { t r } \\left( W ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) \\leq \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) ^ { - 1 } ) \\mathrm { t r } \\left( W ^ { \\top } X \\right) = O ( 1 ) .", + "type": "interline_equation", + "image_path": "f5847230779d421975019dfa8d98de3ee3f410a4e17b78e4f242cba552ba1bdf.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 575, + 323, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 575, + 324, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 167, + 589 + ], + "score": 1.0, + "content": "To sum up, for", + "type": "text" + }, + { + "bbox": [ + 167, + 577, + 200, + 586 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 575, + 218, + 589 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 218, + 576, + 249, + 587 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 575, + 285, + 589 + ], + "score": 1.0, + "content": "we have", + "type": "text" + }, + { + "bbox": [ + 286, + 576, + 320, + 586 + ], + "score": 0.9, + "content": "B \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 575, + 324, + 589 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 107, + 604, + 268, + 617 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 268, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 239, + 618 + ], + "score": 1.0, + "content": "C.7.2 BOUNDED BIAS WHEN", + "type": "text" + }, + { + "bbox": [ + 239, + 605, + 268, + 617 + ], + "score": 0.83, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 623, + 401, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 402, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 131, + 637 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 624, + 157, + 634 + ], + "score": 0.9, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 623, + 402, + 637 + ], + "score": 1.0, + "content": ", the two terms in the expression of the bias can be written as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 116, + 640, + 477, + 689 + ], + "lines": [ + { + "bbox": [ + 116, + 640, + 477, + 689 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 477, + 689 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { Q _ { 1 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } X ^ { \\top } , } \\\\ & { Q _ { 2 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) W ^ { \\top } . } \\end{array}", + "type": "interline_equation", + "image_path": "b65bbc4a7bf15d47244e002c33f957f922700ec73c4a07f13628851548b9fa01.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 116, + 640, + 477, + 656.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 116, + 656.3333333333334, + 477, + 672.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 116, + 672.6666666666667, + 477, + 689.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 693, + 183, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 691, + 183, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 183, + 705 + ], + "score": 1.0, + "content": "Therefore we have", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 707, + 514, + 735 + ], + "lines": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "spans": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "score": 0.88, + "content": "\\operatorname { I } ^ { \\mathrm { { T } } } ( Q _ { 1 } ) = 2 \\mathrm { { t r } } \\left( { \\frac { X ^ { \\top } X } { d } } { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\right)", + "type": "interline_equation", + "image_path": "802e384616b8ee310544d57e8871b3da8cb31d01dc153c6778a231feb4f86630.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 16, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 212, + 504, + 223 + ], + "lines": [ + { + "bbox": [ + 496, + 214, + 504, + 224 + ], + "spans": [ + { + "bbox": [ + 496, + 214, + 504, + 224 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 576, + 504, + 586 + ], + "lines": [ + { + "bbox": [ + 496, + 577, + 504, + 587 + ], + "spans": [ + { + "bbox": [ + 496, + 577, + 504, + 587 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Observe that (85), (86) constitutes a set of implicit functions. Thus differentiating the two functions", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 342, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 167, + 107 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 167, + 96, + 183, + 105 + ], + "score": 0.86, + "content": "\\tau , \\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 93, + 274, + 107 + ], + "score": 1.0, + "content": "and then substitute by", + "type": "text" + }, + { + "bbox": [ + 274, + 94, + 317, + 105 + ], + "score": 0.91, + "content": "\\rho = \\tau = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 93, + 342, + 107 + ], + "score": 1.0, + "content": "gives", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 107 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 110, + 467, + 159 + ], + "lines": [ + { + "bbox": [ + 144, + 110, + 467, + 159 + ], + "spans": [ + { + "bbox": [ + 144, + 110, + 467, + 159 + ], + "score": 0.94, + "content": "q ( \\xi ) = \\frac { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) - \\xi ^ { 2 } \\right) \\left( \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } + r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) } { 2 \\xi ^ { 2 } \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } } .", + "type": "interline_equation", + "image_path": "f3a30b56497382899bcdbcb36652db9ead678eaf172d9441fad1f710afe7a1c4.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 144, + 110, + 467, + 126.33333333333333 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 144, + 126.33333333333333, + 467, + 142.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 144, + 142.66666666666666, + 467, + 159.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 306, + 175 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 307, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 307, + 177 + ], + "score": 1.0, + "content": "Hence by (62) we obtain the asymptotic variance:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 162, + 307, + 177 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 180, + 426, + 208 + ], + "lines": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "spans": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "score": 0.94, + "content": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) = \\operatorname* { l i m } _ { \\xi \\to 0 } \\left( q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } \\right) = \\frac { 1 } { \\gamma _ { 2 } - 1 } .", + "type": "interline_equation", + "image_path": "86eeef0b3abd7ad987a9ea07a9ae6f290701e7389f424b75e6dfa7996e256ba1.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 184, + 180, + 426, + 208 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 385, + 224 + ], + "lines": [ + { + "bbox": [ + 107, + 211, + 385, + 226 + ], + "spans": [ + { + "bbox": [ + 107, + 211, + 214, + 226 + ], + "score": 1.0, + "content": "Combining the case where", + "type": "text" + }, + { + "bbox": [ + 215, + 213, + 244, + 223 + ], + "score": 0.92, + "content": "\\gamma _ { 2 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 211, + 385, + 226 + ], + "score": 1.0, + "content": "in Theorem 4 completes the proof.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 107, + 211, + 385, + 226 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 230, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 229, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 165, + 244 + ], + "score": 1.0, + "content": "Remark. For", + "type": "text" + }, + { + "bbox": [ + 165, + 230, + 242, + 243 + ], + "score": 0.88, + "content": "\\phi ( x ) = \\mathrm { R e L U } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 229, + 247, + 244 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 248, + 231, + 333, + 243 + ], + "score": 0.77, + "content": "c _ { 1 } = 1 / 2 - 1 / ( 2 \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 229, + 337, + 244 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 337, + 231, + 380, + 243 + ], + "score": 0.79, + "content": "c _ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 229, + 403, + 244 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 403, + 230, + 505, + 243 + ], + "score": 0.87, + "content": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 240, + 372, + 254 + ], + "spans": [ + { + "bbox": [ + 107, + 242, + 155, + 254 + ], + "score": 0.91, + "content": "\\log ( 1 + e ^ { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 240, + 273, + 254 + ], + "score": 1.0, + "content": ", numerical integration yields", + "type": "text" + }, + { + "bbox": [ + 273, + 243, + 324, + 253 + ], + "score": 0.83, + "content": "c _ { 1 } \\approx 0 . 2 7 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 240, + 329, + 254 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 329, + 242, + 367, + 254 + ], + "score": 0.89, + "content": "c _ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 240, + 372, + 254 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 229, + 505, + 254 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 241, + 278 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 243, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 243, + 279 + ], + "score": 1.0, + "content": "C.7 PROOF OF COROLLARY 6", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 108, + 286, + 279, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 278, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 251, + 300 + ], + "score": 1.0, + "content": "C.7.1 UNBOUNDED BIAS WHEN", + "type": "text" + }, + { + "bbox": [ + 252, + 288, + 278, + 298 + ], + "score": 0.73, + "content": "h = n", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 304, + 504, + 319 + ], + "lines": [ + { + "bbox": [ + 104, + 302, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 302, + 307, + 321 + ], + "score": 1.0, + "content": "From the bias-variance decomposition (7), the bias", + "type": "text" + }, + { + "bbox": [ + 308, + 307, + 316, + 316 + ], + "score": 0.86, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 302, + 367, + 321 + ], + "score": 1.0, + "content": "is written as", + "type": "text" + }, + { + "bbox": [ + 367, + 304, + 475, + 319 + ], + "score": 0.93, + "content": "\\begin{array} { r } { B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( Q _ { 1 } + Q _ { 2 } + I _ { d } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 302, + 506, + 321 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 302, + 506, + 321 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 160, + 324, + 451, + 339 + ], + "lines": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "spans": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "score": 0.87, + "content": "Q _ { 1 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } ; \\quad Q _ { 2 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } W ^ { \\top } .", + "type": "interline_equation", + "image_path": "d7c0e84d892b7cc2080851eaa9255a8cffe4c12a695558b64413bdd223c190de.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 160, + 324, + 451, + 339 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 507, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 135, + 360 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 135, + 347, + 168, + 356 + ], + "score": 0.89, + "content": "\\textit { h } = \\textit { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 344, + 289, + 360 + ], + "score": 1.0, + "content": ", due to the nonlinearity of", + "type": "text" + }, + { + "bbox": [ + 289, + 347, + 296, + 358 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 344, + 342, + 360 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 342, + 345, + 383, + 358 + ], + "score": 0.92, + "content": "\\phi ( W ^ { \\top } X )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 344, + 506, + 360 + ], + "score": 1.0, + "content": "is full rank a.s., and hence", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 355, + 379, + 371 + ], + "spans": [ + { + "bbox": [ + 107, + 357, + 227, + 370 + ], + "score": 0.9, + "content": "[ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } = [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 355, + 365, + 371 + ], + "score": 1.0, + "content": ". We have the following bound for", + "type": "text" + }, + { + "bbox": [ + 366, + 358, + 379, + 369 + ], + "score": 0.87, + "content": "Q _ { 1 }", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 344, + 506, + 371 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 373, + 514, + 426 + ], + "lines": [ + { + "bbox": [ + 111, + 373, + 514, + 426 + ], + "spans": [ + { + "bbox": [ + 111, + 373, + 514, + 426 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\overset { ! } { \\operatorname { t r } } ( Q _ { 1 } ) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } \\right) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\\\ { \\geq \\displaystyle \\frac { 1 } { d } \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) \\mathrm { t r } \\left( [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) = \\frac { \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } { n } \\mathrm { t r } \\left( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "5766ed0a2fd875e0e35d4aa77091b0681c5f63f7e856db1527c5fff9a055611d.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 373, + 514, + 390.6666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 111, + 390.6666666666667, + 514, + 408.33333333333337 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 408.33333333333337, + 514, + 426.00000000000006 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 506, + 456 + ], + "lines": [ + { + "bbox": [ + 104, + 429, + 507, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 133, + 446 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 133, + 433, + 145, + 443 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 429, + 167, + 446 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 167, + 431, + 195, + 445 + ], + "score": 0.93, + "content": "X / { \\sqrt { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 429, + 214, + 446 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 214, + 432, + 280, + 443 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { d \\times n } ~ = ~ \\mathbb { R } ^ { d \\times p }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 429, + 507, + 446 + ], + "score": 1.0, + "content": "follows the same distribution where each entry i.i.d.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 441, + 155, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 444, + 149, + 456 + ], + "score": 0.93, + "content": "\\mathcal { N } ( 0 , 1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 441, + 155, + 458 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 104, + 429, + 507, + 458 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 460, + 469, + 515 + ], + "lines": [ + { + "bbox": [ + 125, + 460, + 469, + 515 + ], + "spans": [ + { + "bbox": [ + 125, + 460, + 469, + 515 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { d \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } \\mathrm { t r } ( Q _ { 1 } ) \\geq \\frac { 1 } { n } \\mathrm { t r } ( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X ) \\sim \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot W ^ { \\top } W ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot ( I + Q ) ) = \\operatorname* { l i m } _ { \\xi 0 } - \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , x , 0 , x ) \\infty , } \\end{array}", + "type": "interline_equation", + "image_path": "9800d639b0ce67d163d79772a44df6e0175a04082a36777a81e4a888953acb7e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 125, + 460, + 469, + 478.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 125, + 478.3333333333333, + 469, + 496.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 125, + 496.66666666666663, + 469, + 515.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 519, + 506, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 366, + 532 + ], + "score": 1.0, + "content": "where in Section C.5 we have showed (91) is unbounded when", + "type": "text" + }, + { + "bbox": [ + 366, + 522, + 400, + 529 + ], + "score": 0.88, + "content": "n \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 519, + 505, + 532 + ], + "score": 1.0, + "content": ". Moreover, by (154) and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 529, + 500, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 206, + 543 + ], + "score": 1.0, + "content": "Weyl’s theorem we have", + "type": "text" + }, + { + "bbox": [ + 206, + 530, + 286, + 543 + ], + "score": 0.92, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 529, + 326, + 543 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + }, + { + "bbox": [ + 326, + 530, + 372, + 542 + ], + "score": 0.93, + "content": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 1 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 529, + 449, + 543 + ], + "score": 1.0, + "content": "is unbounded. For", + "type": "text" + }, + { + "bbox": [ + 449, + 531, + 495, + 542 + ], + "score": 0.92, + "content": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 2 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 529, + 500, + 543 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 519, + 505, + 543 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 547, + 469, + 571 + ], + "lines": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "spans": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "score": 0.91, + "content": "\\frac { 1 } { d } \\mathrm { t r } ( Q _ { 2 } ) = \\frac { 1 } { d } \\mathrm { t r } \\left( W ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) \\leq \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) ^ { - 1 } ) \\mathrm { t r } \\left( W ^ { \\top } X \\right) = O ( 1 ) .", + "type": "interline_equation", + "image_path": "f5847230779d421975019dfa8d98de3ee3f410a4e17b78e4f242cba552ba1bdf.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 124, + 547, + 469, + 571 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 575, + 323, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 575, + 324, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 167, + 589 + ], + "score": 1.0, + "content": "To sum up, for", + "type": "text" + }, + { + "bbox": [ + 167, + 577, + 200, + 586 + ], + "score": 0.87, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 575, + 218, + 589 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 218, + 576, + 249, + 587 + ], + "score": 0.91, + "content": "\\gamma _ { 2 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 575, + 285, + 589 + ], + "score": 1.0, + "content": "we have", + "type": "text" + }, + { + "bbox": [ + 286, + 576, + 320, + 586 + ], + "score": 0.9, + "content": "B \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 575, + 324, + 589 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 575, + 324, + 589 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 604, + 268, + 617 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 268, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 239, + 618 + ], + "score": 1.0, + "content": "C.7.2 BOUNDED BIAS WHEN", + "type": "text" + }, + { + "bbox": [ + 239, + 605, + 268, + 617 + ], + "score": 0.83, + "content": "\\gamma _ { 2 } > 1", + "type": "inline_equation" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 623, + 401, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 402, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 131, + 637 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 624, + 157, + 634 + ], + "score": 0.9, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 623, + 402, + 637 + ], + "score": 1.0, + "content": ", the two terms in the expression of the bias can be written as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 623, + 402, + 637 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 116, + 640, + 477, + 689 + ], + "lines": [ + { + "bbox": [ + 116, + 640, + 477, + 689 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 477, + 689 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { Q _ { 1 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } X ^ { \\top } , } \\\\ & { Q _ { 2 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) W ^ { \\top } . } \\end{array}", + "type": "interline_equation", + "image_path": "b65bbc4a7bf15d47244e002c33f957f922700ec73c4a07f13628851548b9fa01.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 116, + 640, + 477, + 656.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 116, + 656.3333333333334, + 477, + 672.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 116, + 672.6666666666667, + 477, + 689.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 693, + 183, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 691, + 183, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 183, + 705 + ], + "score": 1.0, + "content": "Therefore we have", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 691, + 183, + 705 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 707, + 514, + 735 + ], + "lines": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "spans": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "score": 0.88, + "content": "\\operatorname { I } ^ { \\mathrm { { T } } } ( Q _ { 1 } ) = 2 \\mathrm { { t r } } \\left( { \\frac { X ^ { \\top } X } { d } } { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\right)", + "type": "interline_equation", + "image_path": "802e384616b8ee310544d57e8871b3da8cb31d01dc153c6778a231feb4f86630.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 111, + 707, + 514, + 735 + ], + "spans": [], + "index": 34 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 144, + 79, + 540, + 181 + ], + "lines": [ + { + "bbox": [ + 144, + 79, + 540, + 181 + ], + "spans": [ + { + "bbox": [ + 144, + 79, + 540, + 181 + ], + "score": 0.95, + "content": "\\begin{array} { r l r } { { \\le 2 \\lambda _ { \\operatorname* { m a x } } ( \\frac { X ^ { \\top } X } { d } ) \\operatorname { t r } ( \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ) } } \\\\ & { = O ( 1 ) \\cdot \\operatorname { t r } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } ) } \\\\ & { \\le O ( 1 ) \\cdot \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) ) \\cdot \\operatorname { t r } ( K _ { W } ) } \\\\ & { = O ( 1 ) \\cdot \\sigma _ { \\operatorname* { m i n } } ^ { - 2 } ( \\phi ( X ^ { \\top } W ) ) \\operatorname { t r } ( K _ { W } ) = O ( 1 ) \\cdot O ( n ^ { - 1 } ) \\cdot O ( n ) = O ( 1 ) , } & { \\quad \\mathrm { ~ \\displaystyle ( 9 5 ) ~ } } \\end{array}", + "type": "interline_equation", + "image_path": "7742145ad610797abc0b4215eb1e361c0f199ca161702f321c5e6d9aba9c8ca2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 144, + 79, + 540, + 113.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 144, + 113.0, + 540, + 147.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 144, + 147.0, + 540, + 181.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 506, + 208 + ], + "lines": [ + { + "bbox": [ + 104, + 183, + 507, + 199 + ], + "spans": [ + { + "bbox": [ + 104, + 183, + 242, + 199 + ], + "score": 1.0, + "content": "in which we used Lemma 11 and", + "type": "text" + }, + { + "bbox": [ + 243, + 185, + 351, + 198 + ], + "score": 0.92, + "content": "\\operatorname { t r } \\left( A B \\right) \\leq \\lambda _ { \\operatorname* { m a x } } ( A ) \\operatorname { t r } \\left( B \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 183, + 456, + 199 + ], + "score": 1.0, + "content": "for positive semi-definite", + "type": "text" + }, + { + "bbox": [ + 457, + 186, + 477, + 197 + ], + "score": 0.9, + "content": "A , B", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 183, + 507, + 199 + ], + "score": 1.0, + "content": ". Simi-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 194, + 129, + 210 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 129, + 210 + ], + "score": 1.0, + "content": "larly", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 213, + 484, + 298 + ], + "lines": [ + { + "bbox": [ + 125, + 213, + 484, + 298 + ], + "spans": [ + { + "bbox": [ + 125, + 213, + 484, + 298 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\frac { 2 } { d } \\mathrm { t r } \\left( Q _ { 2 } \\right) = 2 \\mathrm { t r } \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\cdot \\frac { 1 } { d } W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq \\mathrm { t r } \\left( \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\right) ( \\ldots ) ^ { \\top } \\right) + \\mathrm { t r } \\left( d ^ { - 2 } X ^ { \\top } W W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq n \\cdot \\sigma _ { \\operatorname* { m i n } } \\left( \\phi ( X ^ { \\top } W ) \\right) ^ { - 2 } + \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } \\left( \\frac { 1 } { d } X X ^ { \\top } \\right) \\mathrm { t r } \\left( W W ^ { \\top } \\right) = O ( 1 ) , } \\end{array}", + "type": "interline_equation", + "image_path": "d88b10f67d37e9ed023388efdd27bc012f383dde71ca24414fd2e1bfce283fdb.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 125, + 213, + 484, + 241.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 125, + 241.33333333333334, + 484, + 269.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 125, + 269.6666666666667, + 484, + 298.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 303, + 504, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 177, + 318 + ], + "score": 1.0, + "content": "where we applied", + "type": "text" + }, + { + "bbox": [ + 178, + 304, + 336, + 318 + ], + "score": 0.92, + "content": "\\mathrm { t r } \\left( A B \\right) \\leq ( \\mathrm { t r } \\left( A ^ { \\top } A \\right) + \\mathrm { t r } \\left( B ^ { \\top } B \\right) ) / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 303, + 449, + 318 + ], + "score": 1.0, + "content": ". We therefore conclude that", + "type": "text" + }, + { + "bbox": [ + 450, + 306, + 459, + 315 + ], + "score": 0.85, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 303, + 506, + 318 + ], + "score": 1.0, + "content": "is bounded", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 316, + 504, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 130, + 329 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 130, + 317, + 156, + 327 + ], + "score": 0.9, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 316, + 174, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 175, + 317, + 217, + 328 + ], + "score": 0.9, + "content": "h , n \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 316, + 221, + 329 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 496, + 318, + 504, + 325 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 105, + 335, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "Remark. Concurrent to this work, Mei and Montanari (2019) provides a complete characterization", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 346, + 322, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 322, + 358 + ], + "score": 1.0, + "content": "of the bias term and confirms our observations above.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 230, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 231, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 231, + 402 + ], + "score": 1.0, + "content": "C.8 PROOF OF THEOREM 7", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 506, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 213, + 424 + ], + "score": 1.0, + "content": "For simplicity we assume", + "type": "text" + }, + { + "bbox": [ + 213, + 412, + 240, + 423 + ], + "score": 0.91, + "content": "n , d , h", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 410, + 316, + 424 + ], + "score": 1.0, + "content": "to be even and let", + "type": "text" + }, + { + "bbox": [ + 316, + 411, + 355, + 423 + ], + "score": 0.88, + "content": "d _ { 0 } = d / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 410, + 359, + 424 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 360, + 411, + 401, + 423 + ], + "score": 0.87, + "content": "n _ { 0 } = n / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 410, + 419, + 424 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 420, + 411, + 460, + 423 + ], + "score": 0.91, + "content": "h _ { 0 } = h / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 410, + 506, + 424 + ], + "score": 1.0, + "content": ". Since the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 422, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 104, + 422, + 192, + 437 + ], + "score": 1.0, + "content": "second layer is fixed", + "type": "text" + }, + { + "bbox": [ + 193, + 423, + 307, + 435 + ], + "score": 0.9, + "content": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 422, + 338, + 437 + ], + "score": 1.0, + "content": ", we let", + "type": "text" + }, + { + "bbox": [ + 339, + 423, + 387, + 435 + ], + "score": 0.93, + "content": "a _ { i } = 1 / \\sqrt { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 422, + 406, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 406, + 423, + 476, + 436 + ], + "score": 0.93, + "content": "a _ { i + h _ { 0 } } = - 1 / \\sqrt { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 422, + 506, + 437 + ], + "score": 1.0, + "content": "for all", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 433, + 431, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 152, + 448 + ], + "score": 0.91, + "content": "1 \\leq i \\leq h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 433, + 234, + 450 + ], + "score": 1.0, + "content": ". We therefore write", + "type": "text" + }, + { + "bbox": [ + 234, + 435, + 339, + 448 + ], + "score": 0.9, + "content": "\\pmb { a } ^ { \\top } = h ^ { - 1 / 2 } [ \\mathbf { 1 } _ { h _ { 0 } } , - \\mathbf { 1 } _ { h _ { 0 } } ] ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 433, + 358, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 358, + 437, + 425, + 448 + ], + "score": 0.92, + "content": "W = [ W _ { + } , W _ { - } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 433, + 431, + 450 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 454, + 451, + 481 + ], + "lines": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "spans": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "score": 0.92, + "content": "f ( \\pmb { x } ; W _ { - } , W _ { + } ) = \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) = \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } \\pmb { x } ) - \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } \\pmb { x } ) .", + "type": "interline_equation", + "image_path": "fb080e2c6e923d48b65f31336b269d422b9ee4359ae926993eb1d278d413c92d.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 486, + 270, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 270, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 270, + 499 + ], + "score": 1.0, + "content": "The empirical risk can thus be written as", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 505, + 503, + 535 + ], + "lines": [ + { + "bbox": [ + 111, + 505, + 503, + 535 + ], + "spans": [ + { + "bbox": [ + 111, + 505, + 503, + 535 + ], + "score": 0.91, + "content": "L ( X ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } L ( x ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } \\left[ y - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } x ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } x ) \\right] ^ { 2 } .", + "type": "interline_equation", + "image_path": "29f8c6e1d0bc276ee218a695699ddee6dbe27eec007019c7008a79ded3d7bc9b.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 111, + 505, + 503, + 515.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 111, + 515.0, + 503, + 525.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 525.0, + 503, + 535.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 558, + 266, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 267, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 267, + 572 + ], + "score": 1.0, + "content": "C.8.1 DEFINING GRADIENT FLOWS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 105, + 578, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "In this section we define three gradient flows and show that the three flows are similar in some sense.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 590, + 310, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 121, + 600 + ], + "score": 0.71, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 590, + 310, + 603 + ], + "score": 1.0, + "content": "-Original is the original gradient flow (11), i.e.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 118, + 608, + 476, + 671 + ], + "lines": [ + { + "bbox": [ + 118, + 608, + 476, + 671 + ], + "spans": [ + { + "bbox": [ + 118, + 608, + 476, + 671 + ], + "score": 0.94, + "content": "\\begin{array} { l } { \\displaystyle \\frac { \\partial W _ { + } ^ { O } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { + } ^ { O \\top } \\mathbf { x } _ { i } ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { - } ^ { O \\top } \\mathbf { x } _ { i } ) \\Big ) \\mathbf { x } _ { i } \\boldsymbol { \\phi } ^ { \\prime } ( { \\mathbf { x } } _ { i } ^ { \\top } W _ { + } ^ { O } ) \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\left[ X ( \\boldsymbol { y } - \\boldsymbol { y } ^ { O } ( t ) ) \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\circ \\boldsymbol { \\phi } ^ { \\prime } ( X W _ { + } ^ { O } ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "43969bee080eb8e8ee088b02384d3ebb35bd85c910439835e45cd5f0708a2377.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 118, + 608, + 476, + 629.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 118, + 629.0, + 476, + 650.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 118, + 650.0, + 476, + 671.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 273, + 691 + ], + "score": 1.0, + "content": "starting from the vanishing initialization", + "type": "text" + }, + { + "bbox": [ + 274, + 677, + 379, + 690 + ], + "score": 0.92, + "content": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 676, + 505, + 691 + ], + "score": 1.0, + "content": ". Note that the gradient for the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 688, + 281, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 160, + 703 + ], + "score": 1.0, + "content": "negative part", + "type": "text" + }, + { + "bbox": [ + 160, + 689, + 179, + 702 + ], + "score": 0.88, + "content": "W _ { - } ^ { O }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 688, + 281, + 703 + ], + "score": 1.0, + "content": "can be similarly defined.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 708, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 453, + 721 + ], + "score": 1.0, + "content": "We now define the flow under the same objective but from exact zero initialization", + "type": "text" + }, + { + "bbox": [ + 453, + 707, + 505, + 720 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ^ { D } ( 0 ) = { \\bf 0 }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 718, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 136, + 734 + ], + "score": 1.0, + "content": "termed", + "type": "text" + }, + { + "bbox": [ + 137, + 721, + 151, + 731 + ], + "score": 0.69, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 718, + 448, + 734 + ], + "score": 1.0, + "content": "-Double. Due to zero initialization, a basic observation is that the solution", + "type": "text" + }, + { + "bbox": [ + 448, + 720, + 494, + 733 + ], + "score": 0.92, + "content": "[ W _ { + } ^ { D } , W _ { - } ^ { D } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 718, + 506, + 734 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 144, + 79, + 540, + 181 + ], + "lines": [ + { + "bbox": [ + 144, + 79, + 540, + 181 + ], + "spans": [ + { + "bbox": [ + 144, + 79, + 540, + 181 + ], + "score": 0.95, + "content": "\\begin{array} { r l r } { { \\le 2 \\lambda _ { \\operatorname* { m a x } } ( \\frac { X ^ { \\top } X } { d } ) \\operatorname { t r } ( \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ) } } \\\\ & { = O ( 1 ) \\cdot \\operatorname { t r } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } ) } \\\\ & { \\le O ( 1 ) \\cdot \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) ) \\cdot \\operatorname { t r } ( K _ { W } ) } \\\\ & { = O ( 1 ) \\cdot \\sigma _ { \\operatorname* { m i n } } ^ { - 2 } ( \\phi ( X ^ { \\top } W ) ) \\operatorname { t r } ( K _ { W } ) = O ( 1 ) \\cdot O ( n ^ { - 1 } ) \\cdot O ( n ) = O ( 1 ) , } & { \\quad \\mathrm { ~ \\displaystyle ( 9 5 ) ~ } } \\end{array}", + "type": "interline_equation", + "image_path": "7742145ad610797abc0b4215eb1e361c0f199ca161702f321c5e6d9aba9c8ca2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 144, + 79, + 540, + 113.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 144, + 113.0, + 540, + 147.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 144, + 147.0, + 540, + 181.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 506, + 208 + ], + "lines": [ + { + "bbox": [ + 104, + 183, + 507, + 199 + ], + "spans": [ + { + "bbox": [ + 104, + 183, + 242, + 199 + ], + "score": 1.0, + "content": "in which we used Lemma 11 and", + "type": "text" + }, + { + "bbox": [ + 243, + 185, + 351, + 198 + ], + "score": 0.92, + "content": "\\operatorname { t r } \\left( A B \\right) \\leq \\lambda _ { \\operatorname* { m a x } } ( A ) \\operatorname { t r } \\left( B \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 183, + 456, + 199 + ], + "score": 1.0, + "content": "for positive semi-definite", + "type": "text" + }, + { + "bbox": [ + 457, + 186, + 477, + 197 + ], + "score": 0.9, + "content": "A , B", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 183, + 507, + 199 + ], + "score": 1.0, + "content": ". Simi-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 194, + 129, + 210 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 129, + 210 + ], + "score": 1.0, + "content": "larly", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 183, + 507, + 210 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 213, + 484, + 298 + ], + "lines": [ + { + "bbox": [ + 125, + 213, + 484, + 298 + ], + "spans": [ + { + "bbox": [ + 125, + 213, + 484, + 298 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\frac { 2 } { d } \\mathrm { t r } \\left( Q _ { 2 } \\right) = 2 \\mathrm { t r } \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\cdot \\frac { 1 } { d } W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq \\mathrm { t r } \\left( \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\right) ( \\ldots ) ^ { \\top } \\right) + \\mathrm { t r } \\left( d ^ { - 2 } X ^ { \\top } W W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq n \\cdot \\sigma _ { \\operatorname* { m i n } } \\left( \\phi ( X ^ { \\top } W ) \\right) ^ { - 2 } + \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } \\left( \\frac { 1 } { d } X X ^ { \\top } \\right) \\mathrm { t r } \\left( W W ^ { \\top } \\right) = O ( 1 ) , } \\end{array}", + "type": "interline_equation", + "image_path": "d88b10f67d37e9ed023388efdd27bc012f383dde71ca24414fd2e1bfce283fdb.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 125, + 213, + 484, + 241.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 125, + 241.33333333333334, + 484, + 269.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 125, + 269.6666666666667, + 484, + 298.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 303, + 504, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 177, + 318 + ], + "score": 1.0, + "content": "where we applied", + "type": "text" + }, + { + "bbox": [ + 178, + 304, + 336, + 318 + ], + "score": 0.92, + "content": "\\mathrm { t r } \\left( A B \\right) \\leq ( \\mathrm { t r } \\left( A ^ { \\top } A \\right) + \\mathrm { t r } \\left( B ^ { \\top } B \\right) ) / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 303, + 449, + 318 + ], + "score": 1.0, + "content": ". We therefore conclude that", + "type": "text" + }, + { + "bbox": [ + 450, + 306, + 459, + 315 + ], + "score": 0.85, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 303, + 506, + 318 + ], + "score": 1.0, + "content": "is bounded", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 316, + 504, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 130, + 329 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 130, + 317, + 156, + 327 + ], + "score": 0.9, + "content": "h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 316, + 174, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 175, + 317, + 217, + 328 + ], + "score": 0.9, + "content": "h , n \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 316, + 221, + 329 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 496, + 318, + 504, + 325 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 303, + 506, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 335, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "Remark. Concurrent to this work, Mei and Montanari (2019) provides a complete characterization", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 346, + 322, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 322, + 358 + ], + "score": 1.0, + "content": "of the bias term and confirms our observations above.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 333, + 505, + 358 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 230, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 231, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 231, + 402 + ], + "score": 1.0, + "content": "C.8 PROOF OF THEOREM 7", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 506, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 213, + 424 + ], + "score": 1.0, + "content": "For simplicity we assume", + "type": "text" + }, + { + "bbox": [ + 213, + 412, + 240, + 423 + ], + "score": 0.91, + "content": "n , d , h", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 410, + 316, + 424 + ], + "score": 1.0, + "content": "to be even and let", + "type": "text" + }, + { + "bbox": [ + 316, + 411, + 355, + 423 + ], + "score": 0.88, + "content": "d _ { 0 } = d / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 410, + 359, + 424 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 360, + 411, + 401, + 423 + ], + "score": 0.87, + "content": "n _ { 0 } = n / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 410, + 419, + 424 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 420, + 411, + 460, + 423 + ], + "score": 0.91, + "content": "h _ { 0 } = h / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 410, + 506, + 424 + ], + "score": 1.0, + "content": ". Since the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 422, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 104, + 422, + 192, + 437 + ], + "score": 1.0, + "content": "second layer is fixed", + "type": "text" + }, + { + "bbox": [ + 193, + 423, + 307, + 435 + ], + "score": 0.9, + "content": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 422, + 338, + 437 + ], + "score": 1.0, + "content": ", we let", + "type": "text" + }, + { + "bbox": [ + 339, + 423, + 387, + 435 + ], + "score": 0.93, + "content": "a _ { i } = 1 / \\sqrt { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 422, + 406, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 406, + 423, + 476, + 436 + ], + "score": 0.93, + "content": "a _ { i + h _ { 0 } } = - 1 / \\sqrt { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 422, + 506, + 437 + ], + "score": 1.0, + "content": "for all", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 433, + 431, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 152, + 448 + ], + "score": 0.91, + "content": "1 \\leq i \\leq h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 433, + 234, + 450 + ], + "score": 1.0, + "content": ". We therefore write", + "type": "text" + }, + { + "bbox": [ + 234, + 435, + 339, + 448 + ], + "score": 0.9, + "content": "\\pmb { a } ^ { \\top } = h ^ { - 1 / 2 } [ \\mathbf { 1 } _ { h _ { 0 } } , - \\mathbf { 1 } _ { h _ { 0 } } ] ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 433, + 358, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 358, + 437, + 425, + 448 + ], + "score": 0.92, + "content": "W = [ W _ { + } , W _ { - } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 433, + 431, + 450 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 410, + 506, + 450 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 454, + 451, + 481 + ], + "lines": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "spans": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "score": 0.92, + "content": "f ( \\pmb { x } ; W _ { - } , W _ { + } ) = \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) = \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } \\pmb { x } ) - \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } \\pmb { x } ) .", + "type": "interline_equation", + "image_path": "fb080e2c6e923d48b65f31336b269d422b9ee4359ae926993eb1d278d413c92d.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 159, + 454, + 451, + 481 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 486, + 270, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 270, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 270, + 499 + ], + "score": 1.0, + "content": "The empirical risk can thus be written as", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 106, + 486, + 270, + 499 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 505, + 503, + 535 + ], + "lines": [ + { + "bbox": [ + 111, + 505, + 503, + 535 + ], + "spans": [ + { + "bbox": [ + 111, + 505, + 503, + 535 + ], + "score": 0.91, + "content": "L ( X ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } L ( x ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } \\left[ y - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } x ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } x ) \\right] ^ { 2 } .", + "type": "interline_equation", + "image_path": "29f8c6e1d0bc276ee218a695699ddee6dbe27eec007019c7008a79ded3d7bc9b.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 111, + 505, + 503, + 515.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 111, + 515.0, + 503, + 525.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 525.0, + 503, + 535.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 558, + 266, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 267, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 267, + 572 + ], + "score": 1.0, + "content": "C.8.1 DEFINING GRADIENT FLOWS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "list", + "bbox": [ + 105, + 578, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "In this section we define three gradient flows and show that the three flows are similar in some sense.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 590, + 310, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 121, + 600 + ], + "score": 0.71, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 590, + 310, + 603 + ], + "score": 1.0, + "content": "-Original is the original gradient flow (11), i.e.", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 578, + 506, + 603 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 118, + 608, + 476, + 671 + ], + "lines": [ + { + "bbox": [ + 118, + 608, + 476, + 671 + ], + "spans": [ + { + "bbox": [ + 118, + 608, + 476, + 671 + ], + "score": 0.94, + "content": "\\begin{array} { l } { \\displaystyle \\frac { \\partial W _ { + } ^ { O } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { + } ^ { O \\top } \\mathbf { x } _ { i } ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { - } ^ { O \\top } \\mathbf { x } _ { i } ) \\Big ) \\mathbf { x } _ { i } \\boldsymbol { \\phi } ^ { \\prime } ( { \\mathbf { x } } _ { i } ^ { \\top } W _ { + } ^ { O } ) \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\left[ X ( \\boldsymbol { y } - \\boldsymbol { y } ^ { O } ( t ) ) \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\circ \\boldsymbol { \\phi } ^ { \\prime } ( X W _ { + } ^ { O } ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "43969bee080eb8e8ee088b02384d3ebb35bd85c910439835e45cd5f0708a2377.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 118, + 608, + 476, + 629.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 118, + 629.0, + 476, + 650.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 118, + 650.0, + 476, + 671.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 273, + 691 + ], + "score": 1.0, + "content": "starting from the vanishing initialization", + "type": "text" + }, + { + "bbox": [ + 274, + 677, + 379, + 690 + ], + "score": 0.92, + "content": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 676, + 505, + 691 + ], + "score": 1.0, + "content": ". Note that the gradient for the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 688, + 281, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 160, + 703 + ], + "score": 1.0, + "content": "negative part", + "type": "text" + }, + { + "bbox": [ + 160, + 689, + 179, + 702 + ], + "score": 0.88, + "content": "W _ { - } ^ { O }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 688, + 281, + 703 + ], + "score": 1.0, + "content": "can be similarly defined.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 676, + 505, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 708, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 453, + 721 + ], + "score": 1.0, + "content": "We now define the flow under the same objective but from exact zero initialization", + "type": "text" + }, + { + "bbox": [ + 453, + 707, + 505, + 720 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ^ { D } ( 0 ) = { \\bf 0 }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 718, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 136, + 734 + ], + "score": 1.0, + "content": "termed", + "type": "text" + }, + { + "bbox": [ + 137, + 721, + 151, + 731 + ], + "score": 0.69, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 718, + 448, + 734 + ], + "score": 1.0, + "content": "-Double. Due to zero initialization, a basic observation is that the solution", + "type": "text" + }, + { + "bbox": [ + 448, + 720, + 494, + 733 + ], + "score": 0.92, + "content": "[ W _ { + } ^ { D } , W _ { - } ^ { D } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 718, + 506, + 734 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 707, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 502, + 107 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 420, + 95 + ], + "score": 1.0, + "content": "at most rank-2, and more precisely, the parameters in the flow takes the form of", + "type": "text" + }, + { + "bbox": [ + 421, + 81, + 504, + 96 + ], + "score": 0.91, + "content": "W _ { \\pm } ^ { D } ( t ) = { \\pmb w } _ { \\pm } ^ { D } ( t ) { \\bf 1 } ^ { \\top }", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 291, + 109 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 133, + 109 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 94, + 161, + 108 + ], + "score": 0.93, + "content": "{ \\pmb w } _ { \\pm } ^ { D } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 93, + 291, + 109 + ], + "score": 1.0, + "content": "admits the following dynamics:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 114, + 493, + 148 + ], + "lines": [ + { + "bbox": [ + 117, + 114, + 493, + 148 + ], + "spans": [ + { + "bbox": [ + 117, + 114, + 493, + 148 + ], + "score": 0.94, + "content": "\\frac { \\partial w _ { + } ^ { D } } { \\partial t } = g _ { + } ^ { D } ( \\boldsymbol { w } _ { + } ^ { D } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) + \\sqrt { h } \\phi ( \\boldsymbol { w } _ { - } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\boldsymbol { x } _ { i } \\right] .", + "type": "interline_equation", + "image_path": "d71db0f99a7bbdce8f380ddc4807b59959cb4b1f89dea238fb5baa5ed83b913e.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 117, + 114, + 493, + 125.33333333333333 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 117, + 125.33333333333333, + 493, + 136.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 117, + 136.66666666666666, + 493, + 148.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 399, + 179 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 401, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 191, + 180 + ], + "score": 1.0, + "content": "Lastly, we define the", + "type": "text" + }, + { + "bbox": [ + 191, + 167, + 206, + 177 + ], + "score": 0.74, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 164, + 336, + 180 + ], + "score": 1.0, + "content": "-Single with solution denoted as", + "type": "text" + }, + { + "bbox": [ + 337, + 165, + 394, + 179 + ], + "score": 0.92, + "content": "{ \\pmb w } _ { \\pm } = { \\pmb w } _ { \\pm } ^ { S } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 164, + 401, + 180 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 184, + 474, + 219 + ], + "lines": [ + { + "bbox": [ + 115, + 184, + 474, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 184, + 474, + 219 + ], + "score": 0.94, + "content": "\\frac { \\partial w _ { + } ^ { S } } { \\partial t } = g _ { + } ^ { S } ( w _ { + } ^ { S } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { + } ^ { S \\top } x _ { i } + \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { - } ^ { 1 \\top } x _ { i } \\Big ) \\phi ^ { \\prime } ( 0 ) x _ { i } \\Big ] .", + "type": "interline_equation", + "image_path": "d46d15fbec94e0029916dec56e8ba99890b28de3d83e6f15a2436677356efc44.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 115, + 184, + 474, + 195.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 115, + 195.66666666666666, + 474, + 207.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 115, + 207.33333333333331, + 474, + 218.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 506, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 224, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 206, + 239 + ], + "score": 1.0, + "content": "from zero initialization", + "type": "text" + }, + { + "bbox": [ + 206, + 224, + 258, + 238 + ], + "score": 0.94, + "content": "{ \\pmb w } _ { \\pm } ^ { D } ( 0 ) = { \\bf 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 224, + 462, + 239 + ], + "score": 1.0, + "content": ". This can be seen as replacing the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 462, + 227, + 469, + 237 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 224, + 506, + 239 + ], + "score": 1.0, + "content": "with its", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 236, + 272, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 272, + 249 + ], + "score": 1.0, + "content": "first-order Taylor expansion at the origin.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 107, + 260, + 288, + 272 + ], + "lines": [ + { + "bbox": [ + 106, + 259, + 289, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 289, + 273 + ], + "score": 1.0, + "content": "C.8.2 FROM GF-DOUBLE TO GF-SINGLE", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 104, + 279, + 504, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "score": 1.0, + "content": "Step 1. Solution of GF-single. Among the three flows defined above, only GF-single an explicit", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 291, + 357, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 175, + 303 + ], + "score": 1.0, + "content": "form at any time", + "type": "text" + }, + { + "bbox": [ + 175, + 292, + 180, + 300 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 291, + 357, + 303 + ], + "score": 1.0, + "content": ". Specifically, the solution can be written as:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 307, + 438, + 335 + ], + "lines": [ + { + "bbox": [ + 173, + 307, + 438, + 335 + ], + "spans": [ + { + "bbox": [ + 173, + 307, + 438, + 335 + ], + "score": 0.92, + "content": "w _ { + } ^ { S } ( t ) = - w _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } ,", + "type": "interline_equation", + "image_path": "b762dc632ef5f0943f1c17c6b6373cfd7d3777aa5ffa60d6850ed9156790ace3.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 173, + 307, + 438, + 316.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 173, + 316.3333333333333, + 438, + 325.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 173, + 325.66666666666663, + 438, + 334.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 212, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 212, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 130, + 353 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 341, + 156, + 351 + ], + "score": 0.9, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 339, + 212, + 353 + ], + "score": 1.0, + "content": ", or otherwise", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 356, + 438, + 384 + ], + "lines": [ + { + "bbox": [ + 173, + 356, + 438, + 384 + ], + "spans": [ + { + "bbox": [ + 173, + 356, + 438, + 384 + ], + "score": 0.94, + "content": "\\pmb { w } _ { + } ^ { S } ( t ) = - \\pmb { w } _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } X \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X ^ { \\top } X t } \\right) ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } .", + "type": "interline_equation", + "image_path": "7881650689b109f035de1a165987e743b3b7bd91073898c77af0933159d8d813.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 173, + 356, + 438, + 365.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 173, + 365.3333333333333, + 438, + 374.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 173, + 374.66666666666663, + 438, + 383.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 389, + 506, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 131, + 401 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 390, + 157, + 399 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 389, + 351, + 401 + ], + "score": 1.0, + "content": ". For simplicity we elaborate the proof only for", + "type": "text" + }, + { + "bbox": [ + 351, + 389, + 377, + 399 + ], + "score": 0.89, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 389, + 505, + 401 + ], + "score": 1.0, + "content": ". We provide a condition under", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 401, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 505, + 411 + ], + "score": 1.0, + "content": "which the difference between the trajectories can be controlled, and we first that this condition holds", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 411, + 330, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 266, + 423 + ], + "score": 1.0, + "content": "for the linearized flow GF-single for all", + "type": "text" + }, + { + "bbox": [ + 266, + 412, + 271, + 421 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 411, + 330, + 423 + ], + "score": 1.0, + "content": "in Lemma 17.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 428, + 455, + 456 + ], + "lines": [ + { + "bbox": [ + 156, + 428, + 455, + 456 + ], + "spans": [ + { + "bbox": [ + 156, + 428, + 455, + 456 + ], + "score": 0.91, + "content": "\\mathrm { C o n d i t i o n \\ A : } \\ \\| w ( t ) \\| _ { 2 } = O \\left( { \\frac { 1 } { \\sqrt { d } } } \\right) ; \\left\\| X ^ { \\top } w ( t ) \\right\\| _ { \\infty } = O \\left( { \\frac { \\mathrm { p o l y } \\log d } { \\sqrt { d } } } \\right) .", + "type": "interline_equation", + "image_path": "20e90bda0059164dc605b61170a2f4f17ac694686cc4b9ff7c0ca6d776f91e81.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 156, + 428, + 455, + 437.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 156, + 437.3333333333333, + 455, + 446.66666666666663 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 156, + 446.66666666666663, + 455, + 455.99999999999994 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "Step 2. Bounding the Difference in Gradient Flow Trajectory. Due to low rank property of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 477, + 451, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 451, + 490 + ], + "score": 1.0, + "content": "GF-single and GF-double, in this subsection we slightly abuse the notation and define", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 495, + 439, + 511 + ], + "lines": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "spans": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "score": 0.88, + "content": "f ( \\pmb { x } ; \\pmb { w } _ { \\pm } ) = f ( \\pmb { x } ; \\pmb { w } _ { + } \\pmb { 1 } ^ { \\top } , \\pmb { w } _ { - } \\pmb { 1 } ^ { \\top } ) = \\sqrt { h } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } ) - \\sqrt { h } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } ) .", + "type": "interline_equation", + "image_path": "4763bc2bd2dade9cd1d216f11bf583a0917967d2ff51fe88646e3722206e783c.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 516, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "We now show that the difference between the two trajectories defined above is asymptotically", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 527, + 486, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 162, + 540 + ], + "score": 1.0, + "content": "vanishing for", + "type": "text" + }, + { + "bbox": [ + 162, + 528, + 177, + 539 + ], + "score": 0.84, + "content": "{ \\pmb w } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 527, + 182, + 540 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 182, + 528, + 199, + 539 + ], + "score": 0.69, + "content": "\\mathbf { \\nabla } w _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 527, + 477, + 540 + ], + "score": 1.0, + "content": "follows the same argument). Compare the two trajectories up to time", + "type": "text" + }, + { + "bbox": [ + 477, + 528, + 486, + 537 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 105, + 543, + 500, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 500, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 500, + 678 + ], + "score": 0.94, + "content": "\\begin{array} { r l r } { { \\| { \\boldsymbol w } _ { + } ^ { D } ( T ) - { \\boldsymbol w } _ { + } ^ { S } ( T ) } \\| _ { 2 } = \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) d t \\| _ { 2 } } \\\\ & { \\le \\| \\int _ { 0 } ^ { T } g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } + \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } } \\\\ & { = \\| \\int _ { 0 } ^ { T } \\frac { 1 } { 2 n _ { 0 } } \\frac { 2 n _ { 0 } } { i = 1 } [ ( - \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } + \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } ) \\phi ^ { \\prime } ( 0 ) { \\boldsymbol x } _ { i } ] \\mathrm { d } t \\| _ { 2 } } & \\\\ & { = O ( 1 ) \\int _ { 0 } ^ { T } ( \\| { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) \\| _ { 2 } + \\| { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) \\| _ { 2 } ) \\mathrm { d } t + { \\boldsymbol E } _ { + } , } & { ( 1 0 6 ) } \\end{array}", + "type": "interline_equation", + "image_path": "fd7c61c891b9274eed7e2f7c64f4f67079523a911d7dfdbb65843358a8774117.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 105, + 543, + 500, + 588.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 105, + 588.0, + 500, + 633.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 105, + 633.0, + 500, + 678.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 682, + 267, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 268, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 268, + 694 + ], + "score": 1.0, + "content": "where we have defined the error term as", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 696, + 400, + 730 + ], + "lines": [ + { + "bbox": [ + 210, + 696, + 400, + 730 + ], + "spans": [ + { + "bbox": [ + 210, + 696, + 400, + 730 + ], + "score": 0.93, + "content": "E _ { + } = \\left\\| \\int _ { 0 } ^ { T } \\pmb { g } _ { + } ^ { D } ( \\pmb { w } _ { + } ^ { D } ( s ) ) - \\pmb { g } _ { + } ^ { S } ( \\pmb { w } _ { + } ^ { D } ( s ) ) ~ \\mathrm { d } t \\right\\| _ { 2 } .", + "type": "interline_equation", + "image_path": "a5a4052e4440e0d387c9ef45daad6cf8330a1f9d303bf27b008b3b06d2e63429.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 696, + 400, + 713.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 210, + 713.0, + 400, + 730.0 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "page_idx": 26, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 502, + 107 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 420, + 95 + ], + "score": 1.0, + "content": "at most rank-2, and more precisely, the parameters in the flow takes the form of", + "type": "text" + }, + { + "bbox": [ + 421, + 81, + 504, + 96 + ], + "score": 0.91, + "content": "W _ { \\pm } ^ { D } ( t ) = { \\pmb w } _ { \\pm } ^ { D } ( t ) { \\bf 1 } ^ { \\top }", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 291, + 109 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 133, + 109 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 94, + 161, + 108 + ], + "score": 0.93, + "content": "{ \\pmb w } _ { \\pm } ^ { D } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 93, + 291, + 109 + ], + "score": 1.0, + "content": "admits the following dynamics:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 104, + 81, + 504, + 109 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 114, + 493, + 148 + ], + "lines": [ + { + "bbox": [ + 117, + 114, + 493, + 148 + ], + "spans": [ + { + "bbox": [ + 117, + 114, + 493, + 148 + ], + "score": 0.94, + "content": "\\frac { \\partial w _ { + } ^ { D } } { \\partial t } = g _ { + } ^ { D } ( \\boldsymbol { w } _ { + } ^ { D } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) + \\sqrt { h } \\phi ( \\boldsymbol { w } _ { - } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\boldsymbol { x } _ { i } \\right] .", + "type": "interline_equation", + "image_path": "d71db0f99a7bbdce8f380ddc4807b59959cb4b1f89dea238fb5baa5ed83b913e.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 117, + 114, + 493, + 125.33333333333333 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 117, + 125.33333333333333, + 493, + 136.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 117, + 136.66666666666666, + 493, + 148.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 399, + 179 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 401, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 191, + 180 + ], + "score": 1.0, + "content": "Lastly, we define the", + "type": "text" + }, + { + "bbox": [ + 191, + 167, + 206, + 177 + ], + "score": 0.74, + "content": "\\pmb { G F }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 164, + 336, + 180 + ], + "score": 1.0, + "content": "-Single with solution denoted as", + "type": "text" + }, + { + "bbox": [ + 337, + 165, + 394, + 179 + ], + "score": 0.92, + "content": "{ \\pmb w } _ { \\pm } = { \\pmb w } _ { \\pm } ^ { S } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 164, + 401, + 180 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 164, + 401, + 180 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 184, + 474, + 219 + ], + "lines": [ + { + "bbox": [ + 115, + 184, + 474, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 184, + 474, + 219 + ], + "score": 0.94, + "content": "\\frac { \\partial w _ { + } ^ { S } } { \\partial t } = g _ { + } ^ { S } ( w _ { + } ^ { S } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { + } ^ { S \\top } x _ { i } + \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { - } ^ { 1 \\top } x _ { i } \\Big ) \\phi ^ { \\prime } ( 0 ) x _ { i } \\Big ] .", + "type": "interline_equation", + "image_path": "d46d15fbec94e0029916dec56e8ba99890b28de3d83e6f15a2436677356efc44.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 115, + 184, + 474, + 195.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 115, + 195.66666666666666, + 474, + 207.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 115, + 207.33333333333331, + 474, + 218.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 506, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 224, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 206, + 239 + ], + "score": 1.0, + "content": "from zero initialization", + "type": "text" + }, + { + "bbox": [ + 206, + 224, + 258, + 238 + ], + "score": 0.94, + "content": "{ \\pmb w } _ { \\pm } ^ { D } ( 0 ) = { \\bf 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 224, + 462, + 239 + ], + "score": 1.0, + "content": ". This can be seen as replacing the nonlinearity", + "type": "text" + }, + { + "bbox": [ + 462, + 227, + 469, + 237 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 224, + 506, + 239 + ], + "score": 1.0, + "content": "with its", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 236, + 272, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 272, + 249 + ], + "score": 1.0, + "content": "first-order Taylor expansion at the origin.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 224, + 506, + 249 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 260, + 288, + 272 + ], + "lines": [ + { + "bbox": [ + 106, + 259, + 289, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 289, + 273 + ], + "score": 1.0, + "content": "C.8.2 FROM GF-DOUBLE TO GF-SINGLE", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 104, + 279, + 504, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "score": 1.0, + "content": "Step 1. Solution of GF-single. Among the three flows defined above, only GF-single an explicit", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 291, + 357, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 175, + 303 + ], + "score": 1.0, + "content": "form at any time", + "type": "text" + }, + { + "bbox": [ + 175, + 292, + 180, + 300 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 291, + 357, + 303 + ], + "score": 1.0, + "content": ". Specifically, the solution can be written as:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 280, + 505, + 303 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 307, + 438, + 335 + ], + "lines": [ + { + "bbox": [ + 173, + 307, + 438, + 335 + ], + "spans": [ + { + "bbox": [ + 173, + 307, + 438, + 335 + ], + "score": 0.92, + "content": "w _ { + } ^ { S } ( t ) = - w _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } ,", + "type": "interline_equation", + "image_path": "b762dc632ef5f0943f1c17c6b6373cfd7d3777aa5ffa60d6850ed9156790ace3.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 173, + 307, + 438, + 316.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 173, + 316.3333333333333, + 438, + 325.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 173, + 325.66666666666663, + 438, + 334.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 212, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 212, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 130, + 353 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 341, + 156, + 351 + ], + "score": 0.9, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 339, + 212, + 353 + ], + "score": 1.0, + "content": ", or otherwise", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 339, + 212, + 353 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 356, + 438, + 384 + ], + "lines": [ + { + "bbox": [ + 173, + 356, + 438, + 384 + ], + "spans": [ + { + "bbox": [ + 173, + 356, + 438, + 384 + ], + "score": 0.94, + "content": "\\pmb { w } _ { + } ^ { S } ( t ) = - \\pmb { w } _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } X \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X ^ { \\top } X t } \\right) ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } .", + "type": "interline_equation", + "image_path": "7881650689b109f035de1a165987e743b3b7bd91073898c77af0933159d8d813.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 173, + 356, + 438, + 365.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 173, + 365.3333333333333, + 438, + 374.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 173, + 374.66666666666663, + 438, + 383.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 389, + 506, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 131, + 401 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 390, + 157, + 399 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 389, + 351, + 401 + ], + "score": 1.0, + "content": ". For simplicity we elaborate the proof only for", + "type": "text" + }, + { + "bbox": [ + 351, + 389, + 377, + 399 + ], + "score": 0.89, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 389, + 505, + 401 + ], + "score": 1.0, + "content": ". We provide a condition under", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 401, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 505, + 411 + ], + "score": 1.0, + "content": "which the difference between the trajectories can be controlled, and we first that this condition holds", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 411, + 330, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 266, + 423 + ], + "score": 1.0, + "content": "for the linearized flow GF-single for all", + "type": "text" + }, + { + "bbox": [ + 266, + 412, + 271, + 421 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 411, + 330, + 423 + ], + "score": 1.0, + "content": "in Lemma 17.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 389, + 505, + 423 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 428, + 455, + 456 + ], + "lines": [ + { + "bbox": [ + 156, + 428, + 455, + 456 + ], + "spans": [ + { + "bbox": [ + 156, + 428, + 455, + 456 + ], + "score": 0.91, + "content": "\\mathrm { C o n d i t i o n \\ A : } \\ \\| w ( t ) \\| _ { 2 } = O \\left( { \\frac { 1 } { \\sqrt { d } } } \\right) ; \\left\\| X ^ { \\top } w ( t ) \\right\\| _ { \\infty } = O \\left( { \\frac { \\mathrm { p o l y } \\log d } { \\sqrt { d } } } \\right) .", + "type": "interline_equation", + "image_path": "20e90bda0059164dc605b61170a2f4f17ac694686cc4b9ff7c0ca6d776f91e81.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 156, + 428, + 455, + 437.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 156, + 437.3333333333333, + 455, + 446.66666666666663 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 156, + 446.66666666666663, + 455, + 455.99999999999994 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "Step 2. Bounding the Difference in Gradient Flow Trajectory. Due to low rank property of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 477, + 451, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 451, + 490 + ], + "score": 1.0, + "content": "GF-single and GF-double, in this subsection we slightly abuse the notation and define", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 466, + 506, + 490 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 495, + 439, + 511 + ], + "lines": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "spans": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "score": 0.88, + "content": "f ( \\pmb { x } ; \\pmb { w } _ { \\pm } ) = f ( \\pmb { x } ; \\pmb { w } _ { + } \\pmb { 1 } ^ { \\top } , \\pmb { w } _ { - } \\pmb { 1 } ^ { \\top } ) = \\sqrt { h } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } ) - \\sqrt { h } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } ) .", + "type": "interline_equation", + "image_path": "4763bc2bd2dade9cd1d216f11bf583a0917967d2ff51fe88646e3722206e783c.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 172, + 495, + 439, + 511 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 516, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "We now show that the difference between the two trajectories defined above is asymptotically", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 527, + 486, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 162, + 540 + ], + "score": 1.0, + "content": "vanishing for", + "type": "text" + }, + { + "bbox": [ + 162, + 528, + 177, + 539 + ], + "score": 0.84, + "content": "{ \\pmb w } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 527, + 182, + 540 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 182, + 528, + 199, + 539 + ], + "score": 0.69, + "content": "\\mathbf { \\nabla } w _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 527, + 477, + 540 + ], + "score": 1.0, + "content": "follows the same argument). Compare the two trajectories up to time", + "type": "text" + }, + { + "bbox": [ + 477, + 528, + 486, + 537 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 515, + 505, + 540 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 105, + 543, + 500, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 500, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 500, + 678 + ], + "score": 0.94, + "content": "\\begin{array} { r l r } { { \\| { \\boldsymbol w } _ { + } ^ { D } ( T ) - { \\boldsymbol w } _ { + } ^ { S } ( T ) } \\| _ { 2 } = \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) d t \\| _ { 2 } } \\\\ & { \\le \\| \\int _ { 0 } ^ { T } g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } + \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } } \\\\ & { = \\| \\int _ { 0 } ^ { T } \\frac { 1 } { 2 n _ { 0 } } \\frac { 2 n _ { 0 } } { i = 1 } [ ( - \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } + \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } ) \\phi ^ { \\prime } ( 0 ) { \\boldsymbol x } _ { i } ] \\mathrm { d } t \\| _ { 2 } } & \\\\ & { = O ( 1 ) \\int _ { 0 } ^ { T } ( \\| { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) \\| _ { 2 } + \\| { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) \\| _ { 2 } ) \\mathrm { d } t + { \\boldsymbol E } _ { + } , } & { ( 1 0 6 ) } \\end{array}", + "type": "interline_equation", + "image_path": "fd7c61c891b9274eed7e2f7c64f4f67079523a911d7dfdbb65843358a8774117.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 105, + 543, + 500, + 588.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 105, + 588.0, + 500, + 633.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 105, + 633.0, + 500, + 678.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 682, + 267, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 268, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 268, + 694 + ], + "score": 1.0, + "content": "where we have defined the error term as", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 681, + 268, + 694 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 696, + 400, + 730 + ], + "lines": [ + { + "bbox": [ + 210, + 696, + 400, + 730 + ], + "spans": [ + { + "bbox": [ + 210, + 696, + 400, + 730 + ], + "score": 0.93, + "content": "E _ { + } = \\left\\| \\int _ { 0 } ^ { T } \\pmb { g } _ { + } ^ { D } ( \\pmb { w } _ { + } ^ { D } ( s ) ) - \\pmb { g } _ { + } ^ { S } ( \\pmb { w } _ { + } ^ { D } ( s ) ) ~ \\mathrm { d } t \\right\\| _ { 2 } .", + "type": "interline_equation", + "image_path": "a5a4052e4440e0d387c9ef45daad6cf8330a1f9d303bf27b008b3b06d2e63429.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 696, + 400, + 713.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 210, + 713.0, + 400, + 730.0 + ], + "spans": [], + "index": 37 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 270, + 95 + ], + "score": 1.0, + "content": "To bound the error term, we note that at", + "type": "text" + }, + { + "bbox": [ + 271, + 83, + 294, + 93 + ], + "score": 0.9, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Condition A holds for GF-double. Assume that for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 370, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 130, + 106 + ], + "score": 1.0, + "content": "some", + "type": "text" + }, + { + "bbox": [ + 130, + 94, + 173, + 105 + ], + "score": 0.91, + "content": "0 \\leq t \\leq T", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 93, + 370, + 106 + ], + "score": 1.0, + "content": ", Condition A also holds for GF-double, we have", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 110, + 514, + 342 + ], + "lines": [ + { + "bbox": [ + 111, + 110, + 514, + 342 + ], + "spans": [ + { + "bbox": [ + 111, + 110, + 514, + 342 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { E _ { + } = \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { \\Phi } { u } \\int _ { 0 } ^ { u } [ u , \\frac { \\Phi } { u } ] ( \\boldsymbol { \\cdot } } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } \\\\ & { \\leq \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ \\frac { 1 } { \\sqrt { \\delta } } ( \\boldsymbol { \\cdot } - \\sqrt { \\delta } \\dot { u } ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) ) ( \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } u ( \\boldsymbol { \\cdot } ^ { \\Phi } , \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } \\rangle ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] \\Bigg \\| _ { 0 } ^ { 2 } , } \\\\ & \\quad + \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u _ { + } ^ { T } } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ ( - \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } ( u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] | \\int _ { 0 } ^ { T } \\ \\end{array}", + "type": "interline_equation", + "image_path": "75f36a782e901fa29ab4f484d3f34c84e620969b34c8f808daa14f7f72a17e26.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 110, + 514, + 187.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 187.33333333333331, + 514, + 264.66666666666663 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 264.66666666666663, + 514, + 341.99999999999994 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 507, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 331, + 367 + ], + "score": 1.0, + "content": "where (i) follows from the error of Taylor expansion on", + "type": "text" + }, + { + "bbox": [ + 331, + 354, + 339, + 365 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "and (ii) from Condition A. Therefore, by", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 365, + 277, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 277, + 378 + ], + "score": 1.0, + "content": "Equation (106) and Gronwall’s inequality,", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 379, + 441, + 407 + ], + "lines": [ + { + "bbox": [ + 169, + 379, + 441, + 407 + ], + "spans": [ + { + "bbox": [ + 169, + 379, + 441, + 407 + ], + "score": 0.94, + "content": "\\left\\| w _ { + } ^ { D } ( T ) - w _ { + } ^ { S } ( T ) \\right\\| _ { 2 } \\leq C _ { 1 } \\cdot \\frac { \\log ^ { c } h } { h } e ^ { C _ { 2 } T } = O \\left( \\frac { \\mathrm { p o l y l o g } h } { h } \\right) \\to 0 .", + "type": "interline_equation", + "image_path": "8502fa697ef4301e9370dc17fd11929d08c7c891a2f59fab8fde50582f9d06a8.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 169, + 379, + 441, + 388.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 169, + 388.3333333333333, + 441, + 397.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 169, + 397.66666666666663, + 441, + 406.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 506, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 122, + 423 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 122, + 411, + 196, + 423 + ], + "score": 0.91, + "content": "T \\in O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 410, + 319, + 423 + ], + "score": 1.0, + "content": ". This shows that up to time", + "type": "text" + }, + { + "bbox": [ + 320, + 411, + 328, + 421 + ], + "score": 0.26, + "content": "\\mathrm { T } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 410, + 506, + 423 + ], + "score": 1.0, + "content": ", the difference between the trajectories of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 104, + 420, + 464, + 434 + ], + "score": 1.0, + "content": "GF-single and GF-double vanishes under the the assumption that Condition A holds for", + "type": "text" + }, + { + "bbox": [ + 464, + 421, + 480, + 432 + ], + "score": 0.89, + "content": "\\mathbf { \\Delta } w ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "up to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 431, + 507, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 208, + 445 + ], + "score": 1.0, + "content": "T. Importantly, note that", + "type": "text" + }, + { + "bbox": [ + 208, + 432, + 298, + 445 + ], + "score": 0.93, + "content": "{ \\pmb w } ^ { S } ( 0 ) = { \\pmb w } ^ { D } ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 431, + 430, + 445 + ], + "score": 1.0, + "content": ", and that Condition A holds for", + "type": "text" + }, + { + "bbox": [ + 431, + 432, + 446, + 443 + ], + "score": 0.89, + "content": "\\pmb { w } ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 431, + 475, + 445 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 475, + 433, + 503, + 443 + ], + "score": 0.9, + "content": "T > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 431, + 507, + 445 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 442, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 456 + ], + "score": 1.0, + "content": "Therefore, by a standard contradiction argument (e.g. (Du et al., 2018, Lemma 3.4)), one can show", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 463, + 466 + ], + "score": 1.0, + "content": "that this closeness between the two trajectories implies that Condition A also holds for", + "type": "text" + }, + { + "bbox": [ + 464, + 454, + 480, + 465 + ], + "score": 0.88, + "content": "\\mathbf { \\Delta } w ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "up to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 465, + 488, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 488, + 478 + ], + "score": 1.0, + "content": "time T. With Equation (108) we bound the difference in population risk between the two flows.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 504, + 512 + ], + "lines": [ + { + "bbox": [ + 105, + 487, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 501 + ], + "score": 1.0, + "content": "Step 3. Bounding the Difference in Risk. Now we consider the difference of the population risk", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 497, + 339, + 515 + ], + "spans": [ + { + "bbox": [ + 107, + 499, + 120, + 510 + ], + "score": 0.89, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 497, + 139, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 140, + 499, + 155, + 510 + ], + "score": 0.9, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 497, + 282, + 515 + ], + "score": 1.0, + "content": "of two models with parameters", + "type": "text" + }, + { + "bbox": [ + 282, + 499, + 299, + 513 + ], + "score": 0.91, + "content": "\\pmb { w } _ { \\pm } ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 497, + 317, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 317, + 499, + 334, + 513 + ], + "score": 0.92, + "content": "\\pmb { w } _ { \\pm } ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 497, + 339, + 515 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "interline_equation", + "bbox": [ + 143, + 515, + 465, + 708 + ], + "lines": [ + { + "bbox": [ + 143, + 515, + 465, + 708 + ], + "spans": [ + { + "bbox": [ + 143, + 515, + 465, + 708 + ], + "score": 0.8, + "content": "\\begin{array} { r l } & { \\quad | ( { \\mathcal R } ^ { 3 } - { \\mathcal R } ^ { D } ) | = | \\mathbb { E } _ { { \\mathbf z } } ( x ^ { \\top } ) - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) | ^ { 2 } - \\mathbb { E } _ { { \\mathbf z } _ { \\mathbf z } } ( x ^ { \\top } \\beta - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { D } ) ) ^ { 2 } | } \\\\ & { \\stackrel { ( i ) } { \\le } \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ f ( { \\mathbf x } ; { \\mathbf x } _ { \\mathbf z } ^ { \\top } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } \\mathbb { E } _ { { \\mathbf z } } [ | { \\mathcal R } ^ { 7 } \\cdot f ( { \\mathbf x } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) + \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } } } \\\\ & \\le \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ | { \\mathcal R } | _ { { \\mathbf z } } [ \\langle \\delta | ^ { \\mathcal { R } } \\rangle - \\phi ; \\langle { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ] ^ { 2 } \\mathbb { E } _ { | { \\mathbf z } } ] + | ( { \\mathcal R } \\langle \\mathbf x ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ) [ ^ { 2 } } \\\\ & { \\quad \\cdot \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } ^ { 7 } - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) ] + | \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) | ] ^ { 2 } } } \\\\ & \\stackrel { ( i i ) } { \\le } 2 \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } \\mathbb { R } ^ { 5 } ] - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } ] ^ { 2 } + | \\beta \\langle \\mathbf w _ { \\mathbf z } ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } \\rangle | ^ { 2 } } \\\\ & \\quad \\cdot \\sqrt \\mathbb { E } _ { \\mathbf z } [ | { \\mathcal R } \\end{array}", + "type": "interline_equation", + "image_path": "8a06f95d31d3591849fac9bfce144a9b74f36b2b33c0ea0daa0eabf0899cc45d.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 143, + 515, + 465, + 579.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 143, + 579.3333333333334, + 465, + 643.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 143, + 643.6666666666667, + 465, + 708.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "where (i) is due to Cauchy-Schwarz inequality on norm, (ii) from Jenson’s inequality on squares and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 721, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 734 + ], + "score": 1.0, + "content": "Young’s inequality, and (iii) from the Lipschitz assumption on the activation, and the observation", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + } + ], + "page_idx": 27, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "28", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 270, + 95 + ], + "score": 1.0, + "content": "To bound the error term, we note that at", + "type": "text" + }, + { + "bbox": [ + 271, + 83, + 294, + 93 + ], + "score": 0.9, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Condition A holds for GF-double. Assume that for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 370, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 130, + 106 + ], + "score": 1.0, + "content": "some", + "type": "text" + }, + { + "bbox": [ + 130, + 94, + 173, + 105 + ], + "score": 0.91, + "content": "0 \\leq t \\leq T", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 93, + 370, + 106 + ], + "score": 1.0, + "content": ", Condition A also holds for GF-double, we have", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 110, + 514, + 342 + ], + "lines": [ + { + "bbox": [ + 111, + 110, + 514, + 342 + ], + "spans": [ + { + "bbox": [ + 111, + 110, + 514, + 342 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { E _ { + } = \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { \\Phi } { u } \\int _ { 0 } ^ { u } [ u , \\frac { \\Phi } { u } ] ( \\boldsymbol { \\cdot } } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } \\\\ & { \\leq \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ \\frac { 1 } { \\sqrt { \\delta } } ( \\boldsymbol { \\cdot } - \\sqrt { \\delta } \\dot { u } ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) ) ( \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } u ( \\boldsymbol { \\cdot } ^ { \\Phi } , \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } \\rangle ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] \\Bigg \\| _ { 0 } ^ { 2 } , } \\\\ & \\quad + \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u _ { + } ^ { T } } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ ( - \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } ( u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] | \\int _ { 0 } ^ { T } \\ \\end{array}", + "type": "interline_equation", + "image_path": "75f36a782e901fa29ab4f484d3f34c84e620969b34c8f808daa14f7f72a17e26.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 110, + 514, + 187.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 187.33333333333331, + 514, + 264.66666666666663 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 264.66666666666663, + 514, + 341.99999999999994 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 507, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 331, + 367 + ], + "score": 1.0, + "content": "where (i) follows from the error of Taylor expansion on", + "type": "text" + }, + { + "bbox": [ + 331, + 354, + 339, + 365 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "and (ii) from Condition A. Therefore, by", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 365, + 277, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 277, + 378 + ], + "score": 1.0, + "content": "Equation (106) and Gronwall’s inequality,", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 352, + 505, + 378 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 379, + 441, + 407 + ], + "lines": [ + { + "bbox": [ + 169, + 379, + 441, + 407 + ], + "spans": [ + { + "bbox": [ + 169, + 379, + 441, + 407 + ], + "score": 0.94, + "content": "\\left\\| w _ { + } ^ { D } ( T ) - w _ { + } ^ { S } ( T ) \\right\\| _ { 2 } \\leq C _ { 1 } \\cdot \\frac { \\log ^ { c } h } { h } e ^ { C _ { 2 } T } = O \\left( \\frac { \\mathrm { p o l y l o g } h } { h } \\right) \\to 0 .", + "type": "interline_equation", + "image_path": "8502fa697ef4301e9370dc17fd11929d08c7c891a2f59fab8fde50582f9d06a8.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 169, + 379, + 441, + 388.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 169, + 388.3333333333333, + 441, + 397.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 169, + 397.66666666666663, + 441, + 406.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 506, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 122, + 423 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 122, + 411, + 196, + 423 + ], + "score": 0.91, + "content": "T \\in O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 410, + 319, + 423 + ], + "score": 1.0, + "content": ". This shows that up to time", + "type": "text" + }, + { + "bbox": [ + 320, + 411, + 328, + 421 + ], + "score": 0.26, + "content": "\\mathrm { T } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 410, + 506, + 423 + ], + "score": 1.0, + "content": ", the difference between the trajectories of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 104, + 420, + 464, + 434 + ], + "score": 1.0, + "content": "GF-single and GF-double vanishes under the the assumption that Condition A holds for", + "type": "text" + }, + { + "bbox": [ + 464, + 421, + 480, + 432 + ], + "score": 0.89, + "content": "\\mathbf { \\Delta } w ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "up to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 431, + 507, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 208, + 445 + ], + "score": 1.0, + "content": "T. Importantly, note that", + "type": "text" + }, + { + "bbox": [ + 208, + 432, + 298, + 445 + ], + "score": 0.93, + "content": "{ \\pmb w } ^ { S } ( 0 ) = { \\pmb w } ^ { D } ( 0 ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 431, + 430, + 445 + ], + "score": 1.0, + "content": ", and that Condition A holds for", + "type": "text" + }, + { + "bbox": [ + 431, + 432, + 446, + 443 + ], + "score": 0.89, + "content": "\\pmb { w } ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 431, + 475, + 445 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 475, + 433, + 503, + 443 + ], + "score": 0.9, + "content": "T > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 431, + 507, + 445 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 442, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 456 + ], + "score": 1.0, + "content": "Therefore, by a standard contradiction argument (e.g. (Du et al., 2018, Lemma 3.4)), one can show", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 463, + 466 + ], + "score": 1.0, + "content": "that this closeness between the two trajectories implies that Condition A also holds for", + "type": "text" + }, + { + "bbox": [ + 464, + 454, + 480, + 465 + ], + "score": 0.88, + "content": "\\mathbf { \\Delta } w ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "up to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 465, + 488, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 488, + 478 + ], + "score": 1.0, + "content": "time T. With Equation (108) we bound the difference in population risk between the two flows.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 410, + 507, + 478 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 504, + 512 + ], + "lines": [ + { + "bbox": [ + 105, + 487, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 501 + ], + "score": 1.0, + "content": "Step 3. Bounding the Difference in Risk. Now we consider the difference of the population risk", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 497, + 339, + 515 + ], + "spans": [ + { + "bbox": [ + 107, + 499, + 120, + 510 + ], + "score": 0.89, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 497, + 139, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 140, + 499, + 155, + 510 + ], + "score": 0.9, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 497, + 282, + 515 + ], + "score": 1.0, + "content": "of two models with parameters", + "type": "text" + }, + { + "bbox": [ + 282, + 499, + 299, + 513 + ], + "score": 0.91, + "content": "\\pmb { w } _ { \\pm } ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 497, + 317, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 317, + 499, + 334, + 513 + ], + "score": 0.92, + "content": "\\pmb { w } _ { \\pm } ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 497, + 339, + 515 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 487, + 505, + 515 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 143, + 515, + 465, + 708 + ], + "lines": [ + { + "bbox": [ + 143, + 515, + 465, + 708 + ], + "spans": [ + { + "bbox": [ + 143, + 515, + 465, + 708 + ], + "score": 0.8, + "content": "\\begin{array} { r l } & { \\quad | ( { \\mathcal R } ^ { 3 } - { \\mathcal R } ^ { D } ) | = | \\mathbb { E } _ { { \\mathbf z } } ( x ^ { \\top } ) - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) | ^ { 2 } - \\mathbb { E } _ { { \\mathbf z } _ { \\mathbf z } } ( x ^ { \\top } \\beta - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { D } ) ) ^ { 2 } | } \\\\ & { \\stackrel { ( i ) } { \\le } \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ f ( { \\mathbf x } ; { \\mathbf x } _ { \\mathbf z } ^ { \\top } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } \\mathbb { E } _ { { \\mathbf z } } [ | { \\mathcal R } ^ { 7 } \\cdot f ( { \\mathbf x } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) + \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } } } \\\\ & \\le \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ | { \\mathcal R } | _ { { \\mathbf z } } [ \\langle \\delta | ^ { \\mathcal { R } } \\rangle - \\phi ; \\langle { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ] ^ { 2 } \\mathbb { E } _ { | { \\mathbf z } } ] + | ( { \\mathcal R } \\langle \\mathbf x ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ) [ ^ { 2 } } \\\\ & { \\quad \\cdot \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } ^ { 7 } - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) ] + | \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) | ] ^ { 2 } } } \\\\ & \\stackrel { ( i i ) } { \\le } 2 \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } \\mathbb { R } ^ { 5 } ] - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } ] ^ { 2 } + | \\beta \\langle \\mathbf w _ { \\mathbf z } ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } \\rangle | ^ { 2 } } \\\\ & \\quad \\cdot \\sqrt \\mathbb { E } _ { \\mathbf z } [ | { \\mathcal R } \\end{array}", + "type": "interline_equation", + "image_path": "8a06f95d31d3591849fac9bfce144a9b74f36b2b33c0ea0daa0eabf0899cc45d.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 143, + 515, + 465, + 579.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 143, + 579.3333333333334, + 465, + 643.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 143, + 643.6666666666667, + 465, + 708.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "where (i) is due to Cauchy-Schwarz inequality on norm, (ii) from Jenson’s inequality on squares and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 721, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 734 + ], + "score": 1.0, + "content": "Young’s inequality, and (iii) from the Lipschitz assumption on the activation, and the observation", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 710, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 145, + 96 + ], + "score": 1.0, + "content": "that both", + "type": "text" + }, + { + "bbox": [ + 146, + 82, + 160, + 92 + ], + "score": 0.89, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 80, + 179, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 180, + 82, + 195, + 92 + ], + "score": 0.89, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 80, + 250, + 96 + ], + "score": 1.0, + "content": "are finite for", + "type": "text" + }, + { + "bbox": [ + 251, + 83, + 282, + 94 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "due to the justified Condition A above. Therefore the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 90, + 362, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 90, + 185, + 107 + ], + "score": 1.0, + "content": "difference between", + "type": "text" + }, + { + "bbox": [ + 185, + 93, + 199, + 104 + ], + "score": 0.88, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 90, + 217, + 107 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 218, + 93, + 233, + 104 + ], + "score": 0.91, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 90, + 286, + 107 + ], + "score": 1.0, + "content": "vanishes for", + "type": "text" + }, + { + "bbox": [ + 286, + 93, + 357, + 106 + ], + "score": 0.92, + "content": "T = O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 90, + 362, + 107 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 116, + 504, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 131 + ], + "score": 1.0, + "content": "Step 4. Bounding the Difference from Stationarity We compute the difference in risk between", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 375, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 225, + 141 + ], + "score": 1.0, + "content": "the model at some finite time", + "type": "text" + }, + { + "bbox": [ + 225, + 129, + 230, + 138 + ], + "score": 0.77, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 127, + 342, + 141 + ], + "score": 1.0, + "content": "and the stationary point i.e.", + "type": "text" + }, + { + "bbox": [ + 343, + 129, + 370, + 138 + ], + "score": 0.88, + "content": "t = \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 127, + 375, + 141 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 143, + 505, + 201 + ], + "lines": [ + { + "bbox": [ + 111, + 143, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 111, + 143, + 505, + 201 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\left| R ^ { S } ( t ) - R ^ { S } ( \\infty ) \\right| \\leq C \\sqrt { h } \\cdot \\left\\| w _ { + } ^ { S } ( t ) - w _ { + } ^ { S } ( \\infty ) \\right\\| = C \\sqrt { h } \\left\\| \\frac { 1 } { 2 \\sqrt { h } } e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X y \\right\\| _ { 2 } } \\\\ & { = C \\left\\| e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X X ^ { \\top } \\beta \\right\\| _ { 2 } \\leq C \\exp \\left( - \\phi ^ { \\prime } ( 0 ) ^ { 2 } \\left\\| \\frac { 1 } { n } X X ^ { \\top } \\right\\| _ { 2 } t \\right) \\| \\beta \\| _ { 2 } = C _ { 3 } e ^ { - C _ { 4 } t } . } \\end{array}", + "type": "interline_equation", + "image_path": "34524586fd5ffffc2b924b6ec60eec773e70f4360d1f22d5ce90c5996acb901e.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 111, + 143, + 505, + 162.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 111, + 162.33333333333334, + 505, + 181.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 111, + 181.66666666666669, + 505, + 201.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 424, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 425, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 160, + 229 + ], + "score": 1.0, + "content": "for constants", + "type": "text" + }, + { + "bbox": [ + 160, + 216, + 207, + 227 + ], + "score": 0.92, + "content": "C _ { 3 } , C _ { 4 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 213, + 365, + 229 + ], + "score": 1.0, + "content": ". Combining (110) and (111) yields for", + "type": "text" + }, + { + "bbox": [ + 365, + 216, + 421, + 227 + ], + "score": 0.9, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 213, + 425, + 229 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 230, + 438, + 276 + ], + "lines": [ + { + "bbox": [ + 172, + 230, + 438, + 276 + ], + "spans": [ + { + "bbox": [ + 172, + 230, + 438, + 276 + ], + "score": 0.92, + "content": "\\begin{array} { r } { | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { S } ( T ) - R ^ { D } ( T ) | + | R ^ { S } ( T ) - R ^ { S } ( \\infty ) | } \\\\ { = O ( \\frac { \\mathrm { p o l y l o g } h } { \\sqrt { h } } ) + O ( \\frac { 1 } { \\mathrm { p o l y l o g } h } ) 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "b76e3a6581adc46b42523ed137e642416aa0148e83feb1cd041022bcd34562e3.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 172, + 230, + 438, + 245.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 172, + 245.33333333333334, + 438, + 260.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 172, + 260.6666666666667, + 438, + 276.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 277, + 397, + 290 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 398, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 217, + 291 + ], + "score": 1.0, + "content": "The result in (112) for case", + "type": "text" + }, + { + "bbox": [ + 217, + 279, + 243, + 288 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 277, + 398, + 291 + ], + "score": 1.0, + "content": "follows a similar proof and is omitted.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 108, + 301, + 299, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 300, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 300, + 314 + ], + "score": 1.0, + "content": "C.8.3 FROM GF-ORIGINAL TO GF-DOUBLE", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 506, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "score": 1.0, + "content": "In this section we compare GF-original with GF-double. Note that the two flows differ only at", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 329, + 507, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 258, + 345 + ], + "score": 1.0, + "content": "initialization: vanishing initialization", + "type": "text" + }, + { + "bbox": [ + 258, + 331, + 363, + 344 + ], + "score": 0.91, + "content": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 329, + 454, + 345 + ], + "score": 1.0, + "content": "v.s. zero initialization", + "type": "text" + }, + { + "bbox": [ + 455, + 331, + 503, + 344 + ], + "score": 0.93, + "content": "{ \\pmb w } _ { i } ^ { D } ( 0 ) \\dot { = } { \\bf 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 329, + 507, + 345 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 210, + 356 + ], + "score": 1.0, + "content": "Note that at initialization", + "type": "text" + }, + { + "bbox": [ + 211, + 343, + 310, + 355 + ], + "score": 0.88, + "content": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } = { \\cal O } ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 342, + 506, + 356 + ], + "score": 1.0, + "content": ", and gradient flow decreases the empirical risk;", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "therefore the condition for Lemma 18 is satisfied, and by the Lipschitz condition on the empirical", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 364, + 177, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 177, + 377 + ], + "score": 1.0, + "content": "gradient we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 379, + 488, + 521 + ], + "lines": [ + { + "bbox": [ + 123, + 379, + 488, + 521 + ], + "spans": [ + { + "bbox": [ + 123, + 379, + 488, + 521 + ], + "score": 0.96, + "content": "\\begin{array} { l } { \\displaystyle \\left\\| W ^ { O } ( T ) - W ^ { D } ( T ) \\right\\| _ { F } ^ { 2 } } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) - W ^ { D } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\frac { \\partial \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } } { \\partial t } \\mathrm { d } t } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) ) \\right\\| _ { F } \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } + \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } ^ { 2 } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\left( 1 + L _ { f } \\right) \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } \\mathrm { d } t , } \\end{array}", + "type": "interline_equation", + "image_path": "e70ebf1301b03dada7aa6d1c2a09385f7980c4eb177878ec754e81c383cf9be1.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 123, + 379, + 488, + 426.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 123, + 426.3333333333333, + 488, + 473.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 123, + 473.66666666666663, + 488, + 521.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 523, + 291, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 291, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 291, + 536 + ], + "score": 1.0, + "content": "And hence by Gronwall’s lemma one obtains:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 538, + 452, + 556 + ], + "lines": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "spans": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "score": 0.9, + "content": "\\left\\| { W ^ { O } ( T ) - W ^ { D } ( T ) } \\right\\| _ { F } \\leq \\left\\| { W ^ { O } ( 0 ) } \\right\\| _ { F } e ^ { ( 1 + L _ { f } ) T / 2 } = O ( d ^ { - ( 1 + \\epsilon ) / 2 } ) e ^ { C T } .", + "type": "interline_equation", + "image_path": "9d3d723174790ed8f4e38ffa642641ceec70dfa1b95882eb6d01482d58efe5b5.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 559, + 370, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 370, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 370, + 571 + ], + "score": 1.0, + "content": "Therefore the difference in the function output can be bounded as", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 573, + 480, + 626 + ], + "lines": [ + { + "bbox": [ + 129, + 573, + 480, + 626 + ], + "spans": [ + { + "bbox": [ + 129, + 573, + 480, + 626 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\quad \\left\\| f ( \\pmb { x } , W ^ { D } ( T ) ) - f ( \\pmb { x } , W ^ { O } ( T ) ) \\right\\| _ { 2 } = \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) \\pmb { a } - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\pmb { a } \\right\\| _ { 2 } } \\\\ & { \\leq \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\right\\| _ { 2 } \\left\\| \\pmb { a } \\right\\| _ { 2 } \\leq L _ { \\phi } \\left\\| \\pmb { x } \\right\\| _ { 2 } \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } \\left\\| \\pmb { a } \\right\\| _ { 2 } } \\\\ & { = O ( \\sqrt { d } ) \\cdot \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } = O ( d ^ { - \\epsilon / 2 } ) e ^ { C T } . } \\end{array}", + "type": "interline_equation", + "image_path": "9b91cac8e5da3b0178e57a4079146859214f86ab98ae1bdc8d5e1d73bed50592.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 129, + 573, + 480, + 590.6666666666666 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 129, + 590.6666666666666, + 480, + 608.3333333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 129, + 608.3333333333333, + 480, + 625.9999999999999 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 393, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 394, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 136, + 641 + ], + "score": 1.0, + "content": "Taking", + "type": "text" + }, + { + "bbox": [ + 136, + 628, + 192, + 640 + ], + "score": 0.92, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 627, + 394, + 641 + ], + "score": 1.0, + "content": ", together with the same argument in Step 1 yields", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 643, + 398, + 671 + ], + "lines": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "spans": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "score": 0.94, + "content": "| R ^ { O } ( T ) - R ^ { D } ( T ) | = O \\left( \\frac { \\mathrm { p o l y l o g } h } { d ^ { \\epsilon / 2 } } \\right) \\to 0 .", + "type": "interline_equation", + "image_path": "807813dbde2f926a6b2fbc4cd2d6f038cc133a4c11978bc875be0d4da12fd46c.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 679, + 264, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 265, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 265, + 692 + ], + "score": 1.0, + "content": "C.8.4 PUTTING THINGS TOGETHER", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 318, + 712 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 318, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 258, + 713 + ], + "score": 1.0, + "content": "By (112) and (116), we know that for", + "type": "text" + }, + { + "bbox": [ + 258, + 700, + 315, + 711 + ], + "score": 0.9, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 698, + 318, + 713 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 714, + 447, + 730 + ], + "lines": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "spans": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "score": 0.88, + "content": "| R ^ { O } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { O } ( T ) - R ^ { D } ( T ) | + | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\to 0 .", + "type": "interline_equation", + "image_path": "4049008e35d94ec4525b927c36aa8a0b5c776990c8b8619e261eb269b4f5e87a.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "spans": [], + "index": 31 + } + ] + } + ], + "page_idx": 28, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 145, + 96 + ], + "score": 1.0, + "content": "that both", + "type": "text" + }, + { + "bbox": [ + 146, + 82, + 160, + 92 + ], + "score": 0.89, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 80, + 179, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 180, + 82, + 195, + 92 + ], + "score": 0.89, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 80, + 250, + 96 + ], + "score": 1.0, + "content": "are finite for", + "type": "text" + }, + { + "bbox": [ + 251, + 83, + 282, + 94 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "due to the justified Condition A above. Therefore the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 90, + 362, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 90, + 185, + 107 + ], + "score": 1.0, + "content": "difference between", + "type": "text" + }, + { + "bbox": [ + 185, + 93, + 199, + 104 + ], + "score": 0.88, + "content": "R ^ { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 90, + 217, + 107 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 218, + 93, + 233, + 104 + ], + "score": 0.91, + "content": "R ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 90, + 286, + 107 + ], + "score": 1.0, + "content": "vanishes for", + "type": "text" + }, + { + "bbox": [ + 286, + 93, + 357, + 106 + ], + "score": 0.92, + "content": "T = O ( \\log \\log h )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 90, + 362, + 107 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 104, + 80, + 506, + 107 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 116, + 504, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 131 + ], + "score": 1.0, + "content": "Step 4. Bounding the Difference from Stationarity We compute the difference in risk between", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 375, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 225, + 141 + ], + "score": 1.0, + "content": "the model at some finite time", + "type": "text" + }, + { + "bbox": [ + 225, + 129, + 230, + 138 + ], + "score": 0.77, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 127, + 342, + 141 + ], + "score": 1.0, + "content": "and the stationary point i.e.", + "type": "text" + }, + { + "bbox": [ + 343, + 129, + 370, + 138 + ], + "score": 0.88, + "content": "t = \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 127, + 375, + 141 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 115, + 506, + 141 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 143, + 505, + 201 + ], + "lines": [ + { + "bbox": [ + 111, + 143, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 111, + 143, + 505, + 201 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\left| R ^ { S } ( t ) - R ^ { S } ( \\infty ) \\right| \\leq C \\sqrt { h } \\cdot \\left\\| w _ { + } ^ { S } ( t ) - w _ { + } ^ { S } ( \\infty ) \\right\\| = C \\sqrt { h } \\left\\| \\frac { 1 } { 2 \\sqrt { h } } e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X y \\right\\| _ { 2 } } \\\\ & { = C \\left\\| e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X X ^ { \\top } \\beta \\right\\| _ { 2 } \\leq C \\exp \\left( - \\phi ^ { \\prime } ( 0 ) ^ { 2 } \\left\\| \\frac { 1 } { n } X X ^ { \\top } \\right\\| _ { 2 } t \\right) \\| \\beta \\| _ { 2 } = C _ { 3 } e ^ { - C _ { 4 } t } . } \\end{array}", + "type": "interline_equation", + "image_path": "34524586fd5ffffc2b924b6ec60eec773e70f4360d1f22d5ce90c5996acb901e.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 111, + 143, + 505, + 162.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 111, + 162.33333333333334, + 505, + 181.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 111, + 181.66666666666669, + 505, + 201.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 424, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 425, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 160, + 229 + ], + "score": 1.0, + "content": "for constants", + "type": "text" + }, + { + "bbox": [ + 160, + 216, + 207, + 227 + ], + "score": 0.92, + "content": "C _ { 3 } , C _ { 4 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 213, + 365, + 229 + ], + "score": 1.0, + "content": ". Combining (110) and (111) yields for", + "type": "text" + }, + { + "bbox": [ + 365, + 216, + 421, + 227 + ], + "score": 0.9, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 213, + 425, + 229 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 213, + 425, + 229 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 230, + 438, + 276 + ], + "lines": [ + { + "bbox": [ + 172, + 230, + 438, + 276 + ], + "spans": [ + { + "bbox": [ + 172, + 230, + 438, + 276 + ], + "score": 0.92, + "content": "\\begin{array} { r } { | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { S } ( T ) - R ^ { D } ( T ) | + | R ^ { S } ( T ) - R ^ { S } ( \\infty ) | } \\\\ { = O ( \\frac { \\mathrm { p o l y l o g } h } { \\sqrt { h } } ) + O ( \\frac { 1 } { \\mathrm { p o l y l o g } h } ) 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "b76e3a6581adc46b42523ed137e642416aa0148e83feb1cd041022bcd34562e3.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 172, + 230, + 438, + 245.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 172, + 245.33333333333334, + 438, + 260.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 172, + 260.6666666666667, + 438, + 276.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 277, + 397, + 290 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 398, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 217, + 291 + ], + "score": 1.0, + "content": "The result in (112) for case", + "type": "text" + }, + { + "bbox": [ + 217, + 279, + 243, + 288 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 277, + 398, + 291 + ], + "score": 1.0, + "content": "follows a similar proof and is omitted.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 277, + 398, + 291 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 301, + 299, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 300, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 300, + 314 + ], + "score": 1.0, + "content": "C.8.3 FROM GF-ORIGINAL TO GF-DOUBLE", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 506, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "score": 1.0, + "content": "In this section we compare GF-original with GF-double. Note that the two flows differ only at", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 329, + 507, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 258, + 345 + ], + "score": 1.0, + "content": "initialization: vanishing initialization", + "type": "text" + }, + { + "bbox": [ + 258, + 331, + 363, + 344 + ], + "score": 0.91, + "content": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 329, + 454, + 345 + ], + "score": 1.0, + "content": "v.s. zero initialization", + "type": "text" + }, + { + "bbox": [ + 455, + 331, + 503, + 344 + ], + "score": 0.93, + "content": "{ \\pmb w } _ { i } ^ { D } ( 0 ) \\dot { = } { \\bf 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 329, + 507, + 345 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 210, + 356 + ], + "score": 1.0, + "content": "Note that at initialization", + "type": "text" + }, + { + "bbox": [ + 211, + 343, + 310, + 355 + ], + "score": 0.88, + "content": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } = { \\cal O } ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 342, + 506, + 356 + ], + "score": 1.0, + "content": ", and gradient flow decreases the empirical risk;", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "therefore the condition for Lemma 18 is satisfied, and by the Lipschitz condition on the empirical", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 364, + 177, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 177, + 377 + ], + "score": 1.0, + "content": "gradient we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 320, + 507, + 377 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 379, + 488, + 521 + ], + "lines": [ + { + "bbox": [ + 123, + 379, + 488, + 521 + ], + "spans": [ + { + "bbox": [ + 123, + 379, + 488, + 521 + ], + "score": 0.96, + "content": "\\begin{array} { l } { \\displaystyle \\left\\| W ^ { O } ( T ) - W ^ { D } ( T ) \\right\\| _ { F } ^ { 2 } } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) - W ^ { D } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\frac { \\partial \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } } { \\partial t } \\mathrm { d } t } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) ) \\right\\| _ { F } \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } + \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } ^ { 2 } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\left( 1 + L _ { f } \\right) \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } \\mathrm { d } t , } \\end{array}", + "type": "interline_equation", + "image_path": "e70ebf1301b03dada7aa6d1c2a09385f7980c4eb177878ec754e81c383cf9be1.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 123, + 379, + 488, + 426.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 123, + 426.3333333333333, + 488, + 473.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 123, + 473.66666666666663, + 488, + 521.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 523, + 291, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 291, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 291, + 536 + ], + "score": 1.0, + "content": "And hence by Gronwall’s lemma one obtains:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 523, + 291, + 536 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 538, + 452, + 556 + ], + "lines": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "spans": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "score": 0.9, + "content": "\\left\\| { W ^ { O } ( T ) - W ^ { D } ( T ) } \\right\\| _ { F } \\leq \\left\\| { W ^ { O } ( 0 ) } \\right\\| _ { F } e ^ { ( 1 + L _ { f } ) T / 2 } = O ( d ^ { - ( 1 + \\epsilon ) / 2 } ) e ^ { C T } .", + "type": "interline_equation", + "image_path": "9d3d723174790ed8f4e38ffa642641ceec70dfa1b95882eb6d01482d58efe5b5.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 158, + 538, + 452, + 556 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 559, + 370, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 370, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 370, + 571 + ], + "score": 1.0, + "content": "Therefore the difference in the function output can be bounded as", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 557, + 370, + 571 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 573, + 480, + 626 + ], + "lines": [ + { + "bbox": [ + 129, + 573, + 480, + 626 + ], + "spans": [ + { + "bbox": [ + 129, + 573, + 480, + 626 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\quad \\left\\| f ( \\pmb { x } , W ^ { D } ( T ) ) - f ( \\pmb { x } , W ^ { O } ( T ) ) \\right\\| _ { 2 } = \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) \\pmb { a } - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\pmb { a } \\right\\| _ { 2 } } \\\\ & { \\leq \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\right\\| _ { 2 } \\left\\| \\pmb { a } \\right\\| _ { 2 } \\leq L _ { \\phi } \\left\\| \\pmb { x } \\right\\| _ { 2 } \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } \\left\\| \\pmb { a } \\right\\| _ { 2 } } \\\\ & { = O ( \\sqrt { d } ) \\cdot \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } = O ( d ^ { - \\epsilon / 2 } ) e ^ { C T } . } \\end{array}", + "type": "interline_equation", + "image_path": "9b91cac8e5da3b0178e57a4079146859214f86ab98ae1bdc8d5e1d73bed50592.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 129, + 573, + 480, + 590.6666666666666 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 129, + 590.6666666666666, + 480, + 608.3333333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 129, + 608.3333333333333, + 480, + 625.9999999999999 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 393, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 394, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 136, + 641 + ], + "score": 1.0, + "content": "Taking", + "type": "text" + }, + { + "bbox": [ + 136, + 628, + 192, + 640 + ], + "score": 0.92, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 627, + 394, + 641 + ], + "score": 1.0, + "content": ", together with the same argument in Step 1 yields", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 627, + 394, + 641 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 643, + 398, + 671 + ], + "lines": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "spans": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "score": 0.94, + "content": "| R ^ { O } ( T ) - R ^ { D } ( T ) | = O \\left( \\frac { \\mathrm { p o l y l o g } h } { d ^ { \\epsilon / 2 } } \\right) \\to 0 .", + "type": "interline_equation", + "image_path": "807813dbde2f926a6b2fbc4cd2d6f038cc133a4c11978bc875be0d4da12fd46c.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 213, + 643, + 398, + 671 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 679, + 264, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 265, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 265, + 692 + ], + "score": 1.0, + "content": "C.8.4 PUTTING THINGS TOGETHER", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 318, + 712 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 318, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 258, + 713 + ], + "score": 1.0, + "content": "By (112) and (116), we know that for", + "type": "text" + }, + { + "bbox": [ + 258, + 700, + 315, + 711 + ], + "score": 0.9, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 698, + 318, + 713 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 698, + 318, + 713 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 714, + 447, + 730 + ], + "lines": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "spans": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "score": 0.88, + "content": "| R ^ { O } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { O } ( T ) - R ^ { D } ( T ) | + | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\to 0 .", + "type": "interline_equation", + "image_path": "4049008e35d94ec4525b927c36aa8a0b5c776990c8b8619e261eb269b4f5e87a.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 163, + 714, + 447, + 730 + ], + "spans": [], + "index": 31 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 129 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 296, + 95 + ], + "score": 1.0, + "content": "Finally the proof is completed by observing that", + "type": "text" + }, + { + "bbox": [ + 296, + 82, + 329, + 95 + ], + "score": 0.93, + "content": "R ^ { S } ( \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "is the risk of the minimum-norm solution on", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 330, + 105 + ], + "score": 1.0, + "content": "the input discussed in Section 3. In addition, note that at", + "type": "text" + }, + { + "bbox": [ + 330, + 94, + 387, + 105 + ], + "score": 0.91, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 93, + 506, + 105 + ], + "score": 1.0, + "content": ", from (109)(114) one obtains", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 124, + 118 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 104, + 232, + 118 + ], + "score": 0.94, + "content": "\\left\\| \\mathbf { \\dot { W } } ^ { O } ( t ) - W ^ { S } ( t ) \\right\\| _ { F } \\to 0", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 102, + 379, + 118 + ], + "score": 1.0, + "content": ", and therefore by Lemma 18 we have", + "type": "text" + }, + { + "bbox": [ + 379, + 104, + 502, + 118 + ], + "score": 0.92, + "content": "\\left\\| \\partial L ( X ; W ^ { \\mathcal { O } } ( t ) ) / \\partial W \\right\\| _ { F } \\to 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 102, + 507, + 118 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 295, + 130 + ], + "score": 1.0, + "content": "i.e. the flow on the original objective reaches a", + "type": "text" + }, + { + "bbox": [ + 295, + 117, + 313, + 129 + ], + "score": 0.9, + "content": "o ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 117, + 425, + 130 + ], + "score": 1.0, + "content": "first-order stationary point.", + "type": "text" + }, + { + "bbox": [ + 494, + 117, + 506, + 128 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 230, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 230, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 230, + 159 + ], + "score": 1.0, + "content": "C.9 PROOF OF THEOREM 8", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 166, + 391, + 180 + ], + "lines": [ + { + "bbox": [ + 106, + 166, + 392, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 138, + 182 + ], + "score": 1.0, + "content": "Denote", + "type": "text" + }, + { + "bbox": [ + 138, + 168, + 268, + 180 + ], + "score": 0.92, + "content": "\\omega = \\mathrm { v e c } ( W ) = \\mathrm { v e c } ( [ W _ { + } , W _ { - } ] )", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 166, + 290, + 182 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 290, + 167, + 358, + 180 + ], + "score": 0.91, + "content": "\\omega _ { 0 } = \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 166, + 392, + 182 + ], + "score": 1.0, + "content": ". Define", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 184, + 383, + 214 + ], + "lines": [ + { + "bbox": [ + 227, + 184, + 383, + 214 + ], + "spans": [ + { + "bbox": [ + 227, + 184, + 383, + 214 + ], + "score": 0.94, + "content": "K ( t ) = \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ,", + "type": "interline_equation", + "image_path": "56f7456247d698ede4a0225c28ab28b09060851eec4e2f6c89a7530c305b555a.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 227, + 184, + 383, + 199.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 227, + 199.0, + 383, + 214.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 217, + 506, + 251 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "which is the kernel matrix of the neural tangent kernel Jacot et al. (2018); Du et al. (2018). In the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 228, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 506, + 240 + ], + "score": 1.0, + "content": "following sections we show that under the non-vanishing initialization, the trained two-layer network", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "is well-approximated by the regression model on the NTK, for which we derive the population risk.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 262, + 269, + 274 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 270, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 270, + 275 + ], + "score": 1.0, + "content": "C.9.1 THE KERNEL LINEARIZATION", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 318, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 319, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 131, + 295 + ], + "score": 1.0, + "content": "Write", + "type": "text" + }, + { + "bbox": [ + 132, + 281, + 246, + 294 + ], + "score": 0.93, + "content": "\\pmb { y } _ { N N } ( t ) = f _ { N N } ( \\boldsymbol { X } , t ) \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 280, + 319, + 295 + ], + "score": 1.0, + "content": "and its evolution:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 298, + 383, + 322 + ], + "lines": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "spans": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "score": 0.94, + "content": "\\mathrm { d } { \\pmb y } _ { N N } ( t ) = \\frac { 1 } { n } K ( t ) ( { \\pmb y } - { \\pmb y } _ { N N } ( t ) ) \\ \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "4435a090e161937c1ce992abecbecce3622bbf95c43cfcf615a4ea9b64cd589d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 262, + 337 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 262, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 262, + 339 + ], + "score": 1.0, + "content": "and the corresponding linearized flow:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 341, + 389, + 365 + ], + "lines": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "spans": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "score": 0.94, + "content": "\\mathrm { d } { \\pmb y } _ { N T K } ( t ) = \\frac { 1 } { n } K ( 0 ) ( { \\pmb y } - { \\pmb y } _ { N T K } ( t ) ) \\ \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "2f42e3df666f164d85771eac3a45efe612b52feea3878e2008794d014e4b4753.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 506, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 507, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 507, + 381 + ], + "score": 1.0, + "content": "Previous works (e.g. Du et al. (2018); Oymak and Soltanolkotabi (2019)) have proved (non-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 380, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 506, + 392 + ], + "score": 1.0, + "content": "asymptotically) that the two trajectories (119) and (120) are close if the model is overparameterized,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 121, + 402 + ], + "score": 1.0, + "content": "i.e.", + "type": "text" + }, + { + "bbox": [ + 122, + 391, + 174, + 402 + ], + "score": 0.92, + "content": "h = \\mathrm { p o l y } ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 390, + 505, + 402 + ], + "score": 1.0, + "content": ", under no assumptions on the teacher model. In our asymptotic setup (together with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 402, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "assumptions (A1-3)), we argue that similar conclusion holds without significant overparameterization.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 418, + 504, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 432, + 431 + ], + "score": 1.0, + "content": "We first show the global convergence of the training of two-layer neural network", + "type": "text" + }, + { + "bbox": [ + 433, + 419, + 452, + 430 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 417, + 505, + 431 + ], + "score": 1.0, + "content": ". We employ", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "an argument similar to (Du et al., 2018, Theo. 3.2) by first identifying the condition under which", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 441, + 238, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 238, + 453 + ], + "score": 1.0, + "content": "training converges at linear rate:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 506, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "score": 1.0, + "content": "From Corollary 15 we know that at initialization the lowest eigenvalue of the NTK matrix satisfies", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 490, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 107, + 491, + 193, + 504 + ], + "score": 0.91, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K ( 0 ) ) \\stackrel { } { = } O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 490, + 506, + 505 + ], + "score": 1.0, + "content": ", and thus the kernel regression on the NTK enjoys linear convergence. By", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 501, + 507, + 517 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 229, + 517 + ], + "score": 1.0, + "content": "Lemma 19, we know that for", + "type": "text" + }, + { + "bbox": [ + 229, + 502, + 352, + 515 + ], + "score": 0.91, + "content": "\\lVert W ( t ) - W ( 0 ) \\rVert _ { 2 } = O ( d ^ { 1 - \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 501, + 407, + 517 + ], + "score": 1.0, + "content": ", the order of", + "type": "text" + }, + { + "bbox": [ + 407, + 502, + 491, + 514 + ], + "score": 0.9, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K ( t ) ) \\stackrel { } { = } O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 501, + 507, + 517 + ], + "score": 1.0, + "content": "re-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 512, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 104, + 512, + 333, + 528 + ], + "score": 1.0, + "content": "mains unchanged. Therefore, if we assume that up to time", + "type": "text" + }, + { + "bbox": [ + 334, + 515, + 342, + 524 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 512, + 418, + 528 + ], + "score": 1.0, + "content": "the weights satisfy", + "type": "text" + }, + { + "bbox": [ + 418, + 514, + 505, + 526 + ], + "score": 0.89, + "content": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 524, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 107, + 525, + 147, + 538 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 524, + 278, + 541 + ], + "score": 1.0, + "content": ", then Condition B is satisfied for", + "type": "text" + }, + { + "bbox": [ + 279, + 528, + 300, + 538 + ], + "score": 0.88, + "content": "{ \\pmb y } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 524, + 506, + 541 + ], + "score": 1.0, + "content": "and from (Chizat and Bach, 2018b, Lemma B1) we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 537, + 260, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 260, + 550 + ], + "score": 1.0, + "content": "have the following linear convergence", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 553, + 400, + 569 + ], + "lines": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "spans": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| { \\pmb y } _ { N N } ( t ) - { \\pmb y } \\| _ { 2 } \\le C _ { 1 } \\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } e ^ { - C _ { 2 } t } . } \\end{array}", + "type": "interline_equation", + "image_path": "df0758e4be5c97a12f736387cbe744a7e30743976b9df9d3e125c98d50b4d4e4.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 573, + 506, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 132, + 587 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 573, + 245, + 587 + ], + "score": 0.92, + "content": "\\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } = O ( \\sqrt { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 573, + 343, + 587 + ], + "score": 1.0, + "content": "at initialization, setting", + "type": "text" + }, + { + "bbox": [ + 344, + 573, + 402, + 586 + ], + "score": 0.92, + "content": "T = O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 573, + 506, + 587 + ], + "score": 1.0, + "content": "ensures that the training", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 585, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 125, + 601 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 126, + 588, + 223, + 603 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\frac { 1 } { n } \\| { \\pmb y } _ { N N } ( T ) - { \\pmb y } \\| _ { 2 } ^ { 2 } 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 588, + 235, + 601 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 235, + 591, + 268, + 600 + ], + "score": 0.84, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 588, + 418, + 601 + ], + "score": 1.0, + "content": ". Consequently it is easy to check that", + "type": "text" + }, + { + "bbox": [ + 419, + 585, + 505, + 606 + ], + "score": 0.95, + "content": "\\left\\| \\frac { \\partial { \\cal L } ( X ; W ( T ) ) } { \\partial W ( T ) } \\right\\| _ { 2 } \\to 0", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 603, + 423, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 295, + 617 + ], + "score": 1.0, + "content": "and thus at time T the gradient flow reaches an", + "type": "text" + }, + { + "bbox": [ + 295, + 604, + 313, + 615 + ], + "score": 0.33, + "content": "\\mathsf { o } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 603, + 423, + 617 + ], + "score": 1.0, + "content": "first order stationary point.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 260, + 635 + ], + "score": 1.0, + "content": "We now verify that each weight vector", + "type": "text" + }, + { + "bbox": [ + 261, + 624, + 273, + 633 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 621, + 334, + 635 + ], + "score": 1.0, + "content": "travels at most", + "type": "text" + }, + { + "bbox": [ + 335, + 621, + 375, + 634 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 621, + 506, + 635 + ], + "score": 1.0, + "content": "from initialization. The norm of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 633, + 249, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 156, + 645 + ], + "score": 1.0, + "content": "gradient for", + "type": "text" + }, + { + "bbox": [ + 156, + 635, + 169, + 644 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 633, + 249, + 645 + ], + "score": 1.0, + "content": "can be bounded as:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 116, + 648, + 472, + 704 + ], + "lines": [ + { + "bbox": [ + 116, + 648, + 472, + 704 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 472, + 704 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\left\\| \\frac { \\partial L ( X ; \\boldsymbol { w } _ { i } ( t ) ) } { \\partial \\boldsymbol { w } _ { i } ( t ) } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { n } X \\left[ \\frac { 1 } { \\sqrt { h } } ( \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) ) \\circ \\phi ^ { \\prime } ( X ^ { \\top } \\boldsymbol { w } _ { i } ( t ) ) \\right] \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i ) } { \\leq } O ( 1 ) \\frac { 1 } { d ^ { 1 . 5 } } \\left\\| X \\right\\| _ { 2 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) \\right\\| _ { 2 } \\leq O ( 1 ) d ^ { - 1 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( 0 ) \\right\\| _ { 2 } e ^ { - t } , } \\end{array}", + "type": "interline_equation", + "image_path": "26784e8ec9088b862d2278ac4206066251b182d8ee5c17533cc7a3a6f6917f62.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 116, + 648, + 472, + 666.6666666666666 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 116, + 666.6666666666666, + 472, + 685.3333333333333 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 116, + 685.3333333333333, + 472, + 703.9999999999999 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 707, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 705, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 705, + 280, + 722 + ], + "score": 1.0, + "content": "where (i) follows from the boundedness of", + "type": "text" + }, + { + "bbox": [ + 280, + 708, + 290, + 720 + ], + "score": 0.86, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 705, + 417, + 722 + ], + "score": 1.0, + "content": ". Integrating the gradient yields", + "type": "text" + }, + { + "bbox": [ + 418, + 707, + 505, + 720 + ], + "score": 0.91, + "content": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 718, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 719, + 147, + 733 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 718, + 166, + 733 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 166, + 720, + 273, + 733 + ], + "score": 0.92, + "content": "\\| W ( t ) - W ( 0 ) \\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 718, + 393, + 733 + ], + "score": 1.0, + "content": ". Thus the distance traveled by", + "type": "text" + }, + { + "bbox": [ + 393, + 721, + 405, + 731 + ], + "score": 0.77, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 718, + 505, + 733 + ], + "score": 1.0, + "content": "indeed satisfies the norm", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + } + ], + "page_idx": 29, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 129 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 296, + 95 + ], + "score": 1.0, + "content": "Finally the proof is completed by observing that", + "type": "text" + }, + { + "bbox": [ + 296, + 82, + 329, + 95 + ], + "score": 0.93, + "content": "R ^ { S } ( \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "is the risk of the minimum-norm solution on", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 330, + 105 + ], + "score": 1.0, + "content": "the input discussed in Section 3. In addition, note that at", + "type": "text" + }, + { + "bbox": [ + 330, + 94, + 387, + 105 + ], + "score": 0.91, + "content": "T = \\log \\log h", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 93, + 506, + 105 + ], + "score": 1.0, + "content": ", from (109)(114) one obtains", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 124, + 118 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 104, + 232, + 118 + ], + "score": 0.94, + "content": "\\left\\| \\mathbf { \\dot { W } } ^ { O } ( t ) - W ^ { S } ( t ) \\right\\| _ { F } \\to 0", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 102, + 379, + 118 + ], + "score": 1.0, + "content": ", and therefore by Lemma 18 we have", + "type": "text" + }, + { + "bbox": [ + 379, + 104, + 502, + 118 + ], + "score": 0.92, + "content": "\\left\\| \\partial L ( X ; W ^ { \\mathcal { O } } ( t ) ) / \\partial W \\right\\| _ { F } \\to 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 102, + 507, + 118 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 295, + 130 + ], + "score": 1.0, + "content": "i.e. the flow on the original objective reaches a", + "type": "text" + }, + { + "bbox": [ + 295, + 117, + 313, + 129 + ], + "score": 0.9, + "content": "o ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 117, + 425, + 130 + ], + "score": 1.0, + "content": "first-order stationary point.", + "type": "text" + }, + { + "bbox": [ + 494, + 117, + 506, + 128 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 81, + 507, + 130 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 230, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 230, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 230, + 159 + ], + "score": 1.0, + "content": "C.9 PROOF OF THEOREM 8", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 166, + 391, + 180 + ], + "lines": [ + { + "bbox": [ + 106, + 166, + 392, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 138, + 182 + ], + "score": 1.0, + "content": "Denote", + "type": "text" + }, + { + "bbox": [ + 138, + 168, + 268, + 180 + ], + "score": 0.92, + "content": "\\omega = \\mathrm { v e c } ( W ) = \\mathrm { v e c } ( [ W _ { + } , W _ { - } ] )", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 166, + 290, + 182 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 290, + 167, + 358, + 180 + ], + "score": 0.91, + "content": "\\omega _ { 0 } = \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 166, + 392, + 182 + ], + "score": 1.0, + "content": ". Define", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 166, + 392, + 182 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 184, + 383, + 214 + ], + "lines": [ + { + "bbox": [ + 227, + 184, + 383, + 214 + ], + "spans": [ + { + "bbox": [ + 227, + 184, + 383, + 214 + ], + "score": 0.94, + "content": "K ( t ) = \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ,", + "type": "interline_equation", + "image_path": "56f7456247d698ede4a0225c28ab28b09060851eec4e2f6c89a7530c305b555a.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 227, + 184, + 383, + 199.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 227, + 199.0, + 383, + 214.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 217, + 506, + 251 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "which is the kernel matrix of the neural tangent kernel Jacot et al. (2018); Du et al. (2018). In the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 228, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 506, + 240 + ], + "score": 1.0, + "content": "following sections we show that under the non-vanishing initialization, the trained two-layer network", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "is well-approximated by the regression model on the NTK, for which we derive the population risk.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 217, + 506, + 252 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 262, + 269, + 274 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 270, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 270, + 275 + ], + "score": 1.0, + "content": "C.9.1 THE KERNEL LINEARIZATION", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 318, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 319, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 131, + 295 + ], + "score": 1.0, + "content": "Write", + "type": "text" + }, + { + "bbox": [ + 132, + 281, + 246, + 294 + ], + "score": 0.93, + "content": "\\pmb { y } _ { N N } ( t ) = f _ { N N } ( \\boldsymbol { X } , t ) \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 280, + 319, + 295 + ], + "score": 1.0, + "content": "and its evolution:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 280, + 319, + 295 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 298, + 383, + 322 + ], + "lines": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "spans": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "score": 0.94, + "content": "\\mathrm { d } { \\pmb y } _ { N N } ( t ) = \\frac { 1 } { n } K ( t ) ( { \\pmb y } - { \\pmb y } _ { N N } ( t ) ) \\ \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "4435a090e161937c1ce992abecbecce3622bbf95c43cfcf615a4ea9b64cd589d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 227, + 298, + 383, + 322 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 262, + 337 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 262, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 262, + 339 + ], + "score": 1.0, + "content": "and the corresponding linearized flow:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 325, + 262, + 339 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 341, + 389, + 365 + ], + "lines": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "spans": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "score": 0.94, + "content": "\\mathrm { d } { \\pmb y } _ { N T K } ( t ) = \\frac { 1 } { n } K ( 0 ) ( { \\pmb y } - { \\pmb y } _ { N T K } ( t ) ) \\ \\mathrm { d } t ,", + "type": "interline_equation", + "image_path": "2f42e3df666f164d85771eac3a45efe612b52feea3878e2008794d014e4b4753.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 221, + 341, + 389, + 365 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 506, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 507, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 507, + 381 + ], + "score": 1.0, + "content": "Previous works (e.g. Du et al. (2018); Oymak and Soltanolkotabi (2019)) have proved (non-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 380, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 506, + 392 + ], + "score": 1.0, + "content": "asymptotically) that the two trajectories (119) and (120) are close if the model is overparameterized,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 121, + 402 + ], + "score": 1.0, + "content": "i.e.", + "type": "text" + }, + { + "bbox": [ + 122, + 391, + 174, + 402 + ], + "score": 0.92, + "content": "h = \\mathrm { p o l y } ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 390, + 505, + 402 + ], + "score": 1.0, + "content": ", under no assumptions on the teacher model. In our asymptotic setup (together with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 402, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "assumptions (A1-3)), we argue that similar conclusion holds without significant overparameterization.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 367, + 507, + 415 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 418, + 504, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 432, + 431 + ], + "score": 1.0, + "content": "We first show the global convergence of the training of two-layer neural network", + "type": "text" + }, + { + "bbox": [ + 433, + 419, + 452, + 430 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 417, + 505, + 431 + ], + "score": 1.0, + "content": ". We employ", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "an argument similar to (Du et al., 2018, Theo. 3.2) by first identifying the condition under which", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 441, + 238, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 238, + 453 + ], + "score": 1.0, + "content": "training converges at linear rate:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 417, + 505, + 453 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 506, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "score": 1.0, + "content": "From Corollary 15 we know that at initialization the lowest eigenvalue of the NTK matrix satisfies", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 490, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 107, + 491, + 193, + 504 + ], + "score": 0.91, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K ( 0 ) ) \\stackrel { } { = } O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 490, + 506, + 505 + ], + "score": 1.0, + "content": ", and thus the kernel regression on the NTK enjoys linear convergence. By", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 501, + 507, + 517 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 229, + 517 + ], + "score": 1.0, + "content": "Lemma 19, we know that for", + "type": "text" + }, + { + "bbox": [ + 229, + 502, + 352, + 515 + ], + "score": 0.91, + "content": "\\lVert W ( t ) - W ( 0 ) \\rVert _ { 2 } = O ( d ^ { 1 - \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 501, + 407, + 517 + ], + "score": 1.0, + "content": ", the order of", + "type": "text" + }, + { + "bbox": [ + 407, + 502, + 491, + 514 + ], + "score": 0.9, + "content": "\\lambda _ { \\operatorname* { m i n } } ( K ( t ) ) \\stackrel { } { = } O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 501, + 507, + 517 + ], + "score": 1.0, + "content": "re-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 512, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 104, + 512, + 333, + 528 + ], + "score": 1.0, + "content": "mains unchanged. Therefore, if we assume that up to time", + "type": "text" + }, + { + "bbox": [ + 334, + 515, + 342, + 524 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 512, + 418, + 528 + ], + "score": 1.0, + "content": "the weights satisfy", + "type": "text" + }, + { + "bbox": [ + 418, + 514, + 505, + 526 + ], + "score": 0.89, + "content": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 524, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 107, + 525, + 147, + 538 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 524, + 278, + 541 + ], + "score": 1.0, + "content": ", then Condition B is satisfied for", + "type": "text" + }, + { + "bbox": [ + 279, + 528, + 300, + 538 + ], + "score": 0.88, + "content": "{ \\pmb y } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 524, + 506, + 541 + ], + "score": 1.0, + "content": "and from (Chizat and Bach, 2018b, Lemma B1) we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 537, + 260, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 260, + 550 + ], + "score": 1.0, + "content": "have the following linear convergence", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 479, + 507, + 550 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 553, + 400, + 569 + ], + "lines": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "spans": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| { \\pmb y } _ { N N } ( t ) - { \\pmb y } \\| _ { 2 } \\le C _ { 1 } \\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } e ^ { - C _ { 2 } t } . } \\end{array}", + "type": "interline_equation", + "image_path": "df0758e4be5c97a12f736387cbe744a7e30743976b9df9d3e125c98d50b4d4e4.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 210, + 553, + 400, + 569 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 573, + 506, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 132, + 587 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 573, + 245, + 587 + ], + "score": 0.92, + "content": "\\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } = O ( \\sqrt { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 573, + 343, + 587 + ], + "score": 1.0, + "content": "at initialization, setting", + "type": "text" + }, + { + "bbox": [ + 344, + 573, + 402, + 586 + ], + "score": 0.92, + "content": "T = O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 573, + 506, + 587 + ], + "score": 1.0, + "content": "ensures that the training", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 585, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 125, + 601 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 126, + 588, + 223, + 603 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\frac { 1 } { n } \\| { \\pmb y } _ { N N } ( T ) - { \\pmb y } \\| _ { 2 } ^ { 2 } 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 588, + 235, + 601 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 235, + 591, + 268, + 600 + ], + "score": 0.84, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 588, + 418, + 601 + ], + "score": 1.0, + "content": ". Consequently it is easy to check that", + "type": "text" + }, + { + "bbox": [ + 419, + 585, + 505, + 606 + ], + "score": 0.95, + "content": "\\left\\| \\frac { \\partial { \\cal L } ( X ; W ( T ) ) } { \\partial W ( T ) } \\right\\| _ { 2 } \\to 0", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 603, + 423, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 295, + 617 + ], + "score": 1.0, + "content": "and thus at time T the gradient flow reaches an", + "type": "text" + }, + { + "bbox": [ + 295, + 604, + 313, + 615 + ], + "score": 0.33, + "content": "\\mathsf { o } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 603, + 423, + 617 + ], + "score": 1.0, + "content": "first order stationary point.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 573, + 506, + 617 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 260, + 635 + ], + "score": 1.0, + "content": "We now verify that each weight vector", + "type": "text" + }, + { + "bbox": [ + 261, + 624, + 273, + 633 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 621, + 334, + 635 + ], + "score": 1.0, + "content": "travels at most", + "type": "text" + }, + { + "bbox": [ + 335, + 621, + 375, + 634 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 621, + 506, + 635 + ], + "score": 1.0, + "content": "from initialization. The norm of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 633, + 249, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 156, + 645 + ], + "score": 1.0, + "content": "gradient for", + "type": "text" + }, + { + "bbox": [ + 156, + 635, + 169, + 644 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 633, + 249, + 645 + ], + "score": 1.0, + "content": "can be bounded as:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 621, + 506, + 645 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 116, + 648, + 472, + 704 + ], + "lines": [ + { + "bbox": [ + 116, + 648, + 472, + 704 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 472, + 704 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\left\\| \\frac { \\partial L ( X ; \\boldsymbol { w } _ { i } ( t ) ) } { \\partial \\boldsymbol { w } _ { i } ( t ) } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { n } X \\left[ \\frac { 1 } { \\sqrt { h } } ( \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) ) \\circ \\phi ^ { \\prime } ( X ^ { \\top } \\boldsymbol { w } _ { i } ( t ) ) \\right] \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i ) } { \\leq } O ( 1 ) \\frac { 1 } { d ^ { 1 . 5 } } \\left\\| X \\right\\| _ { 2 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) \\right\\| _ { 2 } \\leq O ( 1 ) d ^ { - 1 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( 0 ) \\right\\| _ { 2 } e ^ { - t } , } \\end{array}", + "type": "interline_equation", + "image_path": "26784e8ec9088b862d2278ac4206066251b182d8ee5c17533cc7a3a6f6917f62.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 116, + 648, + 472, + 666.6666666666666 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 116, + 666.6666666666666, + 472, + 685.3333333333333 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 116, + 685.3333333333333, + 472, + 703.9999999999999 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 707, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 705, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 705, + 280, + 722 + ], + "score": 1.0, + "content": "where (i) follows from the boundedness of", + "type": "text" + }, + { + "bbox": [ + 280, + 708, + 290, + 720 + ], + "score": 0.86, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 705, + 417, + 722 + ], + "score": 1.0, + "content": ". Integrating the gradient yields", + "type": "text" + }, + { + "bbox": [ + 418, + 707, + 505, + 720 + ], + "score": 0.91, + "content": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 718, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 719, + 147, + 733 + ], + "score": 0.92, + "content": "O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 718, + 166, + 733 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 166, + 720, + 273, + 733 + ], + "score": 0.92, + "content": "\\| W ( t ) - W ( 0 ) \\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 718, + 393, + 733 + ], + "score": 1.0, + "content": ". Thus the distance traveled by", + "type": "text" + }, + { + "bbox": [ + 393, + 721, + 405, + 731 + ], + "score": 0.77, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 718, + 505, + 733 + ], + "score": 1.0, + "content": "indeed satisfies the norm", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "assumption above, and following the same argument as (Du et al., 2018, Lemma 3.4) we conclude", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 375, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 312, + 105 + ], + "score": 1.0, + "content": "that Condition B holds true for the gradient flow of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 312, + 94, + 333, + 105 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 333, + 94, + 348, + 105 + ], + "score": 1.0, + "content": "for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 348, + 94, + 371, + 104 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 372, + 94, + 375, + 105 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 705, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "assumption above, and following the same argument as (Du et al., 2018, Lemma 3.4) we conclude", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 375, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 312, + 105 + ], + "score": 1.0, + "content": "that Condition B holds true for the gradient flow of", + "type": "text" + }, + { + "bbox": [ + 312, + 94, + 333, + 105 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 94, + 348, + 105 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 348, + 94, + 371, + 104 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 94, + 375, + 105 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 110, + 507, + 156 + ], + "lines": [ + { + "bbox": [ + 104, + 110, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 104, + 111, + 243, + 125 + ], + "score": 1.0, + "content": "Next we show that for any input", + "type": "text" + }, + { + "bbox": [ + 243, + 113, + 251, + 122 + ], + "score": 0.83, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 111, + 274, + 125 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 274, + 110, + 340, + 124 + ], + "score": 0.93, + "content": "\\| \\hat { \\pmb { x } } \\| _ { 2 } = O ( \\sqrt { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 111, + 506, + 125 + ], + "score": 1.0, + "content": ", the prediction of the neural network is", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 123, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 135 + ], + "score": 1.0, + "content": "uniformly close to the prediction of the prediction of the kernel model, a result similar to (Arora et al.,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 506, + 146 + ], + "score": 1.0, + "content": "2019a, Lemma F.1). Following the notation of Arora et al. (2019a), we write the time derivative of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 144, + 176, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 176, + 157 + ], + "score": 1.0, + "content": "the prediction as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 157, + 506, + 223 + ], + "lines": [ + { + "bbox": [ + 111, + 157, + 506, + 181 + ], + "spans": [ + { + "bbox": [ + 111, + 157, + 506, + 181 + ], + "score": 0.85, + "content": "\\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N N } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N N } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N N } ( t ) } ) ; \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N T K } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N T K } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N T K } ( t ) } ) ,", + "type": "inline_equation", + "image_path": "430edfce49c3ed4da1dfdf4682ea67ccabe1f7cb80b76384940436102f706c3a.jpg" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 181, + 507, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 133, + 204 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 183, + 307, + 202 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { \\pmb u } _ { N N } ( { \\pmb x } , t ) = \\frac { \\partial { \\pmb f } ( X ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } ^ { \\top } \\frac { \\partial { \\pmb f } ( { \\pmb x } ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } \\in \\mathbb { R } ^ { n } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 181, + 326, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 187, + 376, + 200 + ], + "score": 0.94, + "content": "{ \\pmb u } _ { N T K } ( { \\pmb x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 181, + 507, + 204 + ], + "score": 1.0, + "content": "similarly defined on the initial-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 161, + 213 + ], + "score": 1.0, + "content": "ized weights", + "type": "text" + }, + { + "bbox": [ + 161, + 201, + 181, + 213 + ], + "score": 0.86, + "content": "\\omega ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 200, + 405, + 213 + ], + "score": 1.0, + "content": ". We bound the difference between the predictions on", + "type": "text" + }, + { + "bbox": [ + 405, + 201, + 413, + 211 + ], + "score": 0.8, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "up to terminal time T", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 118, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 118, + 224 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 222, + 460, + 398 + ], + "lines": [ + { + "bbox": [ + 150, + 222, + 460, + 398 + ], + "spans": [ + { + "bbox": [ + 150, + 222, + 460, + 398 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad | \\int _ { \\mathbf { N } ^ { \\mathrm { N } } } ( \\hat { x } , t ) - \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) | = | \\int _ { 0 } ^ { T } \\left[ \\frac { \\mathrm { d } } { \\mathrm { d } t } \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) - \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\right] \\mathrm { d } t | } \\\\ & { = \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left[ \\mathbf { a } _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N N } ( t ) ) - u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N T } ( \\hat { x } ) ) \\right] \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } | \\int _ { 0 } ^ { T } u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y _ { N N } ( t ) - y _ { N \\wedge \\mathbf { K } } ( t ) ) \\mathrm { d } t | } \\\\ & { \\quad + \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) ^ { \\top } ( y - y _ { N N } ( t ) ) \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } \\| u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y _ { N N } ( t ) - y _ { N T \\mathbf { K } } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\| u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\operatorname* { m a x } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t . } \\end{array}", + "type": "interline_equation", + "image_path": "7db2b15e344d4697b163e0c2a6c9aedc6405d407e15cde0710d3770ed070662a.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 150, + 222, + 460, + 280.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 280.6666666666667, + 460, + 339.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 150, + 339.33333333333337, + 460, + 398.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 197, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 198, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 198, + 415 + ], + "score": 1.0, + "content": "For the first term have", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 413, + 500, + 500 + ], + "lines": [ + { + "bbox": [ + 104, + 413, + 500, + 500 + ], + "spans": [ + { + "bbox": [ + 104, + 413, + 500, + 500 + ], + "score": 0.91, + "content": "\\begin{array} { r l r } { { \\| { \\boldsymbol y } _ { N N } ( T ) - { \\boldsymbol y } _ { N T K } ( T ) \\| _ { 2 } \\le \\frac { 1 } { n } \\int _ { 0 } ^ { T } \\| K ( t ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) ) - K ( 0 ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N T K } ( t ) ) \\| _ { 2 } \\mathrm { d } t } } \\\\ & { } & { \\le \\frac { 1 } { n 0 < t < T } \\| K ( t ) - K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t + \\frac { 1 } { n } \\| K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { } & { \\overset { ( i ) } { \\le } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) + O ( 1 ) \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t \\overset { ( i i ) } { \\le } O ( d ^ { - \\epsilon ^ { \\prime } } ) , \\quad \\quad \\quad \\quad ( 1 2 4 ) \\mathrm { d } { \\boldsymbol z } . } \\end{array}", + "type": "interline_equation", + "image_path": "f6bf1504965a4e587dabfba74524500e87a4800efd3b2b27f2fef2157cbef0bd.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 104, + 413, + 500, + 442.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 104, + 442.0, + 500, + 471.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 104, + 471.0, + 500, + 500.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 500, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 104, + 498, + 317, + 514 + ], + "score": 1.0, + "content": "where (i) is due to Corollary 15, Lemma 19 for some", + "type": "text" + }, + { + "bbox": [ + 317, + 501, + 343, + 511 + ], + "score": 0.9, + "content": "\\epsilon ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 498, + 462, + 514 + ], + "score": 1.0, + "content": "and the linear convergence of", + "type": "text" + }, + { + "bbox": [ + 463, + 502, + 484, + 512 + ], + "score": 0.88, + "content": "{ \\bf { \\it { \\mathbf { y } } } } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 498, + 506, + 514 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 351, + 523 + ], + "score": 1.0, + "content": "(ii) is due to Gronwall’s inequality. Note that the log factor in", + "type": "text" + }, + { + "bbox": [ + 351, + 512, + 360, + 521 + ], + "score": 0.85, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "is omitted. Similarly, for the second", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 523, + 163, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 163, + 533 + ], + "score": 1.0, + "content": "term we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 532, + 500, + 561 + ], + "lines": [ + { + "bbox": [ + 108, + 532, + 500, + 561 + ], + "spans": [ + { + "bbox": [ + 108, + 532, + 500, + 561 + ], + "score": 0.92, + "content": "\\frac { 1 } { n \\hbar \\epsilon \\mathcal { T } } \\left. u _ { N N } ( \\hat { x } , t ) - u _ { N T K } ( \\hat { x } , t ) \\right. _ { 2 } \\int _ { 0 } ^ { T } \\left. y - y _ { N N } ( t ) \\right. _ { 2 } \\mathrm { d } t \\overset { ( i ) } { \\leq } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) = O ( d ^ { - \\epsilon ^ { \\prime } } ) ,", + "type": "interline_equation", + "image_path": "e069e54b2709bf413fdd3d4367aa285ae74d3f8552429f93a5279ea72e114207.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 108, + 532, + 500, + 541.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 108, + 541.6666666666666, + 500, + 551.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 108, + 551.3333333333333, + 500, + 560.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 573, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 571, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 329, + 587 + ], + "score": 1.0, + "content": "where we used Lemma 19 and the linear convergence of", + "type": "text" + }, + { + "bbox": [ + 330, + 575, + 352, + 586 + ], + "score": 0.9, + "content": "{ \\pmb y } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 571, + 505, + 587 + ], + "score": 1.0, + "content": "in (i). Combining the two cases yields", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 130, + 586, + 458, + 610 + ], + "lines": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "spans": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "score": 0.93, + "content": "| f _ { N N } ( \\pmb { \\hat { x } } , t ) - f _ { N T K } ( \\pmb { \\hat { x } } , t ) | \\leq \\frac { 1 } { n } \\| \\pmb { u } _ { N T K } ( \\pmb { \\hat { x } } , t ) \\| _ { 2 } O ( d ^ { - \\epsilon ^ { \\prime } } ) + O ( d ^ { - \\epsilon ^ { \\prime } } ) \\overset { ( i ) } { = } O ( d ^ { - \\epsilon ^ { \\prime } } ) ,", + "type": "interline_equation", + "image_path": "b92e60630cc091fa04b42cc875a750c4c27f29ff8864975ac2bd6e3e283e65ec.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 609, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 506, + 623 + ], + "score": 1.0, + "content": "where we utilized Corollary 14 in (i). Thus we know that the difference between the population risk", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 619, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 117, + 635 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 621, + 138, + 633 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 619, + 156, + 635 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 156, + 622, + 182, + 633 + ], + "score": 0.91, + "content": "f _ { N T K }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 619, + 506, + 635 + ], + "score": 1.0, + "content": "is also asymptotically vanishing (note that the derivation above is independent of√", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 225, + 645 + ], + "score": 1.0, + "content": "the target function as long as", + "type": "text" + }, + { + "bbox": [ + 225, + 632, + 293, + 644 + ], + "score": 0.92, + "content": "\\bar { | | \\mathbf { y } | | _ { 2 } } = \\bar { O ( \\sqrt { n } ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 632, + 505, + 645 + ], + "score": 1.0, + "content": ". Therefore, in the following subsection we compute", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 641, + 298, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 267, + 657 + ], + "score": 1.0, + "content": "the risk of the linearized (kernel) model", + "type": "text" + }, + { + "bbox": [ + 267, + 644, + 293, + 655 + ], + "score": 0.8, + "content": "f _ { N T K }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 641, + 298, + 657 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 108, + 666, + 277, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 278, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 278, + 678 + ], + "score": 1.0, + "content": "C.9.2 COMPUTING THE KERNEL RISK", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 683, + 503, + 708 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 158, + 698 + ], + "score": 1.0, + "content": "Given input", + "type": "text" + }, + { + "bbox": [ + 158, + 684, + 205, + 695 + ], + "score": 0.92, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 682, + 247, + 698 + ], + "score": 1.0, + "content": "and label", + "type": "text" + }, + { + "bbox": [ + 248, + 684, + 311, + 697 + ], + "score": 0.92, + "content": "\\pmb { y } = \\pmb { \\beta } ^ { \\top } \\pmb { X } + \\pmb { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 682, + 505, + 698 + ], + "score": 1.0, + "content": ", gradient flow on the tangent kernel solves the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 695, + 269, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 256, + 709 + ], + "score": 1.0, + "content": "following equation of the parameters", + "type": "text" + }, + { + "bbox": [ + 256, + 699, + 264, + 706 + ], + "score": 0.77, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 695, + 269, + 709 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 708, + 389, + 736 + ], + "lines": [ + { + "bbox": [ + 221, + 708, + 389, + 736 + ], + "spans": [ + { + "bbox": [ + 221, + 708, + 389, + 736 + ], + "score": 0.94, + "content": "\\pmb { y } = \\pmb { f } ( X ; \\omega ) = \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } ( \\pmb { \\omega } - \\pmb { \\omega } _ { 0 } ) ,", + "type": "interline_equation", + "image_path": "52ed3c953ae3dba6d02df549f4eb4f4f76a89a2dabfe0cf2aca9e621bd12721f.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 708, + 389, + 722.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 221, + 722.0, + 389, + 736.0 + ], + "spans": [], + "index": 33 + } + ] + } + ], + "page_idx": 30, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 106 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 105 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 110, + 507, + 156 + ], + "lines": [ + { + "bbox": [ + 104, + 110, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 104, + 111, + 243, + 125 + ], + "score": 1.0, + "content": "Next we show that for any input", + "type": "text" + }, + { + "bbox": [ + 243, + 113, + 251, + 122 + ], + "score": 0.83, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 111, + 274, + 125 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 274, + 110, + 340, + 124 + ], + "score": 0.93, + "content": "\\| \\hat { \\pmb { x } } \\| _ { 2 } = O ( \\sqrt { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 111, + 506, + 125 + ], + "score": 1.0, + "content": ", the prediction of the neural network is", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 123, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 135 + ], + "score": 1.0, + "content": "uniformly close to the prediction of the prediction of the kernel model, a result similar to (Arora et al.,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 506, + 146 + ], + "score": 1.0, + "content": "2019a, Lemma F.1). Following the notation of Arora et al. (2019a), we write the time derivative of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 144, + 176, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 176, + 157 + ], + "score": 1.0, + "content": "the prediction as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 110, + 506, + 157 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 157, + 506, + 223 + ], + "lines": [ + { + "bbox": [ + 111, + 157, + 506, + 181 + ], + "spans": [ + { + "bbox": [ + 111, + 157, + 506, + 181 + ], + "score": 0.85, + "content": "\\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N N } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N N } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N N } ( t ) } ) ; \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N T K } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N T K } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N T K } ( t ) } ) ,", + "type": "inline_equation", + "image_path": "430edfce49c3ed4da1dfdf4682ea67ccabe1f7cb80b76384940436102f706c3a.jpg" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 181, + 507, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 133, + 204 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 183, + 307, + 202 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { \\pmb u } _ { N N } ( { \\pmb x } , t ) = \\frac { \\partial { \\pmb f } ( X ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } ^ { \\top } \\frac { \\partial { \\pmb f } ( { \\pmb x } ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } \\in \\mathbb { R } ^ { n } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 181, + 326, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 187, + 376, + 200 + ], + "score": 0.94, + "content": "{ \\pmb u } _ { N T K } ( { \\pmb x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 181, + 507, + 204 + ], + "score": 1.0, + "content": "similarly defined on the initial-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 161, + 213 + ], + "score": 1.0, + "content": "ized weights", + "type": "text" + }, + { + "bbox": [ + 161, + 201, + 181, + 213 + ], + "score": 0.86, + "content": "\\omega ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 200, + 405, + 213 + ], + "score": 1.0, + "content": ". We bound the difference between the predictions on", + "type": "text" + }, + { + "bbox": [ + 405, + 201, + 413, + 211 + ], + "score": 0.8, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "up to terminal time T", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 118, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 118, + 224 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 157, + 507, + 224 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 222, + 460, + 398 + ], + "lines": [ + { + "bbox": [ + 150, + 222, + 460, + 398 + ], + "spans": [ + { + "bbox": [ + 150, + 222, + 460, + 398 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad | \\int _ { \\mathbf { N } ^ { \\mathrm { N } } } ( \\hat { x } , t ) - \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) | = | \\int _ { 0 } ^ { T } \\left[ \\frac { \\mathrm { d } } { \\mathrm { d } t } \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) - \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\right] \\mathrm { d } t | } \\\\ & { = \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left[ \\mathbf { a } _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N N } ( t ) ) - u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N T } ( \\hat { x } ) ) \\right] \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } | \\int _ { 0 } ^ { T } u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y _ { N N } ( t ) - y _ { N \\wedge \\mathbf { K } } ( t ) ) \\mathrm { d } t | } \\\\ & { \\quad + \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) ^ { \\top } ( y - y _ { N N } ( t ) ) \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } \\| u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y _ { N N } ( t ) - y _ { N T \\mathbf { K } } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\| u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\operatorname* { m a x } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t . } \\end{array}", + "type": "interline_equation", + "image_path": "7db2b15e344d4697b163e0c2a6c9aedc6405d407e15cde0710d3770ed070662a.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 150, + 222, + 460, + 280.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 280.6666666666667, + 460, + 339.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 150, + 339.33333333333337, + 460, + 398.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 197, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 198, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 198, + 415 + ], + "score": 1.0, + "content": "For the first term have", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 401, + 198, + 415 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 413, + 500, + 500 + ], + "lines": [ + { + "bbox": [ + 104, + 413, + 500, + 500 + ], + "spans": [ + { + "bbox": [ + 104, + 413, + 500, + 500 + ], + "score": 0.91, + "content": "\\begin{array} { r l r } { { \\| { \\boldsymbol y } _ { N N } ( T ) - { \\boldsymbol y } _ { N T K } ( T ) \\| _ { 2 } \\le \\frac { 1 } { n } \\int _ { 0 } ^ { T } \\| K ( t ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) ) - K ( 0 ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N T K } ( t ) ) \\| _ { 2 } \\mathrm { d } t } } \\\\ & { } & { \\le \\frac { 1 } { n 0 < t < T } \\| K ( t ) - K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t + \\frac { 1 } { n } \\| K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { } & { \\overset { ( i ) } { \\le } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) + O ( 1 ) \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t \\overset { ( i i ) } { \\le } O ( d ^ { - \\epsilon ^ { \\prime } } ) , \\quad \\quad \\quad \\quad ( 1 2 4 ) \\mathrm { d } { \\boldsymbol z } . } \\end{array}", + "type": "interline_equation", + "image_path": "f6bf1504965a4e587dabfba74524500e87a4800efd3b2b27f2fef2157cbef0bd.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 104, + 413, + 500, + 442.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 104, + 442.0, + 500, + 471.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 104, + 471.0, + 500, + 500.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 500, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 104, + 498, + 317, + 514 + ], + "score": 1.0, + "content": "where (i) is due to Corollary 15, Lemma 19 for some", + "type": "text" + }, + { + "bbox": [ + 317, + 501, + 343, + 511 + ], + "score": 0.9, + "content": "\\epsilon ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 498, + 462, + 514 + ], + "score": 1.0, + "content": "and the linear convergence of", + "type": "text" + }, + { + "bbox": [ + 463, + 502, + 484, + 512 + ], + "score": 0.88, + "content": "{ \\bf { \\it { \\mathbf { y } } } } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 498, + 506, + 514 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 351, + 523 + ], + "score": 1.0, + "content": "(ii) is due to Gronwall’s inequality. Note that the log factor in", + "type": "text" + }, + { + "bbox": [ + 351, + 512, + 360, + 521 + ], + "score": 0.85, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "is omitted. Similarly, for the second", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 523, + 163, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 163, + 533 + ], + "score": 1.0, + "content": "term we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 104, + 498, + 506, + 533 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 532, + 500, + 561 + ], + "lines": [ + { + "bbox": [ + 108, + 532, + 500, + 561 + ], + "spans": [ + { + "bbox": [ + 108, + 532, + 500, + 561 + ], + "score": 0.92, + "content": "\\frac { 1 } { n \\hbar \\epsilon \\mathcal { T } } \\left. u _ { N N } ( \\hat { x } , t ) - u _ { N T K } ( \\hat { x } , t ) \\right. _ { 2 } \\int _ { 0 } ^ { T } \\left. y - y _ { N N } ( t ) \\right. _ { 2 } \\mathrm { d } t \\overset { ( i ) } { \\leq } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) = O ( d ^ { - \\epsilon ^ { \\prime } } ) ,", + "type": "interline_equation", + "image_path": "e069e54b2709bf413fdd3d4367aa285ae74d3f8552429f93a5279ea72e114207.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 108, + 532, + 500, + 541.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 108, + 541.6666666666666, + 500, + 551.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 108, + 551.3333333333333, + 500, + 560.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 573, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 571, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 329, + 587 + ], + "score": 1.0, + "content": "where we used Lemma 19 and the linear convergence of", + "type": "text" + }, + { + "bbox": [ + 330, + 575, + 352, + 586 + ], + "score": 0.9, + "content": "{ \\pmb y } _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 571, + 505, + 587 + ], + "score": 1.0, + "content": "in (i). Combining the two cases yields", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 571, + 505, + 587 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 130, + 586, + 458, + 610 + ], + "lines": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "spans": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "score": 0.93, + "content": "| f _ { N N } ( \\pmb { \\hat { x } } , t ) - f _ { N T K } ( \\pmb { \\hat { x } } , t ) | \\leq \\frac { 1 } { n } \\| \\pmb { u } _ { N T K } ( \\pmb { \\hat { x } } , t ) \\| _ { 2 } O ( d ^ { - \\epsilon ^ { \\prime } } ) + O ( d ^ { - \\epsilon ^ { \\prime } } ) \\overset { ( i ) } { = } O ( d ^ { - \\epsilon ^ { \\prime } } ) ,", + "type": "interline_equation", + "image_path": "b92e60630cc091fa04b42cc875a750c4c27f29ff8864975ac2bd6e3e283e65ec.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 130, + 586, + 458, + 610 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 609, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 506, + 623 + ], + "score": 1.0, + "content": "where we utilized Corollary 14 in (i). Thus we know that the difference between the population risk", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 619, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 117, + 635 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 621, + 138, + 633 + ], + "score": 0.91, + "content": "f _ { N N }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 619, + 156, + 635 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 156, + 622, + 182, + 633 + ], + "score": 0.91, + "content": "f _ { N T K }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 619, + 506, + 635 + ], + "score": 1.0, + "content": "is also asymptotically vanishing (note that the derivation above is independent of√", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 225, + 645 + ], + "score": 1.0, + "content": "the target function as long as", + "type": "text" + }, + { + "bbox": [ + 225, + 632, + 293, + 644 + ], + "score": 0.92, + "content": "\\bar { | | \\mathbf { y } | | _ { 2 } } = \\bar { O ( \\sqrt { n } ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 632, + 505, + 645 + ], + "score": 1.0, + "content": ". Therefore, in the following subsection we compute", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 641, + 298, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 267, + 657 + ], + "score": 1.0, + "content": "the risk of the linearized (kernel) model", + "type": "text" + }, + { + "bbox": [ + 267, + 644, + 293, + 655 + ], + "score": 0.8, + "content": "f _ { N T K }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 641, + 298, + 657 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 609, + 506, + 657 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 666, + 277, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 278, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 278, + 678 + ], + "score": 1.0, + "content": "C.9.2 COMPUTING THE KERNEL RISK", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 683, + 503, + 708 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 158, + 698 + ], + "score": 1.0, + "content": "Given input", + "type": "text" + }, + { + "bbox": [ + 158, + 684, + 205, + 695 + ], + "score": 0.92, + "content": "X \\in \\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 682, + 247, + 698 + ], + "score": 1.0, + "content": "and label", + "type": "text" + }, + { + "bbox": [ + 248, + 684, + 311, + 697 + ], + "score": 0.92, + "content": "\\pmb { y } = \\pmb { \\beta } ^ { \\top } \\pmb { X } + \\pmb { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 682, + 505, + 698 + ], + "score": 1.0, + "content": ", gradient flow on the tangent kernel solves the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 695, + 269, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 256, + 709 + ], + "score": 1.0, + "content": "following equation of the parameters", + "type": "text" + }, + { + "bbox": [ + 256, + 699, + 264, + 706 + ], + "score": 0.77, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 695, + 269, + 709 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 682, + 505, + 709 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 708, + 389, + 736 + ], + "lines": [ + { + "bbox": [ + 221, + 708, + 389, + 736 + ], + "spans": [ + { + "bbox": [ + 221, + 708, + 389, + 736 + ], + "score": 0.94, + "content": "\\pmb { y } = \\pmb { f } ( X ; \\omega ) = \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } ( \\pmb { \\omega } - \\pmb { \\omega } _ { 0 } ) ,", + "type": "interline_equation", + "image_path": "52ed3c953ae3dba6d02df549f4eb4f4f76a89a2dabfe0cf2aca9e621bd12721f.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 708, + 389, + 722.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 221, + 722.0, + 389, + 736.0 + ], + "spans": [], + "index": 33 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 81, + 506, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 133, + 95 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 82, + 199, + 94 + ], + "score": 0.91, + "content": "\\partial f ( X ; \\omega _ { 0 } ) / \\partial \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 82, + 216, + 95 + ], + "score": 1.0, + "content": "is a", + "type": "text" + }, + { + "bbox": [ + 216, + 83, + 246, + 93 + ], + "score": 0.9, + "content": "d h \\times n", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 82, + 330, + 95 + ], + "score": 1.0, + "content": "matrix with column", + "type": "text" + }, + { + "bbox": [ + 330, + 82, + 395, + 95 + ], + "score": 0.92, + "content": "\\partial f ( \\pmb { x } _ { i } ; \\pmb { \\omega } _ { 0 } ) / \\partial \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 82, + 454, + 95 + ], + "score": 1.0, + "content": ". Note that for", + "type": "text" + }, + { + "bbox": [ + 454, + 84, + 487, + 93 + ], + "score": 0.86, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 92, + 487, + 108 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 171, + 106 + ], + "score": 0.88, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 92, + 174, + 108 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 95, + 206, + 104 + ], + "score": 0.82, + "content": "d h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 92, + 487, + 108 + ], + "score": 1.0, + "content": "trivially holds, and by Corollary 15 we know that solution is given by", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 111, + 448, + 147 + ], + "lines": [ + { + "bbox": [ + 163, + 111, + 448, + 147 + ], + "spans": [ + { + "bbox": [ + 163, + 111, + 448, + 147 + ], + "score": 0.93, + "content": "\\omega _ { 1 } = \\omega _ { 0 } + \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\left( \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } ^ { \\top } \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\right) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) .", + "type": "interline_equation", + "image_path": "a593f61a44d8cf9d15adefc40e6776cd65d466e4235ec038ef7974f64a9b4196.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 163, + 111, + 448, + 123.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 163, + 123.0, + 448, + 135.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 163, + 135.0, + 448, + 147.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 152, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 506, + 165 + ], + "score": 1.0, + "content": "And the population risk can be written as (note that there is a factor of 2 due to the \"doubling trick\" at", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 161, + 252, + 177 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 198, + 177 + ], + "score": 1.0, + "content": "initialization to ensure", + "type": "text" + }, + { + "bbox": [ + 199, + 163, + 244, + 176 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0 .", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 161, + 252, + 177 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 181, + 500, + 332 + ], + "lines": [ + { + "bbox": [ + 104, + 181, + 500, + 332 + ], + "spans": [ + { + "bbox": [ + 104, + 181, + 500, + 332 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { 2 R = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - f ( x ; \\omega _ { 1 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } ( \\omega _ { 1 } - \\omega _ { 0 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial f ( X ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha } [ ( x ^ { \\top } \\beta - \\frac { \\partial \\tilde { f } ( \\kappa ^ { - 1 } X ^ { \\top } \\beta ) } { \\partial \\beta } \\frac { \\partial \\tilde { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial \\tilde { f } ( \\tilde { X } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } ) } { 2 V } \\frac { \\partial ^ { 2 } } { \\partial \\omega } , \\qquad ( 1 2 9 ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "72e1cf33d940d4afe182e3be2eaed1daa330fd38df051ecb9aedf2f1ef1f7d95.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 104, + 181, + 500, + 231.33333333333334 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 104, + 231.33333333333334, + 500, + 281.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 104, + 281.6666666666667, + 500, + 332.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 335, + 426, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 426, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 426, + 350 + ], + "score": 1.0, + "content": "where a bias-variance decomposition is made here, and for simplicity we define", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 353, + 441, + 382 + ], + "lines": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "spans": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "score": 0.94, + "content": "\\hat { \\pmb { u } } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( { \\pmb { x } } ; \\omega _ { 0 } ) } { \\partial \\omega } , \\quad \\hat { K } _ { X } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } .", + "type": "interline_equation", + "image_path": "c1ca5ef4f293f868c59af874b6e441350f64df81c280c8ef9f848f2fb3ee7da6.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 392, + 311, + 405 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 311, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 311, + 405 + ], + "score": 1.0, + "content": "C.9.3 APPROXIMATING THE KERNEL MATRIX", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 411, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 262, + 425 + ], + "score": 1.0, + "content": "In this section we drop the negligible", + "type": "text" + }, + { + "bbox": [ + 263, + 415, + 269, + 422 + ], + "score": 0.75, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 411, + 506, + 425 + ], + "score": 1.0, + "content": "in the initialization. Following Cheng and Singer (2013)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 296, + 435 + ], + "score": 1.0, + "content": "we utilize the orthonormal decomposition of", + "type": "text" + }, + { + "bbox": [ + 296, + 423, + 319, + 435 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 422, + 333, + 435 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 333, + 423, + 378, + 435 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 422, + 419, + 435 + ], + "score": 1.0, + "content": ". Denote", + "type": "text" + }, + { + "bbox": [ + 419, + 423, + 482, + 435 + ], + "score": 0.93, + "content": "b _ { 0 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 422, + 505, + 435 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 433, + 375, + 447 + ], + "spans": [ + { + "bbox": [ + 107, + 434, + 191, + 447 + ], + "score": 0.93, + "content": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 433, + 365, + 447 + ], + "score": 1.0, + "content": ". We have the orthogonal decomposition of", + "type": "text" + }, + { + "bbox": [ + 365, + 435, + 375, + 446 + ], + "score": 0.83, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 452, + 349, + 466 + ], + "lines": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "spans": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "score": 0.93, + "content": "\\phi ^ { \\prime } ( x ) = b _ { 0 } + \\phi _ { \\perp } ^ { \\prime } ( x ) ,", + "type": "interline_equation", + "image_path": "b7dfa8c1de8224d762d70789c4e42b6b546f00eb42b72c16707b34236bcb4e19.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 472, + 470, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 471, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 486 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 472, + 193, + 485 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\phi _ { \\perp } ^ { \\prime } ( G ) ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 471, + 471, + 486 + ], + "score": 1.0, + "content": ". We develop the following lemmas to approximate the kernel matrix.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 489, + 504, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 235, + 504 + ], + "score": 1.0, + "content": "Lemma 13 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 235, + 488, + 267, + 503 + ], + "score": 0.91, + "content": "( \\hat { K } _ { X } ) _ { i j } .", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 488, + 366, + 504 + ], + "score": 1.0, + "content": "). There exist constants", + "type": "text" + }, + { + "bbox": [ + 366, + 490, + 402, + 501 + ], + "score": 0.91, + "content": "c , c ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 488, + 459, + 504 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + }, + { + "bbox": [ + 459, + 490, + 483, + 502 + ], + "score": 0.92, + "content": "i \\neq j", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 488, + 505, + 504 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 501, + 237, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 153, + 518 + ], + "score": 1.0, + "content": "probability", + "type": "text" + }, + { + "bbox": [ + 153, + 502, + 198, + 514 + ], + "score": 0.89, + "content": "1 - e ^ { - c n \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 501, + 237, + 518 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 521, + 404, + 555 + ], + "lines": [ + { + "bbox": [ + 205, + 521, + 404, + 555 + ], + "spans": [ + { + "bbox": [ + 205, + 521, + 404, + 555 + ], + "score": 0.92, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 } ,", + "type": "interline_equation", + "image_path": "ba124e0959c6d29fde9defcbbb7b95836cd045b48b500e30cc88c9bed64fca92.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 205, + 521, + 404, + 538.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 205, + 538.0, + 404, + 555.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 563, + 241, + 575 + ], + "lines": [ + { + "bbox": [ + 104, + 561, + 242, + 577 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 242, + 577 + ], + "score": 1.0, + "content": "and with probability 1 − e−c0nε2 ,", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 581, + 401, + 615 + ], + "lines": [ + { + "bbox": [ + 209, + 581, + 401, + 615 + ], + "spans": [ + { + "bbox": [ + 209, + 581, + 401, + 615 + ], + "score": 0.93, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } - ( b _ { 0 } ^ { 2 } + b _ { 1 } ^ { 2 } ) \\right| < \\varepsilon .", + "type": "interline_equation", + "image_path": "341f89c3c6df5266c8ac725f841033b9c780ac530e50e7a42fb4ba7de26d0e59.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 209, + 581, + 401, + 598.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 209, + 598.0, + 401, + 615.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 626, + 311, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 311, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 164, + 640 + ], + "score": 1.0, + "content": "Proof. When", + "type": "text" + }, + { + "bbox": [ + 164, + 627, + 187, + 639 + ], + "score": 0.9, + "content": "i \\neq j", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 626, + 311, + 640 + ], + "score": 1.0, + "content": "(i.e. Equation (132)), we have", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 644, + 453, + 734 + ], + "lines": [ + { + "bbox": [ + 158, + 644, + 453, + 734 + ], + "spans": [ + { + "bbox": [ + 158, + 644, + 453, + 734 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad \\displaystyle \\frac 1 d [ \\hat { K } _ { X } ] _ { i j } = \\frac 1 d \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } } \\\\ & { = \\displaystyle \\frac 1 d { \\sum _ { k = 1 } ^ { h } } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } = \\frac 1 { d h } { \\displaystyle \\sum _ { k = 1 } ^ { h } } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { j } ) } \\\\ & { \\to \\displaystyle \\frac 1 d { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\mathbb { E } _ { \\pmb w } \\Big [ \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { j } ) \\Big ] = \\frac 1 d H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "3bace22a3d6842b9608a2d205d9d0976acd8db36707e2b16efbf5549b176bf6c.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 158, + 644, + 453, + 674.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 158, + 674.0, + 453, + 704.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 158, + 704.0, + 453, + 734.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "page_idx": 31, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 81, + 506, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 133, + 95 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 82, + 199, + 94 + ], + "score": 0.91, + "content": "\\partial f ( X ; \\omega _ { 0 } ) / \\partial \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 82, + 216, + 95 + ], + "score": 1.0, + "content": "is a", + "type": "text" + }, + { + "bbox": [ + 216, + 83, + 246, + 93 + ], + "score": 0.9, + "content": "d h \\times n", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 82, + 330, + 95 + ], + "score": 1.0, + "content": "matrix with column", + "type": "text" + }, + { + "bbox": [ + 330, + 82, + 395, + 95 + ], + "score": 0.92, + "content": "\\partial f ( \\pmb { x } _ { i } ; \\pmb { \\omega } _ { 0 } ) / \\partial \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 82, + 454, + 95 + ], + "score": 1.0, + "content": ". Note that for", + "type": "text" + }, + { + "bbox": [ + 454, + 84, + 487, + 93 + ], + "score": 0.86, + "content": "n \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 92, + 487, + 108 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 171, + 106 + ], + "score": 0.88, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 92, + 174, + 108 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 95, + 206, + 104 + ], + "score": 0.82, + "content": "d h > n", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 92, + 487, + 108 + ], + "score": 1.0, + "content": "trivially holds, and by Corollary 15 we know that solution is given by", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 108 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 111, + 448, + 147 + ], + "lines": [ + { + "bbox": [ + 163, + 111, + 448, + 147 + ], + "spans": [ + { + "bbox": [ + 163, + 111, + 448, + 147 + ], + "score": 0.93, + "content": "\\omega _ { 1 } = \\omega _ { 0 } + \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\left( \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } ^ { \\top } \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\right) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) .", + "type": "interline_equation", + "image_path": "a593f61a44d8cf9d15adefc40e6776cd65d466e4235ec038ef7974f64a9b4196.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 163, + 111, + 448, + 123.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 163, + 123.0, + 448, + 135.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 163, + 135.0, + 448, + 147.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 152, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 506, + 165 + ], + "score": 1.0, + "content": "And the population risk can be written as (note that there is a factor of 2 due to the \"doubling trick\" at", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 161, + 252, + 177 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 198, + 177 + ], + "score": 1.0, + "content": "initialization to ensure", + "type": "text" + }, + { + "bbox": [ + 199, + 163, + 244, + 176 + ], + "score": 0.93, + "content": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0 .", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 161, + 252, + 177 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 152, + 506, + 177 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 104, + 181, + 500, + 332 + ], + "lines": [ + { + "bbox": [ + 104, + 181, + 500, + 332 + ], + "spans": [ + { + "bbox": [ + 104, + 181, + 500, + 332 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { 2 R = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - f ( x ; \\omega _ { 1 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } ( \\omega _ { 1 } - \\omega _ { 0 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial f ( X ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha } [ ( x ^ { \\top } \\beta - \\frac { \\partial \\tilde { f } ( \\kappa ^ { - 1 } X ^ { \\top } \\beta ) } { \\partial \\beta } \\frac { \\partial \\tilde { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial \\tilde { f } ( \\tilde { X } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } ) } { 2 V } \\frac { \\partial ^ { 2 } } { \\partial \\omega } , \\qquad ( 1 2 9 ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "72e1cf33d940d4afe182e3be2eaed1daa330fd38df051ecb9aedf2f1ef1f7d95.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 104, + 181, + 500, + 231.33333333333334 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 104, + 231.33333333333334, + 500, + 281.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 104, + 281.6666666666667, + 500, + 332.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 335, + 426, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 426, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 426, + 350 + ], + "score": 1.0, + "content": "where a bias-variance decomposition is made here, and for simplicity we define", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 334, + 426, + 350 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 353, + 441, + 382 + ], + "lines": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "spans": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "score": 0.94, + "content": "\\hat { \\pmb { u } } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( { \\pmb { x } } ; \\omega _ { 0 } ) } { \\partial \\omega } , \\quad \\hat { K } _ { X } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } .", + "type": "interline_equation", + "image_path": "c1ca5ef4f293f868c59af874b6e441350f64df81c280c8ef9f848f2fb3ee7da6.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 169, + 353, + 441, + 382 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 392, + 311, + 405 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 311, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 311, + 405 + ], + "score": 1.0, + "content": "C.9.3 APPROXIMATING THE KERNEL MATRIX", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 411, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 262, + 425 + ], + "score": 1.0, + "content": "In this section we drop the negligible", + "type": "text" + }, + { + "bbox": [ + 263, + 415, + 269, + 422 + ], + "score": 0.75, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 411, + 506, + 425 + ], + "score": 1.0, + "content": "in the initialization. Following Cheng and Singer (2013)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 296, + 435 + ], + "score": 1.0, + "content": "we utilize the orthonormal decomposition of", + "type": "text" + }, + { + "bbox": [ + 296, + 423, + 319, + 435 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 422, + 333, + 435 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 333, + 423, + 378, + 435 + ], + "score": 0.92, + "content": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 422, + 419, + 435 + ], + "score": 1.0, + "content": ". Denote", + "type": "text" + }, + { + "bbox": [ + 419, + 423, + 482, + 435 + ], + "score": 0.93, + "content": "b _ { 0 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 422, + 505, + 435 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 433, + 375, + 447 + ], + "spans": [ + { + "bbox": [ + 107, + 434, + 191, + 447 + ], + "score": 0.93, + "content": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 433, + 365, + 447 + ], + "score": 1.0, + "content": ". We have the orthogonal decomposition of", + "type": "text" + }, + { + "bbox": [ + 365, + 435, + 375, + 446 + ], + "score": 0.83, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 411, + 506, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 452, + 349, + 466 + ], + "lines": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "spans": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "score": 0.93, + "content": "\\phi ^ { \\prime } ( x ) = b _ { 0 } + \\phi _ { \\perp } ^ { \\prime } ( x ) ,", + "type": "interline_equation", + "image_path": "b7dfa8c1de8224d762d70789c4e42b6b546f00eb42b72c16707b34236bcb4e19.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 262, + 452, + 349, + 466 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 472, + 470, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 471, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 486 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 472, + 193, + 485 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\phi _ { \\perp } ^ { \\prime } ( G ) ] = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 471, + 471, + 486 + ], + "score": 1.0, + "content": ". We develop the following lemmas to approximate the kernel matrix.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 471, + 471, + 486 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 489, + 504, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 235, + 504 + ], + "score": 1.0, + "content": "Lemma 13 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 235, + 488, + 267, + 503 + ], + "score": 0.91, + "content": "( \\hat { K } _ { X } ) _ { i j } .", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 488, + 366, + 504 + ], + "score": 1.0, + "content": "). There exist constants", + "type": "text" + }, + { + "bbox": [ + 366, + 490, + 402, + 501 + ], + "score": 0.91, + "content": "c , c ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 488, + 459, + 504 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + }, + { + "bbox": [ + 459, + 490, + 483, + 502 + ], + "score": 0.92, + "content": "i \\neq j", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 488, + 505, + 504 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 501, + 237, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 153, + 518 + ], + "score": 1.0, + "content": "probability", + "type": "text" + }, + { + "bbox": [ + 153, + 502, + 198, + 514 + ], + "score": 0.89, + "content": "1 - e ^ { - c n \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 501, + 237, + 518 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 104, + 488, + 505, + 518 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 521, + 404, + 555 + ], + "lines": [ + { + "bbox": [ + 205, + 521, + 404, + 555 + ], + "spans": [ + { + "bbox": [ + 205, + 521, + 404, + 555 + ], + "score": 0.92, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 } ,", + "type": "interline_equation", + "image_path": "ba124e0959c6d29fde9defcbbb7b95836cd045b48b500e30cc88c9bed64fca92.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 205, + 521, + 404, + 538.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 205, + 538.0, + 404, + 555.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 563, + 241, + 575 + ], + "lines": [ + { + "bbox": [ + 104, + 561, + 242, + 577 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 242, + 577 + ], + "score": 1.0, + "content": "and with probability 1 − e−c0nε2 ,", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 104, + 561, + 242, + 577 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 581, + 401, + 615 + ], + "lines": [ + { + "bbox": [ + 209, + 581, + 401, + 615 + ], + "spans": [ + { + "bbox": [ + 209, + 581, + 401, + 615 + ], + "score": 0.93, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } - ( b _ { 0 } ^ { 2 } + b _ { 1 } ^ { 2 } ) \\right| < \\varepsilon .", + "type": "interline_equation", + "image_path": "341f89c3c6df5266c8ac725f841033b9c780ac530e50e7a42fb4ba7de26d0e59.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 209, + 581, + 401, + 598.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 209, + 598.0, + 401, + 615.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 626, + 311, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 311, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 164, + 640 + ], + "score": 1.0, + "content": "Proof. When", + "type": "text" + }, + { + "bbox": [ + 164, + 627, + 187, + 639 + ], + "score": 0.9, + "content": "i \\neq j", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 626, + 311, + 640 + ], + "score": 1.0, + "content": "(i.e. Equation (132)), we have", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 106, + 626, + 311, + 640 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 644, + 453, + 734 + ], + "lines": [ + { + "bbox": [ + 158, + 644, + 453, + 734 + ], + "spans": [ + { + "bbox": [ + 158, + 644, + 453, + 734 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad \\displaystyle \\frac 1 d [ \\hat { K } _ { X } ] _ { i j } = \\frac 1 d \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } } \\\\ & { = \\displaystyle \\frac 1 d { \\sum _ { k = 1 } ^ { h } } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } = \\frac 1 { d h } { \\displaystyle \\sum _ { k = 1 } ^ { h } } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { j } ) } \\\\ & { \\to \\displaystyle \\frac 1 d { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\mathbb { E } _ { \\pmb w } \\Big [ \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { j } ) \\Big ] = \\frac 1 d H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "3bace22a3d6842b9608a2d205d9d0976acd8db36707e2b16efbf5549b176bf6c.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 158, + 644, + 453, + 674.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 158, + 674.0, + 453, + 704.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 158, + 704.0, + 453, + 734.0 + ], + "spans": [], + "index": 28 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 121 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 151, + 97 + ], + "score": 1.0, + "content": "The matrix", + "type": "text" + }, + { + "bbox": [ + 151, + 80, + 339, + 100 + ], + "score": 0.93, + "content": "H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } \\mathbb { E } _ { \\pmb { w } } \\Big [ \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { j } ) \\Big ]", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 81, + 506, + 97 + ], + "score": 1.0, + "content": "can be seen as the expected tangent kernel", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 96, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 96, + 505, + 111 + ], + "score": 1.0, + "content": "of nonlinear activation function studied in Du et al. (2018); Arora et al. (2019b). Moreover, due to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 108, + 411, + 122 + ], + "spans": [ + { + "bbox": [ + 105, + 108, + 223, + 122 + ], + "score": 1.0, + "content": "the assumed boundedness of", + "type": "text" + }, + { + "bbox": [ + 223, + 109, + 246, + 121 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 108, + 411, + 122 + ], + "score": 1.0, + "content": "(A3), by Hoeffding’s inequality we have", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 126, + 460, + 159 + ], + "lines": [ + { + "bbox": [ + 151, + 126, + 460, + 159 + ], + "spans": [ + { + "bbox": [ + 151, + 126, + 460, + 159 + ], + "score": 0.93, + "content": "\\operatorname* { P r } \\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) \\right| < \\frac { 1 } { d } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\varepsilon > 1 - e ^ { - c _ { 1 } h \\varepsilon ^ { 2 } } .", + "type": "interline_equation", + "image_path": "a6a6772afb39175d22e5112efd10259f74ce8b844391960549bb99377de3b746.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 151, + 126, + 460, + 137.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 151, + 137.0, + 460, + 148.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 151, + 148.0, + 460, + 159.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 104, + 163, + 507, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 163, + 258, + 179 + ], + "score": 1.0, + "content": "In addition, by the concentration of", + "type": "text" + }, + { + "bbox": [ + 259, + 165, + 284, + 179 + ], + "score": 0.92, + "content": "\\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 163, + 304, + 179 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 305, + 165, + 330, + 178 + ], + "score": 0.91, + "content": "\\| \\pmb { x } _ { i } \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 163, + 353, + 179 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 353, + 164, + 485, + 179 + ], + "score": 0.92, + "content": "\\mathrm { P r } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d > \\varepsilon < 1 - e ^ { - c _ { 2 } d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 163, + 507, + 179 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 175, + 507, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 255, + 192 + ], + "score": 0.91, + "content": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { i } / d - 1 | < \\varepsilon > 1 - e ^ { - c _ { 3 } d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 175, + 390, + 194 + ], + "score": 1.0, + "content": ", the orthonormal decomposition", + "type": "text" + }, + { + "bbox": [ + 390, + 180, + 480, + 192 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x ) = b _ { 0 } x + \\phi _ { \\perp } ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 175, + 507, + 194 + ], + "score": 1.0, + "content": "leads", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 322, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 312, + 203 + ], + "score": 1.0, + "content": "to the following linear approximation of the matrix", + "type": "text" + }, + { + "bbox": [ + 312, + 191, + 322, + 201 + ], + "score": 0.81, + "content": "H", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 206, + 397, + 231 + ], + "lines": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "spans": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "score": 0.94, + "content": "\\frac { 1 } { d } H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = b _ { 0 } ^ { 2 } \\frac { 1 } { d } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } + O \\big ( ( \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d ) ^ { 2 } \\big ) .", + "type": "interline_equation", + "image_path": "cb26e808d6df3fe8d39c20cd87a656067ebe146a2121b73064dd7d03c832ee6e.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 374, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 374, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 163, + 249 + ], + "score": 1.0, + "content": "and by taking", + "type": "text" + }, + { + "bbox": [ + 163, + 235, + 217, + 249 + ], + "score": 0.93, + "content": "\\varepsilon = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 234, + 374, + 249 + ], + "score": 1.0, + "content": "under the joint event we can show that", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 253, + 403, + 287 + ], + "lines": [ + { + "bbox": [ + 206, + 253, + 403, + 287 + ], + "spans": [ + { + "bbox": [ + 206, + 253, + 403, + 287 + ], + "score": 0.93, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 }", + "type": "interline_equation", + "image_path": "a282ba77a940f70a732d35630effa395391b9966ceb1ea7657b0617c6bb673c5.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 206, + 253, + 403, + 270.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 206, + 270.0, + 403, + 287.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 436, + 306 + ], + "lines": [ + { + "bbox": [ + 103, + 288, + 433, + 309 + ], + "spans": [ + { + "bbox": [ + 103, + 288, + 172, + 309 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 173, + 291, + 217, + 304 + ], + "score": 0.92, + "content": "1 - e ^ { - c d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 288, + 410, + 309 + ], + "score": 1.0, + "content": ". The same argument follows for the case where", + "type": "text" + }, + { + "bbox": [ + 410, + 294, + 433, + 305 + ], + "score": 0.91, + "content": "i = j", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 481, + 327 + ], + "lines": [ + { + "bbox": [ + 104, + 312, + 480, + 328 + ], + "spans": [ + { + "bbox": [ + 104, + 313, + 240, + 328 + ], + "score": 1.0, + "content": "Corollary 14 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 240, + 316, + 248, + 325 + ], + "score": 0.78, + "content": "\\hat { \\textbf { \\textit { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 313, + 329, + 328 + ], + "score": 1.0, + "content": "). For large enough", + "type": "text" + }, + { + "bbox": [ + 329, + 316, + 352, + 326 + ], + "score": 0.88, + "content": "l > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 313, + 421, + 328 + ], + "score": 1.0, + "content": ", with probability", + "type": "text" + }, + { + "bbox": [ + 421, + 312, + 480, + 326 + ], + "score": 0.89, + "content": "1 - d e ^ { - c \\log ^ { l } d }", + "type": "inline_equation" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 330, + 433, + 365 + ], + "lines": [ + { + "bbox": [ + 178, + 330, + 433, + 365 + ], + "spans": [ + { + "bbox": [ + 178, + 330, + 433, + 365 + ], + "score": 0.94, + "content": "\\frac { 1 } { d } \\left\\| \\hat { \\pmb { u } } - \\tilde { \\pmb { u } } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( \\pmb { x } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { \\pmb { u } } \\right\\| _ { 2 } < \\frac { \\log ^ { l } d } { d } ,", + "type": "interline_equation", + "image_path": "4baf9635a33c5c755f86d270576005fb42499035d328fb1c3388a5ac9b07a62e.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 178, + 330, + 433, + 341.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 178, + 341.6666666666667, + 433, + 353.33333333333337 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 178, + 353.33333333333337, + 433, + 365.00000000000006 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 370, + 188, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 189, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 133, + 384 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 370, + 185, + 383 + ], + "score": 0.93, + "content": "\\tilde { \\pmb { u } } = b _ { 0 } ^ { 2 } \\pmb { x } ^ { \\top } \\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 369, + 189, + 384 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 420, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 421, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 168, + 406 + ], + "score": 1.0, + "content": "Proof. Taking", + "type": "text" + }, + { + "bbox": [ + 168, + 391, + 220, + 405 + ], + "score": 0.93, + "content": "\\varepsilon = \\log ^ { l } d / d", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 392, + 421, + 406 + ], + "score": 1.0, + "content": "together with Lemma 13 yields the desired result.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 108, + 412, + 394, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 393, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 240, + 428 + ], + "score": 1.0, + "content": "Corollary 15 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 240, + 413, + 258, + 426 + ], + "score": 0.89, + "content": "{ \\hat { K } } _ { X } { \\mathrm { . } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 412, + 334, + 428 + ], + "score": 1.0, + "content": "). With probability", + "type": "text" + }, + { + "bbox": [ + 334, + 412, + 393, + 426 + ], + "score": 0.82, + "content": "1 - d e ^ { - c \\log ^ { l } d }", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 160, + 430, + 450, + 465 + ], + "lines": [ + { + "bbox": [ + 160, + 430, + 450, + 465 + ], + "spans": [ + { + "bbox": [ + 160, + 430, + 450, + 465 + ], + "score": 0.93, + "content": "\\frac { 1 } { d } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { K } _ { X } \\right\\| _ { F } < \\log ^ { l } d ,", + "type": "interline_equation", + "image_path": "20ccb876ecc17f1c942ed29b2ed64f828b8f11b2e3a94731533694df96a89c6a.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 160, + 430, + 450, + 441.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 160, + 441.6666666666667, + 450, + 453.33333333333337 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 160, + 453.33333333333337, + 450, + 465.00000000000006 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 231, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 232, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 133, + 485 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 469, + 228, + 484 + ], + "score": 0.9, + "content": "\\tilde { K } _ { X } = b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 469, + 232, + 485 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 289, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 289, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 289, + 504 + ], + "score": 1.0, + "content": "Proof. Also by directly applying Lemma 13.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 509, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 104, + 509, + 283, + 525 + ], + "score": 1.0, + "content": "Remark. For initialization larger or equal to", + "type": "text" + }, + { + "bbox": [ + 283, + 511, + 367, + 523 + ], + "score": 0.94, + "content": "{ \\pmb w } _ { i } ( 0 ) \\sim N ( 0 , I _ { d } / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 509, + 506, + 525 + ], + "score": 1.0, + "content": ", the above approximation does not", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 523, + 253, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 253, + 534 + ], + "score": 1.0, + "content": "depend on the scale of initialization.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 506, + 580 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 507, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 164, + 548 + ], + "score": 1.0, + "content": "Remark. For", + "type": "text" + }, + { + "bbox": [ + 164, + 534, + 251, + 547 + ], + "score": 0.86, + "content": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 533, + 254, + 548 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 255, + 534, + 295, + 547 + ], + "score": 0.77, + "content": "b _ { 0 } ^ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 533, + 300, + 548 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 301, + 534, + 363, + 547 + ], + "score": 0.76, + "content": "b _ { 1 } ^ { 2 } = 0 . 0 4 3 3 7 9", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 533, + 387, + 548 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 388, + 535, + 507, + 547 + ], + "score": 0.91, + "content": "\\phi ( x ) = \\mathrm { s i g m o i d } ( x ) = ( 1 +", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 544, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 138, + 558 + ], + "score": 0.89, + "content": "e ^ { - x } ) ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 544, + 142, + 559 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 143, + 546, + 206, + 558 + ], + "score": 0.61, + "content": "b _ { 0 } ^ { 2 } = 0 . 0 4 2 6 9 2", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 544, + 210, + 559 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 210, + 546, + 273, + 558 + ], + "score": 0.82, + "content": "b _ { 1 } ^ { 2 } = 0 . 0 0 2 1 4 4", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 544, + 320, + 559 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 321, + 547, + 351, + 557 + ], + "score": 0.89, + "content": "b _ { 1 } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 544, + 460, + 559 + ], + "score": 1.0, + "content": "for all smooth activations", + "type": "text" + }, + { + "bbox": [ + 461, + 547, + 468, + 558 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 544, + 506, + 559 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 557, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 187, + 570 + ], + "score": 1.0, + "content": "equality holds (i.e.", + "type": "text" + }, + { + "bbox": [ + 188, + 558, + 218, + 568 + ], + "score": 0.89, + "content": "b _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 557, + 280, + 570 + ], + "score": 1.0, + "content": ") if and only if", + "type": "text" + }, + { + "bbox": [ + 281, + 558, + 288, + 568 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 557, + 438, + 570 + ], + "score": 1.0, + "content": "is linear. We comment that smaller", + "type": "text" + }, + { + "bbox": [ + 438, + 558, + 448, + 568 + ], + "score": 0.88, + "content": "b _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 557, + 506, + 570 + ], + "score": 1.0, + "content": "entails larger", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 568, + 387, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 153, + 580 + ], + "score": 1.0, + "content": "variance as", + "type": "text" + }, + { + "bbox": [ + 153, + 569, + 184, + 580 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 568, + 387, + 580 + ], + "score": 1.0, + "content": ", and vice versa, as shown in the following section.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 213, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 214, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 214, + 604 + ], + "score": 1.0, + "content": "C.9.4 THE BIAS TERM", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 366, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 366, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 366, + 624 + ], + "score": 1.0, + "content": "With these approximation above we proceed to calculating (129)", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 627, + 387, + 654 + ], + "lines": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "spans": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "score": 0.94, + "content": "2 B = \\mathbb { E } _ { \\pmb { x } } \\left[ \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - \\hat { \\pmb { u } } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\pmb { \\beta } \\right) ^ { 2 } \\right] .", + "type": "interline_equation", + "image_path": "644d9f31c0a7b3276946727dcfd2dbe64d9e76f80c7ec098763aeb36aea46c38.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 303, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 304, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 263, + 672 + ], + "score": 1.0, + "content": "We first bound the error in substituting", + "type": "text" + }, + { + "bbox": [ + 263, + 659, + 271, + 669 + ], + "score": 0.85, + "content": "\\hat { \\textbf { \\textit { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 657, + 292, + 672 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 293, + 659, + 300, + 668 + ], + "score": 0.83, + "content": "\\tilde { \\mathbf { \\pmb { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 657, + 304, + 672 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 674, + 457, + 731 + ], + "lines": [ + { + "bbox": [ + 131, + 674, + 457, + 731 + ], + "spans": [ + { + "bbox": [ + 131, + 674, + 457, + 731 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\left\\| \\hat { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta - \\tilde { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta \\right\\| _ { 2 } \\leq \\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\boldsymbol X \\right\\| _ { 2 } \\left\\| \\boldsymbol \\beta \\right\\| _ { 2 } } \\\\ & { \\quad \\quad \\quad \\quad \\stackrel { ( i ) } { = } O \\left( \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot \\sqrt { d } \\cdot 1 \\right) = O \\left( \\frac { \\log ^ { l } d } { \\sqrt { d } } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "8827867e5a0b0f22c443292f9987b46fb2daebe9ceaebb0473f431c87e781245.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 131, + 674, + 457, + 693.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 131, + 693.0, + 457, + 712.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 131, + 712.0, + 457, + 731.0 + ], + "spans": [], + "index": 38 + } + ] + } + ], + "page_idx": 32, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "33", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 393, + 505, + 404 + ], + "lines": [ + { + "bbox": [ + 496, + 395, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 496, + 395, + 505, + 404 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 492, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 496, + 494, + 504, + 502 + ], + "spans": [ + { + "bbox": [ + 496, + 494, + 504, + 502 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 294, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 496, + 295, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 496, + 295, + 505, + 304 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 121 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 151, + 97 + ], + "score": 1.0, + "content": "The matrix", + "type": "text" + }, + { + "bbox": [ + 151, + 80, + 339, + 100 + ], + "score": 0.93, + "content": "H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } \\mathbb { E } _ { \\pmb { w } } \\Big [ \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { j } ) \\Big ]", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 81, + 506, + 97 + ], + "score": 1.0, + "content": "can be seen as the expected tangent kernel", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 96, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 96, + 505, + 111 + ], + "score": 1.0, + "content": "of nonlinear activation function studied in Du et al. (2018); Arora et al. (2019b). Moreover, due to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 108, + 411, + 122 + ], + "spans": [ + { + "bbox": [ + 105, + 108, + 223, + 122 + ], + "score": 1.0, + "content": "the assumed boundedness of", + "type": "text" + }, + { + "bbox": [ + 223, + 109, + 246, + 121 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 108, + 411, + 122 + ], + "score": 1.0, + "content": "(A3), by Hoeffding’s inequality we have", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 80, + 506, + 122 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 126, + 460, + 159 + ], + "lines": [ + { + "bbox": [ + 151, + 126, + 460, + 159 + ], + "spans": [ + { + "bbox": [ + 151, + 126, + 460, + 159 + ], + "score": 0.93, + "content": "\\operatorname* { P r } \\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) \\right| < \\frac { 1 } { d } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\varepsilon > 1 - e ^ { - c _ { 1 } h \\varepsilon ^ { 2 } } .", + "type": "interline_equation", + "image_path": "a6a6772afb39175d22e5112efd10259f74ce8b844391960549bb99377de3b746.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 151, + 126, + 460, + 137.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 151, + 137.0, + 460, + 148.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 151, + 148.0, + 460, + 159.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 104, + 163, + 507, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 163, + 258, + 179 + ], + "score": 1.0, + "content": "In addition, by the concentration of", + "type": "text" + }, + { + "bbox": [ + 259, + 165, + 284, + 179 + ], + "score": 0.92, + "content": "\\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 163, + 304, + 179 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 305, + 165, + 330, + 178 + ], + "score": 0.91, + "content": "\\| \\pmb { x } _ { i } \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 163, + 353, + 179 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 353, + 164, + 485, + 179 + ], + "score": 0.92, + "content": "\\mathrm { P r } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d > \\varepsilon < 1 - e ^ { - c _ { 2 } d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 163, + 507, + 179 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 175, + 507, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 255, + 192 + ], + "score": 0.91, + "content": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { i } / d - 1 | < \\varepsilon > 1 - e ^ { - c _ { 3 } d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 175, + 390, + 194 + ], + "score": 1.0, + "content": ", the orthonormal decomposition", + "type": "text" + }, + { + "bbox": [ + 390, + 180, + 480, + 192 + ], + "score": 0.92, + "content": "\\phi ^ { \\prime } ( x ) = b _ { 0 } x + \\phi _ { \\perp } ^ { \\prime } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 175, + 507, + 194 + ], + "score": 1.0, + "content": "leads", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 322, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 312, + 203 + ], + "score": 1.0, + "content": "to the following linear approximation of the matrix", + "type": "text" + }, + { + "bbox": [ + 312, + 191, + 322, + 201 + ], + "score": 0.81, + "content": "H", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 104, + 163, + 507, + 203 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 206, + 397, + 231 + ], + "lines": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "spans": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "score": 0.94, + "content": "\\frac { 1 } { d } H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = b _ { 0 } ^ { 2 } \\frac { 1 } { d } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } + O \\big ( ( \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d ) ^ { 2 } \\big ) .", + "type": "interline_equation", + "image_path": "cb26e808d6df3fe8d39c20cd87a656067ebe146a2121b73064dd7d03c832ee6e.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 213, + 206, + 397, + 231 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 374, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 374, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 163, + 249 + ], + "score": 1.0, + "content": "and by taking", + "type": "text" + }, + { + "bbox": [ + 163, + 235, + 217, + 249 + ], + "score": 0.93, + "content": "\\varepsilon = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 234, + 374, + 249 + ], + "score": 1.0, + "content": "under the joint event we can show that", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 234, + 374, + 249 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 253, + 403, + 287 + ], + "lines": [ + { + "bbox": [ + 206, + 253, + 403, + 287 + ], + "spans": [ + { + "bbox": [ + 206, + 253, + 403, + 287 + ], + "score": 0.93, + "content": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 }", + "type": "interline_equation", + "image_path": "a282ba77a940f70a732d35630effa395391b9966ceb1ea7657b0617c6bb673c5.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 206, + 253, + 403, + 270.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 206, + 270.0, + 403, + 287.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 436, + 306 + ], + "lines": [ + { + "bbox": [ + 103, + 288, + 433, + 309 + ], + "spans": [ + { + "bbox": [ + 103, + 288, + 172, + 309 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 173, + 291, + 217, + 304 + ], + "score": 0.92, + "content": "1 - e ^ { - c d \\varepsilon ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 288, + 410, + 309 + ], + "score": 1.0, + "content": ". The same argument follows for the case where", + "type": "text" + }, + { + "bbox": [ + 410, + 294, + 433, + 305 + ], + "score": 0.91, + "content": "i = j", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 103, + 288, + 433, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 481, + 327 + ], + "lines": [ + { + "bbox": [ + 104, + 312, + 480, + 328 + ], + "spans": [ + { + "bbox": [ + 104, + 313, + 240, + 328 + ], + "score": 1.0, + "content": "Corollary 14 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 240, + 316, + 248, + 325 + ], + "score": 0.78, + "content": "\\hat { \\textbf { \\textit { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 313, + 329, + 328 + ], + "score": 1.0, + "content": "). For large enough", + "type": "text" + }, + { + "bbox": [ + 329, + 316, + 352, + 326 + ], + "score": 0.88, + "content": "l > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 313, + 421, + 328 + ], + "score": 1.0, + "content": ", with probability", + "type": "text" + }, + { + "bbox": [ + 421, + 312, + 480, + 326 + ], + "score": 0.89, + "content": "1 - d e ^ { - c \\log ^ { l } d }", + "type": "inline_equation" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 312, + 480, + 328 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 330, + 433, + 365 + ], + "lines": [ + { + "bbox": [ + 178, + 330, + 433, + 365 + ], + "spans": [ + { + "bbox": [ + 178, + 330, + 433, + 365 + ], + "score": 0.94, + "content": "\\frac { 1 } { d } \\left\\| \\hat { \\pmb { u } } - \\tilde { \\pmb { u } } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( \\pmb { x } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { \\pmb { u } } \\right\\| _ { 2 } < \\frac { \\log ^ { l } d } { d } ,", + "type": "interline_equation", + "image_path": "4baf9635a33c5c755f86d270576005fb42499035d328fb1c3388a5ac9b07a62e.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 178, + 330, + 433, + 341.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 178, + 341.6666666666667, + 433, + 353.33333333333337 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 178, + 353.33333333333337, + 433, + 365.00000000000006 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 370, + 188, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 189, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 133, + 384 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 370, + 185, + 383 + ], + "score": 0.93, + "content": "\\tilde { \\pmb { u } } = b _ { 0 } ^ { 2 } \\pmb { x } ^ { \\top } \\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 369, + 189, + 384 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 369, + 189, + 384 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 420, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 421, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 168, + 406 + ], + "score": 1.0, + "content": "Proof. Taking", + "type": "text" + }, + { + "bbox": [ + 168, + 391, + 220, + 405 + ], + "score": 0.93, + "content": "\\varepsilon = \\log ^ { l } d / d", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 392, + 421, + 406 + ], + "score": 1.0, + "content": "together with Lemma 13 yields the desired result.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 391, + 421, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 412, + 394, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 393, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 240, + 428 + ], + "score": 1.0, + "content": "Corollary 15 (Approximation of", + "type": "text" + }, + { + "bbox": [ + 240, + 413, + 258, + 426 + ], + "score": 0.89, + "content": "{ \\hat { K } } _ { X } { \\mathrm { . } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 412, + 334, + 428 + ], + "score": 1.0, + "content": "). With probability", + "type": "text" + }, + { + "bbox": [ + 334, + 412, + 393, + 426 + ], + "score": 0.82, + "content": "1 - d e ^ { - c \\log ^ { l } d }", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 412, + 393, + 428 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 160, + 430, + 450, + 465 + ], + "lines": [ + { + "bbox": [ + 160, + 430, + 450, + 465 + ], + "spans": [ + { + "bbox": [ + 160, + 430, + 450, + 465 + ], + "score": 0.93, + "content": "\\frac { 1 } { d } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { K } _ { X } \\right\\| _ { F } < \\log ^ { l } d ,", + "type": "interline_equation", + "image_path": "20ccb876ecc17f1c942ed29b2ed64f828b8f11b2e3a94731533694df96a89c6a.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 160, + 430, + 450, + 441.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 160, + 441.6666666666667, + 450, + 453.33333333333337 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 160, + 453.33333333333337, + 450, + 465.00000000000006 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 231, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 232, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 133, + 485 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 469, + 228, + 484 + ], + "score": 0.9, + "content": "\\tilde { K } _ { X } = b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 469, + 232, + 485 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 469, + 232, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 289, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 289, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 289, + 504 + ], + "score": 1.0, + "content": "Proof. Also by directly applying Lemma 13.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 106, + 491, + 289, + 504 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 104, + 509, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 104, + 509, + 283, + 525 + ], + "score": 1.0, + "content": "Remark. For initialization larger or equal to", + "type": "text" + }, + { + "bbox": [ + 283, + 511, + 367, + 523 + ], + "score": 0.94, + "content": "{ \\pmb w } _ { i } ( 0 ) \\sim N ( 0 , I _ { d } / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 509, + 506, + 525 + ], + "score": 1.0, + "content": ", the above approximation does not", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 523, + 253, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 253, + 534 + ], + "score": 1.0, + "content": "depend on the scale of initialization.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 509, + 506, + 534 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 506, + 580 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 507, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 164, + 548 + ], + "score": 1.0, + "content": "Remark. For", + "type": "text" + }, + { + "bbox": [ + 164, + 534, + 251, + 547 + ], + "score": 0.86, + "content": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 533, + 254, + 548 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 255, + 534, + 295, + 547 + ], + "score": 0.77, + "content": "b _ { 0 } ^ { 2 } = 1 / 4", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 533, + 300, + 548 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 301, + 534, + 363, + 547 + ], + "score": 0.76, + "content": "b _ { 1 } ^ { 2 } = 0 . 0 4 3 3 7 9", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 533, + 387, + 548 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 388, + 535, + 507, + 547 + ], + "score": 0.91, + "content": "\\phi ( x ) = \\mathrm { s i g m o i d } ( x ) = ( 1 +", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 544, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 138, + 558 + ], + "score": 0.89, + "content": "e ^ { - x } ) ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 544, + 142, + 559 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 143, + 546, + 206, + 558 + ], + "score": 0.61, + "content": "b _ { 0 } ^ { 2 } = 0 . 0 4 2 6 9 2", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 544, + 210, + 559 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 210, + 546, + 273, + 558 + ], + "score": 0.82, + "content": "b _ { 1 } ^ { 2 } = 0 . 0 0 2 1 4 4", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 544, + 320, + 559 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 321, + 547, + 351, + 557 + ], + "score": 0.89, + "content": "b _ { 1 } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 544, + 460, + 559 + ], + "score": 1.0, + "content": "for all smooth activations", + "type": "text" + }, + { + "bbox": [ + 461, + 547, + 468, + 558 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 544, + 506, + 559 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 557, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 187, + 570 + ], + "score": 1.0, + "content": "equality holds (i.e.", + "type": "text" + }, + { + "bbox": [ + 188, + 558, + 218, + 568 + ], + "score": 0.89, + "content": "b _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 557, + 280, + 570 + ], + "score": 1.0, + "content": ") if and only if", + "type": "text" + }, + { + "bbox": [ + 281, + 558, + 288, + 568 + ], + "score": 0.85, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 557, + 438, + 570 + ], + "score": 1.0, + "content": "is linear. We comment that smaller", + "type": "text" + }, + { + "bbox": [ + 438, + 558, + 448, + 568 + ], + "score": 0.88, + "content": "b _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 557, + 506, + 570 + ], + "score": 1.0, + "content": "entails larger", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 568, + 387, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 153, + 580 + ], + "score": 1.0, + "content": "variance as", + "type": "text" + }, + { + "bbox": [ + 153, + 569, + 184, + 580 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 568, + 387, + 580 + ], + "score": 1.0, + "content": ", and vice versa, as shown in the following section.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 533, + 507, + 580 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 213, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 214, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 214, + 604 + ], + "score": 1.0, + "content": "C.9.4 THE BIAS TERM", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 366, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 366, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 366, + 624 + ], + "score": 1.0, + "content": "With these approximation above we proceed to calculating (129)", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 610, + 366, + 624 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 627, + 387, + 654 + ], + "lines": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "spans": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "score": 0.94, + "content": "2 B = \\mathbb { E } _ { \\pmb { x } } \\left[ \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - \\hat { \\pmb { u } } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\pmb { \\beta } \\right) ^ { 2 } \\right] .", + "type": "interline_equation", + "image_path": "644d9f31c0a7b3276946727dcfd2dbe64d9e76f80c7ec098763aeb36aea46c38.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 224, + 627, + 387, + 654 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 303, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 304, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 263, + 672 + ], + "score": 1.0, + "content": "We first bound the error in substituting", + "type": "text" + }, + { + "bbox": [ + 263, + 659, + 271, + 669 + ], + "score": 0.85, + "content": "\\hat { \\textbf { \\textit { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 657, + 292, + 672 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 293, + 659, + 300, + 668 + ], + "score": 0.83, + "content": "\\tilde { \\mathbf { \\pmb { u } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 657, + 304, + 672 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 657, + 304, + 672 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 674, + 457, + 731 + ], + "lines": [ + { + "bbox": [ + 131, + 674, + 457, + 731 + ], + "spans": [ + { + "bbox": [ + 131, + 674, + 457, + 731 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\left\\| \\hat { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta - \\tilde { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta \\right\\| _ { 2 } \\leq \\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\boldsymbol X \\right\\| _ { 2 } \\left\\| \\boldsymbol \\beta \\right\\| _ { 2 } } \\\\ & { \\quad \\quad \\quad \\quad \\stackrel { ( i ) } { = } O \\left( \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot \\sqrt { d } \\cdot 1 \\right) = O \\left( \\frac { \\log ^ { l } d } { \\sqrt { d } } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "8827867e5a0b0f22c443292f9987b46fb2daebe9ceaebb0473f431c87e781245.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 131, + 674, + 457, + 693.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 131, + 693.0, + 457, + 712.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 131, + 712.0, + 457, + 731.0 + ], + "spans": [], + "index": 38 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 79, + 505, + 111 + ], + "lines": [ + { + "bbox": [ + 102, + 78, + 509, + 101 + ], + "spans": [ + { + "bbox": [ + 102, + 78, + 234, + 101 + ], + "score": 1.0, + "content": "where (i) is due to the fact that", + "type": "text" + }, + { + "bbox": [ + 234, + 80, + 372, + 101 + ], + "score": 0.94, + "content": "\\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } = \\lambda _ { \\operatorname* { m i n } } ^ { - 1 } ( \\hat { K } _ { X } ) = O ( 1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 78, + 390, + 101 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 391, + 83, + 457, + 97 + ], + "score": 0.93, + "content": "\\| X \\| _ { 2 } = O ( { \\sqrt { d } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 78, + 509, + 101 + ], + "score": 1.0, + "content": ". Therefore", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 99, + 205, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 152, + 111 + ], + "score": 1.0, + "content": "we have as", + "type": "text" + }, + { + "bbox": [ + 152, + 99, + 205, + 111 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 114, + 456, + 171 + ], + "lines": [ + { + "bbox": [ + 132, + 114, + 456, + 171 + ], + "spans": [ + { + "bbox": [ + 132, + 114, + 456, + 171 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { 2 B = \\mathbb { E } _ { x } [ ( { \\pmb x } ^ { \\top } \\beta - \\hat { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - \\tilde { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\qquad = \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - b _ { 0 } ^ { 2 } { \\pmb x } ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "c46331efc24df13f917fdbc06e8e67c167e3d3a1c36ca7854f077f2c1962415c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 132, + 114, + 456, + 133.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 132, + 133.0, + 456, + 152.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 132, + 152.0, + 456, + 171.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 502, + 186 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 503, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 216, + 186 + ], + "score": 1.0, + "content": "By taking expectation over", + "type": "text" + }, + { + "bbox": [ + 216, + 176, + 224, + 183 + ], + "score": 0.79, + "content": "_ { \\textbf { \\em x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 172, + 503, + 186 + ], + "score": 1.0, + "content": "and the rotational invariance argument similar to Hastie et al. (2019),", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 189, + 494, + 245 + ], + "lines": [ + { + "bbox": [ + 117, + 189, + 494, + 245 + ], + "spans": [ + { + "bbox": [ + 117, + 189, + 494, + 245 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\mathbb { E } _ { x } \\left[ \\left( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta \\right) ^ { 2 } \\right] } & { = \\mathbb { E } _ { x } \\left[ \\beta ^ { \\top } \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) ^ { 2 } \\beta \\right] } \\\\ & { = \\frac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } \\left( \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "c7cd0607ab6d3cbace39e38a775f5b4d928f82e32ea776a044c5a9b27956d490.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 117, + 189, + 494, + 207.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 117, + 207.66666666666666, + 494, + 226.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 117, + 226.33333333333331, + 494, + 244.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 410, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 411, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 292, + 276 + ], + "score": 1.0, + "content": "In addition, we bound the error in substituting", + "type": "text" + }, + { + "bbox": [ + 293, + 261, + 309, + 273 + ], + "score": 0.91, + "content": "\\hat { K } _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 259, + 323, + 276 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 324, + 261, + 340, + 273 + ], + "score": 0.91, + "content": "\\tilde { K } _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 259, + 411, + 276 + ], + "score": 1.0, + "content": "defined in (139):", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 277, + 471, + 364 + ], + "lines": [ + { + "bbox": [ + 115, + 277, + 471, + 364 + ], + "spans": [ + { + "bbox": [ + 115, + 277, + 471, + 364 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } ( \\hat { K } _ { X } - \\tilde { K } _ { X } ) \\tilde { K } _ { X } ^ { - 1 } \\right) \\right| } \\\\ & { \\qquad < \\displaystyle \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { \\qquad = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "a9ba1caf037d2a81b9fd907732404bfbb489da398eeab5125b2164448ec185a1.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 115, + 277, + 471, + 306.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 115, + 306.0, + 471, + 335.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 115, + 335.0, + 471, + 364.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 163, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 163, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 163, + 380 + ], + "score": 1.0, + "content": "and similarly,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 383, + 459, + 497 + ], + "lines": [ + { + "bbox": [ + 150, + 383, + 459, + 497 + ], + "spans": [ + { + "bbox": [ + 150, + 383, + 459, + 497 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { ~ \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| } \\\\ & { = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } ) X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } ) \\right) \\right| } \\\\ & { < \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "f72f4a59ecfdbef314b4adb1f28a71d1511f198caf239c57254239278a48941b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 150, + 383, + 459, + 421.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 150, + 421.0, + 459, + 459.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 150, + 459.0, + 459, + 497.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 500, + 292, + 512 + ], + "lines": [ + { + "bbox": [ + 106, + 499, + 293, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 293, + 513 + ], + "score": 1.0, + "content": "Combining these two formulas in (143) yields", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 515, + 427, + 598 + ], + "lines": [ + { + "bbox": [ + 183, + 515, + 427, + 598 + ], + "spans": [ + { + "bbox": [ + 183, + 515, + 427, + 598 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { 2 B \\to \\mathbb { E } _ { x } [ ( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\quad \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ) } \\\\ & { \\quad = \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 1 } X ^ { \\top } ) ^ { 2 } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "77a8c4bafb16bbda9850c5d84fd4faecc20df8ada8490b1af26cd21bbf1df9d0.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 183, + 515, + 427, + 531.6 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 183, + 531.6, + 427, + 548.2 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 183, + 548.2, + 427, + 564.8000000000001 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 564.8000000000001, + 427, + 581.4000000000001 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 581.4000000000001, + 427, + 598.0000000000001 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 600, + 361, + 612 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 361, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 361, + 613 + ], + "score": 1.0, + "content": "Utilizing the Marcenko–Pastur law from Section ˇ B.2 we obtain", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 616, + 448, + 650 + ], + "lines": [ + { + "bbox": [ + 163, + 616, + 448, + 650 + ], + "spans": [ + { + "bbox": [ + 163, + 616, + 448, + 650 + ], + "score": 0.95, + "content": "B = \\beta ^ { \\top } \\beta \\left( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } \\right) ,", + "type": "interline_equation", + "image_path": "2fe9501a13b4ebe6aa84d2e71d0eb4933529d5f20e19673ae512ed5ba8b0d83a.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 616, + 448, + 627.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 163, + 627.3333333333334, + 448, + 638.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 163, + 638.6666666666667, + 448, + 650.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 183, + 668 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 184, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 133, + 670 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 654, + 180, + 668 + ], + "score": 0.93, + "content": "m = { b _ { 0 } } ^ { - 2 } { b _ { 1 } } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 653, + 184, + 670 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 107, + 678, + 237, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 237, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 237, + 691 + ], + "score": 1.0, + "content": "C.9.5 THE VARIANCE TERM", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 330, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 330, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 330, + 713 + ], + "score": 1.0, + "content": "Similarly, for the variance we utilize the approximation", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 714, + 368, + 736 + ], + "lines": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "spans": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "score": 0.94, + "content": "2 V = \\mathbb { E } _ { \\pmb { x } } \\left[ \\hat { \\pmb { u } } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { \\pmb { u } } ^ { \\top } \\right] \\sigma ^ { 2 }", + "type": "interline_equation", + "image_path": "528a09dc73ebbe81a976158c9006931939c2d35257e3b5d24c5f1aaca5e31b46.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "spans": [], + "index": 30 + } + ] + } + ], + "page_idx": 33, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 79, + 505, + 111 + ], + "lines": [ + { + "bbox": [ + 102, + 78, + 509, + 101 + ], + "spans": [ + { + "bbox": [ + 102, + 78, + 234, + 101 + ], + "score": 1.0, + "content": "where (i) is due to the fact that", + "type": "text" + }, + { + "bbox": [ + 234, + 80, + 372, + 101 + ], + "score": 0.94, + "content": "\\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } = \\lambda _ { \\operatorname* { m i n } } ^ { - 1 } ( \\hat { K } _ { X } ) = O ( 1 / d )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 78, + 390, + 101 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 391, + 83, + 457, + 97 + ], + "score": 0.93, + "content": "\\| X \\| _ { 2 } = O ( { \\sqrt { d } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 78, + 509, + 101 + ], + "score": 1.0, + "content": ". Therefore", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 99, + 205, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 152, + 111 + ], + "score": 1.0, + "content": "we have as", + "type": "text" + }, + { + "bbox": [ + 152, + 99, + 205, + 111 + ], + "score": 0.9, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 102, + 78, + 509, + 111 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 114, + 456, + 171 + ], + "lines": [ + { + "bbox": [ + 132, + 114, + 456, + 171 + ], + "spans": [ + { + "bbox": [ + 132, + 114, + 456, + 171 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { 2 B = \\mathbb { E } _ { x } [ ( { \\pmb x } ^ { \\top } \\beta - \\hat { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - \\tilde { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\qquad = \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - b _ { 0 } ^ { 2 } { \\pmb x } ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "c46331efc24df13f917fdbc06e8e67c167e3d3a1c36ca7854f077f2c1962415c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 132, + 114, + 456, + 133.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 132, + 133.0, + 456, + 152.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 132, + 152.0, + 456, + 171.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 502, + 186 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 503, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 216, + 186 + ], + "score": 1.0, + "content": "By taking expectation over", + "type": "text" + }, + { + "bbox": [ + 216, + 176, + 224, + 183 + ], + "score": 0.79, + "content": "_ { \\textbf { \\em x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 172, + 503, + 186 + ], + "score": 1.0, + "content": "and the rotational invariance argument similar to Hastie et al. (2019),", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 172, + 503, + 186 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 189, + 494, + 245 + ], + "lines": [ + { + "bbox": [ + 117, + 189, + 494, + 245 + ], + "spans": [ + { + "bbox": [ + 117, + 189, + 494, + 245 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\mathbb { E } _ { x } \\left[ \\left( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta \\right) ^ { 2 } \\right] } & { = \\mathbb { E } _ { x } \\left[ \\beta ^ { \\top } \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) ^ { 2 } \\beta \\right] } \\\\ & { = \\frac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } \\left( \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "c7cd0607ab6d3cbace39e38a775f5b4d928f82e32ea776a044c5a9b27956d490.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 117, + 189, + 494, + 207.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 117, + 207.66666666666666, + 494, + 226.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 117, + 226.33333333333331, + 494, + 244.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 410, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 411, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 292, + 276 + ], + "score": 1.0, + "content": "In addition, we bound the error in substituting", + "type": "text" + }, + { + "bbox": [ + 293, + 261, + 309, + 273 + ], + "score": 0.91, + "content": "\\hat { K } _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 259, + 323, + 276 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 324, + 261, + 340, + 273 + ], + "score": 0.91, + "content": "\\tilde { K } _ { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 259, + 411, + 276 + ], + "score": 1.0, + "content": "defined in (139):", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 259, + 411, + 276 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 277, + 471, + 364 + ], + "lines": [ + { + "bbox": [ + 115, + 277, + 471, + 364 + ], + "spans": [ + { + "bbox": [ + 115, + 277, + 471, + 364 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } ( \\hat { K } _ { X } - \\tilde { K } _ { X } ) \\tilde { K } _ { X } ^ { - 1 } \\right) \\right| } \\\\ & { \\qquad < \\displaystyle \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { \\qquad = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "a9ba1caf037d2a81b9fd907732404bfbb489da398eeab5125b2164448ec185a1.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 115, + 277, + 471, + 306.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 115, + 306.0, + 471, + 335.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 115, + 335.0, + 471, + 364.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 163, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 163, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 163, + 380 + ], + "score": 1.0, + "content": "and similarly,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 366, + 163, + 380 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 383, + 459, + 497 + ], + "lines": [ + { + "bbox": [ + 150, + 383, + 459, + 497 + ], + "spans": [ + { + "bbox": [ + 150, + 383, + 459, + 497 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { ~ \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| } \\\\ & { = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } ) X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } ) \\right) \\right| } \\\\ & { < \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "f72f4a59ecfdbef314b4adb1f28a71d1511f198caf239c57254239278a48941b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 150, + 383, + 459, + 421.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 150, + 421.0, + 459, + 459.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 150, + 459.0, + 459, + 497.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 500, + 292, + 512 + ], + "lines": [ + { + "bbox": [ + 106, + 499, + 293, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 293, + 513 + ], + "score": 1.0, + "content": "Combining these two formulas in (143) yields", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 106, + 499, + 293, + 513 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 515, + 427, + 598 + ], + "lines": [ + { + "bbox": [ + 183, + 515, + 427, + 598 + ], + "spans": [ + { + "bbox": [ + 183, + 515, + 427, + 598 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { 2 B \\to \\mathbb { E } _ { x } [ ( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\quad \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ) } \\\\ & { \\quad = \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 1 } X ^ { \\top } ) ^ { 2 } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "77a8c4bafb16bbda9850c5d84fd4faecc20df8ada8490b1af26cd21bbf1df9d0.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 183, + 515, + 427, + 531.6 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 183, + 531.6, + 427, + 548.2 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 183, + 548.2, + 427, + 564.8000000000001 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 564.8000000000001, + 427, + 581.4000000000001 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 581.4000000000001, + 427, + 598.0000000000001 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 600, + 361, + 612 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 361, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 361, + 613 + ], + "score": 1.0, + "content": "Utilizing the Marcenko–Pastur law from Section ˇ B.2 we obtain", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 600, + 361, + 613 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 616, + 448, + 650 + ], + "lines": [ + { + "bbox": [ + 163, + 616, + 448, + 650 + ], + "spans": [ + { + "bbox": [ + 163, + 616, + 448, + 650 + ], + "score": 0.95, + "content": "B = \\beta ^ { \\top } \\beta \\left( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } \\right) ,", + "type": "interline_equation", + "image_path": "2fe9501a13b4ebe6aa84d2e71d0eb4933529d5f20e19673ae512ed5ba8b0d83a.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 616, + 448, + 627.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 163, + 627.3333333333334, + 448, + 638.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 163, + 638.6666666666667, + 448, + 650.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 183, + 668 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 184, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 133, + 670 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 654, + 180, + 668 + ], + "score": 0.93, + "content": "m = { b _ { 0 } } ^ { - 2 } { b _ { 1 } } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 653, + 184, + 670 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 653, + 184, + 670 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 678, + 237, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 237, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 237, + 691 + ], + "score": 1.0, + "content": "C.9.5 THE VARIANCE TERM", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 330, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 696, + 330, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 330, + 713 + ], + "score": 1.0, + "content": "Similarly, for the variance we utilize the approximation", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 696, + 330, + 713 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 714, + 368, + 736 + ], + "lines": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "spans": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "score": 0.94, + "content": "2 V = \\mathbb { E } _ { \\pmb { x } } \\left[ \\hat { \\pmb { u } } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { \\pmb { u } } ^ { \\top } \\right] \\sigma ^ { 2 }", + "type": "interline_equation", + "image_path": "528a09dc73ebbe81a976158c9006931939c2d35257e3b5d24c5f1aaca5e31b46.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 243, + 714, + 368, + 736 + ], + "spans": [], + "index": 30 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 296, + 95 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 297, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 297, + 96 + ], + "score": 1.0, + "content": "Specifically, we bound the approximation error", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 100, + 460, + 158 + ], + "lines": [ + { + "bbox": [ + 149, + 100, + 460, + 158 + ], + "spans": [ + { + "bbox": [ + 149, + 100, + 460, + 158 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left| \\hat { \\boldsymbol u } \\hat { K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol u } ^ { \\top } - \\tilde { \\boldsymbol u } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\tilde { \\boldsymbol u } ^ { \\top } \\right| \\leq \\left\\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol u } + \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ { = O \\left( \\log ^ { l } d \\cdot \\frac { 1 } { d } \\cdot d \\cdot \\frac { 1 } { d } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "2a3055c96e8a78a1a11d0516f07e7a9d908f2c246fb0e917f13e9d5fed3eb326.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 149, + 100, + 460, + 119.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 149, + 119.33333333333333, + 460, + 138.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 149, + 138.66666666666666, + 460, + 158.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 160, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 161, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 161, + 177 + ], + "score": 1.0, + "content": "and similarly", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 182, + 435, + 306 + ], + "lines": [ + { + "bbox": [ + 174, + 182, + 435, + 306 + ], + "spans": [ + { + "bbox": [ + 174, + 182, + 435, + 306 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\quad \\left| \\mathbb { E } _ { \\mathbf { x } } \\left[ \\tilde { u } \\hat { K } _ { X } ^ { - 1 } \\hat { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } - \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\right] \\right| } \\\\ & { = \\mathrm { t r } \\left( \\left( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } \\right) \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) \\mathbb { E } _ { \\alpha \\tilde { u } \\tilde { u } ^ { \\top } } \\right) } \\\\ & { = \\mathrm { t r } \\left( \\hat { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } - \\tilde { K } _ { X } \\right) \\tilde { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) X ^ { T } X \\right) } \\\\ & { \\leq \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { T } X \\right\\| _ { 2 } } \\\\ & { = O ( d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d ^ { - 1 } \\cdot d ) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "9da21504c8e439b6b5d1544adef1105e0ef1f7169fbe031bb04963489a29671e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 174, + 182, + 435, + 223.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 174, + 223.33333333333334, + 435, + 264.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 174, + 264.6666666666667, + 435, + 306.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 407, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 310, + 407, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 354, + 325 + ], + "score": 1.0, + "content": "By combining the two approximations above we know that as", + "type": "text" + }, + { + "bbox": [ + 355, + 312, + 407, + 323 + ], + "score": 0.9, + "content": "n , d , p \\to \\infty", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 330, + 414, + 364 + ], + "lines": [ + { + "bbox": [ + 196, + 330, + 414, + 364 + ], + "spans": [ + { + "bbox": [ + 196, + 330, + 414, + 364 + ], + "score": 0.92, + "content": "| 2 V - \\mathbb { E } _ { \\pmb { x } } [ \\tilde { \\pmb { u } } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { \\pmb { u } } ^ { \\top } ] \\sigma ^ { 2 } | = O ( \\frac { \\log ^ { l } d } { d } ) 0 .", + "type": "interline_equation", + "image_path": "846c96f34183bd92dc3a78c781ffadc6b8b879cd9cd9cb9d73975cb99118a392.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 330, + 414, + 347.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 196, + 347.0, + 414, + 364.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 370, + 243, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 243, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 243, + 383 + ], + "score": 1.0, + "content": "Therefore the variance is given as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 388, + 434, + 502 + ], + "lines": [ + { + "bbox": [ + 176, + 388, + 434, + 502 + ], + "spans": [ + { + "bbox": [ + 176, + 388, + 434, + 502 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { 2 V \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ b _ { 0 } ^ { 4 } { \\boldsymbol x } ^ { T } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 2 } X ^ { T } { \\boldsymbol x } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\frac { 1 } { d } \\mathrm { t r } ( \\frac { 1 } { d } X ^ { T } X \\cdot ( \\frac { 1 } { d } X ^ { T } X + b _ { 0 } ^ { - 2 } b _ { 1 } ^ { 2 } I ) ^ { - 2 } ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( - \\frac { 1 } { 2 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 2 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) , } \\end{array}", + "type": "interline_equation", + "image_path": "8f32fa4b189ad2c8aa784797458621d67fe9031221b58ed73fef2d6f0b3593a3.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 176, + 388, + 434, + 426.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 176, + 426.0, + 434, + 464.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 176, + 464.0, + 434, + 502.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 507, + 319, + 519 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 320, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 133, + 520 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 509, + 143, + 517 + ], + "score": 0.75, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 506, + 320, + 520 + ], + "score": 1.0, + "content": "is defined in the derivation of the bias term.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 533, + 267, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 267, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 267, + 545 + ], + "score": 1.0, + "content": "C.9.6 PUTTING THINGS TOGETHER", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 352, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 351, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 351, + 565 + ], + "score": 1.0, + "content": "Recall the population risk is the sum of the bias and variance", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 571, + 441, + 641 + ], + "lines": [ + { + "bbox": [ + 170, + 571, + 441, + 641 + ], + "spans": [ + { + "bbox": [ + 170, + 571, + 441, + 641 + ], + "score": 0.94, + "content": "\\begin{array} { c } { { R r ^ { 2 } ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) } } \\\\ { { + \\sigma ^ { 2 } ( - \\frac { 1 } { 4 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) . } } \\end{array}", + "type": "interline_equation", + "image_path": "3facaf2ed650fba8a18513b67d5c668860efd0ce5075b96d09aac330886e0bde.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 170, + 571, + 441, + 594.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 170, + 594.3333333333334, + 441, + 617.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 170, + 617.6666666666667, + 441, + 641.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 506, + 710 + ], + "lines": [ + { + "bbox": [ + 106, + 652, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 317, + 664 + ], + "score": 1.0, + "content": "Observe that the population risk is independent of", + "type": "text" + }, + { + "bbox": [ + 318, + 654, + 329, + 664 + ], + "score": 0.85, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 652, + 505, + 664 + ], + "score": 1.0, + "content": ", i.e. double descent does not occur when", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "the network is overparameterized via changing the width. In addition, the bias is monotonically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 674, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 291, + 687 + ], + "score": 1.0, + "content": "increasing and upper-bounded by the null risk", + "type": "text" + }, + { + "bbox": [ + 291, + 674, + 301, + 684 + ], + "score": 0.86, + "content": "r ^ { \\bar { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "and lower-bounded by the bias of the least squares", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 685, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 226, + 700 + ], + "score": 1.0, + "content": "solution on the input features", + "type": "text" + }, + { + "bbox": [ + 226, + 685, + 267, + 699 + ], + "score": 0.94, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 685, + 455, + 700 + ], + "score": 1.0, + "content": ", whereas the variance remains bounded for all", + "type": "text" + }, + { + "bbox": [ + 455, + 686, + 505, + 699 + ], + "score": 0.93, + "content": "\\gamma _ { 1 } \\in ( 0 , \\infty )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 698, + 255, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 148, + 711 + ], + "score": 1.0, + "content": "as long as", + "type": "text" + }, + { + "bbox": [ + 149, + 699, + 176, + 709 + ], + "score": 0.89, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 698, + 195, + 711 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 195, + 699, + 203, + 710 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 698, + 255, + 711 + ], + "score": 1.0, + "content": "is nonlinear.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + } + ], + "page_idx": 34, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 715, + 504, + 725 + ], + "lines": [ + { + "bbox": [ + 496, + 717, + 504, + 725 + ], + "spans": [ + { + "bbox": [ + 496, + 717, + 504, + 725 + ], + "score": 1.0, + "content": "\u0003", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 296, + 95 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 297, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 297, + 96 + ], + "score": 1.0, + "content": "Specifically, we bound the approximation error", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 107, + 82, + 297, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 100, + 460, + 158 + ], + "lines": [ + { + "bbox": [ + 149, + 100, + 460, + 158 + ], + "spans": [ + { + "bbox": [ + 149, + 100, + 460, + 158 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left| \\hat { \\boldsymbol u } \\hat { K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol u } ^ { \\top } - \\tilde { \\boldsymbol u } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\tilde { \\boldsymbol u } ^ { \\top } \\right| \\leq \\left\\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol u } + \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ { = O \\left( \\log ^ { l } d \\cdot \\frac { 1 } { d } \\cdot d \\cdot \\frac { 1 } { d } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "2a3055c96e8a78a1a11d0516f07e7a9d908f2c246fb0e917f13e9d5fed3eb326.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 149, + 100, + 460, + 119.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 149, + 119.33333333333333, + 460, + 138.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 149, + 138.66666666666666, + 460, + 158.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 160, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 161, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 161, + 177 + ], + "score": 1.0, + "content": "and similarly", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 163, + 161, + 177 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 182, + 435, + 306 + ], + "lines": [ + { + "bbox": [ + 174, + 182, + 435, + 306 + ], + "spans": [ + { + "bbox": [ + 174, + 182, + 435, + 306 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\quad \\left| \\mathbb { E } _ { \\mathbf { x } } \\left[ \\tilde { u } \\hat { K } _ { X } ^ { - 1 } \\hat { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } - \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\right] \\right| } \\\\ & { = \\mathrm { t r } \\left( \\left( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } \\right) \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) \\mathbb { E } _ { \\alpha \\tilde { u } \\tilde { u } ^ { \\top } } \\right) } \\\\ & { = \\mathrm { t r } \\left( \\hat { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } - \\tilde { K } _ { X } \\right) \\tilde { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) X ^ { T } X \\right) } \\\\ & { \\leq \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { T } X \\right\\| _ { 2 } } \\\\ & { = O ( d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d ^ { - 1 } \\cdot d ) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}", + "type": "interline_equation", + "image_path": "9da21504c8e439b6b5d1544adef1105e0ef1f7169fbe031bb04963489a29671e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 174, + 182, + 435, + 223.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 174, + 223.33333333333334, + 435, + 264.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 174, + 264.6666666666667, + 435, + 306.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 407, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 310, + 407, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 354, + 325 + ], + "score": 1.0, + "content": "By combining the two approximations above we know that as", + "type": "text" + }, + { + "bbox": [ + 355, + 312, + 407, + 323 + ], + "score": 0.9, + "content": "n , d , p \\to \\infty", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 310, + 407, + 325 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 330, + 414, + 364 + ], + "lines": [ + { + "bbox": [ + 196, + 330, + 414, + 364 + ], + "spans": [ + { + "bbox": [ + 196, + 330, + 414, + 364 + ], + "score": 0.92, + "content": "| 2 V - \\mathbb { E } _ { \\pmb { x } } [ \\tilde { \\pmb { u } } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { \\pmb { u } } ^ { \\top } ] \\sigma ^ { 2 } | = O ( \\frac { \\log ^ { l } d } { d } ) 0 .", + "type": "interline_equation", + "image_path": "846c96f34183bd92dc3a78c781ffadc6b8b879cd9cd9cb9d73975cb99118a392.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 330, + 414, + 347.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 196, + 347.0, + 414, + 364.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 370, + 243, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 243, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 243, + 383 + ], + "score": 1.0, + "content": "Therefore the variance is given as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 369, + 243, + 383 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 388, + 434, + 502 + ], + "lines": [ + { + "bbox": [ + 176, + 388, + 434, + 502 + ], + "spans": [ + { + "bbox": [ + 176, + 388, + 434, + 502 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { 2 V \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ b _ { 0 } ^ { 4 } { \\boldsymbol x } ^ { T } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 2 } X ^ { T } { \\boldsymbol x } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\frac { 1 } { d } \\mathrm { t r } ( \\frac { 1 } { d } X ^ { T } X \\cdot ( \\frac { 1 } { d } X ^ { T } X + b _ { 0 } ^ { - 2 } b _ { 1 } ^ { 2 } I ) ^ { - 2 } ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( - \\frac { 1 } { 2 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 2 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) , } \\end{array}", + "type": "interline_equation", + "image_path": "8f32fa4b189ad2c8aa784797458621d67fe9031221b58ed73fef2d6f0b3593a3.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 176, + 388, + 434, + 426.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 176, + 426.0, + 434, + 464.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 176, + 464.0, + 434, + 502.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 507, + 319, + 519 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 320, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 133, + 520 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 509, + 143, + 517 + ], + "score": 0.75, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 506, + 320, + 520 + ], + "score": 1.0, + "content": "is defined in the derivation of the bias term.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 506, + 320, + 520 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 533, + 267, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 267, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 267, + 545 + ], + "score": 1.0, + "content": "C.9.6 PUTTING THINGS TOGETHER", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 352, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 351, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 351, + 565 + ], + "score": 1.0, + "content": "Recall the population risk is the sum of the bias and variance", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 553, + 351, + 565 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 571, + 441, + 641 + ], + "lines": [ + { + "bbox": [ + 170, + 571, + 441, + 641 + ], + "spans": [ + { + "bbox": [ + 170, + 571, + 441, + 641 + ], + "score": 0.94, + "content": "\\begin{array} { c } { { R r ^ { 2 } ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) } } \\\\ { { + \\sigma ^ { 2 } ( - \\frac { 1 } { 4 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) . } } \\end{array}", + "type": "interline_equation", + "image_path": "3facaf2ed650fba8a18513b67d5c668860efd0ce5075b96d09aac330886e0bde.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 170, + 571, + 441, + 594.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 170, + 594.3333333333334, + 441, + 617.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 170, + 617.6666666666667, + 441, + 641.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 506, + 710 + ], + "lines": [ + { + "bbox": [ + 106, + 652, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 317, + 664 + ], + "score": 1.0, + "content": "Observe that the population risk is independent of", + "type": "text" + }, + { + "bbox": [ + 318, + 654, + 329, + 664 + ], + "score": 0.85, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 652, + 505, + 664 + ], + "score": 1.0, + "content": ", i.e. double descent does not occur when", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "the network is overparameterized via changing the width. In addition, the bias is monotonically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 674, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 291, + 687 + ], + "score": 1.0, + "content": "increasing and upper-bounded by the null risk", + "type": "text" + }, + { + "bbox": [ + 291, + 674, + 301, + 684 + ], + "score": 0.86, + "content": "r ^ { \\bar { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "and lower-bounded by the bias of the least squares", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 685, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 226, + 700 + ], + "score": 1.0, + "content": "solution on the input features", + "type": "text" + }, + { + "bbox": [ + 226, + 685, + 267, + 699 + ], + "score": 0.94, + "content": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 685, + 455, + 700 + ], + "score": 1.0, + "content": ", whereas the variance remains bounded for all", + "type": "text" + }, + { + "bbox": [ + 455, + 686, + 505, + 699 + ], + "score": 0.93, + "content": "\\gamma _ { 1 } \\in ( 0 , \\infty )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 698, + 255, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 148, + 711 + ], + "score": 1.0, + "content": "as long as", + "type": "text" + }, + { + "bbox": [ + 149, + 699, + 176, + 709 + ], + "score": 0.89, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 698, + 195, + 711 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 195, + 699, + 203, + 710 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 698, + 255, + 711 + ], + "score": 1.0, + "content": "is nonlinear.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 652, + 505, + 711 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 80, + 220, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 221, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 221, + 96 + ], + "score": 1.0, + "content": "D USEFUL LEMMAS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 105, + 274, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 274, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 185, + 120 + ], + "score": 1.0, + "content": "Lemma 16. Given", + "type": "text" + }, + { + "bbox": [ + 186, + 106, + 204, + 118 + ], + "score": 0.89, + "content": "K _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 104, + 274, + 120 + ], + "score": 1.0, + "content": "from (46), define", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 123, + 362, + 138 + ], + "lines": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "spans": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "score": 0.92, + "content": "\\tilde { K } _ { W } = r I _ { h } + s { \\bf 1 } _ { h } { \\bf 1 } _ { h } ^ { \\top } + t Q ,", + "type": "interline_equation", + "image_path": "ae034d2ccf49a28b5f14f5cc55bb7efb8f3d80fb70632a984a4783f2fe0f0252.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 506, + 177 + ], + "lines": [ + { + "bbox": [ + 104, + 142, + 504, + 160 + ], + "spans": [ + { + "bbox": [ + 104, + 142, + 135, + 160 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 144, + 185, + 157 + ], + "score": 0.91, + "content": "Q \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 142, + 208, + 160 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 208, + 144, + 277, + 158 + ], + "score": 0.93, + "content": "Q _ { i \\neq j } \\ = \\ w _ { i } ^ { \\top } w _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 142, + 299, + 160 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 146, + 340, + 158 + ], + "score": 0.91, + "content": "Q _ { i , i } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 142, + 364, + 160 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 365, + 145, + 504, + 158 + ], + "score": 0.86, + "content": "r = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , s =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 156, + 434, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 146, + 173 + ], + "score": 0.82, + "content": "\\mathbb { E } [ \\phi ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 156, + 151, + 177 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 151, + 160, + 214, + 173 + ], + "score": 0.9, + "content": "t = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 156, + 253, + 177 + ], + "score": 1.0, + "content": ". Then as", + "type": "text" + }, + { + "bbox": [ + 264, + 157, + 427, + 178 + ], + "score": 0.62, + "content": "d , h \\to \\infty , \\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d a . s .", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 156, + 434, + 177 + ], + "score": 1.0, + "content": "..", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 186, + 504, + 210 + ], + "lines": [ + { + "bbox": [ + 104, + 185, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 104, + 185, + 242, + 201 + ], + "score": 1.0, + "content": "Proof. Consider the event where", + "type": "text" + }, + { + "bbox": [ + 243, + 186, + 404, + 200 + ], + "score": 0.91, + "content": "\\mathcal { A } _ { \\epsilon } = \\big \\{ | \\| \\pmb { w } _ { i } \\| _ { 2 } - 1 | < \\epsilon , | \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } | < \\epsilon \\big \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 185, + 459, + 201 + ], + "score": 1.0, + "content": ". Under event", + "type": "text" + }, + { + "bbox": [ + 460, + 187, + 472, + 198 + ], + "score": 0.88, + "content": "\\mathcal { A } _ { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 185, + 505, + 201 + ], + "score": 1.0, + "content": ", for the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 198, + 281, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 281, + 211 + ], + "score": 1.0, + "content": "diagonal term of the kernel matrix we have", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 214, + 432, + 254 + ], + "lines": [ + { + "bbox": [ + 177, + 214, + 432, + 254 + ], + "spans": [ + { + "bbox": [ + 177, + 214, + 432, + 254 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { ~ \\Big | [ K _ { W } ] _ { i i } - [ \\tilde { K } _ { W } ] _ { i i } \\Big | = \\Big | \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\Big ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\Big | } \\\\ & { = \\big | \\mathbb { E } [ \\phi ( \\| \\pmb { w } _ { i } \\| _ { 2 } G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\big | = O ( \\epsilon ) . } \\end{array}", + "type": "interline_equation", + "image_path": "093ace88c911b29b6694e360e7207d76f52f55a0da3df8ff86c0b4ae705fe0df.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 177, + 214, + 432, + 227.33333333333334 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 177, + 227.33333333333334, + 432, + 240.66666666666669 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 177, + 240.66666666666669, + 432, + 254.00000000000003 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 442, + 269 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 443, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 443, + 271 + ], + "score": 1.0, + "content": "And for off-diagonal term, by the decomposition introduced in Section B.3 we have", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 274, + 455, + 295 + ], + "lines": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "spans": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "score": 0.92, + "content": "[ K _ { W } ] _ { i j } = \\mathbb { E } _ { \\boldsymbol { x } } \\Big [ \\phi ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) \\Big ] = \\| \\boldsymbol { w } _ { i } \\| _ { 2 } \\| \\boldsymbol { w } _ { j } \\| _ { 2 } + \\mathbb { E } _ { \\boldsymbol { x } } [ \\phi _ { \\bot } ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi _ { \\bot } ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) ] ,", + "type": "interline_equation", + "image_path": "654bfbbe8b3c7dbb706af222a370df5aefbddcc439911a23d215b55210c35966.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 133, + 317 + ], + "score": 1.0, + "content": "hence", + "type": "text" + }, + { + "bbox": [ + 133, + 300, + 270, + 316 + ], + "score": 0.92, + "content": "| [ K _ { W } ] _ { i j } - [ \\tilde { K } _ { W } ] _ { i j } | < ( \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } ) ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 300, + 325, + 317 + ], + "score": 1.0, + "content": ". Notice that", + "type": "text" + }, + { + "bbox": [ + 325, + 303, + 338, + 314 + ], + "score": 0.88, + "content": "\\mathcal { A } _ { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 300, + 399, + 317 + ], + "score": 1.0, + "content": "holds a.s. for", + "type": "text" + }, + { + "bbox": [ + 399, + 301, + 462, + 315 + ], + "score": 0.93, + "content": "\\epsilon = \\log ^ { c } d / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 300, + 505, + 317 + ], + "score": 1.0, + "content": "and large", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 314, + 348, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 138, + 334 + ], + "score": 1.0, + "content": "enough", + "type": "text" + }, + { + "bbox": [ + 138, + 318, + 162, + 329 + ], + "score": 0.87, + "content": "c > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 315, + 240, + 334 + ], + "score": 1.0, + "content": "; we therefore have", + "type": "text" + }, + { + "bbox": [ + 240, + 314, + 344, + 335 + ], + "score": 0.93, + "content": "\\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d .", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 315, + 348, + 334 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 104, + 351, + 504, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 351, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 175, + 366 + ], + "score": 1.0, + "content": "Lemma 17. Let", + "type": "text" + }, + { + "bbox": [ + 175, + 351, + 194, + 364 + ], + "score": 0.9, + "content": "{ \\hat { \\boldsymbol { \\beta } } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 351, + 506, + 366 + ], + "score": 1.0, + "content": "be the solution to the gradient flow at time t defined in (103)(104). Then as", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 363, + 282, + 376 + ], + "spans": [ + { + "bbox": [ + 107, + 364, + 149, + 375 + ], + "score": 0.91, + "content": "n , d \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 363, + 167, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 365, + 197, + 375 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 363, + 282, + 376 + ], + "score": 1.0, + "content": "the following holds.:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 195, + 380, + 416, + 396 + ], + "lines": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "spans": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "score": 0.9, + "content": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O _ { P } ( 1 ) ; \\quad \\| \\boldsymbol { X } ^ { \\top } \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d ) .", + "type": "interline_equation", + "image_path": "882d4cb8e4be17bb4e5db8176a16255e6736d70bc8f0aa83a215155c79205713.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 406, + 492, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 494, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 190, + 419 + ], + "score": 1.0, + "content": "Proof. We consider", + "type": "text" + }, + { + "bbox": [ + 190, + 408, + 216, + 417 + ], + "score": 0.9, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 407, + 494, + 419 + ], + "score": 1.0, + "content": "for simplicity, and result for the other case follows in similar fashion.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 423, + 506, + 467 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 152, + 439 + ], + "score": 1.0, + "content": "In this case", + "type": "text" + }, + { + "bbox": [ + 152, + 424, + 335, + 438 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\hat { \\pmb { \\beta } } ( t ) = \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 423, + 506, + 439 + ], + "score": 1.0, + "content": ". From (Hastie et al., 2019, Corollary 1) we", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 437, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 148, + 454 + ], + "score": 1.0, + "content": "know that", + "type": "text" + }, + { + "bbox": [ + 148, + 439, + 309, + 453 + ], + "score": 0.91, + "content": "\\| \\hat { \\pmb \\beta } ( \\infty ) \\| _ { 2 } = \\left\\| ( X X ^ { \\top } ) ^ { - 1 } X \\pmb y \\right\\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 437, + 324, + 454 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 324, + 440, + 353, + 452 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 437, + 397, + 454 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 397, + 438, + 506, + 453 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { { \\| { \\cal I } - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\| _ { 2 } = } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 452, + 300, + 468 + ], + "spans": [ + { + "bbox": [ + 107, + 454, + 128, + 466 + ], + "score": 0.9, + "content": "O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 452, + 144, + 468 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 144, + 455, + 167, + 465 + ], + "score": 0.89, + "content": "t \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 452, + 228, + 468 + ], + "score": 1.0, + "content": "; it follows that", + "type": "text" + }, + { + "bbox": [ + 229, + 453, + 296, + 467 + ], + "score": 0.91, + "content": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 452, + 300, + 468 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 471, + 506, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 469, + 507, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 269, + 487 + ], + "score": 1.0, + "content": "For the second part, we utilize the SVD", + "type": "text" + }, + { + "bbox": [ + 270, + 472, + 322, + 484 + ], + "score": 0.89, + "content": "\\boldsymbol { X } = \\boldsymbol { U \\Sigma V } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 469, + 354, + 487 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 354, + 472, + 400, + 486 + ], + "score": 0.82, + "content": "\\Sigma = [ \\hat { \\Sigma } ; 0 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 469, + 405, + 487 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 405, + 471, + 448, + 484 + ], + "score": 0.78, + "content": "\\hat { \\Sigma } \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 469, + 466, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 466, + 471, + 503, + 485 + ], + "score": 0.93, + "content": "\\hat { \\Sigma } _ { i i } = \\lambda _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 469, + 507, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 484, + 507, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 143, + 500 + ], + "score": 1.0, + "content": "We have", + "type": "text" + }, + { + "bbox": [ + 143, + 485, + 275, + 500 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) = U \\bar { \\Sigma } U ^ { \\top } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 484, + 303, + 500 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 303, + 486, + 405, + 499 + ], + "score": 0.93, + "content": "\\bar { \\Sigma } _ { i , i } = 1 - \\exp ( - t \\lambda _ { i } ^ { 2 } / n )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 484, + 415, + 500 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 416, + 487, + 439, + 498 + ], + "score": 0.88, + "content": "i \\leq d", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 484, + 507, + 500 + ], + "score": 1.0, + "content": "and 0 otherwise.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 505, + 477, + 606 + ], + "lines": [ + { + "bbox": [ + 113, + 505, + 477, + 606 + ], + "spans": [ + { + "bbox": [ + 113, + 505, + 477, + 606 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\| X ^ { \\top } \\hat { \\pmb \\beta } ( t ) \\| _ { \\infty } = \\left\\| X ^ { \\top } \\left( I - \\exp ( - \\frac t n X X ^ { \\top } ) \\right) \\left( X X ^ { \\top } \\right) ^ { - 1 } X \\pmb y \\right\\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } U ^ { \\top } U \\bar { \\Sigma } U ^ { \\top } U \\hat { \\Sigma } ^ { - 2 } U ^ { \\top } U \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\| \\pmb y \\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) } \\\\ & { \\qquad \\leq \\| V \\| _ { \\infty } \\left\\| \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma \\right\\| _ { \\infty } \\| V ^ { \\top } \\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) \\overset { ( i ) } { \\leq } O _ { P } ( \\mathrm { p o l y } \\log d ) , } \\end{array}", + "type": "interline_equation", + "image_path": "ebcdf9a75a1a71b9f2ae7aeb56343d6144c170cf1d700b8903231f442418092f.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 113, + 505, + 477, + 538.6666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 113, + 538.6666666666666, + 477, + 572.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 113, + 572.3333333333333, + 477, + 605.9999999999999 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 608, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 471, + 621 + ], + "score": 1.0, + "content": "where (i) follows from the concentration of the Gaussian maxima, and the fact that the law of", + "type": "text" + }, + { + "bbox": [ + 472, + 609, + 480, + 619 + ], + "score": 0.83, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 608, + 505, + 621 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 176, + 632 + ], + "score": 1.0, + "content": "Haar measure on", + "type": "text" + }, + { + "bbox": [ + 177, + 620, + 205, + 632 + ], + "score": 0.91, + "content": "S O ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 619, + 321, + 632 + ], + "score": 1.0, + "content": ", and thus for any unit vector", + "type": "text" + }, + { + "bbox": [ + 322, + 622, + 329, + 630 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 619, + 391, + 632 + ], + "score": 1.0, + "content": "independent to", + "type": "text" + }, + { + "bbox": [ + 391, + 620, + 400, + 630 + ], + "score": 0.69, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 619, + 404, + 632 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 404, + 620, + 419, + 630 + ], + "score": 0.73, + "content": "V z", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "is uniform on sphere", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 631, + 309, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 124, + 645 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 632, + 224, + 645 + ], + "score": 0.91, + "content": "\\| V z \\| _ { \\infty } = O ( \\log d / \\sqrt { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 631, + 309, + 645 + ], + "score": 1.0, + "content": ". We wherefore have", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 190, + 650, + 421, + 678 + ], + "lines": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "spans": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "score": 0.94, + "content": "\\left\\| V \\right\\| _ { \\infty } = \\operatorname* { s u p } _ { z } { \\frac { \\left\\| V z \\right\\| _ { \\infty } } { \\left\\| z \\right\\| _ { \\infty } } } = O \\left( { \\frac { \\log d } { \\sqrt { d } } } \\right) { \\frac { \\left\\| z \\right\\| _ { 2 } } { \\left\\| z \\right\\| _ { \\infty } } } = O ( \\log d ) ,", + "type": "interline_equation", + "image_path": "cd7f286c737bf8d4242cb3974f2c3c9b4041f5a47e25fddd0d8b59682b61ae56.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 683, + 406, + 697 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 406, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 255, + 698 + ], + "score": 1.0, + "content": "Note that this result also implies that", + "type": "text" + }, + { + "bbox": [ + 256, + 683, + 403, + 697 + ], + "score": 0.91, + "content": "\\| \\pmb { y } - X ^ { \\top } \\pmb { \\beta } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 682, + 406, + 698 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + } + ], + "page_idx": 35, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 702, + 505, + 712 + ], + "lines": [ + { + "bbox": [ + 496, + 703, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 496, + 703, + 504, + 712 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 80, + 220, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 221, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 221, + 96 + ], + "score": 1.0, + "content": "D USEFUL LEMMAS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 105, + 274, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 274, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 185, + 120 + ], + "score": 1.0, + "content": "Lemma 16. Given", + "type": "text" + }, + { + "bbox": [ + 186, + 106, + 204, + 118 + ], + "score": 0.89, + "content": "K _ { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 104, + 274, + 120 + ], + "score": 1.0, + "content": "from (46), define", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 104, + 274, + 120 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 123, + 362, + 138 + ], + "lines": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "spans": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "score": 0.92, + "content": "\\tilde { K } _ { W } = r I _ { h } + s { \\bf 1 } _ { h } { \\bf 1 } _ { h } ^ { \\top } + t Q ,", + "type": "interline_equation", + "image_path": "ae034d2ccf49a28b5f14f5cc55bb7efb8f3d80fb70632a984a4783f2fe0f0252.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 248, + 123, + 362, + 138 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 506, + 177 + ], + "lines": [ + { + "bbox": [ + 104, + 142, + 504, + 160 + ], + "spans": [ + { + "bbox": [ + 104, + 142, + 135, + 160 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 144, + 185, + 157 + ], + "score": 0.91, + "content": "Q \\in \\mathbb { R } ^ { h \\times h }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 142, + 208, + 160 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 208, + 144, + 277, + 158 + ], + "score": 0.93, + "content": "Q _ { i \\neq j } \\ = \\ w _ { i } ^ { \\top } w _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 142, + 299, + 160 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 146, + 340, + 158 + ], + "score": 0.91, + "content": "Q _ { i , i } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 142, + 364, + 160 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 365, + 145, + 504, + 158 + ], + "score": 0.86, + "content": "r = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , s =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 156, + 434, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 146, + 173 + ], + "score": 0.82, + "content": "\\mathbb { E } [ \\phi ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 156, + 151, + 177 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 151, + 160, + 214, + 173 + ], + "score": 0.9, + "content": "t = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 156, + 253, + 177 + ], + "score": 1.0, + "content": ". Then as", + "type": "text" + }, + { + "bbox": [ + 264, + 157, + 427, + 178 + ], + "score": 0.62, + "content": "d , h \\to \\infty , \\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d a . s .", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 156, + 434, + 177 + ], + "score": 1.0, + "content": "..", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 142, + 504, + 178 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 186, + 504, + 210 + ], + "lines": [ + { + "bbox": [ + 104, + 185, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 104, + 185, + 242, + 201 + ], + "score": 1.0, + "content": "Proof. Consider the event where", + "type": "text" + }, + { + "bbox": [ + 243, + 186, + 404, + 200 + ], + "score": 0.91, + "content": "\\mathcal { A } _ { \\epsilon } = \\big \\{ | \\| \\pmb { w } _ { i } \\| _ { 2 } - 1 | < \\epsilon , | \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } | < \\epsilon \\big \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 185, + 459, + 201 + ], + "score": 1.0, + "content": ". Under event", + "type": "text" + }, + { + "bbox": [ + 460, + 187, + 472, + 198 + ], + "score": 0.88, + "content": "\\mathcal { A } _ { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 185, + 505, + 201 + ], + "score": 1.0, + "content": ", for the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 198, + 281, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 281, + 211 + ], + "score": 1.0, + "content": "diagonal term of the kernel matrix we have", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 185, + 505, + 211 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 214, + 432, + 254 + ], + "lines": [ + { + "bbox": [ + 177, + 214, + 432, + 254 + ], + "spans": [ + { + "bbox": [ + 177, + 214, + 432, + 254 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { ~ \\Big | [ K _ { W } ] _ { i i } - [ \\tilde { K } _ { W } ] _ { i i } \\Big | = \\Big | \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\Big ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\Big | } \\\\ & { = \\big | \\mathbb { E } [ \\phi ( \\| \\pmb { w } _ { i } \\| _ { 2 } G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\big | = O ( \\epsilon ) . } \\end{array}", + "type": "interline_equation", + "image_path": "093ace88c911b29b6694e360e7207d76f52f55a0da3df8ff86c0b4ae705fe0df.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 177, + 214, + 432, + 227.33333333333334 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 177, + 227.33333333333334, + 432, + 240.66666666666669 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 177, + 240.66666666666669, + 432, + 254.00000000000003 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 442, + 269 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 443, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 443, + 271 + ], + "score": 1.0, + "content": "And for off-diagonal term, by the decomposition introduced in Section B.3 we have", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 257, + 443, + 271 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 274, + 455, + 295 + ], + "lines": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "spans": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "score": 0.92, + "content": "[ K _ { W } ] _ { i j } = \\mathbb { E } _ { \\boldsymbol { x } } \\Big [ \\phi ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) \\Big ] = \\| \\boldsymbol { w } _ { i } \\| _ { 2 } \\| \\boldsymbol { w } _ { j } \\| _ { 2 } + \\mathbb { E } _ { \\boldsymbol { x } } [ \\phi _ { \\bot } ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi _ { \\bot } ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) ] ,", + "type": "interline_equation", + "image_path": "654bfbbe8b3c7dbb706af222a370df5aefbddcc439911a23d215b55210c35966.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 133, + 274, + 455, + 295 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 133, + 317 + ], + "score": 1.0, + "content": "hence", + "type": "text" + }, + { + "bbox": [ + 133, + 300, + 270, + 316 + ], + "score": 0.92, + "content": "| [ K _ { W } ] _ { i j } - [ \\tilde { K } _ { W } ] _ { i j } | < ( \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } ) ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 300, + 325, + 317 + ], + "score": 1.0, + "content": ". Notice that", + "type": "text" + }, + { + "bbox": [ + 325, + 303, + 338, + 314 + ], + "score": 0.88, + "content": "\\mathcal { A } _ { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 300, + 399, + 317 + ], + "score": 1.0, + "content": "holds a.s. for", + "type": "text" + }, + { + "bbox": [ + 399, + 301, + 462, + 315 + ], + "score": 0.93, + "content": "\\epsilon = \\log ^ { c } d / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 300, + 505, + 317 + ], + "score": 1.0, + "content": "and large", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 314, + 348, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 138, + 334 + ], + "score": 1.0, + "content": "enough", + "type": "text" + }, + { + "bbox": [ + 138, + 318, + 162, + 329 + ], + "score": 0.87, + "content": "c > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 315, + 240, + 334 + ], + "score": 1.0, + "content": "; we therefore have", + "type": "text" + }, + { + "bbox": [ + 240, + 314, + 344, + 335 + ], + "score": 0.93, + "content": "\\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d .", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 315, + 348, + 334 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 300, + 505, + 335 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 351, + 504, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 351, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 175, + 366 + ], + "score": 1.0, + "content": "Lemma 17. Let", + "type": "text" + }, + { + "bbox": [ + 175, + 351, + 194, + 364 + ], + "score": 0.9, + "content": "{ \\hat { \\boldsymbol { \\beta } } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 351, + 506, + 366 + ], + "score": 1.0, + "content": "be the solution to the gradient flow at time t defined in (103)(104). Then as", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 363, + 282, + 376 + ], + "spans": [ + { + "bbox": [ + 107, + 364, + 149, + 375 + ], + "score": 0.91, + "content": "n , d \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 363, + 167, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 365, + 197, + 375 + ], + "score": 0.91, + "content": "\\gamma _ { 1 } \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 363, + 282, + 376 + ], + "score": 1.0, + "content": "the following holds.:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 351, + 506, + 376 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 195, + 380, + 416, + 396 + ], + "lines": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "spans": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "score": 0.9, + "content": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O _ { P } ( 1 ) ; \\quad \\| \\boldsymbol { X } ^ { \\top } \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d ) .", + "type": "interline_equation", + "image_path": "882d4cb8e4be17bb4e5db8176a16255e6736d70bc8f0aa83a215155c79205713.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 195, + 380, + 416, + 396 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 406, + 492, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 494, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 190, + 419 + ], + "score": 1.0, + "content": "Proof. We consider", + "type": "text" + }, + { + "bbox": [ + 190, + 408, + 216, + 417 + ], + "score": 0.9, + "content": "d < n", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 407, + 494, + 419 + ], + "score": 1.0, + "content": "for simplicity, and result for the other case follows in similar fashion.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 106, + 407, + 494, + 419 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 423, + 506, + 467 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 152, + 439 + ], + "score": 1.0, + "content": "In this case", + "type": "text" + }, + { + "bbox": [ + 152, + 424, + 335, + 438 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\hat { \\pmb { \\beta } } ( t ) = \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 423, + 506, + 439 + ], + "score": 1.0, + "content": ". From (Hastie et al., 2019, Corollary 1) we", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 437, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 148, + 454 + ], + "score": 1.0, + "content": "know that", + "type": "text" + }, + { + "bbox": [ + 148, + 439, + 309, + 453 + ], + "score": 0.91, + "content": "\\| \\hat { \\pmb \\beta } ( \\infty ) \\| _ { 2 } = \\left\\| ( X X ^ { \\top } ) ^ { - 1 } X \\pmb y \\right\\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 437, + 324, + 454 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 324, + 440, + 353, + 452 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 437, + 397, + 454 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 397, + 438, + 506, + 453 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { { \\| { \\cal I } - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\| _ { 2 } = } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 452, + 300, + 468 + ], + "spans": [ + { + "bbox": [ + 107, + 454, + 128, + 466 + ], + "score": 0.9, + "content": "O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 452, + 144, + 468 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 144, + 455, + 167, + 465 + ], + "score": 0.89, + "content": "t \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 452, + 228, + 468 + ], + "score": 1.0, + "content": "; it follows that", + "type": "text" + }, + { + "bbox": [ + 229, + 453, + 296, + 467 + ], + "score": 0.91, + "content": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 452, + 300, + 468 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 423, + 506, + 468 + ] + }, + { + "type": "list", + "bbox": [ + 107, + 471, + 506, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 469, + 507, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 269, + 487 + ], + "score": 1.0, + "content": "For the second part, we utilize the SVD", + "type": "text" + }, + { + "bbox": [ + 270, + 472, + 322, + 484 + ], + "score": 0.89, + "content": "\\boldsymbol { X } = \\boldsymbol { U \\Sigma V } ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 469, + 354, + 487 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 354, + 472, + 400, + 486 + ], + "score": 0.82, + "content": "\\Sigma = [ \\hat { \\Sigma } ; 0 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 469, + 405, + 487 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 405, + 471, + 448, + 484 + ], + "score": 0.78, + "content": "\\hat { \\Sigma } \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 469, + 466, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 466, + 471, + 503, + 485 + ], + "score": 0.93, + "content": "\\hat { \\Sigma } _ { i i } = \\lambda _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 469, + 507, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 484, + 507, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 143, + 500 + ], + "score": 1.0, + "content": "We have", + "type": "text" + }, + { + "bbox": [ + 143, + 485, + 275, + 500 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) = U \\bar { \\Sigma } U ^ { \\top } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 484, + 303, + 500 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 303, + 486, + 405, + 499 + ], + "score": 0.93, + "content": "\\bar { \\Sigma } _ { i , i } = 1 - \\exp ( - t \\lambda _ { i } ^ { 2 } / n )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 484, + 415, + 500 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 416, + 487, + 439, + 498 + ], + "score": 0.88, + "content": "i \\leq d", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 484, + 507, + 500 + ], + "score": 1.0, + "content": "and 0 otherwise.", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 469, + 507, + 500 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 505, + 477, + 606 + ], + "lines": [ + { + "bbox": [ + 113, + 505, + 477, + 606 + ], + "spans": [ + { + "bbox": [ + 113, + 505, + 477, + 606 + ], + "score": 0.96, + "content": "\\begin{array} { r l } & { \\| X ^ { \\top } \\hat { \\pmb \\beta } ( t ) \\| _ { \\infty } = \\left\\| X ^ { \\top } \\left( I - \\exp ( - \\frac t n X X ^ { \\top } ) \\right) \\left( X X ^ { \\top } \\right) ^ { - 1 } X \\pmb y \\right\\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } U ^ { \\top } U \\bar { \\Sigma } U ^ { \\top } U \\hat { \\Sigma } ^ { - 2 } U ^ { \\top } U \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\| \\pmb y \\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) } \\\\ & { \\qquad \\leq \\| V \\| _ { \\infty } \\left\\| \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma \\right\\| _ { \\infty } \\| V ^ { \\top } \\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) \\overset { ( i ) } { \\leq } O _ { P } ( \\mathrm { p o l y } \\log d ) , } \\end{array}", + "type": "interline_equation", + "image_path": "ebcdf9a75a1a71b9f2ae7aeb56343d6144c170cf1d700b8903231f442418092f.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 113, + 505, + 477, + 538.6666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 113, + 538.6666666666666, + 477, + 572.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 113, + 572.3333333333333, + 477, + 605.9999999999999 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 608, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 471, + 621 + ], + "score": 1.0, + "content": "where (i) follows from the concentration of the Gaussian maxima, and the fact that the law of", + "type": "text" + }, + { + "bbox": [ + 472, + 609, + 480, + 619 + ], + "score": 0.83, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 608, + 505, + 621 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 176, + 632 + ], + "score": 1.0, + "content": "Haar measure on", + "type": "text" + }, + { + "bbox": [ + 177, + 620, + 205, + 632 + ], + "score": 0.91, + "content": "S O ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 619, + 321, + 632 + ], + "score": 1.0, + "content": ", and thus for any unit vector", + "type": "text" + }, + { + "bbox": [ + 322, + 622, + 329, + 630 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 619, + 391, + 632 + ], + "score": 1.0, + "content": "independent to", + "type": "text" + }, + { + "bbox": [ + 391, + 620, + 400, + 630 + ], + "score": 0.69, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 619, + 404, + 632 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 404, + 620, + 419, + 630 + ], + "score": 0.73, + "content": "V z", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "is uniform on sphere", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 631, + 309, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 124, + 645 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 632, + 224, + 645 + ], + "score": 0.91, + "content": "\\| V z \\| _ { \\infty } = O ( \\log d / \\sqrt { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 631, + 309, + 645 + ], + "score": 1.0, + "content": ". We wherefore have", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 608, + 505, + 645 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 190, + 650, + 421, + 678 + ], + "lines": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "spans": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "score": 0.94, + "content": "\\left\\| V \\right\\| _ { \\infty } = \\operatorname* { s u p } _ { z } { \\frac { \\left\\| V z \\right\\| _ { \\infty } } { \\left\\| z \\right\\| _ { \\infty } } } = O \\left( { \\frac { \\log d } { \\sqrt { d } } } \\right) { \\frac { \\left\\| z \\right\\| _ { 2 } } { \\left\\| z \\right\\| _ { \\infty } } } = O ( \\log d ) ,", + "type": "interline_equation", + "image_path": "cd7f286c737bf8d4242cb3974f2c3c9b4041f5a47e25fddd0d8b59682b61ae56.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 190, + 650, + 421, + 678 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 683, + 406, + 697 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 406, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 255, + 698 + ], + "score": 1.0, + "content": "Note that this result also implies that", + "type": "text" + }, + { + "bbox": [ + 256, + 683, + 403, + 697 + ], + "score": 0.91, + "content": "\\| \\pmb { y } - X ^ { \\top } \\pmb { \\beta } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 682, + 406, + 698 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 682, + 406, + 698 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 250, + 96 + ], + "score": 1.0, + "content": "Lemma 18. For weight matrices", + "type": "text" + }, + { + "bbox": [ + 251, + 82, + 279, + 93 + ], + "score": 0.87, + "content": "W , W ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 81, + 325, + 96 + ], + "score": 1.0, + "content": "satisfying", + "type": "text" + }, + { + "bbox": [ + 326, + 82, + 431, + 95 + ], + "score": 0.92, + "content": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } \\ = \\ { \\cal O } ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 81, + 466, + 96 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 466, + 82, + 504, + 95 + ], + "score": 0.89, + "content": "f ( X ) \\ =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 94, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 147, + 108 + ], + "score": 0.91, + "content": "\\phi ( X W ) \\mathbf { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 95, + 189, + 108 + ], + "score": 1.0, + "content": "with fixed", + "type": "text" + }, + { + "bbox": [ + 189, + 94, + 302, + 108 + ], + "score": 0.89, + "content": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 95, + 505, + 108 + ], + "score": 1.0, + "content": ", given (A1)-(A3), the gradient of the empirical risk", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 106, + 357, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 236, + 119 + ], + "score": 1.0, + "content": "defined in (11) is Lipschitz w.r.t.", + "type": "text" + }, + { + "bbox": [ + 236, + 108, + 248, + 117 + ], + "score": 0.33, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 106, + 357, + 119 + ], + "score": 1.0, + "content": "in the Frobenius norm, i.e.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 121, + 415, + 150 + ], + "lines": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "spans": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "score": 0.91, + "content": "\\left\\| \\frac { \\partial L ( X ; W ) } { \\partial W } - \\frac { \\partial L ( X ; W ^ { \\prime } ) } { \\partial W } \\right\\| _ { F } \\leq L \\left\\| W - W ^ { \\prime } \\right\\| _ { F } .", + "type": "interline_equation", + "image_path": "49ab82c2bf97ce3826d2691396016681f6da33b1cc29cd818f4ed12d49305ab5.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 171, + 173 + ], + "score": 1.0, + "content": "Proof. Denote", + "type": "text" + }, + { + "bbox": [ + 172, + 159, + 249, + 172 + ], + "score": 0.93, + "content": "\\pmb { y } _ { 1 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 1 } ) \\pmb { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 158, + 269, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 270, + 158, + 347, + 172 + ], + "score": 0.92, + "content": "\\pmb { y } _ { 2 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 2 } ) \\pmb { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 158, + 364, + 173 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 364, + 159, + 396, + 171 + ], + "score": 0.91, + "content": "W _ { 1 } , W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "satisfying the assumption", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 444, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 444, + 184 + ], + "score": 1.0, + "content": "above (which can be seen as a condition on the magnitude of training loss), we have", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 185, + 460, + 323 + ], + "lines": [ + { + "bbox": [ + 149, + 185, + 460, + 323 + ], + "spans": [ + { + "bbox": [ + 149, + 185, + 460, + 323 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad \\displaystyle \\left\\| \\frac { \\partial L ( W _ { 1 } ) } { \\partial W _ { 1 } } - \\frac { \\partial L ( W _ { 2 } ) } { \\partial W _ { 2 } } \\right\\| _ { F } } \\\\ & { = \\displaystyle \\left\\| \\frac { 1 } { n } X \\left[ ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right] - \\frac { 1 } { n } X \\left[ ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right] \\right\\| _ { F } } \\\\ & { \\leq \\displaystyle \\frac { 1 } { n } \\| X \\| _ { 2 } \\left\\| ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\leq \\ O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y _ { 2 } - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right\\| _ { F } } \\\\ & { \\quad + O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y - y _ { 2 } ) a ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) ) \\right\\| _ { F } . } \\end{array}", + "type": "interline_equation", + "image_path": "675e07d8654bb7d135b3468c762772600803e9e009cc73105b701ba83d129d84.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 149, + 185, + 460, + 231.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 149, + 231.0, + 460, + 277.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 149, + 277.0, + 460, + 323.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 276, + 342 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 277, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 277, + 344 + ], + "score": 1.0, + "content": "We upper bound the two terms separately:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 345, + 469, + 413 + ], + "lines": [ + { + "bbox": [ + 119, + 345, + 469, + 413 + ], + "spans": [ + { + "bbox": [ + 119, + 345, + 469, + 413 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\left\\| ( y _ { 2 } - y _ { 1 } ) { \\boldsymbol a } ^ { \\top } \\circ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) \\right\\| _ { 2 } \\overset { ( i ) } { \\leq } \\operatorname* { m a x } \\{ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) _ { i j } \\} \\left\\| y _ { 2 } - y _ { 1 } \\right\\| _ { 2 } \\left\\| { \\boldsymbol a } \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i i ) } { \\leq } O ( 1 ) \\left\\| \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) { \\boldsymbol a } - \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 2 } ) { \\boldsymbol a } \\right\\| _ { F } } \\\\ & { \\qquad \\overset { ( i i i ) } { \\leq } O ( 1 ) \\left\\| \\boldsymbol { X } \\right\\| _ { 2 } \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } , } \\end{array}", + "type": "interline_equation", + "image_path": "822a260f5249dc605d01ded18fe6acd153cea9bb6810ecd8d302d194d47afb06.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 119, + 345, + 469, + 367.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 119, + 367.6666666666667, + 469, + 390.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 119, + 390.33333333333337, + 469, + 413.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 507, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 234, + 430 + ], + "score": 1.0, + "content": "where we applied the inequality", + "type": "text" + }, + { + "bbox": [ + 234, + 416, + 364, + 429 + ], + "score": 0.9, + "content": "\\left\\| A \\circ B \\right\\| _ { F } \\leq \\operatorname* { m a x } \\{ | A _ { i j } | \\} \\left\\| B \\right\\| _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 415, + 453, + 430 + ], + "score": 1.0, + "content": "in (i), boundedness of", + "type": "text" + }, + { + "bbox": [ + 453, + 417, + 462, + 428 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 415, + 506, + 430 + ], + "score": 1.0, + "content": "in (ii) and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 427, + 328, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 167, + 441 + ], + "score": 1.0, + "content": "Lipschitzity of", + "type": "text" + }, + { + "bbox": [ + 168, + 428, + 175, + 439 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 427, + 328, + 441 + ], + "score": 1.0, + "content": "in (iii). Similarly, for the second term", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 441, + 418, + 499 + ], + "lines": [ + { + "bbox": [ + 192, + 441, + 418, + 499 + ], + "spans": [ + { + "bbox": [ + 192, + 441, + 418, + 499 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\qquad \\left\\| ( \\pmb { y } - \\pmb { y } _ { 2 } ) \\pmb { a } ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) ) \\right\\| _ { F } } \\\\ & { \\leq \\operatorname* { m a x } \\{ \\left| a _ { i } \\right| \\} \\left\\| \\pmb { y } - \\pmb { y } _ { 2 } \\right\\| _ { 2 } \\left\\| \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\overset { ( i ) } { \\leq } O ( 1 ) \\left\\| X \\right\\| _ { 2 } \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } , } \\end{array}", + "type": "interline_equation", + "image_path": "25ddc124b105b561a4c9b4d1924e2305540087c765333d036e7014e8397d5684.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 192, + 441, + 418, + 460.3333333333333 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 192, + 460.3333333333333, + 418, + 479.66666666666663 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 192, + 479.66666666666663, + 418, + 498.99999999999994 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 504, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 405, + 514 + ], + "score": 1.0, + "content": "where we used the assumption on the training loss and the Lipscthizity of", + "type": "text" + }, + { + "bbox": [ + 406, + 501, + 415, + 512 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 500, + 505, + 514 + ], + "score": 1.0, + "content": "in (i). Combining the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 512, + 247, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 247, + 523 + ], + "score": 1.0, + "content": "two terms yields the desired result.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 464, + 573 + ], + "score": 1.0, + "content": "Lemma 19. Under assumptions (A1-3) and the non-vanishing initialization, given that", + "type": "text" + }, + { + "bbox": [ + 465, + 560, + 505, + 572 + ], + "score": 0.89, + "content": "\\parallel { \\pmb w } _ { i } ( t ) -", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 570, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 198, + 585 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ( 0 ) \\| _ { 2 } = O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 570, + 229, + 587 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 229, + 574, + 234, + 583 + ], + "score": 0.37, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 570, + 297, + 587 + ], + "score": 1.0, + "content": ", then we have", + "type": "text" + }, + { + "bbox": [ + 298, + 571, + 429, + 585 + ], + "score": 0.91, + "content": "\\| K ( t ) - K ( 0 ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 570, + 506, + 587 + ], + "score": 1.0, + "content": "for some positive", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 107, + 581, + 153, + 599 + ], + "spans": [ + { + "bbox": [ + 107, + 585, + 147, + 596 + ], + "score": 0.9, + "content": "\\epsilon ^ { \\prime } \\in \\Theta ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 581, + 153, + 599 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 604, + 271, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 603, + 272, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 272, + 616 + ], + "score": 1.0, + "content": "Proof. Recall the definition of the NTK:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 122, + 617, + 466, + 651 + ], + "lines": [ + { + "bbox": [ + 122, + 617, + 466, + 651 + ], + "spans": [ + { + "bbox": [ + 122, + 617, + 466, + 651 + ], + "score": 0.94, + "content": "K _ { i j } ( t ) = \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } = { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { j } ) ,", + "type": "interline_equation", + "image_path": "359d9704ed4800e00673e164f155f1f948b69e32172bedc675cbb56736e59bf7.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 122, + 617, + 466, + 628.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 122, + 628.3333333333334, + 466, + 639.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 122, + 639.6666666666667, + 466, + 651.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 234, + 666 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 234, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 234, + 667 + ], + "score": 1.0, + "content": "or equivalently the matrix form", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 668, + 403, + 692 + ], + "lines": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "spans": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "score": 0.94, + "content": "K ( t ) = X ^ { \\top } X \\circ \\frac { 1 } { h } [ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) ] .", + "type": "interline_equation", + "image_path": "f1f4ccd7896642f7f848804fd8d50b3b87ee514c8d4f7696293622181fc7f372.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 695, + 506, + 734 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 506, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 179, + 709 + ], + "score": 1.0, + "content": "At initialization,", + "type": "text" + }, + { + "bbox": [ + 180, + 696, + 245, + 708 + ], + "score": 0.92, + "content": "\\mathbf { { x } } _ { i } ~ \\sim ~ N ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 694, + 268, + 709 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 268, + 696, + 359, + 708 + ], + "score": 0.88, + "content": "{ \\pmb w } _ { k } ( 0 ) ~ \\sim ~ N ( 0 , d ^ { \\epsilon } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 694, + 433, + 709 + ], + "score": 1.0, + "content": ". Thus for fixed", + "type": "text" + }, + { + "bbox": [ + 434, + 698, + 444, + 707 + ], + "score": 0.84, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 694, + 506, + 709 + ], + "score": 1.0, + "content": ", by Gaussian", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 706, + 507, + 721 + ], + "spans": [ + { + "bbox": [ + 104, + 706, + 224, + 721 + ], + "score": 1.0, + "content": "anti-concentration we have", + "type": "text" + }, + { + "bbox": [ + 225, + 708, + 380, + 720 + ], + "score": 0.88, + "content": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } | < \\log d \\ \\leq \\ O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 706, + 425, + 721 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 425, + 709, + 460, + 720 + ], + "score": 0.9, + "content": "\\epsilon _ { 1 } ~ > ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 706, + 507, + 721 + ], + "score": 1.0, + "content": ". In addi-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 172, + 734 + ], + "score": 1.0, + "content": "tion, note that", + "type": "text" + }, + { + "bbox": [ + 172, + 720, + 312, + 733 + ], + "score": 0.91, + "content": "\\lVert \\pmb { w } _ { k } ( t ) - \\pmb { w } _ { k } ( 0 ) \\rVert _ { 2 } ~ = ~ O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 719, + 347, + 734 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 347, + 721, + 353, + 731 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 719, + 437, + 734 + ], + "score": 1.0, + "content": ", and therefore for", + "type": "text" + }, + { + "bbox": [ + 437, + 721, + 461, + 732 + ], + "score": 0.9, + "content": "i , j , k", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + } + ], + "page_idx": 36, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "37", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 529, + 504, + 540 + ], + "lines": [ + { + "bbox": [ + 496, + 531, + 504, + 539 + ], + "spans": [ + { + "bbox": [ + 496, + 531, + 504, + 539 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 250, + 96 + ], + "score": 1.0, + "content": "Lemma 18. For weight matrices", + "type": "text" + }, + { + "bbox": [ + 251, + 82, + 279, + 93 + ], + "score": 0.87, + "content": "W , W ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 81, + 325, + 96 + ], + "score": 1.0, + "content": "satisfying", + "type": "text" + }, + { + "bbox": [ + 326, + 82, + 431, + 95 + ], + "score": 0.92, + "content": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } \\ = \\ { \\cal O } ( { \\sqrt { n } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 81, + 466, + 96 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 466, + 82, + 504, + 95 + ], + "score": 0.89, + "content": "f ( X ) \\ =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 94, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 147, + 108 + ], + "score": 0.91, + "content": "\\phi ( X W ) \\mathbf { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 95, + 189, + 108 + ], + "score": 1.0, + "content": "with fixed", + "type": "text" + }, + { + "bbox": [ + 189, + 94, + 302, + 108 + ], + "score": 0.89, + "content": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 95, + 505, + 108 + ], + "score": 1.0, + "content": ", given (A1)-(A3), the gradient of the empirical risk", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 106, + 357, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 236, + 119 + ], + "score": 1.0, + "content": "defined in (11) is Lipschitz w.r.t.", + "type": "text" + }, + { + "bbox": [ + 236, + 108, + 248, + 117 + ], + "score": 0.33, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 106, + 357, + 119 + ], + "score": 1.0, + "content": "in the Frobenius norm, i.e.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 119 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 121, + 415, + 150 + ], + "lines": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "spans": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "score": 0.91, + "content": "\\left\\| \\frac { \\partial L ( X ; W ) } { \\partial W } - \\frac { \\partial L ( X ; W ^ { \\prime } ) } { \\partial W } \\right\\| _ { F } \\leq L \\left\\| W - W ^ { \\prime } \\right\\| _ { F } .", + "type": "interline_equation", + "image_path": "49ab82c2bf97ce3826d2691396016681f6da33b1cc29cd818f4ed12d49305ab5.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 205, + 121, + 415, + 150 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 171, + 173 + ], + "score": 1.0, + "content": "Proof. Denote", + "type": "text" + }, + { + "bbox": [ + 172, + 159, + 249, + 172 + ], + "score": 0.93, + "content": "\\pmb { y } _ { 1 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 1 } ) \\pmb { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 158, + 269, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 270, + 158, + 347, + 172 + ], + "score": 0.92, + "content": "\\pmb { y } _ { 2 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 2 } ) \\pmb { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 158, + 364, + 173 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 364, + 159, + 396, + 171 + ], + "score": 0.91, + "content": "W _ { 1 } , W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "satisfying the assumption", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 444, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 444, + 184 + ], + "score": 1.0, + "content": "above (which can be seen as a condition on the magnitude of training loss), we have", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 158, + 505, + 184 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 185, + 460, + 323 + ], + "lines": [ + { + "bbox": [ + 149, + 185, + 460, + 323 + ], + "spans": [ + { + "bbox": [ + 149, + 185, + 460, + 323 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\quad \\displaystyle \\left\\| \\frac { \\partial L ( W _ { 1 } ) } { \\partial W _ { 1 } } - \\frac { \\partial L ( W _ { 2 } ) } { \\partial W _ { 2 } } \\right\\| _ { F } } \\\\ & { = \\displaystyle \\left\\| \\frac { 1 } { n } X \\left[ ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right] - \\frac { 1 } { n } X \\left[ ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right] \\right\\| _ { F } } \\\\ & { \\leq \\displaystyle \\frac { 1 } { n } \\| X \\| _ { 2 } \\left\\| ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\leq \\ O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y _ { 2 } - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right\\| _ { F } } \\\\ & { \\quad + O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y - y _ { 2 } ) a ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) ) \\right\\| _ { F } . } \\end{array}", + "type": "interline_equation", + "image_path": "675e07d8654bb7d135b3468c762772600803e9e009cc73105b701ba83d129d84.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 149, + 185, + 460, + 231.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 149, + 231.0, + 460, + 277.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 149, + 277.0, + 460, + 323.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 276, + 342 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 277, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 277, + 344 + ], + "score": 1.0, + "content": "We upper bound the two terms separately:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 329, + 277, + 344 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 345, + 469, + 413 + ], + "lines": [ + { + "bbox": [ + 119, + 345, + 469, + 413 + ], + "spans": [ + { + "bbox": [ + 119, + 345, + 469, + 413 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\left\\| ( y _ { 2 } - y _ { 1 } ) { \\boldsymbol a } ^ { \\top } \\circ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) \\right\\| _ { 2 } \\overset { ( i ) } { \\leq } \\operatorname* { m a x } \\{ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) _ { i j } \\} \\left\\| y _ { 2 } - y _ { 1 } \\right\\| _ { 2 } \\left\\| { \\boldsymbol a } \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i i ) } { \\leq } O ( 1 ) \\left\\| \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) { \\boldsymbol a } - \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 2 } ) { \\boldsymbol a } \\right\\| _ { F } } \\\\ & { \\qquad \\overset { ( i i i ) } { \\leq } O ( 1 ) \\left\\| \\boldsymbol { X } \\right\\| _ { 2 } \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } , } \\end{array}", + "type": "interline_equation", + "image_path": "822a260f5249dc605d01ded18fe6acd153cea9bb6810ecd8d302d194d47afb06.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 119, + 345, + 469, + 367.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 119, + 367.6666666666667, + 469, + 390.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 119, + 390.33333333333337, + 469, + 413.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 507, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 234, + 430 + ], + "score": 1.0, + "content": "where we applied the inequality", + "type": "text" + }, + { + "bbox": [ + 234, + 416, + 364, + 429 + ], + "score": 0.9, + "content": "\\left\\| A \\circ B \\right\\| _ { F } \\leq \\operatorname* { m a x } \\{ | A _ { i j } | \\} \\left\\| B \\right\\| _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 415, + 453, + 430 + ], + "score": 1.0, + "content": "in (i), boundedness of", + "type": "text" + }, + { + "bbox": [ + 453, + 417, + 462, + 428 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 415, + 506, + 430 + ], + "score": 1.0, + "content": "in (ii) and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 427, + 328, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 167, + 441 + ], + "score": 1.0, + "content": "Lipschitzity of", + "type": "text" + }, + { + "bbox": [ + 168, + 428, + 175, + 439 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 427, + 328, + 441 + ], + "score": 1.0, + "content": "in (iii). Similarly, for the second term", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 415, + 506, + 441 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 441, + 418, + 499 + ], + "lines": [ + { + "bbox": [ + 192, + 441, + 418, + 499 + ], + "spans": [ + { + "bbox": [ + 192, + 441, + 418, + 499 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\qquad \\left\\| ( \\pmb { y } - \\pmb { y } _ { 2 } ) \\pmb { a } ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) ) \\right\\| _ { F } } \\\\ & { \\leq \\operatorname* { m a x } \\{ \\left| a _ { i } \\right| \\} \\left\\| \\pmb { y } - \\pmb { y } _ { 2 } \\right\\| _ { 2 } \\left\\| \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\overset { ( i ) } { \\leq } O ( 1 ) \\left\\| X \\right\\| _ { 2 } \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } , } \\end{array}", + "type": "interline_equation", + "image_path": "25ddc124b105b561a4c9b4d1924e2305540087c765333d036e7014e8397d5684.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 192, + 441, + 418, + 460.3333333333333 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 192, + 460.3333333333333, + 418, + 479.66666666666663 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 192, + 479.66666666666663, + 418, + 498.99999999999994 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 504, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 405, + 514 + ], + "score": 1.0, + "content": "where we used the assumption on the training loss and the Lipscthizity of", + "type": "text" + }, + { + "bbox": [ + 406, + 501, + 415, + 512 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 500, + 505, + 514 + ], + "score": 1.0, + "content": "in (i). Combining the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 512, + 247, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 247, + 523 + ], + "score": 1.0, + "content": "two terms yields the desired result.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 500, + 505, + 523 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 464, + 573 + ], + "score": 1.0, + "content": "Lemma 19. Under assumptions (A1-3) and the non-vanishing initialization, given that", + "type": "text" + }, + { + "bbox": [ + 465, + 560, + 505, + 572 + ], + "score": 0.89, + "content": "\\parallel { \\pmb w } _ { i } ( t ) -", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 570, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 198, + 585 + ], + "score": 0.9, + "content": "{ \\pmb w } _ { i } ( 0 ) \\| _ { 2 } = O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 570, + 229, + 587 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 229, + 574, + 234, + 583 + ], + "score": 0.37, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 570, + 297, + 587 + ], + "score": 1.0, + "content": ", then we have", + "type": "text" + }, + { + "bbox": [ + 298, + 571, + 429, + 585 + ], + "score": 0.91, + "content": "\\| K ( t ) - K ( 0 ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 570, + 506, + 587 + ], + "score": 1.0, + "content": "for some positive", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 107, + 581, + 153, + 599 + ], + "spans": [ + { + "bbox": [ + 107, + 585, + 147, + 596 + ], + "score": 0.9, + "content": "\\epsilon ^ { \\prime } \\in \\Theta ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 581, + 153, + 599 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 558, + 506, + 599 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 604, + 271, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 603, + 272, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 272, + 616 + ], + "score": 1.0, + "content": "Proof. Recall the definition of the NTK:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 603, + 272, + 616 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 122, + 617, + 466, + 651 + ], + "lines": [ + { + "bbox": [ + 122, + 617, + 466, + 651 + ], + "spans": [ + { + "bbox": [ + 122, + 617, + 466, + 651 + ], + "score": 0.94, + "content": "K _ { i j } ( t ) = \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } = { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { j } ) ,", + "type": "interline_equation", + "image_path": "359d9704ed4800e00673e164f155f1f948b69e32172bedc675cbb56736e59bf7.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 122, + 617, + 466, + 628.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 122, + 628.3333333333334, + 466, + 639.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 122, + 639.6666666666667, + 466, + 651.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 234, + 666 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 234, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 234, + 667 + ], + "score": 1.0, + "content": "or equivalently the matrix form", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 654, + 234, + 667 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 668, + 403, + 692 + ], + "lines": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "spans": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "score": 0.94, + "content": "K ( t ) = X ^ { \\top } X \\circ \\frac { 1 } { h } [ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) ] .", + "type": "interline_equation", + "image_path": "f1f4ccd7896642f7f848804fd8d50b3b87ee514c8d4f7696293622181fc7f372.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 207, + 668, + 403, + 692 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 695, + 506, + 734 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 506, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 179, + 709 + ], + "score": 1.0, + "content": "At initialization,", + "type": "text" + }, + { + "bbox": [ + 180, + 696, + 245, + 708 + ], + "score": 0.92, + "content": "\\mathbf { { x } } _ { i } ~ \\sim ~ N ( 0 , I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 694, + 268, + 709 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 268, + 696, + 359, + 708 + ], + "score": 0.88, + "content": "{ \\pmb w } _ { k } ( 0 ) ~ \\sim ~ N ( 0 , d ^ { \\epsilon } I _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 694, + 433, + 709 + ], + "score": 1.0, + "content": ". Thus for fixed", + "type": "text" + }, + { + "bbox": [ + 434, + 698, + 444, + 707 + ], + "score": 0.84, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 694, + 506, + 709 + ], + "score": 1.0, + "content": ", by Gaussian", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 706, + 507, + 721 + ], + "spans": [ + { + "bbox": [ + 104, + 706, + 224, + 721 + ], + "score": 1.0, + "content": "anti-concentration we have", + "type": "text" + }, + { + "bbox": [ + 225, + 708, + 380, + 720 + ], + "score": 0.88, + "content": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } | < \\log d \\ \\leq \\ O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 706, + 425, + 721 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 425, + 709, + 460, + 720 + ], + "score": 0.9, + "content": "\\epsilon _ { 1 } ~ > ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 706, + 507, + 721 + ], + "score": 1.0, + "content": ". In addi-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 172, + 734 + ], + "score": 1.0, + "content": "tion, note that", + "type": "text" + }, + { + "bbox": [ + 172, + 720, + 312, + 733 + ], + "score": 0.91, + "content": "\\lVert \\pmb { w } _ { k } ( t ) - \\pmb { w } _ { k } ( 0 ) \\rVert _ { 2 } ~ = ~ O ( d ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 719, + 347, + 734 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 347, + 721, + 353, + 731 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 719, + 437, + 734 + ], + "score": 1.0, + "content": ", and therefore for", + "type": "text" + }, + { + "bbox": [ + 437, + 721, + 461, + 732 + ], + "score": 0.9, + "content": "i , j , k", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 104, + 694, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 505, + 109 + ], + "lines": [ + { + "bbox": [ + 107, + 79, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 81, + 203, + 95 + ], + "score": 0.89, + "content": "| { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) | > O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 79, + 222, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 81, + 318, + 95 + ], + "score": 0.9, + "content": "| x _ { j } ^ { \\top } w _ { k } ( 0 ) | > O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 79, + 380, + 96 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + }, + { + "bbox": [ + 381, + 81, + 505, + 96 + ], + "score": 0.92, + "content": "| \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } ( t ) ) \\phi ^ { \\prime } ( \\pmb { x } _ { j } ^ { \\top } \\pmb { w } _ { k } ( t ) ) -", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 271, + 110 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 267, + 110 + ], + "score": 0.9, + "content": "\\phi ^ { \\prime } ( { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) \\phi ^ { \\prime } ( { \\pmb x } _ { j } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) | = = O ( d ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 94, + 271, + 109 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 114, + 506, + 150 + ], + "lines": [ + { + "bbox": [ + 105, + 114, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 155, + 128 + ], + "score": 1.0, + "content": "Given fixed", + "type": "text" + }, + { + "bbox": [ + 155, + 117, + 166, + 126 + ], + "score": 0.85, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 114, + 197, + 128 + ], + "score": 1.0, + "content": ", define", + "type": "text" + }, + { + "bbox": [ + 198, + 114, + 303, + 127 + ], + "score": 0.92, + "content": "y _ { k } = \\mathbf { 1 } \\{ | x _ { i } ^ { \\top } w _ { k } | < \\log d \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 114, + 434, + 128 + ], + "score": 1.0, + "content": "as the indicator variable that the", + "type": "text" + }, + { + "bbox": [ + 435, + 116, + 441, + 125 + ], + "score": 0.84, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 114, + 506, + 128 + ], + "score": 1.0, + "content": "-th neuron does", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 124, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 214, + 141 + ], + "score": 1.0, + "content": "not saturate. We know that", + "type": "text" + }, + { + "bbox": [ + 214, + 127, + 306, + 140 + ], + "score": 0.91, + "content": "\\mathbb { E } [ y _ { k } ] = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 124, + 327, + 141 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 327, + 127, + 502, + 140 + ], + "score": 0.91, + "content": "\\mathrm { V a r } [ y _ { k } ] = \\mathbb { E } [ y _ { k } ^ { 2 } ] - \\mathbb { E } [ y _ { k } ] ^ { 2 } = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 124, + 507, + 141 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 138, + 210, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 210, + 152 + ], + "score": 1.0, + "content": "By Bernstein’s inequality", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 155, + 417, + 190 + ], + "lines": [ + { + "bbox": [ + 194, + 155, + 417, + 190 + ], + "spans": [ + { + "bbox": [ + 194, + 155, + 417, + 190 + ], + "score": 0.93, + "content": "\\operatorname* { P r } \\left| \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } y _ { k } - \\mathbb { E } [ y _ { k } ] \\right| > \\varepsilon \\leq 2 \\exp \\left( - \\frac { h \\varepsilon ^ { 2 } } { 2 \\sigma ^ { 2 } + 2 \\varepsilon / 3 } \\right) .", + "type": "interline_equation", + "image_path": "558150d5c3bf874f26be75b207fa7c9b3fb7025e9859b3afd27effc941de806c.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 194, + 155, + 417, + 172.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 194, + 172.5, + 417, + 190.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 196, + 384, + 215 + ], + "lines": [ + { + "bbox": [ + 105, + 196, + 385, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 166, + 214 + ], + "score": 1.0, + "content": "Setting ε = q", + "type": "text" + }, + { + "bbox": [ + 137, + 196, + 190, + 216 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\varepsilon = \\sqrt { \\frac { c \\log h } { h ^ { 1 + \\epsilon _ { 2 } } } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 196, + 347, + 216 + ], + "score": 1.0, + "content": ", we know that with probability at least", + "type": "text" + }, + { + "bbox": [ + 347, + 200, + 380, + 211 + ], + "score": 0.91, + "content": "1 - h ^ { - c }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 196, + 385, + 216 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 221, + 394, + 255 + ], + "lines": [ + { + "bbox": [ + 218, + 221, + 394, + 255 + ], + "spans": [ + { + "bbox": [ + 218, + 221, + 394, + 255 + ], + "score": 0.93, + "content": "{ \\frac { 1 } { h } } \\sum _ { k = 1 } ^ { h } y _ { k } \\leq \\varepsilon + \\mathbb { E } [ y _ { k } ] = O \\left( { \\frac { \\mathrm { p o l y l o g } h } { h ^ { 1 / 2 + \\epsilon _ { 3 } } } } \\right) .", + "type": "interline_equation", + "image_path": "b0f2e0ceafc2dbb7e687deebb7c3d3751653f0ad472c4afce719624c23246db7.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 221, + 394, + 238.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 218, + 238.0, + 394, + 255.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 467, + 274 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 468, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 174, + 275 + ], + "score": 1.0, + "content": "Therefore, given", + "type": "text" + }, + { + "bbox": [ + 175, + 264, + 186, + 273 + ], + "score": 0.85, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 261, + 204, + 275 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 204, + 263, + 216, + 274 + ], + "score": 0.86, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { j } }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 261, + 288, + 275 + ], + "score": 1.0, + "content": ", for large enough", + "type": "text" + }, + { + "bbox": [ + 288, + 263, + 298, + 273 + ], + "score": 0.83, + "content": "c _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 261, + 396, + 275 + ], + "score": 1.0, + "content": "with probability at least", + "type": "text" + }, + { + "bbox": [ + 396, + 261, + 430, + 272 + ], + "score": 0.94, + "content": "1 - h ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 261, + 468, + 275 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 279, + 470, + 314 + ], + "lines": [ + { + "bbox": [ + 117, + 279, + 470, + 314 + ], + "spans": [ + { + "bbox": [ + 117, + 279, + 470, + 314 + ], + "score": 0.94, + "content": "\\left| \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { j } ) - \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { j } ) \\right| = O ( h ^ { 1 / 2 - \\epsilon _ { 4 } } ) ,", + "type": "interline_equation", + "image_path": "4cfd4b7c8dbfa482781d8cb7d8385fcf6097f75a9a74645fbe5cc3bbd1c56379.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 117, + 279, + 470, + 290.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 117, + 290.6666666666667, + 470, + 302.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 117, + 302.33333333333337, + 470, + 314.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 319, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 274, + 333 + ], + "score": 1.0, + "content": "in which we utilized the boundedness of", + "type": "text" + }, + { + "bbox": [ + 275, + 320, + 284, + 331 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 318, + 394, + 333 + ], + "score": 1.0, + "content": ". Taking union bound over", + "type": "text" + }, + { + "bbox": [ + 395, + 319, + 405, + 330 + ], + "score": 0.87, + "content": "d ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 318, + 505, + 333 + ], + "score": 1.0, + "content": "elements in the random", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 330, + 192, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 192, + 343 + ], + "score": 1.0, + "content": "feature matrix yields", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 346, + 478, + 439 + ], + "lines": [ + { + "bbox": [ + 133, + 346, + 478, + 439 + ], + "spans": [ + { + "bbox": [ + 133, + 346, + 478, + 439 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { ~ \\| K ( t ) - K ( 0 ) \\| _ { 2 } } \\\\ & { = \\left\\| X ^ { \\top } X \\circ \\frac { 1 } { h } \\left[ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right] \\right\\| } \\\\ & { \\leq \\frac { 1 } { h } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\operatorname* { m a x } \\left\\{ \\left| \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right| _ { i j } \\right\\} } \\\\ & { \\leq \\frac { 1 } { h } O ( d ) O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "4a7ba69b32b73ae7335de1fd33f0fff428bed75d62d2e5bb24831a8a512d0fa8.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 133, + 346, + 478, + 377.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 133, + 377.0, + 478, + 408.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 133, + 408.0, + 478, + 439.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 321, + 464 + ], + "score": 1.0, + "content": "Using the exact same argument, one can derive that", + "type": "text" + }, + { + "bbox": [ + 321, + 449, + 486, + 464 + ], + "score": 0.91, + "content": "\\| { \\pmb u } _ { N N } ( { \\hat { \\pmb x } } ) - { \\pmb u } _ { N T K } ( { \\hat { \\pmb x } } ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 449, + 506, + 464 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 462, + 206, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 206, + 475 + ], + "score": 1.0, + "content": "proof of which we omit.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 513, + 244, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 245, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 245, + 528 + ], + "score": 1.0, + "content": "E ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 357, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 537, + 358, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 358, + 551 + ], + "score": 1.0, + "content": "E.1 RISK OF ReLU NETWORK UNDER SYMMETRIC DATA", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 558, + 243, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 244, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 244, + 571 + ], + "score": 1.0, + "content": "If the dataset is symmetric, that is", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 105, + 594, + 498, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 500, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 500, + 608 + ], + "score": 1.0, + "content": "then population risk of the gradient flow solution can be given explicitly for certain nonlinearities:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 503, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 356, + 623 + ], + "score": 1.0, + "content": "Proposition 20. Given (A1-3)(A5), if the nonlinearity satisfies", + "type": "text" + }, + { + "bbox": [ + 356, + 609, + 444, + 622 + ], + "score": 0.93, + "content": "\\phi ^ { \\prime } ( { \\pmb x } ) + \\phi ^ { \\prime } ( - { \\pmb x } ) = C", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 608, + 496, + 623 + ], + "score": 1.0, + "content": "for constant", + "type": "text" + }, + { + "bbox": [ + 496, + 611, + 503, + 619 + ], + "score": 0.72, + "content": "C", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 619, + 190, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 137, + 633 + ], + "score": 1.0, + "content": "then as", + "type": "text" + }, + { + "bbox": [ + 138, + 621, + 190, + 632 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 637, + 453, + 664 + ], + "lines": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "spans": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "score": 0.94, + "content": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } ( \\hat { f } ) \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } ( \\hat { f } ) = ( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } ) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "73b5af9535895247f602303123b3033fbae9c99e5e9d7bd02edc05404f67e048.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 676, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Note that the requirement on the nonlinearity holds for ReLU and SoftPlus. This expression is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 192, + 700 + ], + "score": 1.0, + "content": "again independent to", + "type": "text" + }, + { + "bbox": [ + 193, + 690, + 204, + 699 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "and aligns with the experimental results in Figure 9 (we only plot the bias", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 473, + 712 + ], + "score": 1.0, + "content": "component for verification). In addition, the bias is upper-bounded by the null risk for all", + "type": "text" + }, + { + "bbox": [ + 474, + 700, + 484, + 711 + ], + "score": 0.83, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 698, + 506, + 712 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "remark that the symmetry assumption does not hold for i.i.d. samples from symmetric distributions,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 720, + 387, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 387, + 732 + ], + "score": 1.0, + "content": "and Figure 9 demonstrates that the additional condition alters the risk.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "page_idx": 37, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "38", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 479, + 505, + 490 + ], + "lines": [ + { + "bbox": [ + 495, + 480, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 495, + 480, + 505, + 491 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 505, + 109 + ], + "lines": [ + { + "bbox": [ + 107, + 79, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 81, + 203, + 95 + ], + "score": 0.89, + "content": "| { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) | > O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 79, + 222, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 81, + 318, + 95 + ], + "score": 0.9, + "content": "| x _ { j } ^ { \\top } w _ { k } ( 0 ) | > O ( \\log d )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 79, + 380, + 96 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + }, + { + "bbox": [ + 381, + 81, + 505, + 96 + ], + "score": 0.92, + "content": "| \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } ( t ) ) \\phi ^ { \\prime } ( \\pmb { x } _ { j } ^ { \\top } \\pmb { w } _ { k } ( t ) ) -", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 271, + 110 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 267, + 110 + ], + "score": 0.9, + "content": "\\phi ^ { \\prime } ( { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) \\phi ^ { \\prime } ( { \\pmb x } _ { j } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) | = = O ( d ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 94, + 271, + 109 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 79, + 505, + 110 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 114, + 506, + 150 + ], + "lines": [ + { + "bbox": [ + 105, + 114, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 155, + 128 + ], + "score": 1.0, + "content": "Given fixed", + "type": "text" + }, + { + "bbox": [ + 155, + 117, + 166, + 126 + ], + "score": 0.85, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 114, + 197, + 128 + ], + "score": 1.0, + "content": ", define", + "type": "text" + }, + { + "bbox": [ + 198, + 114, + 303, + 127 + ], + "score": 0.92, + "content": "y _ { k } = \\mathbf { 1 } \\{ | x _ { i } ^ { \\top } w _ { k } | < \\log d \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 114, + 434, + 128 + ], + "score": 1.0, + "content": "as the indicator variable that the", + "type": "text" + }, + { + "bbox": [ + 435, + 116, + 441, + 125 + ], + "score": 0.84, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 114, + 506, + 128 + ], + "score": 1.0, + "content": "-th neuron does", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 124, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 214, + 141 + ], + "score": 1.0, + "content": "not saturate. We know that", + "type": "text" + }, + { + "bbox": [ + 214, + 127, + 306, + 140 + ], + "score": 0.91, + "content": "\\mathbb { E } [ y _ { k } ] = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 124, + 327, + 141 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 327, + 127, + 502, + 140 + ], + "score": 0.91, + "content": "\\mathrm { V a r } [ y _ { k } ] = \\mathbb { E } [ y _ { k } ^ { 2 } ] - \\mathbb { E } [ y _ { k } ] ^ { 2 } = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 124, + 507, + 141 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 138, + 210, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 210, + 152 + ], + "score": 1.0, + "content": "By Bernstein’s inequality", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 104, + 114, + 507, + 152 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 155, + 417, + 190 + ], + "lines": [ + { + "bbox": [ + 194, + 155, + 417, + 190 + ], + "spans": [ + { + "bbox": [ + 194, + 155, + 417, + 190 + ], + "score": 0.93, + "content": "\\operatorname* { P r } \\left| \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } y _ { k } - \\mathbb { E } [ y _ { k } ] \\right| > \\varepsilon \\leq 2 \\exp \\left( - \\frac { h \\varepsilon ^ { 2 } } { 2 \\sigma ^ { 2 } + 2 \\varepsilon / 3 } \\right) .", + "type": "interline_equation", + "image_path": "558150d5c3bf874f26be75b207fa7c9b3fb7025e9859b3afd27effc941de806c.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 194, + 155, + 417, + 172.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 194, + 172.5, + 417, + 190.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 196, + 384, + 215 + ], + "lines": [ + { + "bbox": [ + 105, + 196, + 385, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 166, + 214 + ], + "score": 1.0, + "content": "Setting ε = q", + "type": "text" + }, + { + "bbox": [ + 137, + 196, + 190, + 216 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\varepsilon = \\sqrt { \\frac { c \\log h } { h ^ { 1 + \\epsilon _ { 2 } } } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 196, + 347, + 216 + ], + "score": 1.0, + "content": ", we know that with probability at least", + "type": "text" + }, + { + "bbox": [ + 347, + 200, + 380, + 211 + ], + "score": 0.91, + "content": "1 - h ^ { - c }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 196, + 385, + 216 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 196, + 385, + 216 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 221, + 394, + 255 + ], + "lines": [ + { + "bbox": [ + 218, + 221, + 394, + 255 + ], + "spans": [ + { + "bbox": [ + 218, + 221, + 394, + 255 + ], + "score": 0.93, + "content": "{ \\frac { 1 } { h } } \\sum _ { k = 1 } ^ { h } y _ { k } \\leq \\varepsilon + \\mathbb { E } [ y _ { k } ] = O \\left( { \\frac { \\mathrm { p o l y l o g } h } { h ^ { 1 / 2 + \\epsilon _ { 3 } } } } \\right) .", + "type": "interline_equation", + "image_path": "b0f2e0ceafc2dbb7e687deebb7c3d3751653f0ad472c4afce719624c23246db7.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 221, + 394, + 238.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 218, + 238.0, + 394, + 255.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 467, + 274 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 468, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 174, + 275 + ], + "score": 1.0, + "content": "Therefore, given", + "type": "text" + }, + { + "bbox": [ + 175, + 264, + 186, + 273 + ], + "score": 0.85, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 261, + 204, + 275 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 204, + 263, + 216, + 274 + ], + "score": 0.86, + "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { j } }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 261, + 288, + 275 + ], + "score": 1.0, + "content": ", for large enough", + "type": "text" + }, + { + "bbox": [ + 288, + 263, + 298, + 273 + ], + "score": 0.83, + "content": "c _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 261, + 396, + 275 + ], + "score": 1.0, + "content": "with probability at least", + "type": "text" + }, + { + "bbox": [ + 396, + 261, + 430, + 272 + ], + "score": 0.94, + "content": "1 - h ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 261, + 468, + 275 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 261, + 468, + 275 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 279, + 470, + 314 + ], + "lines": [ + { + "bbox": [ + 117, + 279, + 470, + 314 + ], + "spans": [ + { + "bbox": [ + 117, + 279, + 470, + 314 + ], + "score": 0.94, + "content": "\\left| \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { j } ) - \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { j } ) \\right| = O ( h ^ { 1 / 2 - \\epsilon _ { 4 } } ) ,", + "type": "interline_equation", + "image_path": "4cfd4b7c8dbfa482781d8cb7d8385fcf6097f75a9a74645fbe5cc3bbd1c56379.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 117, + 279, + 470, + 290.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 117, + 290.6666666666667, + 470, + 302.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 117, + 302.33333333333337, + 470, + 314.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 319, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 274, + 333 + ], + "score": 1.0, + "content": "in which we utilized the boundedness of", + "type": "text" + }, + { + "bbox": [ + 275, + 320, + 284, + 331 + ], + "score": 0.88, + "content": "\\phi ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 318, + 394, + 333 + ], + "score": 1.0, + "content": ". Taking union bound over", + "type": "text" + }, + { + "bbox": [ + 395, + 319, + 405, + 330 + ], + "score": 0.87, + "content": "d ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 318, + 505, + 333 + ], + "score": 1.0, + "content": "elements in the random", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 330, + 192, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 192, + 343 + ], + "score": 1.0, + "content": "feature matrix yields", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 318, + 505, + 343 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 346, + 478, + 439 + ], + "lines": [ + { + "bbox": [ + 133, + 346, + 478, + 439 + ], + "spans": [ + { + "bbox": [ + 133, + 346, + 478, + 439 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { ~ \\| K ( t ) - K ( 0 ) \\| _ { 2 } } \\\\ & { = \\left\\| X ^ { \\top } X \\circ \\frac { 1 } { h } \\left[ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right] \\right\\| } \\\\ & { \\leq \\frac { 1 } { h } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\operatorname* { m a x } \\left\\{ \\left| \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right| _ { i j } \\right\\} } \\\\ & { \\leq \\frac { 1 } { h } O ( d ) O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "4a7ba69b32b73ae7335de1fd33f0fff428bed75d62d2e5bb24831a8a512d0fa8.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 133, + 346, + 478, + 377.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 133, + 377.0, + 478, + 408.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 133, + 408.0, + 478, + 439.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 321, + 464 + ], + "score": 1.0, + "content": "Using the exact same argument, one can derive that", + "type": "text" + }, + { + "bbox": [ + 321, + 449, + 486, + 464 + ], + "score": 0.91, + "content": "\\| { \\pmb u } _ { N N } ( { \\hat { \\pmb x } } ) - { \\pmb u } _ { N T K } ( { \\hat { \\pmb x } } ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 449, + 506, + 464 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 462, + 206, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 206, + 475 + ], + "score": 1.0, + "content": "proof of which we omit.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 449, + 506, + 475 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 513, + 244, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 245, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 245, + 528 + ], + "score": 1.0, + "content": "E ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 357, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 537, + 358, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 358, + 551 + ], + "score": 1.0, + "content": "E.1 RISK OF ReLU NETWORK UNDER SYMMETRIC DATA", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 537, + 358, + 551 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 558, + 243, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 244, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 244, + 571 + ], + "score": 1.0, + "content": "If the dataset is symmetric, that is", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 558, + 244, + 571 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 594, + 498, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 500, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 500, + 608 + ], + "score": 1.0, + "content": "then population risk of the gradient flow solution can be given explicitly for certain nonlinearities:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 594, + 500, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 503, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 356, + 623 + ], + "score": 1.0, + "content": "Proposition 20. Given (A1-3)(A5), if the nonlinearity satisfies", + "type": "text" + }, + { + "bbox": [ + 356, + 609, + 444, + 622 + ], + "score": 0.93, + "content": "\\phi ^ { \\prime } ( { \\pmb x } ) + \\phi ^ { \\prime } ( - { \\pmb x } ) = C", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 608, + 496, + 623 + ], + "score": 1.0, + "content": "for constant", + "type": "text" + }, + { + "bbox": [ + 496, + 611, + 503, + 619 + ], + "score": 0.72, + "content": "C", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 619, + 190, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 137, + 633 + ], + "score": 1.0, + "content": "then as", + "type": "text" + }, + { + "bbox": [ + 138, + 621, + 190, + 632 + ], + "score": 0.91, + "content": "n , d , h \\infty", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 608, + 503, + 633 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 637, + 453, + 664 + ], + "lines": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "spans": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "score": 0.94, + "content": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } ( \\hat { f } ) \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } ( \\hat { f } ) = ( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } ) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "73b5af9535895247f602303123b3033fbae9c99e5e9d7bd02edc05404f67e048.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 137, + 637, + 453, + 664 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 676, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Note that the requirement on the nonlinearity holds for ReLU and SoftPlus. This expression is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 192, + 700 + ], + "score": 1.0, + "content": "again independent to", + "type": "text" + }, + { + "bbox": [ + 193, + 690, + 204, + 699 + ], + "score": 0.86, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "and aligns with the experimental results in Figure 9 (we only plot the bias", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 473, + 712 + ], + "score": 1.0, + "content": "component for verification). In addition, the bias is upper-bounded by the null risk for all", + "type": "text" + }, + { + "bbox": [ + 474, + 700, + 484, + 711 + ], + "score": 0.83, + "content": "\\gamma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 698, + 506, + 712 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "remark that the symmetry assumption does not hold for i.i.d. samples from symmetric distributions,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 720, + 387, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 387, + 732 + ], + "score": 1.0, + "content": "and Figure 9 demonstrates that the additional condition alters the risk.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 677, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 80, + 492, + 223 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 80, + 492, + 223 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 80, + 492, + 223 + ], + "spans": [ + { + "bbox": [ + 117, + 80, + 492, + 223 + ], + "score": 0.969, + "type": "image", + "image_path": "21277368c4f422787876a9038b601c0d6dee6f086e04a2580c02aae5b0b456d2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 80, + 492, + 127.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 127.66666666666666, + 492, + 175.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 175.33333333333331, + 492, + 222.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 232, + 506, + 283 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 232, + 507, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 507, + 243 + ], + "score": 1.0, + "content": "Figure 9: Bias of two-layer ReLU networks with optimized first layer under Gaussian data and linear teacher.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 271, + 254 + ], + "score": 1.0, + "content": "Individual dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 271, + 244, + 281, + 253 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 241, + 506, + 254 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. (a) Vanishing", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 251, + 507, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 507, + 264 + ], + "score": 1.0, + "content": "initialization. The bias under symmetric data is predicted by Proposition 20. (b) Non-vanishing initialization.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "score": 1.0, + "content": "The red and blue lines represent models optimized from i.i.d. and symmetric initialization, respectively. The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 272, + 322, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 322, + 284 + ], + "score": 1.0, + "content": "bias for symmetric initialization is predicted by Theorem 8.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 441, + 315 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 442, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 276, + 316 + ], + "score": 1.0, + "content": "Proof. Without loss of generality assume", + "type": "text" + }, + { + "bbox": [ + 276, + 302, + 342, + 315 + ], + "score": 0.93, + "content": "X = [ X _ { 0 } , - X _ { 0 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 300, + 442, + 316 + ], + "score": 1.0, + "content": ". Then by (100) we have", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 165, + 317, + 445, + 351 + ], + "lines": [ + { + "bbox": [ + 165, + 317, + 445, + 351 + ], + "spans": [ + { + "bbox": [ + 165, + 317, + 445, + 351 + ], + "score": 0.93, + "content": "\\frac { \\partial \\pmb { w } _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) + h _ { 0 } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) \\pmb { x } _ { i } \\right] ,", + "type": "interline_equation", + "image_path": "33ac8cdaea85e29ecbb901d809d84e70a09bb3e446d75ca742d8bece5bc87125.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 165, + 317, + 445, + 328.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 165, + 328.3333333333333, + 445, + 339.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 165, + 339.66666666666663, + 445, + 350.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 505, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 352, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 178, + 366 + ], + "score": 1.0, + "content": "and the flow for", + "type": "text" + }, + { + "bbox": [ + 178, + 355, + 195, + 365 + ], + "score": 0.86, + "content": "{ \\pmb w } _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 352, + 505, + 366 + ], + "score": 1.0, + "content": "follows from symmetry. In this case one can show that from exact zero", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 363, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 268, + 378 + ], + "score": 1.0, + "content": "initialization, for nonlinearity satisfying", + "type": "text" + }, + { + "bbox": [ + 268, + 364, + 347, + 376 + ], + "score": 0.93, + "content": "\\phi ( x ) - \\phi ( - x ) = x", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 363, + 469, + 378 + ], + "score": 1.0, + "content": ", such as ReLU and SoftPlus,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 379, + 514, + 414 + ], + "lines": [ + { + "bbox": [ + 111, + 379, + 514, + 414 + ], + "spans": [ + { + "bbox": [ + 111, + 379, + 514, + 414 + ], + "score": 0.91, + "content": "\\frac { \\partial ( { \\pmb w } _ { + } ) } { \\partial t } + \\frac { \\partial ( { \\pmb w } _ { - } ) } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) + h _ { 0 } \\phi ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) \\Big ) ( \\phi ^ { \\prime } ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) - \\phi ^ { \\prime } ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) ) { \\pmb x } _ { i } \\right] = 0", + "type": "interline_equation", + "image_path": "7c767067d6d62bb4c5d252f691456852e01d21455de7b6eac1d89c8d69a2256b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 111, + 379, + 514, + 390.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 390.6666666666667, + 514, + 402.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 111, + 402.33333333333337, + 514, + 414.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 427, + 271, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 426, + 272, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 245, + 441 + ], + "score": 1.0, + "content": "And therefore the gradient flow of", + "type": "text" + }, + { + "bbox": [ + 245, + 429, + 261, + 439 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 426, + 272, + 441 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 442, + 482, + 543 + ], + "lines": [ + { + "bbox": [ + 129, + 442, + 482, + 543 + ], + "spans": [ + { + "bbox": [ + 129, + 442, + 482, + 543 + ], + "score": 0.96, + "content": "\\begin{array} { l } { \\displaystyle \\frac { \\partial w _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\big ( \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + \\phi ^ { \\prime } \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } w _ { + } ^ { \\top } x _ { i } \\Big ) x _ { i } \\Big ] = \\frac { 1 } { 2 n _ { 0 } } X _ { 0 } y _ { 0 } - \\frac { 1 } { 2 n _ { 0 } } h _ { 0 } X _ { 0 } X _ { 0 } ^ { \\top } w _ { + } . } \\end{array}", + "type": "interline_equation", + "image_path": "b7c8baedb48e72bc23843067a4fb0641ddca959153ac7b35a4dc5f49c656340d.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 129, + 442, + 482, + 475.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 129, + 475.6666666666667, + 482, + 509.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 129, + 509.33333333333337, + 482, + 543.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 504, + 567 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 156, + 558 + ], + "score": 1.0, + "content": "The flow of", + "type": "text" + }, + { + "bbox": [ + 157, + 546, + 173, + 556 + ], + "score": 0.85, + "content": "{ \\pmb w } _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "follows from symmetry. Solving for the stationary points (i.e. gradient becomes", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 555, + 195, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 195, + 568 + ], + "score": 1.0, + "content": "zero), it the clear that", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 568, + 422, + 630 + ], + "lines": [ + { + "bbox": [ + 189, + 568, + 422, + 630 + ], + "spans": [ + { + "bbox": [ + 189, + 568, + 422, + 630 + ], + "score": 0.95, + "content": "\\pmb { w } _ { + } ^ { ( t = \\infty ) } = - \\pmb { w } _ { - } ^ { ( t = \\infty ) } = \\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { 1 } { h _ { 0 } } ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } , } & { \\gamma _ { 1 } < 0 . 5 , } \\\\ { \\displaystyle } \\\\ { \\displaystyle \\frac { 1 } { h _ { 0 } } X ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } , } & { \\gamma _ { 1 } > 0 . 5 . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "e60181587ccd6e28ebe9de98be2cab5499ff2872bea90a945f533c3a143331e7.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 568, + 422, + 583.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 189, + 583.5, + 422, + 599.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 189, + 599.0, + 422, + 614.5 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 189, + 614.5, + 422, + 630.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 631, + 239, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 239, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 239, + 644 + ], + "score": 1.0, + "content": "And hence the asymptotic risk is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 645, + 450, + 673 + ], + "lines": [ + { + "bbox": [ + 161, + 645, + 450, + 673 + ], + "spans": [ + { + "bbox": [ + 161, + 645, + 450, + 673 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } \\to \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } = \\left( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } \\right) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "a984cf11779b8fa79f7dec70b5480784869009784eaab7ffb7690d4dc8b6472f.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 161, + 645, + 450, + 654.3333333333334 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 161, + 654.3333333333334, + 450, + 663.6666666666667 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 161, + 663.6666666666667, + 450, + 673.0000000000001 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 680, + 505, + 726 + ], + "lines": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "The same conclusion holds for vanishing initialization if we assume that the trajectory stays close to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 704 + ], + "score": 1.0, + "content": "that of exact zero initialization. Note that although the prediction aligns well with the experimental", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "score": 1.0, + "content": "results, the argument in Theorem 7 does not directly apply due to the undefined derivative of ReLU", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 714, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 714, + 342, + 726 + ], + "score": 1.0, + "content": "at the origin, and thus this result is not rigorously justified.", + "type": "text" + }, + { + "bbox": [ + 494, + 714, + 505, + 725 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + } + ], + "page_idx": 38, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "39", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 80, + 492, + 223 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 80, + 492, + 223 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 80, + 492, + 223 + ], + "spans": [ + { + "bbox": [ + 117, + 80, + 492, + 223 + ], + "score": 0.969, + "type": "image", + "image_path": "21277368c4f422787876a9038b601c0d6dee6f086e04a2580c02aae5b0b456d2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 80, + 492, + 127.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 127.66666666666666, + 492, + 175.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 175.33333333333331, + 492, + 222.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 232, + 506, + 283 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 232, + 507, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 507, + 243 + ], + "score": 1.0, + "content": "Figure 9: Bias of two-layer ReLU networks with optimized first layer under Gaussian data and linear teacher.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 271, + 254 + ], + "score": 1.0, + "content": "Individual dotted lines correspond to different", + "type": "text" + }, + { + "bbox": [ + 271, + 244, + 281, + 253 + ], + "score": 0.84, + "content": "\\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 241, + 506, + 254 + ], + "score": 1.0, + "content": "(from 0.2 to 2) which is independent to the risk. (a) Vanishing", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 251, + 507, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 507, + 264 + ], + "score": 1.0, + "content": "initialization. The bias under symmetric data is predicted by Proposition 20. (b) Non-vanishing initialization.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "score": 1.0, + "content": "The red and blue lines represent models optimized from i.i.d. and symmetric initialization, respectively. The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 272, + 322, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 322, + 284 + ], + "score": 1.0, + "content": "bias for symmetric initialization is predicted by Theorem 8.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 441, + 315 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 442, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 276, + 316 + ], + "score": 1.0, + "content": "Proof. Without loss of generality assume", + "type": "text" + }, + { + "bbox": [ + 276, + 302, + 342, + 315 + ], + "score": 0.93, + "content": "X = [ X _ { 0 } , - X _ { 0 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 300, + 442, + 316 + ], + "score": 1.0, + "content": ". Then by (100) we have", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 300, + 442, + 316 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 165, + 317, + 445, + 351 + ], + "lines": [ + { + "bbox": [ + 165, + 317, + 445, + 351 + ], + "spans": [ + { + "bbox": [ + 165, + 317, + 445, + 351 + ], + "score": 0.93, + "content": "\\frac { \\partial \\pmb { w } _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) + h _ { 0 } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) \\pmb { x } _ { i } \\right] ,", + "type": "interline_equation", + "image_path": "33ac8cdaea85e29ecbb901d809d84e70a09bb3e446d75ca742d8bece5bc87125.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 165, + 317, + 445, + 328.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 165, + 328.3333333333333, + 445, + 339.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 165, + 339.66666666666663, + 445, + 350.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 505, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 352, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 178, + 366 + ], + "score": 1.0, + "content": "and the flow for", + "type": "text" + }, + { + "bbox": [ + 178, + 355, + 195, + 365 + ], + "score": 0.86, + "content": "{ \\pmb w } _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 352, + 505, + 366 + ], + "score": 1.0, + "content": "follows from symmetry. In this case one can show that from exact zero", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 363, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 268, + 378 + ], + "score": 1.0, + "content": "initialization, for nonlinearity satisfying", + "type": "text" + }, + { + "bbox": [ + 268, + 364, + 347, + 376 + ], + "score": 0.93, + "content": "\\phi ( x ) - \\phi ( - x ) = x", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 363, + 469, + 378 + ], + "score": 1.0, + "content": ", such as ReLU and SoftPlus,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 352, + 505, + 378 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 379, + 514, + 414 + ], + "lines": [ + { + "bbox": [ + 111, + 379, + 514, + 414 + ], + "spans": [ + { + "bbox": [ + 111, + 379, + 514, + 414 + ], + "score": 0.91, + "content": "\\frac { \\partial ( { \\pmb w } _ { + } ) } { \\partial t } + \\frac { \\partial ( { \\pmb w } _ { - } ) } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) + h _ { 0 } \\phi ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) \\Big ) ( \\phi ^ { \\prime } ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) - \\phi ^ { \\prime } ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) ) { \\pmb x } _ { i } \\right] = 0", + "type": "interline_equation", + "image_path": "7c767067d6d62bb4c5d252f691456852e01d21455de7b6eac1d89c8d69a2256b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 111, + 379, + 514, + 390.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 390.6666666666667, + 514, + 402.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 111, + 402.33333333333337, + 514, + 414.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 427, + 271, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 426, + 272, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 245, + 441 + ], + "score": 1.0, + "content": "And therefore the gradient flow of", + "type": "text" + }, + { + "bbox": [ + 245, + 429, + 261, + 439 + ], + "score": 0.87, + "content": "{ \\pmb w } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 426, + 272, + 441 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 426, + 272, + 441 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 442, + 482, + 543 + ], + "lines": [ + { + "bbox": [ + 129, + 442, + 482, + 543 + ], + "spans": [ + { + "bbox": [ + 129, + 442, + 482, + 543 + ], + "score": 0.96, + "content": "\\begin{array} { l } { \\displaystyle \\frac { \\partial w _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\big ( \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + \\phi ^ { \\prime } \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } w _ { + } ^ { \\top } x _ { i } \\Big ) x _ { i } \\Big ] = \\frac { 1 } { 2 n _ { 0 } } X _ { 0 } y _ { 0 } - \\frac { 1 } { 2 n _ { 0 } } h _ { 0 } X _ { 0 } X _ { 0 } ^ { \\top } w _ { + } . } \\end{array}", + "type": "interline_equation", + "image_path": "b7c8baedb48e72bc23843067a4fb0641ddca959153ac7b35a4dc5f49c656340d.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 129, + 442, + 482, + 475.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 129, + 475.6666666666667, + 482, + 509.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 129, + 509.33333333333337, + 482, + 543.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 504, + 567 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 156, + 558 + ], + "score": 1.0, + "content": "The flow of", + "type": "text" + }, + { + "bbox": [ + 157, + 546, + 173, + 556 + ], + "score": 0.85, + "content": "{ \\pmb w } _ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "follows from symmetry. Solving for the stationary points (i.e. gradient becomes", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 555, + 195, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 195, + 568 + ], + "score": 1.0, + "content": "zero), it the clear that", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 543, + 506, + 568 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 568, + 422, + 630 + ], + "lines": [ + { + "bbox": [ + 189, + 568, + 422, + 630 + ], + "spans": [ + { + "bbox": [ + 189, + 568, + 422, + 630 + ], + "score": 0.95, + "content": "\\pmb { w } _ { + } ^ { ( t = \\infty ) } = - \\pmb { w } _ { - } ^ { ( t = \\infty ) } = \\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { 1 } { h _ { 0 } } ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } , } & { \\gamma _ { 1 } < 0 . 5 , } \\\\ { \\displaystyle } \\\\ { \\displaystyle \\frac { 1 } { h _ { 0 } } X ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } , } & { \\gamma _ { 1 } > 0 . 5 . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "e60181587ccd6e28ebe9de98be2cab5499ff2872bea90a945f533c3a143331e7.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 568, + 422, + 583.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 189, + 583.5, + 422, + 599.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 189, + 599.0, + 422, + 614.5 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 189, + 614.5, + 422, + 630.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 631, + 239, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 239, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 239, + 644 + ], + "score": 1.0, + "content": "And hence the asymptotic risk is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 631, + 239, + 644 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 645, + 450, + 673 + ], + "lines": [ + { + "bbox": [ + 161, + 645, + 450, + 673 + ], + "spans": [ + { + "bbox": [ + 161, + 645, + 450, + 673 + ], + "score": 0.92, + "content": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } \\to \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } = \\left( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } \\right) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } .", + "type": "interline_equation", + "image_path": "a984cf11779b8fa79f7dec70b5480784869009784eaab7ffb7690d4dc8b6472f.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 161, + 645, + 450, + 654.3333333333334 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 161, + 654.3333333333334, + 450, + 663.6666666666667 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 161, + 663.6666666666667, + 450, + 673.0000000000001 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 680, + 505, + 726 + ], + "lines": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "The same conclusion holds for vanishing initialization if we assume that the trajectory stays close to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 704 + ], + "score": 1.0, + "content": "that of exact zero initialization. Note that although the prediction aligns well with the experimental", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "score": 1.0, + "content": "results, the argument in Theorem 7 does not directly apply due to the undefined derivative of ReLU", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 714, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 714, + 342, + 726 + ], + "score": 1.0, + "content": "at the origin, and thus this result is not rigorously justified.", + "type": "text" + }, + { + "bbox": [ + 494, + 714, + 505, + 725 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 680, + 506, + 726 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 232, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 232, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 232, + 95 + ], + "score": 1.0, + "content": "F EXPERIMENT SETUP", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 152 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 119 + ], + "score": 1.0, + "content": "Optimizing the Second Layer. We compute the minimum-norm solution by directly solving", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 222, + 130 + ], + "score": 1.0, + "content": "the pseudo-inverse. We set", + "type": "text" + }, + { + "bbox": [ + 223, + 118, + 267, + 127 + ], + "score": 0.87, + "content": "n = 1 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 117, + 307, + 130 + ], + "score": 1.0, + "content": "and vary", + "type": "text" + }, + { + "bbox": [ + 308, + 119, + 333, + 129 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 117, + 505, + 130 + ], + "score": 1.0, + "content": "from 0.1 to 3. The linear teacher model", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 107, + 128, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 107, + 128, + 164, + 142 + ], + "score": 0.92, + "content": "F ( { \\pmb x } ) = { \\pmb x } ^ { \\top } \\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 128, + 208, + 143 + ], + "score": 1.0, + "content": "is fixed as", + "type": "text" + }, + { + "bbox": [ + 209, + 128, + 268, + 142 + ], + "score": 0.93, + "content": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 128, + 311, + 143 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 311, + 129, + 343, + 142 + ], + "score": 0.93, + "content": "( \\gamma _ { 1 } , \\gamma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 128, + 507, + 143 + ], + "score": 1.0, + "content": "we average across 50 random draws of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 129, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 129, + 153 + ], + "score": 1.0, + "content": "data.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 212 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "Optimizing the First Layer. For both initializations, we use gradient descent with small step size", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 109, + 175, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 109, + 177, + 144, + 189 + ], + "score": 0.85, + "content": "( \\eta = 0 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 176, + 366, + 190 + ], + "score": 1.0, + "content": ") and train the model for minimally 25000 steps and till", + "type": "text" + }, + { + "bbox": [ + 366, + 175, + 472, + 189 + ], + "score": 0.92, + "content": "\\| \\nabla _ { W } f ( X , W ) \\| _ { F } ^ { 2 } < 1 0 ^ { - 6 }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 176, + 506, + 190 + ], + "score": 1.0, + "content": ". We fix", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 189, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 142, + 201 + ], + "score": 0.88, + "content": "n = 3 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 189, + 179, + 202 + ], + "score": 1.0, + "content": "and vary", + "type": "text" + }, + { + "bbox": [ + 180, + 192, + 204, + 202 + ], + "score": 0.81, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 189, + 398, + 202 + ], + "score": 1.0, + "content": "from 0.1 to 3 with the same linear teacher model", + "type": "text" + }, + { + "bbox": [ + 399, + 189, + 457, + 202 + ], + "score": 0.94, + "content": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 189, + 505, + 202 + ], + "score": 1.0, + "content": ". The risk is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 201, + 364, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 364, + 213 + ], + "score": 1.0, + "content": "averaged across 20 models trained from different initializations.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + } + ], + "page_idx": 39, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "40", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 232, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 232, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 232, + 95 + ], + "score": 1.0, + "content": "F EXPERIMENT SETUP", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 152 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 119 + ], + "score": 1.0, + "content": "Optimizing the Second Layer. We compute the minimum-norm solution by directly solving", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 222, + 130 + ], + "score": 1.0, + "content": "the pseudo-inverse. We set", + "type": "text" + }, + { + "bbox": [ + 223, + 118, + 267, + 127 + ], + "score": 0.87, + "content": "n = 1 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 117, + 307, + 130 + ], + "score": 1.0, + "content": "and vary", + "type": "text" + }, + { + "bbox": [ + 308, + 119, + 333, + 129 + ], + "score": 0.89, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 117, + 505, + 130 + ], + "score": 1.0, + "content": "from 0.1 to 3. The linear teacher model", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 107, + 128, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 107, + 128, + 164, + 142 + ], + "score": 0.92, + "content": "F ( { \\pmb x } ) = { \\pmb x } ^ { \\top } \\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 128, + 208, + 143 + ], + "score": 1.0, + "content": "is fixed as", + "type": "text" + }, + { + "bbox": [ + 209, + 128, + 268, + 142 + ], + "score": 0.93, + "content": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 128, + 311, + 143 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 311, + 129, + 343, + 142 + ], + "score": 0.93, + "content": "( \\gamma _ { 1 } , \\gamma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 128, + 507, + 143 + ], + "score": 1.0, + "content": "we average across 50 random draws of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 129, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 129, + 153 + ], + "score": 1.0, + "content": "data.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 105, + 507, + 153 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 212 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "Optimizing the First Layer. For both initializations, we use gradient descent with small step size", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 109, + 175, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 109, + 177, + 144, + 189 + ], + "score": 0.85, + "content": "( \\eta = 0 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 176, + 366, + 190 + ], + "score": 1.0, + "content": ") and train the model for minimally 25000 steps and till", + "type": "text" + }, + { + "bbox": [ + 366, + 175, + 472, + 189 + ], + "score": 0.92, + "content": "\\| \\nabla _ { W } f ( X , W ) \\| _ { F } ^ { 2 } < 1 0 ^ { - 6 }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 176, + 506, + 190 + ], + "score": 1.0, + "content": ". We fix", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 189, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 142, + 201 + ], + "score": 0.88, + "content": "n = 3 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 189, + 179, + 202 + ], + "score": 1.0, + "content": "and vary", + "type": "text" + }, + { + "bbox": [ + 180, + 192, + 204, + 202 + ], + "score": 0.81, + "content": "\\gamma _ { 1 } , \\gamma _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 189, + 398, + 202 + ], + "score": 1.0, + "content": "from 0.1 to 3 with the same linear teacher model", + "type": "text" + }, + { + "bbox": [ + 399, + 189, + 457, + 202 + ], + "score": 0.94, + "content": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 189, + 505, + 202 + ], + "score": 1.0, + "content": ". The risk is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 201, + 364, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 364, + 213 + ], + "score": 1.0, + "content": "averaged across 20 models trained from different initializations.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 163, + 506, + 213 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/H1gBsgBYwH/H1gBsgBYwH_model.json b/parse/train/H1gBsgBYwH/H1gBsgBYwH_model.json new file mode 100644 index 0000000000000000000000000000000000000000..1eb68af2cfb4ac0a2d8c0a43b5320eae03422af3 --- /dev/null +++ b/parse/train/H1gBsgBYwH/H1gBsgBYwH_model.json @@ -0,0 +1,60008 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 645, + 1302, + 645, + 1302, + 1042, + 398, + 1042 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1391, + 1404, + 1391, + 1404, + 1666, + 298, + 1666 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1161, + 1404, + 1161, + 1404, + 1375, + 298, + 1375 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1681, + 1403, + 1681, + 1403, + 1926, + 298, + 1926 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 300, + 1943, + 1403, + 1943, + 1403, + 2034, + 300, + 2034 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 299, + 219, + 1409, + 219, + 1409, + 324, + 299, + 324 + ], + "score": 0.959 + }, + { + "category_id": 0, + "poly": [ + 302, + 1094, + 573, + 1094, + 573, + 1128, + 302, + 1128 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 104, + 299, + 104 + ], + "score": 0.882 + }, + { + "category_id": 0, + "poly": [ + 773, + 582, + 926, + 582, + 926, + 615, + 773, + 615 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 856, + 2088, + 856, + 2112, + 841, + 2112 + ], + "score": 0.723 + }, + { + "category_id": 1, + "poly": [ + 309, + 370, + 1423, + 370, + 1423, + 502, + 309, + 502 + ], + "score": 0.683 + }, + { + "category_id": 13, + "poly": [ + 574, + 860, + 657, + 860, + 657, + 888, + 574, + 888 + ], + "score": 0.9, + "latex": "h \\approx n" + }, + { + "category_id": 13, + "poly": [ + 408, + 374, + 469, + 374, + 469, + 406, + 408, + 406 + ], + "score": 0.85, + "latex": "\\mathbf { B a } ^ { 1 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 390, + 1804, + 438, + 1804, + 438, + 1835, + 390, + 1835 + ], + "score": 0.84, + "latex": "n , d ." + }, + { + "category_id": 13, + "poly": [ + 1008, + 373, + 1092, + 373, + 1092, + 406, + 1008, + 406 + ], + "score": 0.84, + "latex": "\\mathbf { W _ { u } } 1 , 2 , 4" + }, + { + "category_id": 13, + "poly": [ + 397, + 707, + 416, + 707, + 416, + 734, + 397, + 734 + ], + "score": 0.77, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1132, + 678, + 1150, + 678, + 1150, + 704, + 1132, + 704 + ], + "score": 0.74, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 1310, + 2009, + 1329, + 2009, + 1329, + 2031, + 1310, + 2031 + ], + "score": 0.69, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1004, + 682, + 1024, + 682, + 1024, + 704, + 1004, + 704 + ], + "score": 0.66, + "latex": "n" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 219.0, + 1412.0, + 219.0, + 1412.0, + 271.0, + 297.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 274.0, + 1082.0, + 274.0, + 1082.0, + 328.0, + 292.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1090.0, + 579.0, + 1090.0, + 579.0, + 1136.0, + 294.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 581.0, + 932.0, + 581.0, + 932.0, + 618.0, + 769.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 857.0, + 2088.0, + 857.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 647.0, + 1303.0, + 647.0, + 1303.0, + 679.0, + 396.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 676.0, + 1003.0, + 676.0, + 1003.0, + 711.0, + 394.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 676.0, + 1131.0, + 676.0, + 1131.0, + 711.0, + 1025.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 676.0, + 1306.0, + 676.0, + 1306.0, + 711.0, + 1151.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 707.0, + 396.0, + 707.0, + 396.0, + 742.0, + 393.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 707.0, + 1305.0, + 707.0, + 1305.0, + 742.0, + 417.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 738.0, + 1306.0, + 738.0, + 1306.0, + 773.0, + 393.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 767.0, + 1305.0, + 767.0, + 1305.0, + 802.0, + 393.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 800.0, + 1305.0, + 800.0, + 1305.0, + 832.0, + 395.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 829.0, + 1306.0, + 829.0, + 1306.0, + 864.0, + 394.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 860.0, + 573.0, + 860.0, + 573.0, + 892.0, + 393.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 860.0, + 1305.0, + 860.0, + 1305.0, + 892.0, + 658.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 891.0, + 1306.0, + 891.0, + 1306.0, + 923.0, + 394.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 919.0, + 1305.0, + 919.0, + 1305.0, + 953.0, + 394.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 951.0, + 1304.0, + 951.0, + 1304.0, + 983.0, + 394.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 982.0, + 1305.0, + 982.0, + 1305.0, + 1014.0, + 394.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1012.0, + 1243.0, + 1012.0, + 1243.0, + 1047.0, + 394.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1389.0, + 1408.0, + 1389.0, + 1408.0, + 1424.0, + 295.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 1406.0, + 1418.0, + 1406.0, + 1454.0, + 293.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1452.0, + 1405.0, + 1452.0, + 1405.0, + 1484.0, + 297.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1483.0, + 1408.0, + 1483.0, + 1408.0, + 1515.0, + 294.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1513.0, + 1406.0, + 1513.0, + 1406.0, + 1545.0, + 296.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1543.0, + 1405.0, + 1543.0, + 1405.0, + 1575.0, + 296.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1573.0, + 1405.0, + 1573.0, + 1405.0, + 1605.0, + 296.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1603.0, + 1405.0, + 1603.0, + 1405.0, + 1639.0, + 294.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1633.0, + 926.0, + 1633.0, + 926.0, + 1669.0, + 295.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1156.0, + 1405.0, + 1156.0, + 1405.0, + 1198.0, + 293.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1194.0, + 1404.0, + 1194.0, + 1404.0, + 1225.0, + 296.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1223.0, + 1404.0, + 1223.0, + 1404.0, + 1257.0, + 294.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1255.0, + 1404.0, + 1255.0, + 1404.0, + 1285.0, + 296.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1283.0, + 1406.0, + 1283.0, + 1406.0, + 1317.0, + 293.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1310.0, + 1409.0, + 1310.0, + 1409.0, + 1350.0, + 292.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1344.0, + 840.0, + 1344.0, + 840.0, + 1377.0, + 296.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1681.0, + 1404.0, + 1681.0, + 1404.0, + 1715.0, + 296.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1714.0, + 1405.0, + 1714.0, + 1405.0, + 1745.0, + 293.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1741.0, + 1407.0, + 1741.0, + 1407.0, + 1776.0, + 293.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 1407.0, + 1774.0, + 1407.0, + 1807.0, + 294.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 389.0, + 1802.0, + 389.0, + 1836.0, + 294.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1802.0, + 1404.0, + 1802.0, + 1404.0, + 1836.0, + 439.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1833.0, + 1407.0, + 1833.0, + 1407.0, + 1871.0, + 292.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1865.0, + 1405.0, + 1865.0, + 1405.0, + 1899.0, + 293.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 814.0, + 1896.0, + 814.0, + 1929.0, + 294.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1408.0, + 1942.0, + 1408.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1407.0, + 1972.0, + 1407.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1309.0, + 2004.0, + 1309.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 2004.0, + 1404.0, + 2004.0, + 1404.0, + 2037.0, + 1330.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 371.0, + 407.0, + 371.0, + 407.0, + 409.0, + 316.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 371.0, + 1007.0, + 371.0, + 1007.0, + 409.0, + 470.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 371.0, + 1331.0, + 371.0, + 1331.0, + 409.0, + 1093.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 406.0, + 1422.0, + 406.0, + 1422.0, + 443.0, + 311.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 441.0, + 924.0, + 441.0, + 924.0, + 473.0, + 315.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 467.0, + 1215.0, + 467.0, + 1215.0, + 506.0, + 310.0, + 506.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1769, + 1403, + 1769, + 1403, + 2044, + 298, + 2044 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1408, + 1404, + 1408, + 1404, + 1746, + 298, + 1746 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1080, + 1405, + 1080, + 1405, + 1386, + 297, + 1386 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1405, + 229, + 1405, + 474, + 298, + 474 + ], + "score": 0.979 + }, + { + "category_id": 3, + "poly": [ + 1026, + 493, + 1397, + 493, + 1397, + 771, + 1026, + 771 + ], + "score": 0.969 + }, + { + "category_id": 4, + "poly": [ + 1022, + 790, + 1403, + 790, + 1403, + 1014, + 1022, + 1014 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 301, + 1021, + 577, + 1021, + 577, + 1052, + 301, + 1052 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 816, + 76, + 816, + 104, + 300, + 104 + ], + "score": 0.897 + }, + { + "category_id": 1, + "poly": [ + 298, + 489, + 948, + 489, + 948, + 522, + 298, + 522 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 369, + 550, + 1003, + 550, + 1003, + 979, + 369, + 979 + ], + "score": 0.818 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.703 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.23 + }, + { + "category_id": 13, + "poly": [ + 1190, + 874, + 1303, + 874, + 1303, + 905, + 1190, + 905 + ], + "score": 0.92, + "latex": "\\gamma _ { 1 } = d / \\bar { n }" + }, + { + "category_id": 13, + "poly": [ + 859, + 1174, + 950, + 1174, + 950, + 1201, + 859, + 1201 + ], + "score": 0.91, + "latex": "h \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 503, + 323, + 577, + 323, + 577, + 353, + 503, + 353 + ], + "score": 0.91, + "latex": "n , d , h" + }, + { + "category_id": 13, + "poly": [ + 692, + 261, + 742, + 261, + 742, + 295, + 692, + 295 + ], + "score": 0.91, + "latex": "h / n" + }, + { + "category_id": 13, + "poly": [ + 1300, + 1172, + 1346, + 1172, + 1346, + 1205, + 1300, + 1205 + ], + "score": 0.89, + "latex": "1 / h" + }, + { + "category_id": 13, + "poly": [ + 533, + 295, + 623, + 295, + 623, + 320, + 533, + 320 + ], + "score": 0.88, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 569, + 232, + 589, + 232, + 589, + 258, + 569, + 258 + ], + "score": 0.8, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1127, + 848, + 1238, + 848, + 1238, + 874, + 1127, + 874 + ], + "score": 0.79, + "latex": "( \\mathrm { S N R } = 1 6 " + }, + { + "category_id": 13, + "poly": [ + 395, + 232, + 413, + 232, + 413, + 259, + 395, + 259 + ], + "score": 0.76, + "latex": "d" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 487.0, + 1075.0, + 487.0, + 1075.0, + 514.0, + 1040.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 534.0, + 1073.0, + 534.0, + 1073.0, + 561.0, + 1040.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 536.0, + 1259.0, + 536.0, + 1259.0, + 554.0, + 1234.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 581.0, + 1074.0, + 581.0, + 1074.0, + 638.0, + 1023.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 581.0, + 1142.0, + 581.0, + 1142.0, + 616.0, + 1128.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 596.0, + 1259.0, + 596.0, + 1259.0, + 610.0, + 1235.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 610.0, + 1303.0, + 610.0, + 1303.0, + 618.0, + 1295.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 606.0, + 1350.0, + 606.0, + 1350.0, + 620.0, + 1339.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 629.0, + 1073.0, + 629.0, + 1073.0, + 656.0, + 1040.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 677.0, + 1073.0, + 677.0, + 1073.0, + 700.0, + 1038.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 681.0, + 1198.0, + 681.0, + 1198.0, + 707.0, + 1119.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 703.0, + 1204.0, + 703.0, + 1204.0, + 727.0, + 1118.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 723.0, + 1074.0, + 723.0, + 1074.0, + 748.0, + 1040.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 732.0, + 1126.0, + 732.0, + 1126.0, + 758.0, + 1091.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 733.0, + 1188.0, + 733.0, + 1188.0, + 758.0, + 1152.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 732.0, + 1249.0, + 732.0, + 1249.0, + 758.0, + 1212.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 733.0, + 1310.0, + 733.0, + 1310.0, + 758.0, + 1273.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 732.0, + 1370.0, + 732.0, + 1370.0, + 758.0, + 1334.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 751.0, + 1266.0, + 751.0, + 1266.0, + 773.0, + 1198.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 636.5, + 1157.0, + 636.5, + 1157.0, + 663.5, + 1054.0, + 663.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 791.0, + 1404.0, + 791.0, + 1404.0, + 822.0, + 1022.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 820.0, + 1403.0, + 820.0, + 1403.0, + 848.0, + 1023.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 847.0, + 1126.0, + 847.0, + 1126.0, + 876.0, + 1022.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 847.0, + 1404.0, + 847.0, + 1404.0, + 876.0, + 1239.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 874.0, + 1189.0, + 874.0, + 1189.0, + 905.0, + 1020.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 874.0, + 1405.0, + 874.0, + 1405.0, + 905.0, + 1304.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 903.0, + 1404.0, + 903.0, + 1404.0, + 931.0, + 1022.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 930.0, + 1403.0, + 930.0, + 1403.0, + 959.0, + 1023.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 958.0, + 1405.0, + 958.0, + 1405.0, + 989.0, + 1022.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 984.0, + 1371.0, + 984.0, + 1371.0, + 1019.0, + 1020.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1018.0, + 581.0, + 1018.0, + 581.0, + 1057.0, + 296.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1767.0, + 1404.0, + 1767.0, + 1404.0, + 1802.0, + 296.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1796.0, + 1405.0, + 1796.0, + 1405.0, + 1834.0, + 294.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1831.0, + 1404.0, + 1831.0, + 1404.0, + 1864.0, + 294.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1858.0, + 1407.0, + 1858.0, + 1407.0, + 1895.0, + 294.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1887.0, + 1405.0, + 1887.0, + 1405.0, + 1925.0, + 294.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1922.0, + 1404.0, + 1922.0, + 1404.0, + 1955.0, + 296.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1952.0, + 1404.0, + 1952.0, + 1404.0, + 1985.0, + 296.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1979.0, + 1407.0, + 1979.0, + 1407.0, + 2017.0, + 293.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2012.0, + 913.0, + 2012.0, + 913.0, + 2048.0, + 294.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1408.0, + 1405.0, + 1408.0, + 1405.0, + 1443.0, + 296.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1438.0, + 1404.0, + 1438.0, + 1404.0, + 1472.0, + 296.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1469.0, + 1405.0, + 1469.0, + 1405.0, + 1504.0, + 294.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1502.0, + 1404.0, + 1502.0, + 1404.0, + 1533.0, + 296.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1531.0, + 1405.0, + 1531.0, + 1405.0, + 1565.0, + 293.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1562.0, + 1405.0, + 1562.0, + 1405.0, + 1593.0, + 296.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1591.0, + 1408.0, + 1591.0, + 1408.0, + 1627.0, + 293.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1623.0, + 1401.0, + 1623.0, + 1401.0, + 1654.0, + 296.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1650.0, + 1405.0, + 1650.0, + 1405.0, + 1688.0, + 295.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1683.0, + 1405.0, + 1683.0, + 1405.0, + 1718.0, + 296.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1713.0, + 1323.0, + 1713.0, + 1323.0, + 1750.0, + 293.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1078.0, + 1406.0, + 1078.0, + 1406.0, + 1114.0, + 296.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1110.0, + 1407.0, + 1110.0, + 1407.0, + 1143.0, + 292.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1142.0, + 1403.0, + 1142.0, + 1403.0, + 1173.0, + 296.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1173.0, + 858.0, + 1173.0, + 858.0, + 1205.0, + 296.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1173.0, + 1299.0, + 1173.0, + 1299.0, + 1205.0, + 951.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1173.0, + 1403.0, + 1173.0, + 1403.0, + 1205.0, + 1347.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1200.0, + 1405.0, + 1200.0, + 1405.0, + 1237.0, + 294.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1233.0, + 1407.0, + 1233.0, + 1407.0, + 1266.0, + 294.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1263.0, + 1405.0, + 1263.0, + 1405.0, + 1295.0, + 296.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1293.0, + 1406.0, + 1293.0, + 1406.0, + 1328.0, + 294.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1357.0, + 296.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1355.0, + 1299.0, + 1355.0, + 1299.0, + 1388.0, + 292.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 231.0, + 394.0, + 231.0, + 394.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 231.0, + 568.0, + 231.0, + 568.0, + 264.0, + 414.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 231.0, + 1407.0, + 231.0, + 1407.0, + 264.0, + 590.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 258.0, + 691.0, + 258.0, + 691.0, + 299.0, + 291.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 258.0, + 1406.0, + 258.0, + 1406.0, + 299.0, + 743.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 290.0, + 532.0, + 290.0, + 532.0, + 325.0, + 293.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 290.0, + 1407.0, + 290.0, + 1407.0, + 325.0, + 624.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 322.0, + 502.0, + 322.0, + 502.0, + 355.0, + 295.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 322.0, + 1405.0, + 322.0, + 1405.0, + 355.0, + 578.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 353.0, + 1405.0, + 353.0, + 1405.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 381.0, + 1403.0, + 381.0, + 1403.0, + 416.0, + 292.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 413.0, + 1403.0, + 413.0, + 1403.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 442.0, + 755.0, + 442.0, + 755.0, + 476.0, + 295.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 486.0, + 952.0, + 486.0, + 952.0, + 529.0, + 294.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 550.0, + 1001.0, + 550.0, + 1001.0, + 586.0, + 372.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 581.0, + 1001.0, + 581.0, + 1001.0, + 614.0, + 394.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 611.0, + 941.0, + 611.0, + 941.0, + 646.0, + 394.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 655.0, + 1007.0, + 655.0, + 1007.0, + 693.0, + 379.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 690.0, + 1005.0, + 690.0, + 1005.0, + 720.0, + 395.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 719.0, + 1001.0, + 719.0, + 1001.0, + 752.0, + 394.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 750.0, + 1001.0, + 750.0, + 1001.0, + 782.0, + 394.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 780.0, + 888.0, + 780.0, + 888.0, + 812.0, + 394.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 825.0, + 1002.0, + 825.0, + 1002.0, + 857.0, + 372.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 855.0, + 1003.0, + 855.0, + 1003.0, + 890.0, + 394.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 886.0, + 1005.0, + 886.0, + 1005.0, + 918.0, + 395.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 917.0, + 1006.0, + 917.0, + 1006.0, + 947.0, + 395.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 949.0, + 978.0, + 949.0, + 978.0, + 978.0, + 394.0, + 978.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 229, + 1405, + 229, + 1405, + 627, + 298, + 627 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1147, + 1406, + 1147, + 1406, + 1365, + 297, + 1365 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 643, + 1406, + 643, + 1406, + 890, + 297, + 890 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1671, + 1404, + 1671, + 1404, + 1794, + 297, + 1794 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1849, + 1404, + 1849, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.979 + }, + { + "category_id": 8, + "poly": [ + 701, + 1049, + 998, + 1049, + 998, + 1137, + 701, + 1137 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 293, + 1549, + 1400, + 1549, + 1400, + 1616, + 293, + 1616 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 295, + 1378, + 1401, + 1378, + 1401, + 1443, + 295, + 1443 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 519, + 1625, + 1178, + 1625, + 1178, + 1664, + 519, + 1664 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 395, + 1455, + 1299, + 1455, + 1299, + 1540, + 395, + 1540 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 493, + 1803, + 1203, + 1803, + 1203, + 1841, + 493, + 1841 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 816, + 76, + 816, + 104, + 299, + 104 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1075, + 1400, + 1075, + 1400, + 1106, + 1366, + 1106 + ], + "score": 0.888 + }, + { + "category_id": 0, + "poly": [ + 298, + 932, + 1064, + 932, + 1064, + 970, + 298, + 970 + ], + "score": 0.868 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1808, + 1400, + 1808, + 1400, + 1837, + 1366, + 1837 + ], + "score": 0.868 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1479, + 1400, + 1479, + 1400, + 1509, + 1366, + 1509 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 297, + 999, + 1315, + 999, + 1315, + 1034, + 297, + 1034 + ], + "score": 0.804 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.676 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.353 + }, + { + "category_id": 14, + "poly": [ + 700, + 1043, + 997, + 1043, + 997, + 1138, + 700, + 1138 + ], + "score": 0.95, + "latex": "f ( \\pmb { x } ) = \\sum _ { i = 1 } ^ { h } a _ { i } \\phi ( \\langle \\pmb { x } , \\pmb { w } _ { i } \\rangle ) ," + }, + { + "category_id": 13, + "poly": [ + 950, + 999, + 1087, + 999, + 1087, + 1034, + 950, + 1034 + ], + "score": 0.92, + "latex": "f : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }" + }, + { + "category_id": 13, + "poly": [ + 612, + 1148, + 718, + 1148, + 718, + 1181, + 612, + 1181 + ], + "score": 0.92, + "latex": "\\boldsymbol { w } _ { i } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 14, + "poly": [ + 398, + 1450, + 1298, + 1450, + 1298, + 1542, + 398, + 1542 + ], + "score": 0.92, + "latex": "( { \\pmb x } _ { i } , \\varepsilon _ { i } ) \\overset { \\mathrm { i . i . d . } } { \\sim } P _ { { \\pmb x } } \\times P _ { \\varepsilon } , \\quad y _ { i } = F ( { \\pmb x } _ { i } ) + \\varepsilon _ { i } , \\quad L ( { \\boldsymbol X } ; f ) = \\frac { 1 } { 2 n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - f ( { \\pmb x } _ { i } ) \\right) ^ { 2 } ," + }, + { + "category_id": 13, + "poly": [ + 373, + 1148, + 465, + 1148, + 465, + 1179, + 373, + 1179 + ], + "score": 0.91, + "latex": "\\pmb { x } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1377, + 1345, + 1377, + 1345, + 1407, + 1203, + 1407 + ], + "score": 0.91, + "latex": "F : { \\mathbb { R } ^ { d } } \\to { \\mathbb { R } }" + }, + { + "category_id": 13, + "poly": [ + 882, + 1274, + 970, + 1274, + 970, + 1304, + 882, + 1304 + ], + "score": 0.91, + "latex": "\\ b { y } \\in \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1152, + 1283, + 1152, + 1283, + 1181, + 1199, + 1181 + ], + "score": 0.91, + "latex": "a _ { i } \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 958, + 1244, + 1191, + 1244, + 1191, + 1275, + 958, + 1275 + ], + "score": 0.9, + "latex": "\\pmb { a } = [ a _ { 1 } , . . . a _ { h } ] \\in \\mathbb { R } ^ { h }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1733, + 383, + 1733, + 383, + 1765, + 296, + 1765 + ], + "score": 0.9, + "latex": "\\gamma _ { 2 } > 1" + }, + { + "category_id": 13, + "poly": [ + 348, + 1304, + 616, + 1304, + 616, + 1336, + 348, + 1336 + ], + "score": 0.9, + "latex": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }" + }, + { + "category_id": 13, + "poly": [ + 703, + 1182, + 832, + 1182, + 832, + 1210, + 703, + 1210 + ], + "score": 0.9, + "latex": "\\phi : \\mathbb { R } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1005, + 1211, + 1154, + 1211, + 1154, + 1243, + 1005, + 1243 + ], + "score": 0.89, + "latex": "G \\sim \\mathcal { N } ( 0 , 1 )" + }, + { + "category_id": 14, + "poly": [ + 518, + 1624, + 1177, + 1624, + 1177, + 1663, + 518, + 1663 + ], + "score": 0.89, + "latex": "n , d , h \\infty ; \\quad d / n \\gamma _ { 1 } , h / n \\gamma _ { 2 } ; \\quad \\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty ) ," + }, + { + "category_id": 13, + "poly": [ + 402, + 1240, + 697, + 1240, + 697, + 1272, + 402, + 1272 + ], + "score": 0.89, + "latex": "W = [ { \\pmb w } _ { 1 } , . . . { \\pmb w } _ { h } ] \\in \\mathbb { R } ^ { d \\times h }" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1703, + 1128, + 1703, + 1128, + 1734, + 1012, + 1734 + ], + "score": 0.89, + "latex": "\\gamma _ { 2 } 1 ; 2" + }, + { + "category_id": 14, + "poly": [ + 497, + 1803, + 1201, + 1803, + 1201, + 1841, + 497, + 1841 + ], + "score": 0.88, + "latex": "\\mathrm { d } W ( t ) = - \\nabla _ { W } L ( X ; f ) \\mathrm { d } t \\quad \\mathrm { o r } \\quad \\mathrm { d } a ( t ) = - \\nabla _ { a } L ( X ; f ) \\mathrm { d } t ," + }, + { + "category_id": 13, + "poly": [ + 297, + 1584, + 644, + 1584, + 644, + 1618, + 297, + 1618 + ], + "score": 0.87, + "latex": "R ( f ) = \\mathbb { E } _ { P _ { x } } [ ( F ( { \\pmb x } ) - f ( { \\pmb x } ) ) ^ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 787, + 1550, + 942, + 1550, + 942, + 1586, + 787, + 1586 + ], + "score": 0.87, + "latex": "\\mathrm { V a r } ( \\varepsilon _ { i } ) = \\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 379, + 1273, + 656, + 1273, + 656, + 1302, + 379, + 1302 + ], + "score": 0.87, + "latex": "X = [ { \\pmb x } _ { 1 } , . . . { \\pmb x } _ { n } ] \\in \\mathbb { R } ^ { d \\times n }" + }, + { + "category_id": 13, + "poly": [ + 372, + 1552, + 487, + 1552, + 487, + 1585, + 372, + 1585 + ], + "score": 0.85, + "latex": "\\mathbb { E } [ { \\pmb x } _ { i } ] = 0" + }, + { + "category_id": 13, + "poly": [ + 1314, + 1304, + 1332, + 1304, + 1332, + 1335, + 1314, + 1335 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 856, + 1212, + 960, + 1212, + 960, + 1244, + 856, + 1244 + ], + "score": 0.84, + "latex": "\\forall k \\in \\mathbb { Z } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 957, + 1677, + 987, + 1677, + 987, + 1704, + 957, + 1704 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 498, + 1552, + 655, + 1552, + 655, + 1584, + 498, + 1584 + ], + "score": 0.83, + "latex": "\\mathrm { C o v } ( { \\pmb x } _ { i } ) = \\Sigma" + }, + { + "category_id": 13, + "poly": [ + 668, + 1212, + 842, + 1212, + 842, + 1245, + 668, + 1245 + ], + "score": 0.83, + "latex": "\\mathbb { E } [ \\phi ( G ) ^ { k } ] < \\infty" + }, + { + "category_id": 13, + "poly": [ + 1147, + 1003, + 1167, + 1003, + 1167, + 1029, + 1147, + 1029 + ], + "score": 0.82, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1768, + 1175, + 1768, + 1175, + 1790, + 1156, + 1790 + ], + "score": 0.79, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 666, + 1552, + 776, + 1552, + 776, + 1586, + 666, + 1586 + ], + "score": 0.75, + "latex": "\\mathbb { E } [ \\varepsilon _ { i } ] = 0" + }, + { + "category_id": 13, + "poly": [ + 902, + 1763, + 934, + 1763, + 934, + 1790, + 902, + 1790 + ], + "score": 0.74, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 1357, + 1153, + 1369, + 1153, + 1369, + 1178, + 1357, + 1178 + ], + "score": 0.72, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 689, + 1917, + 710, + 1917, + 710, + 1939, + 689, + 1939 + ], + "score": 0.7, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 1174, + 1154, + 1186, + 1154, + 1186, + 1178, + 1174, + 1178 + ], + "score": 0.66, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 415, + 1974, + 447, + 1974, + 447, + 2001, + 415, + 2001 + ], + "score": 0.61, + "latex": "W" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 929.0, + 1068.0, + 929.0, + 1068.0, + 975.0, + 291.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 860.0, + 2085.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 263.0, + 295.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 259.0, + 1408.0, + 259.0, + 1408.0, + 297.0, + 293.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 292.0, + 1405.0, + 292.0, + 1405.0, + 324.0, + 296.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 319.0, + 1407.0, + 319.0, + 1407.0, + 355.0, + 293.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 353.0, + 1403.0, + 353.0, + 1403.0, + 385.0, + 297.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 380.0, + 1405.0, + 380.0, + 1405.0, + 416.0, + 295.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 414.0, + 1403.0, + 414.0, + 1403.0, + 446.0, + 296.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 444.0, + 1403.0, + 444.0, + 1403.0, + 476.0, + 296.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 475.0, + 1405.0, + 475.0, + 1405.0, + 507.0, + 296.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 505.0, + 1403.0, + 505.0, + 1403.0, + 537.0, + 296.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 532.0, + 1402.0, + 532.0, + 1402.0, + 568.0, + 295.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 564.0, + 1406.0, + 564.0, + 1406.0, + 600.0, + 296.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 594.0, + 1187.0, + 594.0, + 1187.0, + 626.0, + 296.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1145.0, + 372.0, + 1145.0, + 372.0, + 1185.0, + 292.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1145.0, + 611.0, + 1145.0, + 611.0, + 1185.0, + 466.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1145.0, + 1173.0, + 1145.0, + 1173.0, + 1185.0, + 719.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1145.0, + 1198.0, + 1145.0, + 1198.0, + 1185.0, + 1187.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1145.0, + 1356.0, + 1145.0, + 1356.0, + 1185.0, + 1284.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 1145.0, + 1406.0, + 1145.0, + 1406.0, + 1185.0, + 1370.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1181.0, + 702.0, + 1181.0, + 702.0, + 1213.0, + 296.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1181.0, + 1403.0, + 1181.0, + 1403.0, + 1213.0, + 833.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1206.0, + 667.0, + 1206.0, + 667.0, + 1249.0, + 292.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1206.0, + 855.0, + 1206.0, + 855.0, + 1249.0, + 843.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1206.0, + 1004.0, + 1206.0, + 1004.0, + 1249.0, + 961.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1206.0, + 1408.0, + 1206.0, + 1408.0, + 1249.0, + 1155.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1232.0, + 401.0, + 1232.0, + 401.0, + 1281.0, + 290.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1232.0, + 957.0, + 1232.0, + 957.0, + 1281.0, + 698.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1232.0, + 1409.0, + 1232.0, + 1409.0, + 1281.0, + 1192.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1265.0, + 378.0, + 1265.0, + 378.0, + 1311.0, + 291.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1265.0, + 881.0, + 1265.0, + 881.0, + 1311.0, + 657.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1265.0, + 1409.0, + 1265.0, + 1409.0, + 1311.0, + 971.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1297.0, + 347.0, + 1297.0, + 347.0, + 1338.0, + 292.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1297.0, + 1313.0, + 1297.0, + 1313.0, + 1338.0, + 617.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 1297.0, + 1406.0, + 1297.0, + 1406.0, + 1338.0, + 1333.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1333.0, + 649.0, + 1333.0, + 649.0, + 1367.0, + 296.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 641.0, + 1408.0, + 641.0, + 1408.0, + 677.0, + 295.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 673.0, + 1407.0, + 673.0, + 1407.0, + 711.0, + 292.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 705.0, + 1404.0, + 705.0, + 1404.0, + 739.0, + 295.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 738.0, + 1404.0, + 738.0, + 1404.0, + 768.0, + 297.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 765.0, + 1406.0, + 765.0, + 1406.0, + 799.0, + 295.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 797.0, + 1406.0, + 797.0, + 1406.0, + 830.0, + 295.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 827.0, + 1406.0, + 827.0, + 1406.0, + 860.0, + 294.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 857.0, + 778.0, + 857.0, + 778.0, + 891.0, + 295.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1671.0, + 956.0, + 1671.0, + 956.0, + 1707.0, + 292.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1671.0, + 1406.0, + 1671.0, + 1406.0, + 1707.0, + 988.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1701.0, + 1011.0, + 1701.0, + 1011.0, + 1737.0, + 294.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1701.0, + 1407.0, + 1701.0, + 1407.0, + 1737.0, + 1129.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1733.0, + 295.0, + 1733.0, + 295.0, + 1766.0, + 292.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1733.0, + 1407.0, + 1733.0, + 1407.0, + 1766.0, + 384.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1760.0, + 901.0, + 1760.0, + 901.0, + 1800.0, + 294.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1760.0, + 1155.0, + 1760.0, + 1155.0, + 1800.0, + 935.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1760.0, + 1190.0, + 1760.0, + 1190.0, + 1800.0, + 1176.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1885.0, + 295.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1881.0, + 1404.0, + 1881.0, + 1404.0, + 1916.0, + 295.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1910.0, + 688.0, + 1910.0, + 688.0, + 1946.0, + 295.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1910.0, + 1406.0, + 1910.0, + 1406.0, + 1946.0, + 711.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1939.0, + 1404.0, + 1939.0, + 1404.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 414.0, + 1971.0, + 414.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2007.0, + 448.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 616.0, + 2003.0, + 616.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1550.0, + 371.0, + 1550.0, + 371.0, + 1586.0, + 296.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1550.0, + 497.0, + 1550.0, + 497.0, + 1586.0, + 488.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1550.0, + 665.0, + 1550.0, + 665.0, + 1586.0, + 656.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1550.0, + 786.0, + 1550.0, + 786.0, + 1586.0, + 777.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1550.0, + 1405.0, + 1550.0, + 1405.0, + 1586.0, + 943.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1581.0, + 1354.0, + 1581.0, + 1354.0, + 1622.0, + 645.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1374.0, + 1202.0, + 1374.0, + 1202.0, + 1413.0, + 294.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1374.0, + 1404.0, + 1374.0, + 1404.0, + 1413.0, + 1346.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1410.0, + 1104.0, + 1410.0, + 1104.0, + 1443.0, + 297.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 998.0, + 949.0, + 998.0, + 949.0, + 1037.0, + 295.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 998.0, + 1146.0, + 998.0, + 1146.0, + 1037.0, + 1088.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 998.0, + 1314.0, + 998.0, + 1314.0, + 1037.0, + 1168.0, + 1037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1849, + 1405, + 1849, + 1405, + 2034, + 296, + 2034 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1234, + 1404, + 1234, + 1404, + 1396, + 297, + 1396 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1705, + 1405, + 1705, + 1405, + 1837, + 297, + 1837 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1493, + 1405, + 1493, + 1405, + 1594, + 297, + 1594 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 291, + 1404, + 291, + 1404, + 388, + 297, + 388 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 416, + 1406, + 416, + 1406, + 519, + 298, + 519 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 524, + 1405, + 524, + 1405, + 618, + 298, + 618 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 313, + 984, + 1384, + 984, + 1384, + 1149, + 313, + 1149 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 507, + 1599, + 1191, + 1599, + 1191, + 1670, + 507, + 1670 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 1425, + 1402, + 1425, + 1402, + 1489, + 298, + 1489 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 417, + 792, + 1284, + 792, + 1284, + 880, + 417, + 880 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 302, + 718, + 1400, + 718, + 1400, + 785, + 302, + 785 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 298, + 1180, + 1008, + 1180, + 1008, + 1218, + 298, + 1218 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 291, + 893, + 1298, + 893, + 1298, + 930, + 291, + 930 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 313, + 637, + 1296, + 637, + 1296, + 670, + 313, + 670 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 817, + 74, + 817, + 105, + 299, + 105 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 331, + 674, + 1349, + 674, + 1349, + 710, + 331, + 710 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 297, + 937, + 1227, + 937, + 1227, + 978, + 297, + 978 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1616, + 1400, + 1616, + 1400, + 1646, + 1366, + 1646 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1365, + 804, + 1401, + 804, + 1401, + 833, + 1365, + 833 + ], + "score": 0.853 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1143, + 1401, + 1143, + 1401, + 1172, + 1365, + 1172 + ], + "score": 0.853 + }, + { + "category_id": 0, + "poly": [ + 298, + 225, + 787, + 225, + 787, + 262, + 298, + 262 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.789 + }, + { + "category_id": 13, + "poly": [ + 1109, + 1330, + 1260, + 1330, + 1260, + 1366, + 1109, + 1366 + ], + "score": 0.95, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } _ { \\operatorname* { m i n } } = { \\boldsymbol { X } } ^ { \\dagger } { \\boldsymbol { y } }" + }, + { + "category_id": 13, + "poly": [ + 729, + 717, + 904, + 717, + 904, + 757, + 729, + 757 + ], + "score": 0.95, + "latex": "f ( \\pmb { x } ) = \\langle \\pmb { x } , \\hat { \\beta } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1246, + 449, + 1398, + 449, + 1398, + 494, + 1246, + 494 + ], + "score": 0.94, + "latex": "\\left| \\left| \\pmb { a } ^ { \\top } \\Phi - \\pmb { y } \\right| \\right| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 642, + 675, + 799, + 675, + 799, + 710, + 642, + 710 + ], + "score": 0.94, + "latex": "\\pmb { x } _ { i } \\sim \\mathcal { N } ( 0 , I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 372, + 889, + 549, + 889, + 549, + 932, + 372, + 932 + ], + "score": 0.93, + "latex": "\\left\\| \\pmb { x } \\right\\| _ { \\Sigma } ^ { 2 } = \\pmb { x } ^ { \\top } \\Sigma \\pmb { x }" + }, + { + "category_id": 13, + "poly": [ + 1227, + 1527, + 1351, + 1527, + 1351, + 1565, + 1227, + 1565 + ], + "score": 0.93, + "latex": "\\boldsymbol { y } = \\boldsymbol { X } ^ { \\intercal } \\hat { \\boldsymbol { \\beta } }" + }, + { + "category_id": 13, + "poly": [ + 994, + 717, + 1110, + 717, + 1110, + 755, + 994, + 755 + ], + "score": 0.92, + "latex": "\\hat { \\boldsymbol { \\beta } } = W \\hat { \\mathbf { a } }" + }, + { + "category_id": 14, + "poly": [ + 311, + 983, + 1386, + 983, + 1386, + 1150, + 311, + 1150 + ], + "score": 0.92, + "latex": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < \\gamma _ { 1 } , } \\\\ { \\frac { \\gamma _ { 1 } } { g _ { 1 } } \\sigma ^ { 2 } , } & R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { \\gamma _ { 2 } } { g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } < 1 , } \\\\ { ~ } & { R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\{ \\begin{array} { l l } { \\frac { \\gamma _ { 2 } g _ { 1 } } { \\gamma _ { 1 } g _ { 2 } } r ^ { 2 } + \\frac { g _ { 1 } + g _ { 2 } } { g _ { 1 } g _ { 2 } } \\sigma ^ { 2 } , } & { \\gamma _ { 2 } > 1 . } \\end{array} } \\end{array} } \\end{array} \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 801, + 1492, + 907, + 1492, + 907, + 1527, + 801, + 1527 + ], + "score": 0.92, + "latex": "\\pm \\ : 0" + }, + { + "category_id": 13, + "poly": [ + 768, + 294, + 882, + 294, + 882, + 328, + 768, + 328 + ], + "score": 0.92, + "latex": "\\phi ( { \\pmb x } ) = { \\pmb x }" + }, + { + "category_id": 14, + "poly": [ + 507, + 1597, + 1191, + 1597, + 1191, + 1669, + 507, + 1669 + ], + "score": 0.92, + "latex": "R _ { ( \\gamma _ { 1 } < 1 ) } \\to \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } > 1 ) } \\to \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } r ^ { 2 } + \\frac { 1 } { \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 391, + 1768, + 451, + 1768, + 451, + 1806, + 391, + 1806 + ], + "score": 0.92, + "latex": "\\hat { \\beta } _ { \\mathrm { m i n } }" + }, + { + "category_id": 13, + "poly": [ + 935, + 291, + 1069, + 291, + 1069, + 323, + 935, + 323 + ], + "score": 0.92, + "latex": "\\Phi = W ^ { \\top } X" + }, + { + "category_id": 13, + "poly": [ + 634, + 1270, + 723, + 1270, + 723, + 1298, + 634, + 1298 + ], + "score": 0.92, + "latex": "\\gamma _ { 2 } 1" + }, + { + "category_id": 13, + "poly": [ + 632, + 1526, + 744, + 1526, + 744, + 1565, + 632, + 1565 + ], + "score": 0.92, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } = { \\widehat { \\boldsymbol { W } } } \\mathbf { a }" + }, + { + "category_id": 13, + "poly": [ + 645, + 583, + 750, + 583, + 750, + 616, + 645, + 616 + ], + "score": 0.91, + "latex": "\\mathbf { \\bar { a } } = \\Phi ^ { \\dagger } \\mathbf { y }" + }, + { + "category_id": 14, + "poly": [ + 415, + 787, + 1283, + 787, + 1283, + 886, + 415, + 886 + ], + "score": 0.91, + "latex": "R = \\mathbb { E } _ { { \\mathbf { x } } \\sim P _ { \\mathbf { x } } } [ \\Vert \\hat { \\beta } - \\beta \\Vert _ { \\Sigma } ^ { 2 } \\vert { \\cal X } , { \\cal W } ] = \\underbrace { \\Vert \\mathbb { E } [ \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ] - \\beta \\Vert _ { 2 } ^ { 2 } } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathrm { t r } \\left( \\mathrm { C o v } ( \\hat { \\beta } \\vert { \\cal X } , { \\cal W } ) \\right) } _ { { \\cal V } = \\mathrm { v a r i a n c e } } ," + }, + { + "category_id": 13, + "poly": [ + 555, + 1493, + 682, + 1493, + 682, + 1528, + 555, + 1528 + ], + "score": 0.91, + "latex": "W ( 0 ) = 0" + }, + { + "category_id": 13, + "poly": [ + 844, + 1182, + 998, + 1182, + 998, + 1217, + 844, + 1217 + ], + "score": 0.91, + "latex": "g _ { 2 } = | \\gamma _ { 2 } - 1 |" + }, + { + "category_id": 13, + "poly": [ + 298, + 553, + 456, + 553, + 456, + 588, + 298, + 588 + ], + "score": 0.91, + "latex": "\\langle \\phi ( \\pmb { x } ^ { \\top } W ) , \\hat { \\pmb { a } } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1309, + 524, + 1403, + 524, + 1403, + 558, + 1309, + 558 + ], + "score": 0.9, + "latex": "f ( { \\pmb x } ) =" + }, + { + "category_id": 13, + "poly": [ + 887, + 1298, + 962, + 1298, + 962, + 1327, + 887, + 1327 + ], + "score": 0.9, + "latex": "h > d" + }, + { + "category_id": 13, + "poly": [ + 759, + 1298, + 835, + 1298, + 835, + 1327, + 759, + 1327 + ], + "score": 0.9, + "latex": "n > d" + }, + { + "category_id": 13, + "poly": [ + 538, + 1238, + 609, + 1238, + 609, + 1266, + 538, + 1266 + ], + "score": 0.9, + "latex": "d > n" + }, + { + "category_id": 13, + "poly": [ + 986, + 946, + 1131, + 946, + 1131, + 977, + 986, + 977 + ], + "score": 0.9, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1299, + 1219, + 1299, + 1219, + 1330, + 1022, + 1330 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } < \\operatorname* { m i n } ( 1 , \\gamma _ { 2 } ) )" + }, + { + "category_id": 13, + "poly": [ + 1057, + 1495, + 1130, + 1495, + 1130, + 1524, + 1057, + 1524 + ], + "score": 0.89, + "latex": "t > 0" + }, + { + "category_id": 13, + "poly": [ + 798, + 941, + 944, + 941, + 944, + 979, + 798, + 979 + ], + "score": 0.87, + "latex": "\\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 661, + 1239, + 743, + 1239, + 743, + 1268, + 661, + 1268 + ], + "score": 0.87, + "latex": "\\gamma _ { 1 } > 1" + }, + { + "category_id": 13, + "poly": [ + 1109, + 1268, + 1201, + 1268, + 1201, + 1298, + 1109, + 1298 + ], + "score": 0.85, + "latex": "( \\gamma _ { 2 } > 1 )" + }, + { + "category_id": 13, + "poly": [ + 630, + 1182, + 784, + 1182, + 784, + 1217, + 630, + 1217 + ], + "score": 0.85, + "latex": "g _ { 1 } = | \\gamma _ { 1 } - 1 |" + }, + { + "category_id": 13, + "poly": [ + 820, + 1713, + 850, + 1713, + 850, + 1740, + 820, + 1740 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1164, + 723, + 1186, + 723, + 1186, + 752, + 1164, + 752 + ], + "score": 0.82, + "latex": "\\hat { \\textbf { \\textit a } }" + }, + { + "category_id": 13, + "poly": [ + 1076, + 675, + 1240, + 675, + 1240, + 710, + 1076, + 710 + ], + "score": 0.82, + "latex": "F ( { \\pmb x } ) = \\langle { \\pmb x } , { \\pmb \\beta } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 835, + 587, + 851, + 587, + 851, + 617, + 835, + 617 + ], + "score": 0.8, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 342, + 1531, + 407, + 1531, + 407, + 1566, + 342, + 1566 + ], + "score": 0.78, + "latex": "W ( t )" + }, + { + "category_id": 13, + "poly": [ + 1120, + 1459, + 1152, + 1459, + 1152, + 1485, + 1120, + 1485 + ], + "score": 0.76, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 707, + 945, + 743, + 945, + 743, + 976, + 707, + 976 + ], + "score": 0.73, + "latex": "{ \\pmb w } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1051, + 462, + 1072, + 462, + 1072, + 484, + 1051, + 484 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 822, + 556, + 853, + 556, + 853, + 582, + 822, + 582 + ], + "score": 0.68, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 931, + 526, + 954, + 526, + 954, + 552, + 931, + 552 + ], + "score": 0.62, + "latex": "X _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1254, + 675, + 1344, + 675, + 1344, + 711, + 1254, + 711 + ], + "score": 0.51, + "latex": "\\| { \\boldsymbol { \\beta } } \\| = r" + }, + { + "category_id": 13, + "poly": [ + 371, + 1181, + 786, + 1181, + 786, + 1217, + 371, + 1217 + ], + "score": 0.48, + "latex": "d / n \\to \\gamma _ { 1 } , h / n \\to \\gamma _ { 2 } , g _ { 1 } = | \\gamma _ { 1 } - 1 | ," + }, + { + "category_id": 13, + "poly": [ + 371, + 1182, + 488, + 1182, + 488, + 1216, + 371, + 1216 + ], + "score": 0.46, + "latex": "d / n \\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 499, + 1182, + 617, + 1182, + 617, + 1216, + 499, + 1216 + ], + "score": 0.45, + "latex": "h / n \\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1531, + 330, + 1531, + 330, + 1561, + 297, + 1561 + ], + "score": 0.4, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 1153, + 531, + 1175, + 531, + 1175, + 557, + 1153, + 557 + ], + "score": 0.37, + "latex": "\\textbf { { y } }" + }, + { + "category_id": 13, + "poly": [ + 1228, + 562, + 1248, + 562, + 1248, + 582, + 1228, + 582 + ], + "score": 0.32, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 791.0, + 221.0, + 791.0, + 267.0, + 291.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 2086.0, + 861.0, + 2086.0, + 861.0, + 2118.0, + 837.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1883.0, + 295.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 1407.0, + 1883.0, + 1407.0, + 1915.0, + 295.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1403.0, + 1912.0, + 1403.0, + 1944.0, + 295.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1944.0, + 1403.0, + 1944.0, + 1403.0, + 1976.0, + 296.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 745.0, + 2003.0, + 745.0, + 2034.0, + 295.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1236.0, + 537.0, + 1236.0, + 537.0, + 1270.0, + 296.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1236.0, + 660.0, + 1236.0, + 660.0, + 1270.0, + 610.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1236.0, + 1404.0, + 1236.0, + 1404.0, + 1270.0, + 744.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1265.0, + 633.0, + 1265.0, + 633.0, + 1303.0, + 294.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1265.0, + 1108.0, + 1265.0, + 1108.0, + 1303.0, + 724.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1265.0, + 1405.0, + 1265.0, + 1405.0, + 1303.0, + 1202.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1293.0, + 758.0, + 1293.0, + 758.0, + 1334.0, + 294.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1293.0, + 886.0, + 1293.0, + 886.0, + 1334.0, + 836.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1293.0, + 1021.0, + 1293.0, + 1021.0, + 1334.0, + 963.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1293.0, + 1404.0, + 1293.0, + 1404.0, + 1334.0, + 1220.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1326.0, + 1108.0, + 1326.0, + 1108.0, + 1372.0, + 291.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1372.0, + 1261.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1362.0, + 398.0, + 1362.0, + 398.0, + 1397.0, + 294.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1706.0, + 819.0, + 1706.0, + 819.0, + 1743.0, + 294.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1706.0, + 1408.0, + 1706.0, + 1408.0, + 1743.0, + 851.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1736.0, + 1405.0, + 1736.0, + 1405.0, + 1771.0, + 294.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1769.0, + 390.0, + 1769.0, + 390.0, + 1811.0, + 291.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1769.0, + 1407.0, + 1769.0, + 1407.0, + 1811.0, + 452.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1805.0, + 643.0, + 1805.0, + 643.0, + 1839.0, + 295.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1487.0, + 554.0, + 1487.0, + 554.0, + 1532.0, + 291.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1487.0, + 800.0, + 1487.0, + 800.0, + 1532.0, + 683.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1487.0, + 1056.0, + 1487.0, + 1056.0, + 1532.0, + 908.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1487.0, + 1406.0, + 1487.0, + 1406.0, + 1532.0, + 1131.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1527.0, + 341.0, + 1527.0, + 341.0, + 1570.0, + 331.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1527.0, + 631.0, + 1527.0, + 631.0, + 1570.0, + 408.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1527.0, + 1226.0, + 1527.0, + 1226.0, + 1570.0, + 745.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1527.0, + 1406.0, + 1527.0, + 1406.0, + 1570.0, + 1352.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1561.0, + 1109.0, + 1561.0, + 1109.0, + 1599.0, + 291.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 290.0, + 767.0, + 290.0, + 767.0, + 330.0, + 292.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 290.0, + 934.0, + 290.0, + 934.0, + 330.0, + 883.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 330.0, + 1070.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 326.0, + 1402.0, + 326.0, + 1402.0, + 357.0, + 296.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 356.0, + 1117.0, + 356.0, + 1117.0, + 390.0, + 296.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 417.0, + 1404.0, + 417.0, + 1404.0, + 452.0, + 296.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 448.0, + 1050.0, + 448.0, + 1050.0, + 500.0, + 290.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 448.0, + 1245.0, + 448.0, + 1245.0, + 500.0, + 1073.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 448.0, + 1409.0, + 448.0, + 1409.0, + 500.0, + 1399.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 487.0, + 1058.0, + 487.0, + 1058.0, + 522.0, + 295.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 520.0, + 930.0, + 520.0, + 930.0, + 561.0, + 292.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 520.0, + 1152.0, + 520.0, + 1152.0, + 561.0, + 955.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 520.0, + 1308.0, + 520.0, + 1308.0, + 561.0, + 1176.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 554.0, + 821.0, + 554.0, + 821.0, + 592.0, + 457.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 554.0, + 1227.0, + 554.0, + 1227.0, + 592.0, + 854.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 554.0, + 1406.0, + 554.0, + 1406.0, + 592.0, + 1249.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 584.0, + 644.0, + 584.0, + 644.0, + 621.0, + 292.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 584.0, + 834.0, + 584.0, + 834.0, + 621.0, + 751.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 584.0, + 1280.0, + 584.0, + 1280.0, + 621.0, + 852.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1425.0, + 1404.0, + 1425.0, + 1404.0, + 1462.0, + 296.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1457.0, + 1119.0, + 1457.0, + 1119.0, + 1490.0, + 294.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1457.0, + 1163.0, + 1457.0, + 1163.0, + 1490.0, + 1153.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 719.0, + 728.0, + 719.0, + 728.0, + 759.0, + 296.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 719.0, + 993.0, + 719.0, + 993.0, + 759.0, + 905.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 719.0, + 1163.0, + 719.0, + 1163.0, + 759.0, + 1111.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 719.0, + 1405.0, + 719.0, + 1405.0, + 759.0, + 1187.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 752.0, + 1348.0, + 752.0, + 1348.0, + 788.0, + 297.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1176.0, + 370.0, + 1176.0, + 370.0, + 1223.0, + 294.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1176.0, + 843.0, + 1176.0, + 843.0, + 1223.0, + 787.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1176.0, + 1009.0, + 1176.0, + 1009.0, + 1223.0, + 999.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 887.0, + 371.0, + 887.0, + 371.0, + 934.0, + 293.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 887.0, + 1305.0, + 887.0, + 1305.0, + 934.0, + 550.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 634.0, + 1299.0, + 634.0, + 1299.0, + 677.0, + 304.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 673.0, + 641.0, + 673.0, + 641.0, + 713.0, + 345.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 673.0, + 1075.0, + 673.0, + 1075.0, + 713.0, + 800.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 673.0, + 1253.0, + 673.0, + 1253.0, + 713.0, + 1241.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 673.0, + 1354.0, + 673.0, + 1354.0, + 713.0, + 1345.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 924.0, + 706.0, + 924.0, + 706.0, + 989.0, + 288.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 924.0, + 797.0, + 924.0, + 797.0, + 989.0, + 744.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 924.0, + 985.0, + 924.0, + 985.0, + 989.0, + 945.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 924.0, + 1239.0, + 924.0, + 1239.0, + 989.0, + 1132.0, + 989.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1910, + 1403, + 1910, + 1403, + 2036, + 298, + 2036 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 327, + 914, + 1373, + 914, + 1373, + 1075, + 327, + 1075 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 313, + 224, + 1379, + 224, + 1379, + 519, + 313, + 519 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 804, + 1403, + 804, + 1403, + 900, + 299, + 900 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 321, + 1230, + 1374, + 1230, + 1374, + 1397, + 321, + 1397 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 295, + 538, + 1408, + 538, + 1408, + 680, + 295, + 680 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 296, + 1102, + 1401, + 1102, + 1401, + 1165, + 296, + 1165 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 1822, + 1402, + 1822, + 1402, + 1889, + 297, + 1889 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 574, + 1682, + 1127, + 1682, + 1127, + 1724, + 574, + 1724 + ], + "score": 0.926 + }, + { + "category_id": 0, + "poly": [ + 299, + 734, + 1141, + 734, + 1141, + 772, + 299, + 772 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 297, + 1171, + 1193, + 1171, + 1193, + 1214, + 297, + 1214 + ], + "score": 0.918 + }, + { + "category_id": 8, + "poly": [ + 335, + 1461, + 1332, + 1461, + 1332, + 1540, + 335, + 1540 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 293, + 1737, + 951, + 1737, + 951, + 1772, + 293, + 1772 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 295, + 1410, + 1228, + 1410, + 1228, + 1447, + 295, + 1447 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 296, + 1779, + 907, + 1779, + 907, + 1813, + 296, + 1813 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1688, + 1400, + 1688, + 1400, + 1719, + 1352, + 1719 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 300, + 1636, + 1334, + 1636, + 1334, + 1671, + 300, + 1671 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1568, + 1400, + 1568, + 1400, + 1597, + 1366, + 1597 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1365, + 928, + 1400, + 928, + 1400, + 959, + 1365, + 959 + ], + "score": 0.853 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1487, + 1400, + 1487, + 1400, + 1516, + 1366, + 1516 + ], + "score": 0.849 + }, + { + "category_id": 8, + "poly": [ + 335, + 1546, + 865, + 1546, + 865, + 1623, + 335, + 1623 + ], + "score": 0.835 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.75 + }, + { + "category_id": 9, + "poly": [ + 1282, + 1243, + 1371, + 1243, + 1371, + 1275, + 1282, + 1275 + ], + "score": 0.47 + }, + { + "category_id": 14, + "poly": [ + 326, + 912, + 1372, + 912, + 1372, + 1078, + 326, + 1078 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { R = \\mathbb { E } _ { \\boldsymbol { x } \\sim P _ { \\boldsymbol { x } } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } | \\boldsymbol { X } , \\boldsymbol { W } ] } \\\\ & { \\quad = \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } | \\boldsymbol { X } , \\boldsymbol { W } ] - \\boldsymbol { F } ( \\boldsymbol { x } ) \\| _ { 2 } ^ { 2 } ] } _ { B = \\mathrm { b i a s } } + \\underbrace { \\mathbb { E } _ { \\boldsymbol { x } } [ \\| \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } - \\mathbb { E } [ \\phi ( \\boldsymbol { x } ^ { \\top } \\boldsymbol { W } ) \\hat { \\boldsymbol { a } } ] \\| _ { 2 } ^ { 2 } \\big | \\boldsymbol { X } , \\boldsymbol { W } ] } _ { \\boldsymbol { V } = \\mathrm { v a r i a n c e } } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 333, + 1458, + 1330, + 1458, + 1330, + 1627, + 333, + 1627 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { m _ { 1 } ^ { - 1 } = - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - c _ { 1 } m _ { 2 } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - 2 \\tau c _ { 2 } m _ { 1 } m _ { 2 } + c _ { 2 } ^ { 2 } m _ { 1 } m _ { 2 } ^ { 2 } } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\\\ & { m _ { 2 } ^ { - 1 } = - \\xi - r \\gamma _ { 2 } m _ { 1 } + \\frac { \\gamma _ { 2 } c _ { 2 } m _ { 1 } ^ { 2 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) } { m _ { 1 } \\left( c _ { 2 } m _ { 2 } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 337, + 1738, + 486, + 1738, + 486, + 1772, + 337, + 1772 + ], + "score": 0.93, + "latex": "G \\sim \\mathcal { N } ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 849, + 565, + 976, + 565, + 976, + 597, + 849, + 597 + ], + "score": 0.92, + "latex": "r ^ { 2 } / \\sigma ^ { 2 } \\overset { \\cdot } { = } 1 6" + }, + { + "category_id": 13, + "poly": [ + 496, + 866, + 601, + 866, + 601, + 900, + 496, + 900 + ], + "score": 0.92, + "latex": "\\hat { \\mathbf { a } } = \\Phi ^ { \\dagger } \\mathbf { y }" + }, + { + "category_id": 13, + "poly": [ + 945, + 1410, + 1189, + 1410, + 1189, + 1447, + 945, + 1447 + ], + "score": 0.92, + "latex": "\\{ | m _ { 1 } | , | m _ { 2 } | < 1 / { \\mathfrak { T } } \\xi \\}" + }, + { + "category_id": 13, + "poly": [ + 835, + 651, + 916, + 651, + 916, + 679, + 835, + 679 + ], + "score": 0.92, + "latex": "\\gamma _ { 2 } 1" + }, + { + "category_id": 13, + "poly": [ + 938, + 1181, + 1083, + 1181, + 1083, + 1213, + 938, + 1213 + ], + "score": 0.91, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 481, + 1639, + 552, + 1639, + 552, + 1670, + 481, + 1670 + ], + "score": 0.91, + "latex": "\\xi , \\rho , \\tau" + }, + { + "category_id": 13, + "poly": [ + 372, + 1853, + 773, + 1853, + 773, + 1889, + 372, + 1889 + ], + "score": 0.91, + "latex": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\sigma ^ { 2 } \\mathrm { m i n } \\{ \\gamma _ { 2 } , 1 \\} / | 1 - \\gamma _ { 2 } |" + }, + { + "category_id": 13, + "poly": [ + 1010, + 626, + 1108, + 626, + 1108, + 651, + 1010, + 651 + ], + "score": 0.9, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 407, + 1782, + 496, + 1782, + 496, + 1813, + 407, + 1813 + ], + "score": 0.9, + "latex": "c _ { 1 } \\geq c _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 650, + 1637, + 738, + 1637, + 738, + 1670, + 650, + 1670 + ], + "score": 0.9, + "latex": "\\Im \\xi > 0" + }, + { + "category_id": 14, + "poly": [ + 318, + 1232, + 1376, + 1232, + 1376, + 1398, + 318, + 1398 + ], + "score": 0.89, + "latex": "V = \\{ \\begin{array} { l l } { \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } & { \\gamma _ { 2 } < 1 , } \\\\ { \\quad } & { \\gamma _ { 2 } < 1 \\leq r \\leq r , } \\\\ { \\sigma ^ { 2 } \\displaystyle \\operatorname* { l i m } _ { \\xi 0 } - [ \\gamma _ { 2 } \\frac { \\partial } { \\partial x } m _ { 1 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\partial } { \\partial x } m _ { 2 } ( \\xi , c _ { 1 } x , c _ { 2 } x ) | _ { x = 0 } + \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } ] } & { \\gamma _ { 2 } > 1 . } \\end{array} " + }, + { + "category_id": 13, + "poly": [ + 1142, + 1642, + 1205, + 1642, + 1205, + 1670, + 1142, + 1670 + ], + "score": 0.89, + "latex": "c _ { 1 } , c _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 568, + 1095, + 568, + 1095, + 596, + 1019, + 596 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } < 1" + }, + { + "category_id": 13, + "poly": [ + 1047, + 599, + 1142, + 599, + 1142, + 624, + 1047, + 624 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1108, + 568, + 1189, + 568, + 1189, + 596, + 1108, + 596 + ], + "score": 0.88, + "latex": "( \\gamma _ { 1 } > 1" + }, + { + "category_id": 13, + "poly": [ + 784, + 1782, + 804, + 1782, + 804, + 1812, + 784, + 1812 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 540, + 1739, + 576, + 1739, + 576, + 1771, + 540, + 1771 + ], + "score": 0.86, + "latex": "\\Im \\xi" + }, + { + "category_id": 14, + "poly": [ + 574, + 1681, + 1123, + 1681, + 1123, + 1723, + 574, + 1723 + ], + "score": 0.86, + "latex": "c _ { 1 } = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , \\quad c _ { 2 } = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } ," + }, + { + "category_id": 13, + "poly": [ + 545, + 1826, + 646, + 1826, + 646, + 1853, + 545, + 1853 + ], + "score": 0.85, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 948, + 2007, + 977, + 2007, + 977, + 2035, + 948, + 2035 + ], + "score": 0.84, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 509, + 570, + 537, + 570, + 537, + 596, + 509, + 596 + ], + "score": 0.83, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 510, + 838, + 530, + 838, + 530, + 868, + 510, + 868 + ], + "score": 0.83, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 772, + 1638, + 840, + 1638, + 840, + 1670, + 772, + 1670 + ], + "score": 0.82, + "latex": "\\xi < 0" + }, + { + "category_id": 13, + "poly": [ + 925, + 1740, + 942, + 1740, + 942, + 1771, + 925, + 1771 + ], + "score": 0.81, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 849, + 1639, + 970, + 1639, + 970, + 1670, + 849, + 1670 + ], + "score": 0.78, + "latex": "\\rho > \\tau > 0" + }, + { + "category_id": 13, + "poly": [ + 623, + 1173, + 862, + 1173, + 862, + 1214, + 623, + 1214 + ], + "score": 0.77, + "latex": "{ \\pmb w } _ { i } \\overset { \\mathrm { i . i . d . } } { \\sim } \\mathcal { N } ( 0 , d ^ { - 1 } I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 931, + 813, + 952, + 813, + 952, + 834, + 931, + 834 + ], + "score": 0.75, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 539, + 1412, + 666, + 1412, + 666, + 1446, + 539, + 1446 + ], + "score": 0.71, + "latex": "m _ { 2 } ( \\xi , \\rho , \\tau )" + }, + { + "category_id": 13, + "poly": [ + 1195, + 808, + 1228, + 808, + 1228, + 835, + 1195, + 835 + ], + "score": 0.69, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 399, + 1412, + 526, + 1412, + 526, + 1446, + 399, + 1446 + ], + "score": 0.6, + "latex": "m _ { 1 } ( \\xi , \\rho , \\tau )" + }, + { + "category_id": 14, + "poly": [ + 1280, + 1240, + 1370, + 1240, + 1370, + 1277, + 1280, + 1277 + ], + "score": 0.58, + "latex": "\\gamma _ { 2 } < 1 ," + }, + { + "category_id": 13, + "poly": [ + 623, + 1175, + 659, + 1175, + 659, + 1212, + 623, + 1212 + ], + "score": 0.38, + "latex": "{ \\pmb w } _ { i }" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 244.0, + 358.0, + 244.0, + 358.0, + 267.0, + 327.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 242.0, + 643.0, + 242.0, + 643.0, + 265.0, + 569.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 225.0, + 716.0, + 225.0, + 716.0, + 247.0, + 687.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 242.0, + 987.0, + 242.0, + 987.0, + 265.0, + 910.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 225.0, + 1074.0, + 225.0, + 1074.0, + 247.0, + 1045.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 233.0, + 1193.0, + 233.0, + 1193.0, + 325.0, + 1149.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 246.0, + 1258.0, + 246.0, + 1258.0, + 259.0, + 1241.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 241.0, + 1343.0, + 241.0, + 1343.0, + 265.0, + 1267.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 262.0, + 650.0, + 262.0, + 650.0, + 284.0, + 566.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 274.0, + 715.0, + 274.0, + 715.0, + 295.0, + 697.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 262.0, + 1008.0, + 262.0, + 1008.0, + 285.0, + 912.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 269.0, + 1073.0, + 269.0, + 1073.0, + 290.0, + 1055.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 262.0, + 1366.0, + 262.0, + 1366.0, + 285.0, + 1267.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 283.0, + 359.0, + 283.0, + 359.0, + 305.0, + 326.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 299.0, + 631.0, + 299.0, + 631.0, + 316.0, + 616.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 310.0, + 694.0, + 310.0, + 694.0, + 379.0, + 674.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 315.0, + 1071.0, + 315.0, + 1071.0, + 332.0, + 1057.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 315.0, + 1207.0, + 315.0, + 1207.0, + 328.0, + 1198.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 321.0, + 359.0, + 321.0, + 359.0, + 365.0, + 310.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 325.0, + 712.0, + 325.0, + 712.0, + 343.0, + 698.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 325.0, + 1054.0, + 325.0, + 1054.0, + 365.0, + 1032.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 336.0, + 1212.0, + 336.0, + 1212.0, + 347.0, + 1203.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 344.0, + 1148.0, + 344.0, + 1148.0, + 354.0, + 1140.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 349.0, + 1218.0, + 349.0, + 1218.0, + 359.0, + 1208.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 359.0, + 359.0, + 359.0, + 359.0, + 383.0, + 326.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 359.0, + 1071.0, + 359.0, + 1071.0, + 375.0, + 1057.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 373.0, + 712.0, + 373.0, + 712.0, + 390.0, + 699.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 397.0, + 358.0, + 397.0, + 358.0, + 421.0, + 326.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 405.0, + 1069.0, + 405.0, + 1069.0, + 418.0, + 1060.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 390.0, + 1192.0, + 390.0, + 1192.0, + 428.0, + 1176.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 437.0, + 359.0, + 437.0, + 359.0, + 459.0, + 327.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 436.0, + 530.0, + 436.0, + 530.0, + 442.0, + 524.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 423.0, + 712.0, + 423.0, + 712.0, + 441.0, + 698.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 442.0, + 1096.0, + 442.0, + 1096.0, + 477.0, + 1052.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 443.0, + 1151.0, + 443.0, + 1151.0, + 455.0, + 1129.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 444.0, + 1263.0, + 444.0, + 1263.0, + 457.0, + 1203.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 453.0, + 422.0, + 453.0, + 422.0, + 476.0, + 389.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 454.0, + 491.0, + 454.0, + 491.0, + 475.0, + 457.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 454.0, + 557.0, + 454.0, + 557.0, + 476.0, + 524.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 453.0, + 624.0, + 453.0, + 624.0, + 476.0, + 591.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 453.0, + 733.0, + 453.0, + 733.0, + 474.0, + 694.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 454.0, + 785.0, + 454.0, + 785.0, + 475.0, + 744.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 454.0, + 837.0, + 454.0, + 837.0, + 475.0, + 796.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 454.0, + 890.0, + 454.0, + 890.0, + 475.0, + 846.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 454.0, + 941.0, + 454.0, + 941.0, + 475.0, + 899.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 454.0, + 993.0, + 454.0, + 993.0, + 475.0, + 952.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 453.0, + 1144.0, + 453.0, + 1144.0, + 475.0, + 1103.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 455.0, + 1195.0, + 455.0, + 1195.0, + 473.0, + 1157.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 454.0, + 1248.0, + 454.0, + 1248.0, + 475.0, + 1208.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 453.0, + 1301.0, + 453.0, + 1301.0, + 474.0, + 1259.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 454.0, + 1350.0, + 454.0, + 1350.0, + 475.0, + 1310.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 467.0, + 540.0, + 467.0, + 540.0, + 491.0, + 471.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 469.0, + 894.0, + 469.0, + 894.0, + 491.0, + 830.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 469.0, + 1253.0, + 469.0, + 1253.0, + 491.0, + 1189.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 486.0, + 581.0, + 486.0, + 581.0, + 519.0, + 396.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 488.0, + 960.0, + 488.0, + 960.0, + 520.0, + 738.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 485.0, + 1296.0, + 485.0, + 1296.0, + 521.0, + 1116.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 285.0, + 617.0, + 285.0, + 617.0, + 298.5, + 599.0, + 298.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.75, + 380.0, + 879.75, + 380.0, + 879.75, + 388.0, + 871.75, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.25, + 442.0, + 1100.25, + 442.0, + 1100.25, + 455.5, + 1070.25, + 455.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 536.0, + 1406.0, + 536.0, + 1406.0, + 570.0, + 294.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 564.0, + 508.0, + 564.0, + 508.0, + 599.0, + 293.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 564.0, + 848.0, + 564.0, + 848.0, + 599.0, + 538.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 564.0, + 1018.0, + 564.0, + 1018.0, + 599.0, + 977.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 564.0, + 1107.0, + 564.0, + 1107.0, + 599.0, + 1096.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 564.0, + 1407.0, + 564.0, + 1407.0, + 599.0, + 1190.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 593.0, + 1046.0, + 593.0, + 1046.0, + 628.0, + 293.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 593.0, + 1409.0, + 593.0, + 1409.0, + 628.0, + 1143.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 619.0, + 1009.0, + 619.0, + 1009.0, + 653.0, + 292.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 619.0, + 1407.0, + 619.0, + 1407.0, + 653.0, + 1109.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 649.0, + 834.0, + 649.0, + 834.0, + 680.0, + 294.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 649.0, + 1364.0, + 649.0, + 1364.0, + 680.0, + 917.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 734.0, + 1143.0, + 734.0, + 1143.0, + 775.0, + 294.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2119.0, + 838.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1905.0, + 1409.0, + 1905.0, + 1409.0, + 1950.0, + 291.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1975.0, + 294.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 947.0, + 2001.0, + 947.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 2001.0, + 1407.0, + 2001.0, + 1407.0, + 2038.0, + 978.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 805.0, + 930.0, + 805.0, + 930.0, + 840.0, + 294.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 805.0, + 1194.0, + 805.0, + 1194.0, + 840.0, + 953.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 805.0, + 1405.0, + 805.0, + 1405.0, + 840.0, + 1229.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 839.0, + 509.0, + 839.0, + 509.0, + 869.0, + 295.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 839.0, + 1404.0, + 839.0, + 1404.0, + 869.0, + 531.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 864.0, + 495.0, + 864.0, + 495.0, + 903.0, + 293.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 864.0, + 1309.0, + 864.0, + 1309.0, + 903.0, + 602.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1100.0, + 1403.0, + 1100.0, + 1403.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1135.0, + 916.0, + 1135.0, + 916.0, + 1167.0, + 297.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1819.0, + 544.0, + 1819.0, + 544.0, + 1858.0, + 296.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1819.0, + 1406.0, + 1819.0, + 1406.0, + 1858.0, + 647.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1852.0, + 371.0, + 1852.0, + 371.0, + 1892.0, + 293.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1852.0, + 784.0, + 1852.0, + 784.0, + 1892.0, + 774.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1159.0, + 622.0, + 1159.0, + 622.0, + 1225.0, + 288.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1159.0, + 937.0, + 1159.0, + 937.0, + 1225.0, + 863.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1159.0, + 1198.0, + 1159.0, + 1198.0, + 1225.0, + 1084.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1730.0, + 336.0, + 1730.0, + 336.0, + 1781.0, + 289.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1730.0, + 539.0, + 1730.0, + 539.0, + 1781.0, + 487.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1730.0, + 924.0, + 1730.0, + 924.0, + 1781.0, + 577.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1730.0, + 952.0, + 1730.0, + 952.0, + 1781.0, + 943.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1405.0, + 398.0, + 1405.0, + 398.0, + 1453.0, + 291.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1405.0, + 538.0, + 1405.0, + 538.0, + 1453.0, + 527.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1405.0, + 944.0, + 1405.0, + 944.0, + 1453.0, + 667.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 1405.0, + 1226.0, + 1405.0, + 1226.0, + 1453.0, + 1190.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1777.0, + 406.0, + 1777.0, + 406.0, + 1817.0, + 293.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1777.0, + 783.0, + 1777.0, + 783.0, + 1817.0, + 497.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 1777.0, + 908.0, + 1777.0, + 908.0, + 1817.0, + 805.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1631.0, + 480.0, + 1631.0, + 480.0, + 1677.0, + 293.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1631.0, + 649.0, + 1631.0, + 649.0, + 1677.0, + 553.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1631.0, + 771.0, + 1631.0, + 771.0, + 1677.0, + 739.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 1631.0, + 848.0, + 1631.0, + 848.0, + 1677.0, + 841.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1631.0, + 1141.0, + 1631.0, + 1141.0, + 1677.0, + 971.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1631.0, + 1334.0, + 1631.0, + 1334.0, + 1677.0, + 1206.0, + 1677.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1337, + 1406, + 1337, + 1406, + 1603, + 297, + 1603 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 479, + 1405, + 479, + 1405, + 634, + 297, + 634 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 750, + 1405, + 750, + 1405, + 850, + 298, + 850 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 970, + 1406, + 970, + 1406, + 1072, + 296, + 1072 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1910, + 1406, + 1910, + 1406, + 2035, + 298, + 2035 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1698, + 1404, + 1698, + 1404, + 1824, + 297, + 1824 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 301, + 1076, + 1399, + 1076, + 1399, + 1170, + 301, + 1170 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 294, + 229, + 1403, + 229, + 1403, + 293, + 294, + 293 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 295, + 1189, + 1402, + 1189, + 1402, + 1251, + 295, + 1251 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 433, + 868, + 1265, + 868, + 1265, + 954, + 433, + 954 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 294, + 307, + 1403, + 307, + 1403, + 370, + 294, + 370 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 381, + 1401, + 381, + 1401, + 454, + 297, + 454 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 317, + 1839, + 1370, + 1839, + 1370, + 1878, + 317, + 1878 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.909 + }, + { + "category_id": 8, + "poly": [ + 360, + 1267, + 1332, + 1267, + 1332, + 1308, + 360, + 1308 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1352, + 894, + 1399, + 894, + 1399, + 924, + 1352, + 924 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 300, + 1642, + 701, + 1642, + 701, + 1673, + 300, + 1673 + ], + "score": 0.889 + }, + { + "category_id": 0, + "poly": [ + 299, + 679, + 1105, + 679, + 1105, + 717, + 299, + 717 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.798 + }, + { + "category_id": 14, + "poly": [ + 435, + 865, + 1264, + 865, + 1264, + 957, + 435, + 957 + ], + "score": 0.93, + "latex": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { \\partial L ( X ; W ) } { \\partial W } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left[ y _ { i } - \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } _ { i } ) \\right] \\pmb { x } _ { i } [ \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } W ) \\circ \\pmb { a } ] ," + }, + { + "category_id": 13, + "poly": [ + 541, + 1497, + 895, + 1497, + 895, + 1538, + 541, + 1538 + ], + "score": 0.93, + "latex": "\\lVert W ( 0 ) - \\widehat { W } \\rVert _ { F } / \\lVert W ( 0 ) \\rVert _ { F } \\gg 1" + }, + { + "category_id": 13, + "poly": [ + 802, + 1106, + 975, + 1106, + 975, + 1140, + 802, + 1140 + ], + "score": 0.93, + "latex": "\\pmb { x } \\sim \\mathcal { N } ( 0 , I / d )" + }, + { + "category_id": 13, + "poly": [ + 623, + 1371, + 694, + 1371, + 694, + 1409, + 623, + 1409 + ], + "score": 0.92, + "latex": "1 / { \\sqrt { h } }" + }, + { + "category_id": 13, + "poly": [ + 430, + 812, + 728, + 812, + 728, + 851, + 430, + 851 + ], + "score": 0.92, + "latex": "a _ { i } \\sim \\operatorname { U n i f } \\{ - 1 / \\sqrt { h } , 1 \\sqrt { h } \\}" + }, + { + "category_id": 13, + "poly": [ + 1306, + 1001, + 1396, + 1001, + 1396, + 1036, + 1306, + 1036 + ], + "score": 0.92, + "latex": "W ^ { \\mathrm { i n i t } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1020, + 1569, + 1088, + 1569, + 1088, + 1604, + 1020, + 1604 + ], + "score": 0.92, + "latex": "W ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 1057, + 340, + 1144, + 340, + 1144, + 371, + 1057, + 371 + ], + "score": 0.91, + "latex": "\\gamma _ { 2 } 1" + }, + { + "category_id": 13, + "poly": [ + 335, + 1700, + 451, + 1700, + 451, + 1731, + 335, + 1731 + ], + "score": 0.91, + "latex": "d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 425, + 423, + 504, + 423, + 504, + 454, + 425, + 454 + ], + "score": 0.91, + "latex": "\\gamma _ { 2 } > 1" + }, + { + "category_id": 13, + "poly": [ + 399, + 232, + 477, + 232, + 477, + 259, + 399, + 259 + ], + "score": 0.9, + "latex": "h = n" + }, + { + "category_id": 13, + "poly": [ + 501, + 1269, + 813, + 1269, + 813, + 1306, + 501, + 1306 + ], + "score": 0.9, + "latex": "{ \\pmb w } _ { i } ^ { \\mathrm { V a n } } ( 0 ) \\sim \\mathcal { N } ( 0 , I _ { d } / d h ^ { 1 + \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 1067, + 482, + 1142, + 482, + 1142, + 509, + 1067, + 509 + ], + "score": 0.9, + "latex": "h = n" + }, + { + "category_id": 13, + "poly": [ + 815, + 515, + 922, + 515, + 922, + 542, + 815, + 542 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1030, + 236, + 1136, + 236, + 1136, + 262, + 1030, + 262 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 583, + 1080, + 673, + 1080, + 673, + 1105, + 583, + 1105 + ], + "score": 0.88, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 577, + 1033, + 610, + 1033, + 610, + 1068, + 577, + 1068 + ], + "score": 0.87, + "latex": "\\widehat { W }" + }, + { + "category_id": 13, + "poly": [ + 452, + 1566, + 485, + 1566, + 485, + 1598, + 452, + 1598 + ], + "score": 0.86, + "latex": "\\widehat { W }" + }, + { + "category_id": 13, + "poly": [ + 1042, + 1268, + 1324, + 1268, + 1324, + 1306, + 1042, + 1306 + ], + "score": 0.86, + "latex": "{ \\pmb w } _ { i } ^ { \\mathrm { N V } } ( 0 ) \\sim \\mathcal { N } ( 0 , I _ { d } / d ^ { - \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 1099, + 1913, + 1120, + 1913, + 1120, + 1944, + 1099, + 1944 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 963, + 391, + 1060, + 391, + 1060, + 419, + 963, + 419 + ], + "score": 0.85, + "latex": "B \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1094, + 393, + 1181, + 393, + 1181, + 422, + 1094, + 422 + ], + "score": 0.85, + "latex": "\\gamma _ { 2 } 1" + }, + { + "category_id": 13, + "poly": [ + 1006, + 1035, + 1025, + 1035, + 1025, + 1072, + 1006, + 1072 + ], + "score": 0.84, + "latex": "\\hat { f }" + }, + { + "category_id": 13, + "poly": [ + 640, + 236, + 671, + 236, + 671, + 263, + 640, + 263 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1235, + 313, + 1303, + 313, + 1303, + 341, + 1235, + 341 + ], + "score": 0.83, + "latex": "\\gamma _ { 1 } , \\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1223, + 1376, + 1243, + 1376, + 1243, + 1403, + 1223, + 1403 + ], + "score": 0.81, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 399, + 1845, + 419, + 1845, + 419, + 1875, + 399, + 1875 + ], + "score": 0.8, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 938, + 817, + 971, + 817, + 971, + 844, + 938, + 844 + ], + "score": 0.79, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 837, + 1006, + 850, + 1006, + 850, + 1030, + 837, + 1030 + ], + "score": 0.79, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 871, + 1838, + 1370, + 1838, + 1370, + 1878, + 871, + 1878 + ], + "score": 0.78, + "latex": "\\phi ^ { \\prime } ( 0 ) \\neq 0 ; | \\phi ^ { \\prime } ( \\pm x ) - \\phi ^ { \\prime } ( \\pm \\infty ) | = O ( e ^ { - x } ) ." + }, + { + "category_id": 13, + "poly": [ + 795, + 387, + 896, + 387, + 896, + 424, + 795, + 424 + ], + "score": 0.76, + "latex": "\\mathcal { N } ( 0 , I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 1350, + 392, + 1375, + 392, + 1375, + 418, + 1350, + 418 + ], + "score": 0.75, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 702, + 394, + 739, + 394, + 739, + 421, + 702, + 421 + ], + "score": 0.48, + "latex": "{ \\pmb w } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 875, + 1840, + 988, + 1840, + 988, + 1877, + 875, + 1877 + ], + "score": 0.4, + "latex": "\\phi ^ { \\prime } ( 0 ) \\neq 0" + }, + { + "category_id": 13, + "poly": [ + 548, + 390, + 650, + 390, + 650, + 422, + 548, + 422 + ], + "score": 0.27, + "latex": "( A I ) ( A 2 )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1639.0, + 703.0, + 1639.0, + 703.0, + 1678.0, + 294.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 679.0, + 1107.0, + 679.0, + 1107.0, + 720.0, + 293.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1337.0, + 1406.0, + 1337.0, + 1406.0, + 1379.0, + 292.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1372.0, + 622.0, + 1372.0, + 622.0, + 1412.0, + 292.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1372.0, + 1222.0, + 1372.0, + 1222.0, + 1412.0, + 695.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1372.0, + 1406.0, + 1372.0, + 1406.0, + 1412.0, + 1244.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1405.0, + 1408.0, + 1405.0, + 1408.0, + 1441.0, + 295.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 1407.0, + 1436.0, + 1407.0, + 1471.0, + 294.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1462.0, + 1406.0, + 1462.0, + 1406.0, + 1504.0, + 292.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1500.0, + 540.0, + 1500.0, + 540.0, + 1540.0, + 294.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1500.0, + 1404.0, + 1500.0, + 1404.0, + 1540.0, + 896.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1535.0, + 1404.0, + 1535.0, + 1404.0, + 1570.0, + 295.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1568.0, + 451.0, + 1568.0, + 451.0, + 1606.0, + 296.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1568.0, + 1019.0, + 1568.0, + 1019.0, + 1606.0, + 486.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1568.0, + 1098.0, + 1568.0, + 1098.0, + 1606.0, + 1089.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 478.0, + 1066.0, + 478.0, + 1066.0, + 514.0, + 295.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 478.0, + 1405.0, + 478.0, + 1405.0, + 514.0, + 1143.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 509.0, + 814.0, + 509.0, + 814.0, + 546.0, + 295.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 509.0, + 1406.0, + 509.0, + 1406.0, + 546.0, + 923.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 541.0, + 1407.0, + 541.0, + 1407.0, + 575.0, + 291.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 571.0, + 1405.0, + 571.0, + 1405.0, + 604.0, + 296.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 602.0, + 1255.0, + 602.0, + 1255.0, + 635.0, + 296.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 751.0, + 1403.0, + 751.0, + 1403.0, + 786.0, + 296.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 781.0, + 1407.0, + 781.0, + 1407.0, + 817.0, + 292.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 814.0, + 429.0, + 814.0, + 429.0, + 854.0, + 293.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 814.0, + 937.0, + 814.0, + 937.0, + 854.0, + 729.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 814.0, + 1405.0, + 814.0, + 1405.0, + 854.0, + 972.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 970.0, + 1408.0, + 970.0, + 1408.0, + 1006.0, + 295.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 997.0, + 836.0, + 997.0, + 836.0, + 1040.0, + 294.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 997.0, + 1305.0, + 997.0, + 1305.0, + 1040.0, + 851.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 997.0, + 1409.0, + 997.0, + 1409.0, + 1040.0, + 1397.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1034.0, + 576.0, + 1034.0, + 576.0, + 1077.0, + 293.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1034.0, + 1005.0, + 1034.0, + 1005.0, + 1077.0, + 611.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1034.0, + 1039.0, + 1034.0, + 1039.0, + 1077.0, + 1026.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1908.0, + 1098.0, + 1908.0, + 1098.0, + 1948.0, + 292.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1908.0, + 1408.0, + 1908.0, + 1408.0, + 1948.0, + 1121.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1409.0, + 1942.0, + 1409.0, + 1977.0, + 292.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2002.0, + 946.0, + 2002.0, + 946.0, + 2037.0, + 292.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1700.0, + 334.0, + 1700.0, + 334.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1700.0, + 1404.0, + 1700.0, + 1404.0, + 1733.0, + 452.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1730.0, + 1405.0, + 1730.0, + 1405.0, + 1766.0, + 292.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1762.0, + 1404.0, + 1762.0, + 1404.0, + 1795.0, + 294.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1793.0, + 1203.0, + 1793.0, + 1203.0, + 1826.0, + 295.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1073.0, + 582.0, + 1073.0, + 582.0, + 1112.0, + 295.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1073.0, + 1404.0, + 1073.0, + 1404.0, + 1112.0, + 674.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1104.0, + 801.0, + 1104.0, + 801.0, + 1142.0, + 293.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1104.0, + 1407.0, + 1104.0, + 1407.0, + 1142.0, + 976.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1135.0, + 1016.0, + 1135.0, + 1016.0, + 1173.0, + 295.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 398.0, + 229.0, + 398.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 229.0, + 639.0, + 229.0, + 639.0, + 265.0, + 478.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 229.0, + 1029.0, + 229.0, + 1029.0, + 265.0, + 672.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 1137.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 867.0, + 261.0, + 867.0, + 296.0, + 294.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1188.0, + 1405.0, + 1188.0, + 1405.0, + 1224.0, + 295.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1220.0, + 695.0, + 1220.0, + 695.0, + 1254.0, + 294.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 307.0, + 1234.0, + 307.0, + 1234.0, + 342.0, + 296.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 307.0, + 1406.0, + 307.0, + 1406.0, + 342.0, + 1304.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 337.0, + 1056.0, + 337.0, + 1056.0, + 372.0, + 293.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 337.0, + 1157.0, + 337.0, + 1157.0, + 372.0, + 1145.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 369.0, + 547.0, + 369.0, + 547.0, + 436.0, + 286.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 369.0, + 701.0, + 369.0, + 701.0, + 436.0, + 651.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 369.0, + 794.0, + 369.0, + 794.0, + 436.0, + 740.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 369.0, + 962.0, + 369.0, + 962.0, + 436.0, + 897.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 369.0, + 1093.0, + 369.0, + 1093.0, + 436.0, + 1061.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 369.0, + 1349.0, + 369.0, + 1349.0, + 436.0, + 1182.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 369.0, + 1415.0, + 369.0, + 1415.0, + 436.0, + 1376.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 420.0, + 424.0, + 420.0, + 424.0, + 455.0, + 292.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 420.0, + 516.0, + 420.0, + 516.0, + 455.0, + 505.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1838.0, + 398.0, + 1838.0, + 398.0, + 1881.0, + 325.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1838.0, + 870.0, + 1838.0, + 870.0, + 1881.0, + 420.0, + 1881.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 435, + 1405, + 435, + 1405, + 622, + 297, + 622 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 712, + 1404, + 712, + 1404, + 866, + 297, + 866 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 920, + 1404, + 920, + 1404, + 1076, + 297, + 1076 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1622, + 1404, + 1622, + 1404, + 1747, + 299, + 1747 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1090, + 1405, + 1090, + 1405, + 1183, + 297, + 1183 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 225, + 1406, + 225, + 1406, + 323, + 297, + 323 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 455, + 1383, + 1243, + 1383, + 1243, + 1564, + 455, + 1564 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 299, + 1197, + 1403, + 1197, + 1403, + 1290, + 299, + 1290 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 369, + 1766, + 1405, + 1766, + 1405, + 2024, + 369, + 2024 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 575, + 326, + 1122, + 326, + 1122, + 403, + 575, + 403 + ], + "score": 0.954 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 817, + 74, + 817, + 105, + 299, + 105 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 300, + 655, + 762, + 655, + 762, + 687, + 300, + 687 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1352, + 877, + 1399, + 877, + 1399, + 907, + 1352, + 907 + ], + "score": 0.892 + }, + { + "category_id": 8, + "poly": [ + 333, + 871, + 1307, + 871, + 1307, + 914, + 333, + 914 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 291, + 1343, + 1356, + 1343, + 1356, + 1377, + 291, + 1377 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 390, + 1296, + 1298, + 1296, + 1298, + 1335, + 390, + 1335 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1352, + 347, + 1400, + 347, + 1400, + 377, + 1352, + 377 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1504, + 1400, + 1504, + 1400, + 1534, + 1352, + 1534 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 300, + 1567, + 1149, + 1567, + 1149, + 1606, + 300, + 1606 + ], + "score": 0.817 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.765 + }, + { + "category_id": 13, + "poly": [ + 574, + 1799, + 678, + 1799, + 678, + 1832, + 574, + 1832 + ], + "score": 0.93, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( K )" + }, + { + "category_id": 14, + "poly": [ + 575, + 324, + 1124, + 324, + 1124, + 402, + 575, + 402 + ], + "score": 0.93, + "latex": "R ( \\hat { f } ) \\operatorname* { m a x } \\{ 0 , \\frac { \\gamma _ { 1 } - 1 } { \\gamma _ { 1 } } \\} r ^ { 2 } + \\frac { \\operatorname* { m i n } \\{ \\gamma _ { 1 } , 1 \\} } { | 1 - \\gamma _ { 1 } | } \\sigma ^ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 1055, + 1228, + 1181, + 1228, + 1181, + 1262, + 1055, + 1262 + ], + "score": 0.93, + "latex": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0" + }, + { + "category_id": 13, + "poly": [ + 763, + 1989, + 854, + 1989, + 854, + 2023, + 763, + 2023 + ], + "score": 0.93, + "latex": "f ^ { \\mathrm { i n i t } } ( X )" + }, + { + "category_id": 13, + "poly": [ + 369, + 918, + 534, + 918, + 534, + 950, + 369, + 950 + ], + "score": 0.93, + "latex": "J \\in \\mathbb { R } ^ { n \\times ( d \\times h ) }" + }, + { + "category_id": 13, + "poly": [ + 1176, + 433, + 1290, + 433, + 1290, + 469, + 1176, + 469 + ], + "score": 0.92, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }" + }, + { + "category_id": 13, + "poly": [ + 681, + 1346, + 825, + 1346, + 825, + 1377, + 681, + 1377 + ], + "score": 0.92, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1311, + 1898, + 1403, + 1898, + 1403, + 1932, + 1311, + 1932 + ], + "score": 0.92, + "latex": "f ^ { \\mathrm { i n i t } } ( X )" + }, + { + "category_id": 13, + "poly": [ + 909, + 1683, + 998, + 1683, + 998, + 1718, + 909, + 1718 + ], + "score": 0.92, + "latex": "( J J ^ { \\top } ) ^ { \\dagger }" + }, + { + "category_id": 13, + "poly": [ + 908, + 226, + 1153, + 226, + 1153, + 264, + 908, + 264 + ], + "score": 0.92, + "latex": "\\hat { f } ( \\cdot ) = f ^ { \\nu a n } ( \\cdot , W ( T ) )" + }, + { + "category_id": 13, + "poly": [ + 712, + 1770, + 799, + 1770, + 799, + 1800, + 712, + 1800 + ], + "score": 0.92, + "latex": "\\gamma _ { 1 } 1" + }, + { + "category_id": 13, + "poly": [ + 654, + 229, + 853, + 229, + 853, + 264, + 654, + 264 + ], + "score": 0.92, + "latex": "T = O ( \\log \\log h )" + }, + { + "category_id": 13, + "poly": [ + 1107, + 261, + 1366, + 261, + 1366, + 296, + 1107, + 296 + ], + "score": 0.91, + "latex": "\\| \\partial W ( T ) / \\partial t \\| _ { F } \\in o ( 1 ) ," + }, + { + "category_id": 13, + "poly": [ + 1203, + 1196, + 1294, + 1196, + 1294, + 1232, + 1203, + 1232 + ], + "score": 0.91, + "latex": "f ^ { \\mathrm { i n i t } } ( X )" + }, + { + "category_id": 13, + "poly": [ + 468, + 805, + 874, + 805, + 874, + 839, + 468, + 839 + ], + "score": 0.89, + "latex": "k ( { \\pmb x } , { \\pmb y } ) = \\langle \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb x } ) , \\nabla _ { W ^ { \\mathrm { i n i t } } } f ( { \\pmb y } ) \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 1716, + 391, + 1716, + 391, + 1745, + 298, + 1745 + ], + "score": 0.89, + "latex": "d h \\gg n" + }, + { + "category_id": 14, + "poly": [ + 338, + 869, + 1310, + 869, + 1310, + 913, + 338, + 913 + ], + "score": 0.88, + "latex": "\\mathrm { v e c } ( W ^ { * } ) \\approx \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) + \\Delta ; \\quad \\Delta = J ^ { \\dagger } ( { \\pmb y } - f ^ { \\mathrm { i n i t } } ( { \\pmb X } ) ) ; \\quad J _ { [ i , j ] } = \\nabla _ { \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } ) _ { j } } f ^ { \\mathrm { i n i t } } ( { \\pmb x } _ { i } ) ," + }, + { + "category_id": 13, + "poly": [ + 601, + 262, + 652, + 262, + 652, + 294, + 601, + 294 + ], + "score": 0.87, + "latex": "o ( 1 )" + }, + { + "category_id": 14, + "poly": [ + 453, + 1381, + 1247, + 1381, + 1247, + 1567, + 453, + 1567 + ], + "score": 0.87, + "latex": "\\begin{array} { c } { { R ( \\hat { f } ) ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) r ^ { 2 } } } \\\\ { { + ( \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } - \\frac { 1 } { 4 } ) \\sigma ^ { 2 } , } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1253, + 232, + 1398, + 232, + 1398, + 262, + 1253, + 262 + ], + "score": 0.87, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1076, + 1295, + 1289, + 1295, + 1289, + 1336, + 1076, + 1336 + ], + "score": 0.86, + "latex": "a _ { i } \\mathbf { w } _ { i } ^ { \\mathrm { i n i t } } = - a _ { j } \\mathbf { w } _ { j } ^ { \\mathrm { i n i t } }" + }, + { + "category_id": 13, + "poly": [ + 1068, + 1873, + 1098, + 1873, + 1098, + 1901, + 1068, + 1901 + ], + "score": 0.86, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 876, + 1630, + 906, + 1630, + 906, + 1656, + 876, + 1656 + ], + "score": 0.86, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 765, + 1298, + 886, + 1298, + 886, + 1333, + 765, + 1333 + ], + "score": 0.86, + "latex": "\\forall i \\in [ 1 , h ]" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1156, + 1398, + 1156, + 1398, + 1183, + 1368, + 1183 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1368, + 595, + 1398, + 595, + 1398, + 622, + 1368, + 622 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1092, + 1341, + 1112, + 1341, + 1112, + 1378, + 1092, + 1378 + ], + "score": 0.84, + "latex": "\\hat { f }" + }, + { + "category_id": 13, + "poly": [ + 508, + 1568, + 685, + 1568, + 685, + 1605, + 508, + 1605 + ], + "score": 0.84, + "latex": "b _ { 0 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ] ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 938, + 471, + 969, + 471, + 969, + 500, + 938, + 500 + ], + "score": 0.83, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1377, + 1091, + 1404, + 1091, + 1404, + 1118, + 1377, + 1118 + ], + "score": 0.79, + "latex": "\\Delta" + }, + { + "category_id": 13, + "poly": [ + 919, + 1298, + 1026, + 1298, + 1026, + 1333, + 919, + 1333 + ], + "score": 0.79, + "latex": "j \\in [ 1 , h ]" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1965, + 1348, + 1965, + 1348, + 1991, + 1327, + 1991 + ], + "score": 0.78, + "latex": "\\textbf { { y } }" + }, + { + "category_id": 13, + "poly": [ + 746, + 1568, + 980, + 1568, + 980, + 1605, + 746, + 1605 + ], + "score": 0.74, + "latex": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1568, + 496, + 1568, + 496, + 1605, + 371, + 1605 + ], + "score": 0.64, + "latex": "m = b _ { 1 } ^ { 2 } / b _ { 0 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1037, + 265, + 1058, + 265, + 1058, + 290, + 1037, + 290 + ], + "score": 0.63, + "latex": "T ," + }, + { + "category_id": 13, + "poly": [ + 961, + 561, + 984, + 561, + 984, + 587, + 961, + 587 + ], + "score": 0.45, + "latex": "\\mathrm { E }" + }, + { + "category_id": 13, + "poly": [ + 993, + 1569, + 1141, + 1569, + 1141, + 1604, + 993, + 1604 + ], + "score": 0.27, + "latex": "G \\sim \\mathcal { N } ( 0 , 1 )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 651.0, + 767.0, + 651.0, + 767.0, + 693.0, + 293.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 435.0, + 1175.0, + 435.0, + 1175.0, + 471.0, + 295.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 435.0, + 1405.0, + 435.0, + 1405.0, + 471.0, + 1291.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 467.0, + 937.0, + 467.0, + 937.0, + 503.0, + 295.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 467.0, + 1403.0, + 467.0, + 1403.0, + 503.0, + 970.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 498.0, + 1406.0, + 498.0, + 1406.0, + 532.0, + 292.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 526.0, + 1405.0, + 526.0, + 1405.0, + 564.0, + 292.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 559.0, + 960.0, + 559.0, + 960.0, + 591.0, + 295.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 559.0, + 1403.0, + 559.0, + 1403.0, + 591.0, + 985.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 586.0, + 1367.0, + 586.0, + 1367.0, + 628.0, + 292.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 586.0, + 1410.0, + 586.0, + 1410.0, + 628.0, + 1399.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 712.0, + 1405.0, + 712.0, + 1405.0, + 749.0, + 295.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 741.0, + 1406.0, + 741.0, + 1406.0, + 780.0, + 292.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 772.0, + 1405.0, + 772.0, + 1405.0, + 810.0, + 292.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 805.0, + 467.0, + 805.0, + 467.0, + 842.0, + 295.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 805.0, + 1405.0, + 805.0, + 1405.0, + 842.0, + 875.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 837.0, + 756.0, + 837.0, + 756.0, + 870.0, + 296.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 915.0, + 368.0, + 915.0, + 368.0, + 957.0, + 294.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 915.0, + 1407.0, + 915.0, + 1407.0, + 957.0, + 535.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 950.0, + 1404.0, + 950.0, + 1404.0, + 987.0, + 295.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 981.0, + 1406.0, + 981.0, + 1406.0, + 1018.0, + 295.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1011.0, + 1405.0, + 1011.0, + 1405.0, + 1047.0, + 294.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1041.0, + 1156.0, + 1041.0, + 1156.0, + 1079.0, + 294.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1624.0, + 875.0, + 1624.0, + 875.0, + 1656.0, + 295.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1624.0, + 1404.0, + 1624.0, + 1404.0, + 1656.0, + 907.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1655.0, + 1404.0, + 1655.0, + 1404.0, + 1688.0, + 295.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1683.0, + 908.0, + 1683.0, + 908.0, + 1720.0, + 294.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1683.0, + 1405.0, + 1683.0, + 1405.0, + 1720.0, + 999.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1714.0, + 297.0, + 1714.0, + 297.0, + 1749.0, + 294.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1714.0, + 831.0, + 1714.0, + 831.0, + 1749.0, + 392.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1088.0, + 1376.0, + 1088.0, + 1376.0, + 1124.0, + 294.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1152.0, + 296.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1144.0, + 1367.0, + 1144.0, + 1367.0, + 1191.0, + 291.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1144.0, + 1411.0, + 1144.0, + 1411.0, + 1191.0, + 1399.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 222.0, + 653.0, + 222.0, + 653.0, + 268.0, + 291.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 222.0, + 907.0, + 222.0, + 907.0, + 268.0, + 854.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 222.0, + 1252.0, + 222.0, + 1252.0, + 268.0, + 1154.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 222.0, + 1409.0, + 222.0, + 1409.0, + 268.0, + 1399.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 258.0, + 600.0, + 258.0, + 600.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 258.0, + 1036.0, + 258.0, + 1036.0, + 296.0, + 653.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 258.0, + 1106.0, + 258.0, + 1106.0, + 296.0, + 1059.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 258.0, + 1406.0, + 258.0, + 1406.0, + 296.0, + 1367.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 290.0, + 775.0, + 290.0, + 775.0, + 327.0, + 295.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1192.0, + 1202.0, + 1192.0, + 1202.0, + 1234.0, + 292.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 1192.0, + 1405.0, + 1192.0, + 1405.0, + 1234.0, + 1295.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1222.0, + 1054.0, + 1222.0, + 1054.0, + 1263.0, + 293.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1222.0, + 1404.0, + 1222.0, + 1404.0, + 1263.0, + 1182.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1257.0, + 930.0, + 1257.0, + 930.0, + 1294.0, + 294.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1767.0, + 711.0, + 1767.0, + 711.0, + 1806.0, + 368.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1767.0, + 1405.0, + 1767.0, + 1405.0, + 1806.0, + 800.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1796.0, + 573.0, + 1796.0, + 573.0, + 1837.0, + 391.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1796.0, + 1404.0, + 1796.0, + 1404.0, + 1837.0, + 679.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1838.0, + 1405.0, + 1838.0, + 1405.0, + 1873.0, + 383.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1869.0, + 1067.0, + 1869.0, + 1067.0, + 1904.0, + 395.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1869.0, + 1405.0, + 1869.0, + 1405.0, + 1904.0, + 1099.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1894.0, + 1310.0, + 1894.0, + 1310.0, + 1936.0, + 392.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1894.0, + 1409.0, + 1894.0, + 1409.0, + 1936.0, + 1404.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1929.0, + 1405.0, + 1929.0, + 1405.0, + 1962.0, + 392.0, + 1962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1960.0, + 1326.0, + 1960.0, + 1326.0, + 1994.0, + 394.0, + 1994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1960.0, + 1406.0, + 1960.0, + 1406.0, + 1994.0, + 1349.0, + 1994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1987.0, + 762.0, + 1987.0, + 762.0, + 2026.0, + 394.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1987.0, + 1404.0, + 1987.0, + 1404.0, + 2026.0, + 855.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1340.0, + 680.0, + 1340.0, + 680.0, + 1384.0, + 294.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1340.0, + 1091.0, + 1340.0, + 1091.0, + 1384.0, + 826.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1340.0, + 1361.0, + 1340.0, + 1361.0, + 1384.0, + 1113.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1282.0, + 764.0, + 1282.0, + 764.0, + 1347.0, + 394.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1282.0, + 918.0, + 1282.0, + 918.0, + 1347.0, + 887.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1282.0, + 1075.0, + 1282.0, + 1075.0, + 1347.0, + 1027.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 1282.0, + 1303.0, + 1282.0, + 1303.0, + 1347.0, + 1290.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1562.0, + 370.0, + 1562.0, + 370.0, + 1611.0, + 294.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1562.0, + 507.0, + 1562.0, + 507.0, + 1611.0, + 497.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1562.0, + 745.0, + 1562.0, + 745.0, + 1611.0, + 686.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1562.0, + 992.0, + 1562.0, + 992.0, + 1611.0, + 981.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1562.0, + 1152.0, + 1562.0, + 1152.0, + 1611.0, + 1142.0, + 1611.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1261, + 1405, + 1261, + 1405, + 1508, + 298, + 1508 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1619, + 1404, + 1619, + 1404, + 1805, + 299, + 1805 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1820, + 1402, + 1820, + 1402, + 2034, + 298, + 2034 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1093, + 1403, + 1093, + 1403, + 1246, + 298, + 1246 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 315, + 226, + 1374, + 226, + 1374, + 638, + 315, + 638 + ], + "score": 0.971 + }, + { + "category_id": 4, + "poly": [ + 297, + 660, + 1405, + 660, + 1405, + 745, + 297, + 745 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 299, + 858, + 1400, + 858, + 1400, + 953, + 299, + 953 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 298, + 961, + 1401, + 961, + 1401, + 1035, + 298, + 1035 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 298, + 1041, + 1203, + 1041, + 1203, + 1075, + 298, + 1075 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 301, + 1550, + 844, + 1550, + 844, + 1586, + 301, + 1586 + ], + "score": 0.908 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.889 + }, + { + "category_id": 0, + "poly": [ + 302, + 802, + 778, + 802, + 778, + 832, + 302, + 832 + ], + "score": 0.837 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 858, + 2089, + 858, + 2112, + 840, + 2112 + ], + "score": 0.792 + }, + { + "category_id": 13, + "poly": [ + 911, + 960, + 1181, + 960, + 1181, + 999, + 911, + 999 + ], + "score": 0.94, + "latex": "B ( \\hat { f } ^ { V a n } ) \\le B ( \\hat { f } ^ { N V } ) \\le 1 ." + }, + { + "category_id": 13, + "poly": [ + 755, + 997, + 847, + 997, + 847, + 1036, + 755, + 1036 + ], + "score": 0.93, + "latex": "V ( \\hat { f } ^ { V a n } )" + }, + { + "category_id": 13, + "poly": [ + 546, + 965, + 685, + 965, + 685, + 999, + 546, + 999 + ], + "score": 0.92, + "latex": "\\gamma _ { 1 } \\in ( 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 461, + 998, + 644, + 998, + 644, + 1036, + 461, + 1036 + ], + "score": 0.91, + "latex": "V ( \\hat { f } ^ { N V } ) = { \\cal { O } } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 407, + 1043, + 487, + 1043, + 487, + 1073, + 407, + 1073 + ], + "score": 0.9, + "latex": "m \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 1181, + 1852, + 1284, + 1852, + 1284, + 1882, + 1181, + 1882 + ], + "score": 0.9, + "latex": "\\gamma _ { 2 } ~ ~ 1" + }, + { + "category_id": 13, + "poly": [ + 812, + 1975, + 899, + 1975, + 899, + 2005, + 812, + 2005 + ], + "score": 0.89, + "latex": "\\gamma _ { 2 } 0" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1003, + 1237, + 1003, + 1237, + 1034, + 1151, + 1034 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } 1" + }, + { + "category_id": 13, + "poly": [ + 373, + 1001, + 450, + 1001, + 450, + 1032, + 373, + 1032 + ], + "score": 0.87, + "latex": "m > 0" + }, + { + "category_id": 13, + "poly": [ + 719, + 1852, + 747, + 1852, + 747, + 1882, + 719, + 1882 + ], + "score": 0.86, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 937, + 891, + 983, + 891, + 983, + 922, + 937, + 922 + ], + "score": 0.86, + "latex": "( \\gamma _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1082, + 1043, + 1102, + 1043, + 1102, + 1074, + 1082, + 1074 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 645, + 691, + 673, + 691, + 673, + 717, + 645, + 717 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 778, + 1043, + 797, + 1043, + 797, + 1074, + 778, + 1074 + ], + "score": 0.83, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 881, + 965, + 901, + 965, + 901, + 998, + 881, + 998 + ], + "score": 0.6, + "latex": "\\phi" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 228.0, + 374.0, + 228.0, + 374.0, + 261.0, + 331.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 250.0, + 413.0, + 250.0, + 413.0, + 263.0, + 397.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 241.0, + 613.0, + 241.0, + 613.0, + 272.0, + 427.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 231.0, + 918.0, + 231.0, + 918.0, + 260.0, + 878.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 241.0, + 1319.0, + 241.0, + 1319.0, + 272.0, + 1133.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 279.0, + 374.0, + 279.0, + 374.0, + 313.0, + 331.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 277.0, + 407.0, + 277.0, + 407.0, + 287.0, + 398.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 266.0, + 654.0, + 266.0, + 654.0, + 322.0, + 424.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 274.0, + 918.0, + 274.0, + 918.0, + 306.0, + 877.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 266.0, + 1361.0, + 266.0, + 1361.0, + 320.0, + 1128.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 319.0, + 919.0, + 319.0, + 919.0, + 351.0, + 878.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 332.0, + 375.0, + 332.0, + 375.0, + 364.0, + 331.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 370.0, + 374.0, + 370.0, + 374.0, + 427.0, + 314.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 352.0, + 918.0, + 352.0, + 918.0, + 444.0, + 858.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 409.0, + 918.0, + 409.0, + 918.0, + 441.0, + 876.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 436.0, + 375.0, + 436.0, + 375.0, + 468.0, + 331.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 454.0, + 919.0, + 454.0, + 919.0, + 486.0, + 877.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 487.0, + 375.0, + 487.0, + 375.0, + 519.0, + 331.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 500.0, + 919.0, + 500.0, + 919.0, + 532.0, + 877.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 540.0, + 376.0, + 540.0, + 376.0, + 571.0, + 331.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 543.0, + 921.0, + 543.0, + 921.0, + 575.0, + 877.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 562.0, + 424.0, + 562.0, + 424.0, + 592.0, + 369.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 562.0, + 486.0, + 562.0, + 486.0, + 593.0, + 433.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 562.0, + 550.0, + 562.0, + 550.0, + 592.0, + 496.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 562.0, + 611.0, + 562.0, + 611.0, + 592.0, + 558.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 565.0, + 674.0, + 565.0, + 674.0, + 591.0, + 622.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 565.0, + 738.0, + 565.0, + 738.0, + 591.0, + 686.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 565.0, + 799.0, + 565.0, + 799.0, + 591.0, + 748.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 562.0, + 969.0, + 562.0, + 969.0, + 592.0, + 917.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 562.0, + 1032.0, + 562.0, + 1032.0, + 592.0, + 979.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 565.0, + 1093.0, + 565.0, + 1093.0, + 591.0, + 1044.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 562.0, + 1158.0, + 562.0, + 1158.0, + 592.0, + 1104.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 565.0, + 1220.0, + 565.0, + 1220.0, + 591.0, + 1168.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 565.0, + 1283.0, + 565.0, + 1283.0, + 591.0, + 1233.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 565.0, + 1344.0, + 565.0, + 1344.0, + 591.0, + 1295.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 582.0, + 638.0, + 582.0, + 638.0, + 615.0, + 557.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 582.0, + 1185.0, + 582.0, + 1185.0, + 615.0, + 1103.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 604.0, + 625.0, + 604.0, + 625.0, + 645.0, + 527.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 607.0, + 1191.0, + 607.0, + 1191.0, + 642.0, + 1051.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 272.5, + 1145.0, + 272.5, + 1145.0, + 321.5, + 1099.0, + 321.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 328.5, + 858.0, + 328.5, + 858.0, + 393.5, + 507.0, + 393.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 657.0, + 1406.0, + 657.0, + 1406.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 687.0, + 644.0, + 687.0, + 644.0, + 716.0, + 296.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 687.0, + 1403.0, + 687.0, + 1403.0, + 716.0, + 674.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 714.0, + 972.0, + 714.0, + 972.0, + 745.0, + 294.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1547.0, + 848.0, + 1547.0, + 848.0, + 1592.0, + 292.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 798.0, + 782.0, + 798.0, + 782.0, + 838.0, + 293.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2117.0, + 838.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1260.0, + 1405.0, + 1260.0, + 1405.0, + 1297.0, + 293.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1294.0, + 1406.0, + 1294.0, + 1406.0, + 1328.0, + 296.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1324.0, + 1406.0, + 1324.0, + 1406.0, + 1358.0, + 296.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1354.0, + 1407.0, + 1354.0, + 1407.0, + 1388.0, + 293.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1382.0, + 1407.0, + 1382.0, + 1407.0, + 1421.0, + 292.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1450.0, + 295.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1446.0, + 1405.0, + 1446.0, + 1405.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1477.0, + 755.0, + 1477.0, + 755.0, + 1511.0, + 296.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1615.0, + 1408.0, + 1615.0, + 1408.0, + 1658.0, + 292.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 1404.0, + 1653.0, + 1404.0, + 1685.0, + 295.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1679.0, + 1408.0, + 1679.0, + 1408.0, + 1717.0, + 293.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1745.0, + 295.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1743.0, + 1406.0, + 1743.0, + 1406.0, + 1776.0, + 293.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1773.0, + 1110.0, + 1773.0, + 1110.0, + 1808.0, + 296.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1406.0, + 1818.0, + 1406.0, + 1856.0, + 293.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1849.0, + 718.0, + 1849.0, + 718.0, + 1885.0, + 292.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1849.0, + 1180.0, + 1849.0, + 1180.0, + 1885.0, + 748.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1885.0, + 1285.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1404.0, + 1880.0, + 1404.0, + 1915.0, + 294.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1403.0, + 1911.0, + 1403.0, + 1945.0, + 294.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 811.0, + 1972.0, + 811.0, + 2004.0, + 293.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2004.0, + 900.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2001.0, + 1406.0, + 2001.0, + 1406.0, + 2037.0, + 292.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1093.0, + 1404.0, + 1093.0, + 1404.0, + 1129.0, + 294.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1160.0, + 293.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1155.0, + 1404.0, + 1155.0, + 1404.0, + 1188.0, + 293.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1183.0, + 1405.0, + 1183.0, + 1405.0, + 1222.0, + 293.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1216.0, + 815.0, + 1216.0, + 815.0, + 1249.0, + 296.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 858.0, + 1406.0, + 858.0, + 1406.0, + 895.0, + 293.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 888.0, + 936.0, + 888.0, + 936.0, + 926.0, + 294.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 888.0, + 1406.0, + 888.0, + 1406.0, + 926.0, + 984.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 920.0, + 1250.0, + 920.0, + 1250.0, + 955.0, + 293.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 957.0, + 545.0, + 957.0, + 545.0, + 1003.0, + 293.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 957.0, + 880.0, + 957.0, + 880.0, + 1003.0, + 686.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 957.0, + 910.0, + 957.0, + 910.0, + 1003.0, + 902.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 957.0, + 1408.0, + 957.0, + 1408.0, + 1003.0, + 1182.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 994.0, + 372.0, + 994.0, + 372.0, + 1040.0, + 292.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 994.0, + 460.0, + 994.0, + 460.0, + 1040.0, + 451.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 994.0, + 754.0, + 994.0, + 754.0, + 1040.0, + 645.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 994.0, + 1150.0, + 994.0, + 1150.0, + 1040.0, + 848.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 994.0, + 1250.0, + 994.0, + 1250.0, + 1040.0, + 1238.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1038.0, + 406.0, + 1038.0, + 406.0, + 1079.0, + 292.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1038.0, + 777.0, + 1038.0, + 777.0, + 1079.0, + 488.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1038.0, + 1081.0, + 1038.0, + 1081.0, + 1079.0, + 798.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1038.0, + 1206.0, + 1038.0, + 1206.0, + 1079.0, + 1103.0, + 1079.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 309, + 1405, + 309, + 1405, + 613, + 298, + 613 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 710, + 1403, + 710, + 1403, + 862, + 299, + 862 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 294, + 229, + 1400, + 229, + 1400, + 292, + 294, + 292 + ], + "score": 0.946 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 816, + 76, + 816, + 104, + 300, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 300, + 947, + 487, + 947, + 487, + 981, + 300, + 981 + ], + "score": 0.835 + }, + { + "category_id": 0, + "poly": [ + 301, + 655, + 558, + 655, + 558, + 682, + 301, + 682 + ], + "score": 0.819 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 292, + 987, + 1408, + 987, + 1408, + 2036, + 292, + 2036 + ], + "score": 0.614 + }, + { + "category_id": 13, + "poly": [ + 467, + 429, + 515, + 429, + 515, + 463, + 467, + 463 + ], + "score": 0.86, + "latex": "1 / h" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 947.0, + 491.0, + 947.0, + 491.0, + 984.0, + 297.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 656.0, + 561.0, + 656.0, + 561.0, + 684.0, + 299.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 307.0, + 1408.0, + 307.0, + 1408.0, + 342.0, + 295.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 337.0, + 1407.0, + 337.0, + 1407.0, + 373.0, + 295.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 366.0, + 1406.0, + 366.0, + 1406.0, + 402.0, + 295.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 400.0, + 1406.0, + 400.0, + 1406.0, + 432.0, + 295.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 431.0, + 466.0, + 431.0, + 466.0, + 462.0, + 296.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 431.0, + 1407.0, + 431.0, + 1407.0, + 462.0, + 516.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 460.0, + 1407.0, + 460.0, + 1407.0, + 495.0, + 295.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 490.0, + 1407.0, + 490.0, + 1407.0, + 525.0, + 295.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 523.0, + 1402.0, + 523.0, + 1402.0, + 554.0, + 296.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 549.0, + 1406.0, + 549.0, + 1406.0, + 587.0, + 292.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 581.0, + 897.0, + 581.0, + 897.0, + 617.0, + 292.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 709.0, + 1407.0, + 709.0, + 1407.0, + 749.0, + 294.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 743.0, + 1404.0, + 743.0, + 1404.0, + 776.0, + 295.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 772.0, + 1403.0, + 772.0, + 1403.0, + 808.0, + 294.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 799.0, + 1407.0, + 799.0, + 1407.0, + 838.0, + 292.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 832.0, + 445.0, + 832.0, + 445.0, + 866.0, + 294.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 1267.0, + 262.0, + 1267.0, + 296.0, + 294.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 997.0, + 1407.0, + 997.0, + 1407.0, + 1041.0, + 293.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1032.0, + 893.0, + 1032.0, + 893.0, + 1068.0, + 323.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1084.0, + 1405.0, + 1084.0, + 1405.0, + 1124.0, + 293.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1117.0, + 706.0, + 1117.0, + 706.0, + 1151.0, + 319.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1169.0, + 1407.0, + 1169.0, + 1407.0, + 1210.0, + 291.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1201.0, + 1268.0, + 1201.0, + 1268.0, + 1241.0, + 322.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1252.0, + 1410.0, + 1252.0, + 1410.0, + 1299.0, + 290.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1290.0, + 992.0, + 1290.0, + 992.0, + 1324.0, + 322.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1340.0, + 1406.0, + 1340.0, + 1406.0, + 1381.0, + 293.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1372.0, + 1212.0, + 1372.0, + 1212.0, + 1411.0, + 321.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1428.0, + 1405.0, + 1428.0, + 1405.0, + 1468.0, + 294.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1459.0, + 1382.0, + 1459.0, + 1382.0, + 1497.0, + 321.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 1406.0, + 1511.0, + 1406.0, + 1554.0, + 293.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1545.0, + 1405.0, + 1545.0, + 1405.0, + 1585.0, + 322.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1576.0, + 622.0, + 1576.0, + 622.0, + 1612.0, + 322.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1627.0, + 1409.0, + 1627.0, + 1409.0, + 1669.0, + 293.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1662.0, + 635.0, + 1662.0, + 635.0, + 1697.0, + 322.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1755.0, + 291.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1745.0, + 1411.0, + 1745.0, + 1411.0, + 1783.0, + 321.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1800.0, + 1407.0, + 1800.0, + 1407.0, + 1844.0, + 291.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1833.0, + 906.0, + 1833.0, + 906.0, + 1871.0, + 320.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1885.0, + 1405.0, + 1885.0, + 1405.0, + 1927.0, + 291.0, + 1927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1919.0, + 1129.0, + 1919.0, + 1129.0, + 1955.0, + 324.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1969.0, + 1406.0, + 1969.0, + 1406.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2007.0, + 707.0, + 2007.0, + 707.0, + 2038.0, + 322.0, + 2038.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.834 + }, + { + "category_id": 1, + "poly": [ + 299, + 1971, + 1401, + 1971, + 1401, + 2033, + 299, + 2033 + ], + "score": 0.765 + }, + { + "category_id": 1, + "poly": [ + 292, + 914, + 1402, + 914, + 1402, + 980, + 292, + 980 + ], + "score": 0.76 + }, + { + "category_id": 1, + "poly": [ + 296, + 714, + 1400, + 714, + 1400, + 780, + 296, + 780 + ], + "score": 0.751 + }, + { + "category_id": 1, + "poly": [ + 297, + 513, + 1400, + 513, + 1400, + 579, + 297, + 579 + ], + "score": 0.747 + }, + { + "category_id": 1, + "poly": [ + 296, + 1685, + 1405, + 1685, + 1405, + 1781, + 296, + 1781 + ], + "score": 0.733 + }, + { + "category_id": 1, + "poly": [ + 298, + 1084, + 1402, + 1084, + 1402, + 1180, + 298, + 1180 + ], + "score": 0.733 + }, + { + "category_id": 1, + "poly": [ + 298, + 398, + 1402, + 398, + 1402, + 493, + 298, + 493 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 296, + 1284, + 1399, + 1284, + 1399, + 1351, + 296, + 1351 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 295, + 799, + 1402, + 799, + 1402, + 893, + 295, + 893 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 289, + 1000, + 1403, + 1000, + 1403, + 1065, + 289, + 1065 + ], + "score": 0.717 + }, + { + "category_id": 1, + "poly": [ + 293, + 1800, + 1403, + 1800, + 1403, + 1866, + 293, + 1866 + ], + "score": 0.716 + }, + { + "category_id": 1, + "poly": [ + 292, + 229, + 1404, + 229, + 1404, + 294, + 292, + 294 + ], + "score": 0.711 + }, + { + "category_id": 1, + "poly": [ + 297, + 1885, + 1401, + 1885, + 1401, + 1951, + 297, + 1951 + ], + "score": 0.705 + }, + { + "category_id": 1, + "poly": [ + 292, + 1600, + 1402, + 1600, + 1402, + 1666, + 292, + 1666 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 297, + 1199, + 1398, + 1199, + 1398, + 1266, + 297, + 1266 + ], + "score": 0.697 + }, + { + "category_id": 1, + "poly": [ + 297, + 1484, + 1406, + 1484, + 1406, + 1580, + 297, + 1580 + ], + "score": 0.687 + }, + { + "category_id": 1, + "poly": [ + 292, + 314, + 1402, + 314, + 1402, + 378, + 292, + 378 + ], + "score": 0.674 + }, + { + "category_id": 1, + "poly": [ + 298, + 1368, + 1406, + 1368, + 1406, + 1464, + 298, + 1464 + ], + "score": 0.652 + }, + { + "category_id": 1, + "poly": [ + 296, + 598, + 1407, + 598, + 1407, + 694, + 296, + 694 + ], + "score": 0.606 + }, + { + "category_id": 13, + "poly": [ + 1018, + 605, + 1047, + 605, + 1047, + 629, + 1018, + 629 + ], + "score": 0.69, + "latex": "\\approx" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 830.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1968.0, + 1407.0, + 1968.0, + 1407.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1998.0, + 395.0, + 1998.0, + 395.0, + 2037.0, + 318.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 911.0, + 1408.0, + 911.0, + 1408.0, + 954.0, + 292.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 942.0, + 461.0, + 942.0, + 461.0, + 979.0, + 319.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 715.0, + 1404.0, + 715.0, + 1404.0, + 752.0, + 295.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 747.0, + 1192.0, + 747.0, + 1192.0, + 781.0, + 321.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 508.0, + 1404.0, + 508.0, + 1404.0, + 555.0, + 292.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 546.0, + 788.0, + 546.0, + 788.0, + 580.0, + 322.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1683.0, + 1406.0, + 1683.0, + 1406.0, + 1723.0, + 293.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1713.0, + 1407.0, + 1713.0, + 1407.0, + 1758.0, + 319.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1749.0, + 533.0, + 1749.0, + 533.0, + 1780.0, + 324.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1084.0, + 1407.0, + 1084.0, + 1407.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1117.0, + 1406.0, + 1117.0, + 1406.0, + 1152.0, + 323.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1147.0, + 1028.0, + 1147.0, + 1028.0, + 1181.0, + 323.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 397.0, + 1406.0, + 397.0, + 1406.0, + 436.0, + 292.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 428.0, + 1407.0, + 428.0, + 1407.0, + 469.0, + 321.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 461.0, + 544.0, + 461.0, + 544.0, + 493.0, + 325.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1282.0, + 1403.0, + 1282.0, + 1403.0, + 1323.0, + 294.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1318.0, + 1191.0, + 1318.0, + 1191.0, + 1351.0, + 321.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 799.0, + 1405.0, + 799.0, + 1405.0, + 838.0, + 294.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 833.0, + 1404.0, + 833.0, + 1404.0, + 867.0, + 323.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 861.0, + 612.0, + 861.0, + 612.0, + 893.0, + 322.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 998.0, + 1408.0, + 998.0, + 1408.0, + 1038.0, + 295.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1031.0, + 1019.0, + 1031.0, + 1019.0, + 1065.0, + 321.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1799.0, + 1404.0, + 1799.0, + 1404.0, + 1836.0, + 295.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1833.0, + 707.0, + 1833.0, + 707.0, + 1867.0, + 320.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 228.0, + 1407.0, + 228.0, + 1407.0, + 267.0, + 296.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 260.0, + 1167.0, + 260.0, + 1167.0, + 296.0, + 320.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1884.0, + 1403.0, + 1884.0, + 1403.0, + 1921.0, + 296.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1917.0, + 1274.0, + 1917.0, + 1274.0, + 1953.0, + 324.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1598.0, + 1407.0, + 1598.0, + 1407.0, + 1638.0, + 294.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1632.0, + 1330.0, + 1632.0, + 1330.0, + 1668.0, + 324.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1196.0, + 1404.0, + 1196.0, + 1404.0, + 1239.0, + 293.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1231.0, + 1228.0, + 1231.0, + 1228.0, + 1268.0, + 322.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1483.0, + 1406.0, + 1483.0, + 1406.0, + 1522.0, + 294.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1514.0, + 1406.0, + 1514.0, + 1406.0, + 1555.0, + 320.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1545.0, + 533.0, + 1545.0, + 533.0, + 1580.0, + 323.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 313.0, + 1404.0, + 313.0, + 1404.0, + 349.0, + 296.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 345.0, + 977.0, + 345.0, + 977.0, + 381.0, + 320.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1372.0, + 1407.0, + 1372.0, + 1407.0, + 1403.0, + 297.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1400.0, + 1406.0, + 1400.0, + 1406.0, + 1436.0, + 322.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1433.0, + 610.0, + 1433.0, + 610.0, + 1462.0, + 322.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 598.0, + 1017.0, + 598.0, + 1017.0, + 636.0, + 295.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 598.0, + 1405.0, + 598.0, + 1405.0, + 636.0, + 1048.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 631.0, + 1404.0, + 631.0, + 1404.0, + 666.0, + 323.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 663.0, + 707.0, + 663.0, + 707.0, + 692.0, + 320.0, + 692.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 229, + 1403, + 229, + 1403, + 322, + 296, + 322 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 299, + 343, + 1404, + 343, + 1404, + 437, + 299, + 437 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 299, + 1630, + 1404, + 1630, + 1404, + 1724, + 299, + 1724 + ], + "score": 0.883 + }, + { + "category_id": 1, + "poly": [ + 300, + 902, + 1402, + 902, + 1402, + 968, + 300, + 968 + ], + "score": 0.881 + }, + { + "category_id": 1, + "poly": [ + 296, + 737, + 1401, + 737, + 1401, + 801, + 296, + 801 + ], + "score": 0.875 + }, + { + "category_id": 1, + "poly": [ + 288, + 1546, + 1404, + 1546, + 1404, + 1611, + 288, + 1611 + ], + "score": 0.863 + }, + { + "category_id": 1, + "poly": [ + 298, + 820, + 1403, + 820, + 1403, + 885, + 298, + 885 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1403, + 1941, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.858 + }, + { + "category_id": 1, + "poly": [ + 291, + 1744, + 1403, + 1744, + 1403, + 1809, + 291, + 1809 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 299, + 455, + 1407, + 455, + 1407, + 549, + 299, + 549 + ], + "score": 0.846 + }, + { + "category_id": 1, + "poly": [ + 297, + 985, + 1400, + 985, + 1400, + 1051, + 297, + 1051 + ], + "score": 0.845 + }, + { + "category_id": 1, + "poly": [ + 297, + 652, + 1400, + 652, + 1400, + 718, + 297, + 718 + ], + "score": 0.842 + }, + { + "category_id": 1, + "poly": [ + 297, + 570, + 1399, + 570, + 1399, + 635, + 297, + 635 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 299, + 1265, + 1400, + 1265, + 1400, + 1332, + 299, + 1332 + ], + "score": 0.833 + }, + { + "category_id": 1, + "poly": [ + 299, + 1182, + 1400, + 1182, + 1400, + 1249, + 299, + 1249 + ], + "score": 0.825 + }, + { + "category_id": 1, + "poly": [ + 296, + 1349, + 1403, + 1349, + 1403, + 1415, + 296, + 1415 + ], + "score": 0.821 + }, + { + "category_id": 1, + "poly": [ + 300, + 1826, + 1401, + 1826, + 1401, + 1921, + 300, + 1921 + ], + "score": 0.821 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 296, + 1432, + 1401, + 1432, + 1401, + 1527, + 296, + 1527 + ], + "score": 0.815 + }, + { + "category_id": 1, + "poly": [ + 298, + 1069, + 1407, + 1069, + 1407, + 1163, + 298, + 1163 + ], + "score": 0.755 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 226.0, + 1408.0, + 226.0, + 1408.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 261.0, + 1408.0, + 261.0, + 1408.0, + 295.0, + 323.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 286.0, + 410.0, + 286.0, + 410.0, + 327.0, + 320.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 343.0, + 1406.0, + 343.0, + 1406.0, + 380.0, + 295.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 373.0, + 1409.0, + 373.0, + 1409.0, + 409.0, + 321.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 405.0, + 462.0, + 405.0, + 462.0, + 436.0, + 322.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1628.0, + 1406.0, + 1628.0, + 1406.0, + 1667.0, + 294.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1662.0, + 1408.0, + 1662.0, + 1408.0, + 1694.0, + 321.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1690.0, + 396.0, + 1690.0, + 396.0, + 1724.0, + 321.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 903.0, + 1407.0, + 903.0, + 1407.0, + 939.0, + 296.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 935.0, + 737.0, + 935.0, + 737.0, + 967.0, + 323.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 736.0, + 1405.0, + 736.0, + 1405.0, + 775.0, + 294.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 768.0, + 838.0, + 768.0, + 838.0, + 804.0, + 323.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1546.0, + 1405.0, + 1546.0, + 1405.0, + 1582.0, + 294.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1579.0, + 709.0, + 1579.0, + 709.0, + 1609.0, + 319.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 820.0, + 1405.0, + 820.0, + 1405.0, + 856.0, + 296.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 851.0, + 991.0, + 851.0, + 991.0, + 887.0, + 322.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1937.0, + 1408.0, + 1937.0, + 1408.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1972.0, + 1403.0, + 1972.0, + 1403.0, + 2006.0, + 322.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 2003.0, + 708.0, + 2003.0, + 708.0, + 2036.0, + 320.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1784.0, + 292.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1774.0, + 1084.0, + 1774.0, + 1084.0, + 1810.0, + 322.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 456.0, + 1407.0, + 456.0, + 1407.0, + 494.0, + 297.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 489.0, + 1408.0, + 489.0, + 1408.0, + 523.0, + 322.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 515.0, + 397.0, + 515.0, + 397.0, + 552.0, + 318.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 984.0, + 1404.0, + 984.0, + 1404.0, + 1023.0, + 295.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1018.0, + 1213.0, + 1018.0, + 1213.0, + 1052.0, + 322.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 650.0, + 1404.0, + 650.0, + 1404.0, + 692.0, + 293.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 685.0, + 774.0, + 685.0, + 774.0, + 717.0, + 322.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 566.0, + 1404.0, + 566.0, + 1404.0, + 609.0, + 296.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 603.0, + 1251.0, + 603.0, + 1251.0, + 636.0, + 322.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1266.0, + 1404.0, + 1266.0, + 1404.0, + 1302.0, + 297.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1297.0, + 932.0, + 1297.0, + 932.0, + 1332.0, + 324.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1183.0, + 1404.0, + 1183.0, + 1404.0, + 1218.0, + 295.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1214.0, + 1346.0, + 1214.0, + 1346.0, + 1249.0, + 320.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1350.0, + 1404.0, + 1350.0, + 1404.0, + 1386.0, + 295.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1381.0, + 1291.0, + 1381.0, + 1291.0, + 1416.0, + 321.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1825.0, + 1405.0, + 1825.0, + 1405.0, + 1864.0, + 293.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1856.0, + 1405.0, + 1856.0, + 1405.0, + 1894.0, + 321.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1889.0, + 533.0, + 1889.0, + 533.0, + 1921.0, + 324.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1433.0, + 1405.0, + 1433.0, + 1405.0, + 1467.0, + 295.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1464.0, + 1403.0, + 1464.0, + 1403.0, + 1498.0, + 324.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1494.0, + 611.0, + 1494.0, + 611.0, + 1528.0, + 321.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1071.0, + 1407.0, + 1071.0, + 1407.0, + 1105.0, + 296.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1101.0, + 1409.0, + 1101.0, + 1409.0, + 1135.0, + 321.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1129.0, + 397.0, + 1129.0, + 397.0, + 1163.0, + 321.0, + 1163.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 300, + 1006, + 1399, + 1006, + 1399, + 1071, + 300, + 1071 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 299, + 1224, + 1401, + 1224, + 1401, + 1291, + 299, + 1291 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 292, + 592, + 1403, + 592, + 1403, + 656, + 292, + 656 + ], + "score": 0.856 + }, + { + "category_id": 1, + "poly": [ + 299, + 923, + 1401, + 923, + 1401, + 988, + 299, + 988 + ], + "score": 0.856 + }, + { + "category_id": 1, + "poly": [ + 292, + 1141, + 1402, + 1141, + 1402, + 1207, + 292, + 1207 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 295, + 1090, + 1396, + 1090, + 1396, + 1125, + 295, + 1125 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 303, + 727, + 1400, + 727, + 1400, + 822, + 303, + 822 + ], + "score": 0.854 + }, + { + "category_id": 1, + "poly": [ + 302, + 477, + 1404, + 477, + 1404, + 572, + 302, + 572 + ], + "score": 0.854 + }, + { + "category_id": 1, + "poly": [ + 293, + 229, + 1403, + 229, + 1403, + 294, + 293, + 294 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 300, + 840, + 1400, + 840, + 1400, + 905, + 300, + 905 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 293, + 394, + 1403, + 394, + 1403, + 459, + 293, + 459 + ], + "score": 0.835 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 294, + 312, + 1402, + 312, + 1402, + 377, + 294, + 377 + ], + "score": 0.81 + }, + { + "category_id": 1, + "poly": [ + 294, + 674, + 1368, + 674, + 1368, + 710, + 294, + 710 + ], + "score": 0.808 + }, + { + "category_id": 1, + "poly": [ + 299, + 1308, + 1404, + 1308, + 1404, + 1404, + 299, + 1404 + ], + "score": 0.775 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1008.0, + 1404.0, + 1008.0, + 1404.0, + 1044.0, + 295.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1039.0, + 1254.0, + 1039.0, + 1254.0, + 1072.0, + 321.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1221.0, + 1406.0, + 1221.0, + 1406.0, + 1265.0, + 294.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1258.0, + 1323.0, + 1258.0, + 1323.0, + 1291.0, + 323.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 591.0, + 1406.0, + 591.0, + 1406.0, + 629.0, + 294.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 624.0, + 1237.0, + 624.0, + 1237.0, + 657.0, + 323.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 923.0, + 1405.0, + 923.0, + 1405.0, + 961.0, + 294.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 954.0, + 1400.0, + 954.0, + 1400.0, + 989.0, + 324.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1138.0, + 1407.0, + 1138.0, + 1407.0, + 1181.0, + 294.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1174.0, + 970.0, + 1174.0, + 970.0, + 1208.0, + 321.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1084.0, + 1396.0, + 1084.0, + 1396.0, + 1130.0, + 291.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 727.0, + 1405.0, + 727.0, + 1405.0, + 765.0, + 298.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 760.0, + 1404.0, + 760.0, + 1404.0, + 794.0, + 323.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 788.0, + 1249.0, + 788.0, + 1249.0, + 826.0, + 323.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 477.0, + 1408.0, + 477.0, + 1408.0, + 515.0, + 296.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 511.0, + 1408.0, + 511.0, + 1408.0, + 545.0, + 322.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 540.0, + 777.0, + 540.0, + 777.0, + 574.0, + 323.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 231.0, + 1403.0, + 231.0, + 1403.0, + 263.0, + 296.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 259.0, + 941.0, + 259.0, + 941.0, + 295.0, + 322.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 842.0, + 1404.0, + 842.0, + 1404.0, + 878.0, + 296.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 873.0, + 894.0, + 873.0, + 894.0, + 906.0, + 321.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 394.0, + 1408.0, + 394.0, + 1408.0, + 430.0, + 295.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 427.0, + 775.0, + 427.0, + 775.0, + 461.0, + 322.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 312.0, + 1404.0, + 312.0, + 1404.0, + 348.0, + 297.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 345.0, + 877.0, + 345.0, + 877.0, + 377.0, + 322.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 672.0, + 1367.0, + 672.0, + 1367.0, + 712.0, + 295.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1306.0, + 1405.0, + 1306.0, + 1405.0, + 1346.0, + 293.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1338.0, + 1405.0, + 1338.0, + 1405.0, + 1378.0, + 321.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1371.0, + 1080.0, + 1371.0, + 1080.0, + 1405.0, + 322.0, + 1405.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 297, + 1490, + 1252, + 1490, + 1252, + 1593, + 297, + 1593 + ], + "score": 0.972, + "html": "
Singularity in2nd Layer Trained (RF)Vanishing Init.Non-vanishing Init.
Bias1: No; Y2: Yesγ1: No; γ2: Noγ1: No; 2: No
VarianceY1: No; 2: Yes1: Yes; Y2: NoY1: No; 2: No
" + }, + { + "category_id": 3, + "poly": [ + 333, + 876, + 1356, + 876, + 1356, + 1294, + 333, + 1294 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 323, + 309, + 1373, + 309, + 1373, + 689, + 323, + 689 + ], + "score": 0.965 + }, + { + "category_id": 4, + "poly": [ + 296, + 712, + 1406, + 712, + 1406, + 826, + 296, + 826 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 291, + 1313, + 1407, + 1313, + 1407, + 1379, + 291, + 1379 + ], + "score": 0.939 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 816, + 76, + 816, + 104, + 299, + 104 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.845 + }, + { + "category_id": 6, + "poly": [ + 295, + 1439, + 1056, + 1439, + 1056, + 1472, + 295, + 1472 + ], + "score": 0.594 + }, + { + "category_id": 0, + "poly": [ + 300, + 228, + 850, + 228, + 850, + 262, + 300, + 262 + ], + "score": 0.355 + }, + { + "category_id": 0, + "poly": [ + 295, + 1439, + 1056, + 1439, + 1056, + 1472, + 295, + 1472 + ], + "score": 0.348 + }, + { + "category_id": 2, + "poly": [ + 300, + 228, + 850, + 228, + 850, + 262, + 300, + 262 + ], + "score": 0.244 + }, + { + "category_id": 13, + "poly": [ + 771, + 742, + 891, + 742, + 891, + 772, + 771, + 772 + ], + "score": 0.89, + "latex": "\\mathrm { S N R } { = } 1 / 4 \\mathrm { \\Omega }" + }, + { + "category_id": 13, + "poly": [ + 1088, + 1351, + 1118, + 1351, + 1118, + 1378, + 1088, + 1378 + ], + "score": 0.84, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1284, + 1314, + 1402, + 1314, + 1402, + 1344, + 1284, + 1344 + ], + "score": 0.72, + "latex": "S _ { \\mathrm { { N R = } } 2 5 / 1 6 }" + }, + { + "category_id": 13, + "poly": [ + 894, + 1527, + 925, + 1527, + 925, + 1556, + 894, + 1556 + ], + "score": 0.68, + "latex": "\\overline { { \\gamma _ { 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 638, + 1528, + 669, + 1528, + 669, + 1556, + 638, + 1556 + ], + "score": 0.67, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 542, + 1529, + 572, + 1529, + 572, + 1556, + 542, + 1556 + ], + "score": 0.58, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1527, + 1161, + 1527, + 1161, + 1556, + 1130, + 1556 + ], + "score": 0.57, + "latex": "\\overline { { \\gamma _ { 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 799, + 1529, + 828, + 1529, + 828, + 1556, + 799, + 1556 + ], + "score": 0.55, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 902, + 1560, + 931, + 1560, + 931, + 1588, + 902, + 1588 + ], + "score": 0.45, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 638, + 1561, + 668, + 1561, + 668, + 1588, + 638, + 1588 + ], + "score": 0.44, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 799, + 1562, + 828, + 1562, + 828, + 1588, + 799, + 1588 + ], + "score": 0.31, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1031, + 1562, + 1060, + 1562, + 1060, + 1588, + 1031, + 1588 + ], + "score": 0.28, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 542, + 1561, + 571, + 1561, + 571, + 1588, + 542, + 1588 + ], + "score": 0.27, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 876.0, + 402.0, + 876.0, + 402.0, + 909.0, + 353.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 877.0, + 909.0, + 877.0, + 909.0, + 901.0, + 890.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 913.0, + 440.0, + 913.0, + 440.0, + 922.0, + 430.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 899.0, + 570.0, + 899.0, + 570.0, + 934.0, + 462.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 898.0, + 1332.0, + 898.0, + 1332.0, + 931.0, + 1226.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 928.0, + 401.0, + 928.0, + 401.0, + 962.0, + 352.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 930.0, + 575.0, + 930.0, + 575.0, + 962.0, + 461.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 943.0, + 909.0, + 943.0, + 909.0, + 967.0, + 890.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 928.0, + 1341.0, + 928.0, + 1341.0, + 959.0, + 1224.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 981.0, + 401.0, + 981.0, + 401.0, + 1016.0, + 352.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1003.0, + 1114.0, + 1003.0, + 1114.0, + 1012.0, + 1105.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1017.0, + 407.0, + 1017.0, + 407.0, + 1084.0, + 324.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1009.0, + 911.0, + 1009.0, + 911.0, + 1076.0, + 864.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1072.0, + 910.0, + 1072.0, + 910.0, + 1100.0, + 889.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1091.0, + 399.0, + 1091.0, + 399.0, + 1121.0, + 356.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1142.0, + 401.0, + 1142.0, + 401.0, + 1175.0, + 353.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1138.0, + 910.0, + 1138.0, + 910.0, + 1166.0, + 889.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1196.0, + 400.0, + 1196.0, + 400.0, + 1226.0, + 354.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1206.0, + 494.0, + 1206.0, + 494.0, + 1240.0, + 446.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1209.0, + 591.0, + 1209.0, + 591.0, + 1239.0, + 545.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1206.0, + 689.0, + 1206.0, + 689.0, + 1240.0, + 641.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1209.0, + 784.0, + 1209.0, + 784.0, + 1239.0, + 739.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1209.0, + 983.0, + 1209.0, + 983.0, + 1239.0, + 939.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1209.0, + 1067.0, + 1209.0, + 1067.0, + 1239.0, + 1021.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1206.0, + 1152.0, + 1206.0, + 1152.0, + 1240.0, + 1102.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 1209.0, + 1232.0, + 1209.0, + 1232.0, + 1239.0, + 1185.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1208.0, + 1316.0, + 1208.0, + 1316.0, + 1241.0, + 1268.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1231.0, + 662.0, + 1231.0, + 662.0, + 1263.0, + 567.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1231.0, + 1177.0, + 1231.0, + 1177.0, + 1261.0, + 1081.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1258.0, + 656.0, + 1258.0, + 656.0, + 1296.0, + 523.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1258.0, + 1179.0, + 1258.0, + 1179.0, + 1295.0, + 1045.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 972.5, + 715.0, + 972.5, + 715.0, + 982.0, + 699.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 334.0, + 395.0, + 334.0, + 395.0, + 364.0, + 335.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 327.0, + 559.0, + 327.0, + 559.0, + 350.0, + 450.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 325.0, + 931.0, + 325.0, + 931.0, + 355.0, + 880.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 341.0, + 1183.0, + 341.0, + 1183.0, + 353.0, + 1171.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 332.0, + 1247.0, + 332.0, + 1247.0, + 341.0, + 1234.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 350.0, + 534.0, + 350.0, + 534.0, + 375.0, + 450.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 373.0, + 929.0, + 373.0, + 929.0, + 399.0, + 882.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 378.0, + 987.0, + 378.0, + 987.0, + 391.0, + 971.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 400.0, + 393.0, + 400.0, + 393.0, + 425.0, + 336.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 415.0, + 931.0, + 415.0, + 931.0, + 445.0, + 882.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 435.0, + 1078.0, + 435.0, + 1078.0, + 449.0, + 1062.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 449.0, + 1109.0, + 449.0, + 1109.0, + 459.0, + 1097.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 463.0, + 393.0, + 463.0, + 393.0, + 489.0, + 336.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 460.0, + 931.0, + 460.0, + 931.0, + 490.0, + 882.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 458.0, + 1023.0, + 458.0, + 1023.0, + 470.0, + 1012.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 483.0, + 1110.0, + 483.0, + 1110.0, + 493.0, + 1101.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 485.0, + 1344.0, + 485.0, + 1344.0, + 495.0, + 1334.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 500.0, + 965.0, + 500.0, + 965.0, + 516.0, + 951.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 496.0, + 1139.0, + 496.0, + 1139.0, + 513.0, + 1108.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 508.0, + 929.0, + 508.0, + 929.0, + 534.0, + 872.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 512.0, + 994.0, + 512.0, + 994.0, + 524.0, + 983.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 512.0, + 1106.0, + 512.0, + 1106.0, + 527.0, + 1091.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 525.0, + 393.0, + 525.0, + 393.0, + 554.0, + 325.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 535.0, + 1192.0, + 535.0, + 1192.0, + 545.0, + 1181.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 542.0, + 1007.0, + 542.0, + 1007.0, + 559.0, + 982.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 553.0, + 929.0, + 553.0, + 929.0, + 578.0, + 872.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 583.0, + 967.0, + 583.0, + 967.0, + 594.0, + 956.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 575.0, + 1095.0, + 575.0, + 1095.0, + 599.0, + 986.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 589.0, + 394.0, + 589.0, + 394.0, + 618.0, + 327.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 595.0, + 931.0, + 595.0, + 931.0, + 624.0, + 868.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 606.0, + 967.0, + 606.0, + 967.0, + 618.0, + 956.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 598.0, + 1071.0, + 598.0, + 1071.0, + 627.0, + 985.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 634.0, + 841.0, + 634.0, + 841.0, + 662.0, + 381.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 633.0, + 1041.0, + 633.0, + 1041.0, + 662.0, + 979.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 633.0, + 1119.0, + 633.0, + 1119.0, + 662.0, + 1055.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 634.0, + 1189.0, + 634.0, + 1189.0, + 660.0, + 1139.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 635.0, + 1265.0, + 635.0, + 1265.0, + 658.0, + 1216.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.0, + 635.0, + 1344.0, + 635.0, + 1344.0, + 660.0, + 1297.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 658.0, + 726.0, + 658.0, + 726.0, + 693.0, + 444.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 659.0, + 1288.0, + 659.0, + 1288.0, + 690.0, + 962.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 482.5, + 1085.0, + 482.5, + 1085.0, + 499.5, + 1053.0, + 499.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 511.5, + 1147.0, + 511.5, + 1147.0, + 525.5, + 1112.0, + 525.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 601.5, + 1175.0, + 601.5, + 1175.0, + 620.5, + 1158.0, + 620.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 713.0, + 1404.0, + 713.0, + 1404.0, + 746.0, + 296.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 741.0, + 770.0, + 741.0, + 770.0, + 770.0, + 296.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 741.0, + 1402.0, + 741.0, + 1402.0, + 770.0, + 892.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 769.0, + 1404.0, + 769.0, + 1404.0, + 802.0, + 294.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 794.0, + 435.0, + 794.0, + 435.0, + 828.0, + 294.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1314.0, + 1283.0, + 1314.0, + 1283.0, + 1349.0, + 294.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1314.0, + 1406.0, + 1314.0, + 1406.0, + 1349.0, + 1403.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1341.0, + 1087.0, + 1341.0, + 1087.0, + 1385.0, + 293.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1341.0, + 1130.0, + 1341.0, + 1130.0, + 1385.0, + 1119.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1439.0, + 1058.0, + 1439.0, + 1058.0, + 1474.0, + 296.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 226.0, + 853.0, + 226.0, + 853.0, + 266.0, + 296.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1439.0, + 1058.0, + 1439.0, + 1058.0, + 1474.0, + 296.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 226.0, + 853.0, + 226.0, + 853.0, + 266.0, + 296.0, + 266.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 1938, + 1404, + 1938, + 1404, + 2031, + 299, + 2031 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 343, + 223, + 1351, + 223, + 1351, + 603, + 343, + 603 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 341, + 1257, + 1352, + 1257, + 1352, + 1635, + 341, + 1635 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 340, + 737, + 1351, + 737, + 1351, + 1121, + 340, + 1121 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 297, + 1124, + 1404, + 1124, + 1404, + 1210, + 297, + 1210 + ], + "score": 0.957 + }, + { + "category_id": 4, + "poly": [ + 296, + 1641, + 1405, + 1641, + 1405, + 1726, + 296, + 1726 + ], + "score": 0.951 + }, + { + "category_id": 4, + "poly": [ + 296, + 609, + 1404, + 609, + 1404, + 693, + 296, + 693 + ], + "score": 0.948 + }, + { + "category_id": 0, + "poly": [ + 301, + 1795, + 562, + 1795, + 562, + 1830, + 301, + 1830 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 300, + 1874, + 682, + 1874, + 682, + 1906, + 300, + 1906 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.849 + }, + { + "category_id": 13, + "poly": [ + 418, + 1153, + 484, + 1153, + 484, + 1178, + 418, + 1178 + ], + "score": 0.9, + "latex": "h = d" + }, + { + "category_id": 13, + "poly": [ + 947, + 1156, + 974, + 1156, + 974, + 1181, + 947, + 1181 + ], + "score": 0.85, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 960, + 1974, + 987, + 1974, + 987, + 1998, + 960, + 1998 + ], + "score": 0.85, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1303, + 1693, + 1330, + 1693, + 1330, + 1721, + 1303, + 1721 + ], + "score": 0.84, + "latex": "r ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 670, + 1125, + 719, + 1125, + 719, + 1154, + 670, + 1154 + ], + "score": 0.84, + "latex": "1 / d )" + }, + { + "category_id": 13, + "poly": [ + 462, + 2002, + 489, + 2002, + 489, + 2028, + 462, + 2028 + ], + "score": 0.84, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 645, + 640, + 673, + 640, + 673, + 664, + 645, + 664 + ], + "score": 0.83, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1326, + 1973, + 1350, + 1973, + 1350, + 1998, + 1326, + 1998 + ], + "score": 0.83, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 1134, + 1669, + 1176, + 1669, + 1176, + 1697, + 1134, + 1697 + ], + "score": 0.82, + "latex": "\\left( \\gamma _ { 2 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 541, + 2001, + 590, + 2001, + 590, + 2029, + 541, + 2029 + ], + "score": 0.78, + "latex": "U X" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 227.0, + 397.0, + 227.0, + 397.0, + 258.0, + 356.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 243.0, + 433.0, + 243.0, + 433.0, + 261.0, + 408.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 238.0, + 621.0, + 238.0, + 621.0, + 266.0, + 443.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 228.0, + 931.0, + 228.0, + 931.0, + 258.0, + 892.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 231.0, + 1303.0, + 231.0, + 1303.0, + 270.0, + 1094.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 270.0, + 399.0, + 270.0, + 399.0, + 309.0, + 355.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 266.0, + 437.0, + 266.0, + 437.0, + 283.0, + 410.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 262.0, + 654.0, + 262.0, + 654.0, + 286.0, + 443.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 270.0, + 1110.0, + 270.0, + 1110.0, + 279.0, + 1099.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 262.0, + 1337.0, + 262.0, + 1337.0, + 286.0, + 1124.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 285.0, + 537.0, + 285.0, + 537.0, + 311.0, + 444.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 275.0, + 932.0, + 275.0, + 932.0, + 305.0, + 892.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 284.0, + 1218.0, + 284.0, + 1218.0, + 311.0, + 1121.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 323.0, + 398.0, + 323.0, + 398.0, + 353.0, + 356.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 323.0, + 930.0, + 323.0, + 930.0, + 350.0, + 893.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 349.0, + 405.0, + 349.0, + 405.0, + 419.0, + 333.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 341.0, + 930.0, + 341.0, + 930.0, + 426.0, + 877.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 416.0, + 397.0, + 416.0, + 397.0, + 446.0, + 356.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 415.0, + 930.0, + 415.0, + 930.0, + 442.0, + 893.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 464.0, + 396.0, + 464.0, + 396.0, + 494.0, + 355.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 461.0, + 928.0, + 461.0, + 928.0, + 488.0, + 893.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 480.0, + 1330.0, + 480.0, + 1330.0, + 501.0, + 1262.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 510.0, + 399.0, + 510.0, + 399.0, + 540.0, + 356.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 506.0, + 932.0, + 506.0, + 932.0, + 536.0, + 891.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 534.0, + 442.0, + 534.0, + 442.0, + 562.0, + 391.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 536.0, + 498.0, + 536.0, + 498.0, + 561.0, + 451.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 535.0, + 557.0, + 535.0, + 557.0, + 560.0, + 509.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 536.0, + 674.0, + 536.0, + 674.0, + 560.0, + 567.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 536.0, + 733.0, + 536.0, + 733.0, + 561.0, + 683.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 536.0, + 788.0, + 536.0, + 788.0, + 561.0, + 739.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 535.0, + 976.0, + 535.0, + 976.0, + 560.0, + 928.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 536.0, + 1033.0, + 536.0, + 1033.0, + 561.0, + 985.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 537.0, + 1091.0, + 537.0, + 1091.0, + 558.0, + 1042.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 534.0, + 1211.0, + 534.0, + 1211.0, + 561.0, + 1102.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 536.0, + 1268.0, + 536.0, + 1268.0, + 561.0, + 1219.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 536.0, + 1323.0, + 536.0, + 1323.0, + 561.0, + 1274.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 553.0, + 641.0, + 553.0, + 641.0, + 580.0, + 564.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 553.0, + 1177.0, + 553.0, + 1177.0, + 583.0, + 1100.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 573.0, + 630.0, + 573.0, + 630.0, + 611.0, + 533.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 576.0, + 1184.0, + 576.0, + 1184.0, + 609.0, + 1047.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.75, + 291.5, + 831.75, + 291.5, + 831.75, + 342.5, + 561.75, + 342.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.75, + 395.5, + 581.75, + 395.5, + 581.75, + 410.5, + 547.75, + 410.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 469.5, + 494.0, + 469.5, + 494.0, + 478.0, + 472.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 511.0, + 537.0, + 511.0, + 537.0, + 524.0, + 515.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1264.0, + 406.0, + 1264.0, + 406.0, + 1299.0, + 357.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1278.0, + 438.0, + 1278.0, + 438.0, + 1291.0, + 422.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1270.0, + 616.0, + 1270.0, + 616.0, + 1299.0, + 449.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1263.0, + 939.0, + 1263.0, + 939.0, + 1298.0, + 894.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1269.0, + 1284.0, + 1269.0, + 1284.0, + 1298.0, + 1118.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1294.0, + 670.0, + 1294.0, + 670.0, + 1320.0, + 448.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1301.0, + 1103.0, + 1301.0, + 1103.0, + 1309.0, + 1094.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1293.0, + 1331.0, + 1293.0, + 1331.0, + 1317.0, + 1121.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1312.0, + 775.0, + 1312.0, + 775.0, + 1321.0, + 765.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1316.0, + 677.0, + 1316.0, + 677.0, + 1342.0, + 447.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1316.0, + 1337.0, + 1316.0, + 1337.0, + 1341.0, + 1118.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1335.0, + 406.0, + 1335.0, + 406.0, + 1368.0, + 358.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1333.0, + 940.0, + 1333.0, + 940.0, + 1367.0, + 894.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1383.0, + 410.0, + 1383.0, + 410.0, + 1444.0, + 332.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1381.0, + 945.0, + 1381.0, + 945.0, + 1443.0, + 866.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1472.0, + 404.0, + 1472.0, + 404.0, + 1506.0, + 357.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1474.0, + 937.0, + 1474.0, + 937.0, + 1505.0, + 895.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1537.0, + 1065.0, + 1537.0, + 1065.0, + 1546.0, + 1057.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1543.0, + 409.0, + 1543.0, + 409.0, + 1572.0, + 359.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1541.0, + 959.0, + 1541.0, + 959.0, + 1574.0, + 895.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1560.0, + 484.0, + 1560.0, + 484.0, + 1594.0, + 439.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1561.0, + 586.0, + 1561.0, + 586.0, + 1591.0, + 542.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 1560.0, + 689.0, + 1560.0, + 689.0, + 1593.0, + 643.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1562.0, + 791.0, + 1562.0, + 791.0, + 1592.0, + 747.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1561.0, + 1018.0, + 1561.0, + 1018.0, + 1592.0, + 975.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1561.0, + 1122.0, + 1561.0, + 1122.0, + 1591.0, + 1078.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1561.0, + 1224.0, + 1561.0, + 1224.0, + 1592.0, + 1181.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1562.0, + 1328.0, + 1562.0, + 1328.0, + 1591.0, + 1284.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1583.0, + 647.0, + 1583.0, + 647.0, + 1612.0, + 558.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1583.0, + 1184.0, + 1583.0, + 1184.0, + 1612.0, + 1095.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1608.0, + 703.0, + 1608.0, + 703.0, + 1641.0, + 458.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1608.0, + 1235.0, + 1608.0, + 1235.0, + 1640.0, + 996.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.75, + 1395.5, + 1369.75, + 1395.5, + 1369.75, + 1442.5, + 1097.75, + 1442.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.75, + 1506.5, + 1119.75, + 1506.5, + 1119.75, + 1518.5, + 1084.75, + 1518.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 737.0, + 409.0, + 737.0, + 409.0, + 770.0, + 358.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 737.0, + 945.0, + 737.0, + 945.0, + 770.0, + 893.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 751.0, + 610.0, + 751.0, + 610.0, + 811.0, + 576.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 756.0, + 586.0, + 756.0, + 586.0, + 770.0, + 569.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 758.0, + 607.0, + 758.0, + 607.0, + 768.0, + 598.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 761.0, + 591.0, + 761.0, + 591.0, + 802.0, + 565.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 763.0, + 554.0, + 763.0, + 554.0, + 772.0, + 545.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 767.0, + 608.0, + 767.0, + 608.0, + 789.0, + 597.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 772.0, + 555.0, + 772.0, + 555.0, + 782.0, + 545.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 796.0, + 585.0, + 796.0, + 585.0, + 806.0, + 577.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 787.0, + 942.0, + 787.0, + 942.0, + 816.0, + 895.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 800.0, + 409.0, + 800.0, + 409.0, + 832.0, + 358.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 806.0, + 611.0, + 806.0, + 611.0, + 815.0, + 596.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 836.0, + 944.0, + 836.0, + 944.0, + 866.0, + 906.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 863.0, + 409.0, + 863.0, + 409.0, + 925.0, + 339.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 877.0, + 903.0, + 877.0, + 903.0, + 923.0, + 873.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 886.0, + 942.0, + 886.0, + 942.0, + 914.0, + 907.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 926.0, + 409.0, + 926.0, + 409.0, + 958.0, + 358.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 934.0, + 943.0, + 934.0, + 943.0, + 965.0, + 906.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 991.0, + 404.0, + 991.0, + 404.0, + 1017.0, + 358.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 981.0, + 944.0, + 981.0, + 944.0, + 1013.0, + 906.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1031.0, + 979.0, + 1031.0, + 979.0, + 1059.0, + 918.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1045.0, + 496.0, + 1045.0, + 496.0, + 1077.0, + 455.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1047.0, + 585.0, + 1047.0, + 585.0, + 1073.0, + 540.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1047.0, + 672.0, + 1047.0, + 672.0, + 1076.0, + 630.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1045.0, + 760.0, + 1045.0, + 760.0, + 1076.0, + 718.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1045.0, + 1032.0, + 1045.0, + 1032.0, + 1076.0, + 989.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1047.0, + 1120.0, + 1047.0, + 1120.0, + 1076.0, + 1078.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 1047.0, + 1208.0, + 1047.0, + 1208.0, + 1076.0, + 1166.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1047.0, + 1294.0, + 1047.0, + 1294.0, + 1074.0, + 1251.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1067.0, + 650.0, + 1067.0, + 650.0, + 1097.0, + 563.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1067.0, + 1186.0, + 1067.0, + 1186.0, + 1097.0, + 1097.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1091.0, + 687.0, + 1091.0, + 687.0, + 1124.0, + 472.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1088.0, + 1222.0, + 1088.0, + 1222.0, + 1128.0, + 1009.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1122.0, + 669.0, + 1122.0, + 669.0, + 1158.0, + 295.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1122.0, + 1407.0, + 1122.0, + 1407.0, + 1158.0, + 720.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1150.0, + 417.0, + 1150.0, + 417.0, + 1183.0, + 292.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1150.0, + 946.0, + 1150.0, + 946.0, + 1183.0, + 485.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1150.0, + 1404.0, + 1150.0, + 1404.0, + 1183.0, + 975.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1177.0, + 1331.0, + 1177.0, + 1331.0, + 1212.0, + 294.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1640.0, + 1405.0, + 1640.0, + 1405.0, + 1672.0, + 295.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1667.0, + 1133.0, + 1667.0, + 1133.0, + 1699.0, + 294.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 1667.0, + 1403.0, + 1667.0, + 1403.0, + 1699.0, + 1177.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1693.0, + 1302.0, + 1693.0, + 1302.0, + 1727.0, + 291.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 1693.0, + 1342.0, + 1693.0, + 1342.0, + 1727.0, + 1331.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 606.0, + 1405.0, + 606.0, + 1405.0, + 638.0, + 295.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 635.0, + 644.0, + 635.0, + 644.0, + 667.0, + 294.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 635.0, + 1405.0, + 635.0, + 1405.0, + 667.0, + 674.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 660.0, + 1102.0, + 660.0, + 1102.0, + 696.0, + 293.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1792.0, + 568.0, + 1792.0, + 568.0, + 1838.0, + 291.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1872.0, + 686.0, + 1872.0, + 686.0, + 1910.0, + 293.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1934.0, + 1409.0, + 1934.0, + 1409.0, + 1977.0, + 292.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 959.0, + 1970.0, + 959.0, + 2003.0, + 294.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1970.0, + 1325.0, + 1970.0, + 1325.0, + 2003.0, + 988.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1970.0, + 1403.0, + 1970.0, + 1403.0, + 2003.0, + 1351.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1999.0, + 461.0, + 1999.0, + 461.0, + 2034.0, + 296.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1999.0, + 540.0, + 1999.0, + 540.0, + 2034.0, + 490.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1999.0, + 745.0, + 1999.0, + 745.0, + 2034.0, + 591.0, + 2034.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 986, + 1406, + 986, + 1406, + 1118, + 295, + 1118 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1858, + 1407, + 1858, + 1407, + 1958, + 297, + 1958 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 1314, + 1407, + 1314, + 1407, + 1439, + 296, + 1439 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 1201, + 1406, + 1201, + 1406, + 1301, + 296, + 1301 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 295, + 427, + 1406, + 427, + 1406, + 527, + 295, + 527 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 677, + 895, + 1022, + 895, + 1022, + 979, + 677, + 979 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 604, + 1447, + 1094, + 1447, + 1094, + 1517, + 604, + 1517 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 1703, + 1406, + 1703, + 1406, + 1768, + 297, + 1768 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 298, + 630, + 1410, + 630, + 1410, + 696, + 298, + 696 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 649, + 1966, + 1052, + 1966, + 1052, + 2032, + 649, + 2032 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 608, + 303, + 1087, + 303, + 1087, + 363, + 608, + 363 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 294, + 225, + 1402, + 225, + 1402, + 294, + 294, + 294 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 314, + 533, + 1332, + 533, + 1332, + 621, + 314, + 621 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 298, + 1537, + 1332, + 1537, + 1332, + 1571, + 298, + 1571 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 298, + 849, + 1261, + 849, + 1261, + 885, + 298, + 885 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 318, + 1580, + 1375, + 1580, + 1375, + 1666, + 318, + 1666 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 288, + 371, + 1307, + 371, + 1307, + 408, + 288, + 408 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 725, + 1112, + 725, + 1112, + 761, + 297, + 761 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 300, + 1801, + 723, + 1801, + 723, + 1836, + 300, + 1836 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 353, + 1126, + 1294, + 1126, + 1294, + 1192, + 353, + 1192 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 300, + 793, + 686, + 793, + 686, + 826, + 300, + 826 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1464, + 1400, + 1464, + 1400, + 1496, + 1351, + 1496 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1664, + 1400, + 1664, + 1400, + 1694, + 1351, + 1694 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1142, + 1400, + 1142, + 1400, + 1174, + 1351, + 1174 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1351, + 318, + 1401, + 318, + 1401, + 349, + 1351, + 349 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1983, + 1401, + 1983, + 1401, + 2014, + 1351, + 2014 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1353, + 560, + 1401, + 560, + 1401, + 591, + 1353, + 591 + ], + "score": 0.853 + }, + { + "category_id": 9, + "poly": [ + 1353, + 229, + 1403, + 229, + 1403, + 263, + 1353, + 263 + ], + "score": 0.138 + }, + { + "category_id": 14, + "poly": [ + 677, + 890, + 1020, + 890, + 1020, + 982, + 677, + 982 + ], + "score": 0.95, + "latex": "\\mu _ { A } ( d \\lambda ) = \\frac { 1 } { p } \\sum _ { i = 1 } ^ { p } \\delta _ { \\lambda _ { i } ( A ) } ( \\lambda ) d \\lambda ," + }, + { + "category_id": 14, + "poly": [ + 602, + 1445, + 1095, + 1445, + 1095, + 1517, + 602, + 1517 + ], + "score": 0.95, + "latex": "\\int _ { S } f ( \\lambda ) \\cdot \\mu _ { W } ( d \\lambda ) \\to \\int f _ { S } ( \\lambda ) \\cdot \\mu _ { M P } ( d \\lambda ) ." + }, + { + "category_id": 14, + "poly": [ + 610, + 298, + 1088, + 298, + 1088, + 365, + 610, + 365 + ], + "score": 0.94, + "latex": "\\operatorname { \\mathbb { E } } _ { X } [ \\beta ^ { T } A ( X ) \\beta ] = { \\frac { 1 } { d } } \\beta ^ { T } \\beta \\operatorname { \\mathbb { E } } _ { X } [ \\operatorname { t r } \\left( A ( X ) \\right) ] ." + }, + { + "category_id": 14, + "poly": [ + 647, + 1965, + 1052, + 1965, + 1052, + 2031, + 647, + 2031 + ], + "score": 0.94, + "latex": "H _ { n } ( x ) = ( - 1 ) ^ { n } e ^ { - x ^ { 2 } / 2 } \\frac { \\partial ^ { n } } { \\partial x ^ { n } } e ^ { - x ^ { 2 } / 2 } ," + }, + { + "category_id": 14, + "poly": [ + 322, + 1575, + 1374, + 1575, + 1374, + 1669, + 322, + 1669 + ], + "score": 0.94, + "latex": "\\operatorname { t r } ( X ^ { \\top } X ) = \\operatorname { t r } ( { \\frac { 1 } { p } } W _ { p } ^ { - 1 } ) = { \\frac { 1 } { p } } \\sum _ { i = 1 } ^ { p } { \\frac { 1 } { \\lambda _ { i } ( W _ { p } ) } } = \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { W _ { p } } ( d \\lambda ) \\int _ { S } { \\frac { 1 } { \\lambda } } \\mu _ { \\operatorname { M P } ( \\gamma ) } ( d \\lambda ) = { \\frac { 1 } { 1 - \\gamma } } ." + }, + { + "category_id": 13, + "poly": [ + 486, + 1406, + 637, + 1406, + 637, + 1440, + 486, + 1440 + ], + "score": 0.94, + "latex": "f ( \\lambda ) \\in C ( S )" + }, + { + "category_id": 13, + "poly": [ + 1090, + 629, + 1347, + 629, + 1347, + 665, + 1090, + 665 + ], + "score": 0.93, + "latex": "A ( X U ) = U A ( X ) U ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 537, + 372, + 626, + 372, + 626, + 405, + 537, + 405 + ], + "score": 0.93, + "latex": "\\beta \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 589, + 1084, + 694, + 1084, + 694, + 1120, + 589, + 1120 + ], + "score": 0.93, + "latex": "\\mu _ { W _ { p } } ( d \\lambda )" + }, + { + "category_id": 13, + "poly": [ + 875, + 631, + 944, + 631, + 944, + 665, + 875, + 665 + ], + "score": 0.93, + "latex": "A ( X )" + }, + { + "category_id": 13, + "poly": [ + 771, + 428, + 866, + 428, + 866, + 466, + 771, + 466 + ], + "score": 0.93, + "latex": "\\{ U _ { i } \\} _ { i = 1 } ^ { d }" + }, + { + "category_id": 14, + "poly": [ + 317, + 527, + 1338, + 527, + 1338, + 623, + 317, + 623 + ], + "score": 0.93, + "latex": "\\mathbb { E } [ \\beta ^ { T } A ( X ) \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } \\mathbb { E } [ \\beta ^ { T } U _ { i } A ( X ) U _ { i } ^ { \\top } \\beta ] = \\frac { 1 } { d } \\sum _ { i = 1 } ^ { d } e _ { i } ^ { T } \\mathbb { E } [ A ( X ) ] e _ { i } = \\frac { \\beta ^ { T } \\beta } { d } \\mathbb { E } [ \\mathrm { t r } ( A ( X ) ) ] ." + }, + { + "category_id": 13, + "poly": [ + 443, + 259, + 698, + 259, + 698, + 295, + 443, + 295 + ], + "score": 0.93, + "latex": "A ( U X ) = U A ( X ) U ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1052, + 460, + 1052, + 460, + 1085, + 298, + 1085 + ], + "score": 0.92, + "latex": "X _ { i } \\sim \\mathcal { N } ( \\mathbf { 0 } , I )" + }, + { + "category_id": 13, + "poly": [ + 856, + 1407, + 956, + 1407, + 956, + 1440, + 856, + 1440 + ], + "score": 0.92, + "latex": "p / n = \\gamma" + }, + { + "category_id": 13, + "poly": [ + 423, + 1895, + 545, + 1895, + 545, + 1930, + 423, + 1930 + ], + "score": 0.92, + "latex": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )" + }, + { + "category_id": 13, + "poly": [ + 1006, + 988, + 1256, + 988, + 1256, + 1023, + 1006, + 1023 + ], + "score": 0.92, + "latex": "A = W _ { p } \\sim W _ { p } ( I , n )" + }, + { + "category_id": 13, + "poly": [ + 1119, + 372, + 1303, + 372, + 1303, + 408, + 1119, + 408 + ], + "score": 0.92, + "latex": "X _ { i j } \\sim \\mathcal { N } ( 0 , \\sigma ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1143, + 1052, + 1352, + 1052, + 1352, + 1085, + 1143, + 1085 + ], + "score": 0.92, + "latex": "p / n = \\gamma \\in ( 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 374, + 988, + 583, + 988, + 583, + 1021, + 374, + 1021 + ], + "score": 0.92, + "latex": "\\delta _ { a } ( x ) = \\delta ( x - a )" + }, + { + "category_id": 13, + "poly": [ + 1132, + 1086, + 1247, + 1086, + 1247, + 1120, + 1132, + 1120 + ], + "score": 0.92, + "latex": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( \\lambda )" + }, + { + "category_id": 13, + "poly": [ + 1062, + 1707, + 1139, + 1707, + 1139, + 1737, + 1062, + 1737 + ], + "score": 0.91, + "latex": "\\gamma 1" + }, + { + "category_id": 13, + "poly": [ + 679, + 1021, + 812, + 1021, + 812, + 1051, + 679, + 1051 + ], + "score": 0.91, + "latex": "\\ b { X } \\in \\mathbb { R } ^ { n \\times p }" + }, + { + "category_id": 13, + "poly": [ + 1173, + 228, + 1298, + 228, + 1298, + 261, + 1173, + 261 + ], + "score": 0.91, + "latex": "X \\in \\mathbb { R } ^ { d \\times n }" + }, + { + "category_id": 13, + "poly": [ + 983, + 428, + 1163, + 428, + 1163, + 464, + 983, + 464 + ], + "score": 0.91, + "latex": "U _ { i } ^ { \\top } \\beta = \\| \\beta \\| e _ { i }" + }, + { + "category_id": 13, + "poly": [ + 705, + 852, + 822, + 852, + 822, + 881, + 705, + 881 + ], + "score": 0.91, + "latex": "A \\in \\mathbb { R } ^ { p \\times p }" + }, + { + "category_id": 13, + "poly": [ + 439, + 1022, + 622, + 1022, + 622, + 1056, + 439, + 1056 + ], + "score": 0.91, + "latex": "W _ { p } = X ^ { \\top } X / n" + }, + { + "category_id": 13, + "poly": [ + 784, + 227, + 952, + 227, + 952, + 263, + 784, + 263 + ], + "score": 0.91, + "latex": "A ( X ) \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 14, + "poly": [ + 354, + 1124, + 1300, + 1124, + 1300, + 1195, + 354, + 1195 + ], + "score": 0.91, + "latex": "\\mu _ { \\mathrm { M P } ( \\gamma ) } ( d \\lambda ) = [ 1 - \\gamma ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\frac { 1 } { 2 \\pi \\gamma \\lambda } \\sqrt { ( ( 1 + \\sqrt \\gamma ) ^ { 2 } - \\lambda ) ( \\lambda - ( 1 - \\sqrt \\gamma ) ^ { 2 } ) } d \\lambda ." + }, + { + "category_id": 13, + "poly": [ + 1254, + 1540, + 1323, + 1540, + 1323, + 1571, + 1254, + 1571 + ], + "score": 0.91, + "latex": "\\gamma < 1" + }, + { + "category_id": 13, + "poly": [ + 585, + 1235, + 975, + 1235, + 975, + 1273, + 585, + 1273 + ], + "score": 0.91, + "latex": "S = \\{ 0 \\} \\cup [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 + \\sqrt { \\gamma } ) ^ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 640, + 465, + 760, + 465, + 760, + 496, + 640, + 496 + ], + "score": 0.91, + "latex": "U _ { i } X \\sim X" + }, + { + "category_id": 13, + "poly": [ + 1082, + 1270, + 1151, + 1270, + 1151, + 1301, + 1082, + 1301 + ], + "score": 0.9, + "latex": "\\gamma \\neq 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 1236, + 370, + 1236, + 370, + 1274, + 298, + 1274 + ], + "score": 0.9, + "latex": "\\sqrt { \\gamma } ) ^ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 1334, + 229, + 1403, + 229, + 1403, + 264, + 1334, + 264 + ], + "score": 0.9, + "latex": "A ( X )" + }, + { + "category_id": 13, + "poly": [ + 791, + 1890, + 1130, + 1890, + 1130, + 1930, + 791, + 1930 + ], + "score": 0.89, + "latex": "\\mu _ { G } ( d x ) = ( \\sqrt { 2 \\pi } ) ^ { - 1 } e ^ { - x ^ { 2 } / 2 } d x" + }, + { + "category_id": 13, + "poly": [ + 422, + 1239, + 543, + 1239, + 543, + 1270, + 422, + 1270 + ], + "score": 0.89, + "latex": "0 < \\gamma < 1" + }, + { + "category_id": 13, + "poly": [ + 974, + 1057, + 1091, + 1057, + 1091, + 1084, + 974, + 1084 + ], + "score": 0.89, + "latex": "n , p \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 381, + 1208, + 459, + 1208, + 459, + 1239, + 381, + 1239 + ], + "score": 0.89, + "latex": "\\mu _ { \\mathrm { M P } ( \\gamma ) }" + }, + { + "category_id": 13, + "poly": [ + 679, + 1410, + 796, + 1410, + 796, + 1439, + 679, + 1439 + ], + "score": 0.89, + "latex": "n , p \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 342, + 1928, + 412, + 1928, + 412, + 1957, + 342, + 1957 + ], + "score": 0.88, + "latex": "n \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 1027, + 1238, + 1097, + 1238, + 1097, + 1269, + 1027, + 1269 + ], + "score": 0.88, + "latex": "\\gamma \\geq 1" + }, + { + "category_id": 13, + "poly": [ + 872, + 465, + 1351, + 465, + 1351, + 499, + 872, + 499 + ], + "score": 0.87, + "latex": "\\mathbb { E } [ A ( X ) ] = \\mathbb { E } [ A ( U _ { i } X ) ] = U _ { i } \\mathbb { E } [ A ( X ) ] U _ { i } ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 520, + 463, + 555, + 463, + 555, + 493, + 520, + 493 + ], + "score": 0.87, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1158, + 1200, + 1410, + 1200, + 1410, + 1239, + 1158, + 1239 + ], + "score": 0.86, + "latex": "S = [ ( 1 - \\sqrt { \\gamma } ) ^ { 2 } , ( 1 +" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1862, + 1354, + 1862, + 1354, + 1893, + 1333, + 1893 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1253, + 436, + 1281, + 436, + 1281, + 462, + 1253, + 462 + ], + "score": 0.84, + "latex": "e _ { i }" + }, + { + "category_id": 13, + "poly": [ + 851, + 374, + 879, + 374, + 879, + 401, + 851, + 401 + ], + "score": 0.84, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 697, + 1272, + 720, + 1272, + 720, + 1296, + 697, + 1296 + ], + "score": 0.83, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 690, + 664, + 713, + 664, + 713, + 690, + 690, + 690 + ], + "score": 0.82, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 866, + 264, + 889, + 264, + 889, + 290, + 866, + 290 + ], + "score": 0.8, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 631, + 1895, + 779, + 1895, + 779, + 1930, + 631, + 1930 + ], + "score": 0.77, + "latex": "G \\sim \\mathcal { N } ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 1357, + 434, + 1369, + 434, + 1369, + 459, + 1357, + 459 + ], + "score": 0.76, + "latex": "i" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1799.0, + 726.0, + 1799.0, + 726.0, + 1839.0, + 293.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 791.0, + 688.0, + 791.0, + 688.0, + 829.0, + 294.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 986.0, + 373.0, + 986.0, + 373.0, + 1023.0, + 294.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 986.0, + 1005.0, + 986.0, + 1005.0, + 1023.0, + 584.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 986.0, + 1407.0, + 986.0, + 1407.0, + 1023.0, + 1257.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1018.0, + 438.0, + 1018.0, + 438.0, + 1056.0, + 293.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1018.0, + 678.0, + 1018.0, + 678.0, + 1056.0, + 623.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1018.0, + 1408.0, + 1018.0, + 1408.0, + 1056.0, + 813.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1051.0, + 973.0, + 1051.0, + 973.0, + 1088.0, + 461.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1051.0, + 1142.0, + 1051.0, + 1142.0, + 1088.0, + 1092.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1051.0, + 1406.0, + 1051.0, + 1406.0, + 1088.0, + 1353.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1081.0, + 588.0, + 1081.0, + 588.0, + 1122.0, + 293.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1081.0, + 1131.0, + 1081.0, + 1131.0, + 1122.0, + 695.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1081.0, + 1262.0, + 1081.0, + 1262.0, + 1122.0, + 1248.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1855.0, + 1332.0, + 1855.0, + 1332.0, + 1898.0, + 294.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1855.0, + 1407.0, + 1855.0, + 1407.0, + 1898.0, + 1355.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1887.0, + 422.0, + 1887.0, + 422.0, + 1936.0, + 291.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1887.0, + 630.0, + 1887.0, + 630.0, + 1936.0, + 546.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1887.0, + 790.0, + 1887.0, + 790.0, + 1936.0, + 780.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1887.0, + 1413.0, + 1887.0, + 1413.0, + 1936.0, + 1131.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1925.0, + 341.0, + 1925.0, + 341.0, + 1959.0, + 295.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1925.0, + 780.0, + 1925.0, + 780.0, + 1959.0, + 413.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1314.0, + 1404.0, + 1314.0, + 1404.0, + 1350.0, + 293.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1345.0, + 1407.0, + 1345.0, + 1407.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1378.0, + 1409.0, + 1378.0, + 1409.0, + 1410.0, + 295.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 485.0, + 1405.0, + 485.0, + 1443.0, + 293.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1405.0, + 678.0, + 1405.0, + 678.0, + 1443.0, + 638.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 1405.0, + 855.0, + 1405.0, + 855.0, + 1443.0, + 797.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1405.0, + 1121.0, + 1405.0, + 1121.0, + 1443.0, + 957.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1197.0, + 380.0, + 1197.0, + 380.0, + 1240.0, + 291.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1197.0, + 1157.0, + 1197.0, + 1157.0, + 1240.0, + 460.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1234.0, + 421.0, + 1234.0, + 421.0, + 1273.0, + 371.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1234.0, + 584.0, + 1234.0, + 584.0, + 1273.0, + 544.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1234.0, + 1026.0, + 1234.0, + 1026.0, + 1273.0, + 976.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1234.0, + 1407.0, + 1234.0, + 1407.0, + 1273.0, + 1098.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1268.0, + 696.0, + 1268.0, + 696.0, + 1303.0, + 296.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1268.0, + 1081.0, + 1268.0, + 1081.0, + 1303.0, + 721.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1268.0, + 1164.0, + 1268.0, + 1164.0, + 1303.0, + 1152.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 424.0, + 770.0, + 424.0, + 770.0, + 471.0, + 290.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 424.0, + 982.0, + 424.0, + 982.0, + 471.0, + 867.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 424.0, + 1252.0, + 424.0, + 1252.0, + 471.0, + 1164.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 424.0, + 1356.0, + 424.0, + 1356.0, + 471.0, + 1282.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 424.0, + 1408.0, + 424.0, + 1408.0, + 471.0, + 1370.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 462.0, + 519.0, + 462.0, + 519.0, + 501.0, + 293.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 462.0, + 639.0, + 462.0, + 639.0, + 501.0, + 556.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 462.0, + 871.0, + 462.0, + 871.0, + 501.0, + 761.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 462.0, + 1406.0, + 462.0, + 1406.0, + 501.0, + 1352.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 493.0, + 373.0, + 493.0, + 373.0, + 530.0, + 291.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1703.0, + 1061.0, + 1703.0, + 1061.0, + 1739.0, + 295.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1703.0, + 1403.0, + 1703.0, + 1403.0, + 1739.0, + 1140.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1734.0, + 694.0, + 1734.0, + 694.0, + 1769.0, + 295.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 627.0, + 874.0, + 627.0, + 874.0, + 667.0, + 293.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 627.0, + 1089.0, + 627.0, + 1089.0, + 667.0, + 945.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 627.0, + 1407.0, + 627.0, + 1407.0, + 667.0, + 1348.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 660.0, + 689.0, + 660.0, + 689.0, + 698.0, + 293.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 660.0, + 1128.0, + 660.0, + 1128.0, + 698.0, + 714.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 662.0, + 1407.0, + 662.0, + 1407.0, + 695.0, + 1374.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 217.0, + 783.0, + 217.0, + 783.0, + 270.0, + 291.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 217.0, + 1172.0, + 217.0, + 1172.0, + 270.0, + 953.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 217.0, + 1333.0, + 217.0, + 1333.0, + 270.0, + 1299.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 217.0, + 1408.0, + 217.0, + 1408.0, + 270.0, + 1404.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 260.0, + 442.0, + 260.0, + 442.0, + 296.0, + 294.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 260.0, + 865.0, + 260.0, + 865.0, + 296.0, + 699.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 260.0, + 957.0, + 260.0, + 957.0, + 296.0, + 890.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1534.0, + 1253.0, + 1534.0, + 1253.0, + 1575.0, + 293.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1534.0, + 1336.0, + 1534.0, + 1336.0, + 1575.0, + 1324.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 845.0, + 704.0, + 845.0, + 704.0, + 894.0, + 290.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 845.0, + 1267.0, + 845.0, + 1267.0, + 894.0, + 823.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 367.0, + 536.0, + 367.0, + 536.0, + 413.0, + 288.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 367.0, + 850.0, + 367.0, + 850.0, + 413.0, + 627.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 367.0, + 1118.0, + 367.0, + 1118.0, + 413.0, + 880.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 367.0, + 1314.0, + 367.0, + 1314.0, + 413.0, + 1304.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 723.0, + 1114.0, + 723.0, + 1114.0, + 764.0, + 293.0, + 764.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 861, + 1407, + 861, + 1407, + 961, + 297, + 961 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1602, + 1403, + 1602, + 1403, + 1698, + 299, + 1698 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 686, + 1238, + 1012, + 1238, + 1012, + 1308, + 686, + 1308 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 492, + 454, + 1205, + 454, + 1205, + 543, + 492, + 543 + ], + "score": 0.956 + }, + { + "category_id": 8, + "poly": [ + 696, + 1980, + 1002, + 1980, + 1002, + 2045, + 696, + 2045 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 301, + 1140, + 1401, + 1140, + 1401, + 1216, + 301, + 1216 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 294, + 1897, + 1408, + 1897, + 1408, + 1961, + 294, + 1961 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 534, + 283, + 1161, + 283, + 1161, + 355, + 534, + 355 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 296, + 577, + 1407, + 577, + 1407, + 642, + 296, + 642 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 300, + 373, + 1402, + 373, + 1402, + 439, + 300, + 439 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 636, + 1819, + 1061, + 1819, + 1061, + 1863, + 636, + 1863 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 717, + 1406, + 717, + 1406, + 782, + 297, + 782 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 297, + 1734, + 1404, + 1734, + 1404, + 1800, + 297, + 1800 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 694, + 662, + 1004, + 662, + 1004, + 701, + 694, + 701 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 554, + 801, + 1143, + 801, + 1143, + 845, + 554, + 845 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 541, + 1375, + 1152, + 1375, + 1152, + 1433, + 541, + 1433 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 295, + 1321, + 1324, + 1321, + 1324, + 1357, + 295, + 1357 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 299, + 1452, + 897, + 1452, + 897, + 1487, + 299, + 1487 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 300, + 229, + 810, + 229, + 810, + 263, + 300, + 263 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 299, + 1542, + 641, + 1542, + 641, + 1575, + 299, + 1575 + ], + "score": 0.928 + }, + { + "category_id": 0, + "poly": [ + 300, + 1008, + 729, + 1008, + 729, + 1045, + 300, + 1045 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 299, + 1081, + 610, + 1081, + 610, + 1115, + 299, + 1115 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.921 + }, + { + "category_id": 9, + "poly": [ + 1351, + 807, + 1400, + 807, + 1400, + 838, + 1351, + 838 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1351, + 301, + 1400, + 301, + 1400, + 333, + 1351, + 333 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1826, + 1400, + 1826, + 1400, + 1858, + 1351, + 1858 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1258, + 1400, + 1258, + 1400, + 1289, + 1351, + 1289 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1352, + 664, + 1400, + 664, + 1400, + 695, + 1352, + 695 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1389, + 1400, + 1389, + 1400, + 1421, + 1351, + 1421 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1992, + 1401, + 1992, + 1401, + 2024, + 1351, + 2024 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 480, + 1400, + 480, + 1400, + 512, + 1351, + 512 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1454, + 1402, + 1454, + 1402, + 1482, + 1374, + 1482 + ], + "score": 0.788 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1454, + 1402, + 1454, + 1402, + 1483, + 1374, + 1483 + ], + "score": 0.184 + }, + { + "category_id": 14, + "poly": [ + 492, + 451, + 1207, + 451, + 1207, + 545, + 492, + 545 + ], + "score": 0.95, + "latex": "\\phi ( x ) = \\sum _ { i = 0 } ^ { \\infty } c _ { i } H _ { i } ( x ) = \\sum _ { i = 0 } ^ { \\infty } \\left( \\frac { 1 } { k ! } \\int _ { \\mathbb { R } } \\phi ( a ) H _ { i } ( a ) \\mu _ { G } ( d a ) \\right) H _ { i } ( x ) ." + }, + { + "category_id": 14, + "poly": [ + 535, + 280, + 1163, + 280, + 1163, + 355, + 535, + 355 + ], + "score": 0.95, + "latex": "\\mathbb { E } [ H _ { j } ( G ) H _ { k } ( G ) ] = \\int _ { \\mathbb { R } } H _ { j } ( x ) H _ { k } ( x ) \\mu _ { G } ( d x ) = j ! \\cdot \\delta _ { j k } ." + }, + { + "category_id": 14, + "poly": [ + 684, + 1235, + 1018, + 1235, + 1018, + 1307, + 684, + 1307 + ], + "score": 0.94, + "latex": "\\frac { \\partial \\pmb { \\theta } ( t ) } { \\partial t } = \\frac { 1 } { n } X ( y - X ^ { \\top } \\pmb { \\theta } ( t ) ) ." + }, + { + "category_id": 14, + "poly": [ + 695, + 1979, + 1005, + 1979, + 1005, + 2045, + 695, + 2045 + ], + "score": 0.94, + "latex": "B 0 ; \\quad V \\frac { \\gamma _ { 1 } } { 1 - \\gamma _ { 1 } } \\sigma ^ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 442, + 1173, + 599, + 1173, + 599, + 1218, + 442, + 1218 + ], + "score": 0.93, + "latex": "\\left\\| \\pmb { y } - X ^ { \\top } \\pmb { \\theta } \\right\\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 881, + 374, + 1003, + 374, + 1003, + 410, + 881, + 410 + ], + "score": 0.93, + "latex": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )" + }, + { + "category_id": 14, + "poly": [ + 544, + 1374, + 1155, + 1374, + 1155, + 1435, + 544, + 1435 + ], + "score": 0.93, + "latex": "\\pmb { \\theta } ( t ) = e ^ { - \\frac { t } { n } X X ^ { \\top } } \\pmb { \\theta } _ { 0 } + ( X X ^ { \\top } ) ^ { \\dagger } \\left( I - e ^ { - \\frac { t } { n } X X ^ { \\top } } \\right) X \\pmb { y } ." + }, + { + "category_id": 13, + "poly": [ + 1155, + 377, + 1210, + 377, + 1210, + 410, + 1155, + 410 + ], + "score": 0.92, + "latex": "\\phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 473, + 1636, + 516, + 1636, + 516, + 1665, + 473, + 1665 + ], + "score": 0.92, + "latex": "\\Phi _ { X }" + }, + { + "category_id": 13, + "poly": [ + 444, + 899, + 620, + 899, + 620, + 931, + 444, + 931 + ], + "score": 0.92, + "latex": "\\mathbb { E } [ \\phi _ { \\bot } ( G ) ^ { 2 } ] = 0" + }, + { + "category_id": 14, + "poly": [ + 636, + 1818, + 1064, + 1818, + 1064, + 1862, + 636, + 1862 + ], + "score": 0.92, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } = W ( X ^ { \\top } W ) ^ { \\dagger } { \\boldsymbol { y } } = ( X X ^ { \\top } ) ^ { - 1 } X { \\boldsymbol { y } } ." + }, + { + "category_id": 13, + "poly": [ + 367, + 376, + 502, + 376, + 502, + 411, + 367, + 411 + ], + "score": 0.92, + "latex": "\\{ H _ { i } ( x ) \\} _ { i = 0 } ^ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 389, + 1734, + 466, + 1734, + 466, + 1765, + 389, + 1765 + ], + "score": 0.92, + "latex": "W ^ { \\top } X" + }, + { + "category_id": 14, + "poly": [ + 694, + 660, + 1003, + 660, + 1003, + 700, + 694, + 700 + ], + "score": 0.92, + "latex": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x + \\phi _ { \\perp } ( x ) ," + }, + { + "category_id": 13, + "poly": [ + 299, + 749, + 420, + 749, + 420, + 783, + 299, + 783 + ], + "score": 0.92, + "latex": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )" + }, + { + "category_id": 13, + "poly": [ + 727, + 1739, + 915, + 1739, + 915, + 1769, + 727, + 1769 + ], + "score": 0.92, + "latex": "\\gamma _ { 1 } < 1 , \\gamma _ { 2 } > \\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 840, + 609, + 896, + 609, + 896, + 643, + 840, + 643 + ], + "score": 0.92, + "latex": "\\phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 494, + 1929, + 638, + 1929, + 638, + 1960, + 494, + 1960 + ], + "score": 0.92, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 686, + 1140, + 776, + 1140, + 776, + 1175, + 686, + 1175 + ], + "score": 0.92, + "latex": "\\pmb { y } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 979, + 1637, + 1022, + 1637, + 1022, + 1665, + 979, + 1665 + ], + "score": 0.92, + "latex": "\\Phi _ { X }" + }, + { + "category_id": 13, + "poly": [ + 366, + 1454, + 448, + 1454, + 448, + 1484, + 366, + 1484 + ], + "score": 0.91, + "latex": "\\pmb { \\theta } _ { 0 } = 0" + }, + { + "category_id": 13, + "poly": [ + 471, + 1140, + 599, + 1140, + 599, + 1171, + 471, + 1171 + ], + "score": 0.91, + "latex": "X \\in \\mathbb { R } ^ { d \\times n }" + }, + { + "category_id": 13, + "poly": [ + 673, + 899, + 858, + 899, + 858, + 931, + 673, + 931 + ], + "score": 0.91, + "latex": "\\phi ( x ) = c _ { 0 } + c _ { 1 } x" + }, + { + "category_id": 14, + "poly": [ + 551, + 800, + 1138, + 800, + 1138, + 843, + 551, + 843 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] = \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } + \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 558, + 719, + 746, + 719, + 746, + 753, + 558, + 753 + ], + "score": 0.9, + "latex": "c _ { 1 } = \\mathbb { E } [ G \\phi ( G ) ]" + }, + { + "category_id": 13, + "poly": [ + 535, + 1456, + 619, + 1456, + 619, + 1482, + 535, + 1482 + ], + "score": 0.9, + "latex": "t \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 516, + 609, + 642, + 609, + 642, + 643, + 516, + 643 + ], + "score": 0.89, + "latex": "H _ { 0 } ( x ) = 1" + }, + { + "category_id": 13, + "poly": [ + 562, + 1325, + 593, + 1325, + 593, + 1355, + 562, + 1355 + ], + "score": 0.89, + "latex": "\\pmb { \\theta } _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 653, + 609, + 782, + 609, + 782, + 643, + 653, + 643 + ], + "score": 0.88, + "latex": "H _ { 1 } ( x ) = x" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1736, + 1397, + 1736, + 1397, + 1770, + 1127, + 1770 + ], + "score": 0.87, + "latex": "( \\Phi _ { X } ) = d < \\operatorname* { m i n } ( n , h )" + }, + { + "category_id": 13, + "poly": [ + 471, + 864, + 1087, + 864, + 1087, + 900, + 471, + 900 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } - \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 } = \\mathbb { E } [ \\phi _ { \\perp } ( G ) ^ { 2 } ] \\ge 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 375, + 719, + 542, + 719, + 542, + 753, + 375, + 753 + ], + "score": 0.84, + "latex": "c _ { 0 } = \\mathbb { E } [ \\phi ( G ) ]" + }, + { + "category_id": 13, + "poly": [ + 1040, + 1144, + 1059, + 1144, + 1059, + 1170, + 1040, + 1170 + ], + "score": 0.79, + "latex": "\\pmb \\theta" + }, + { + "category_id": 13, + "poly": [ + 1304, + 1144, + 1322, + 1144, + 1322, + 1170, + 1304, + 1170 + ], + "score": 0.79, + "latex": "\\pmb \\theta" + }, + { + "category_id": 13, + "poly": [ + 965, + 1327, + 979, + 1327, + 979, + 1351, + 965, + 1351 + ], + "score": 0.78, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 926, + 1610, + 995, + 1610, + 995, + 1636, + 926, + 1636 + ], + "score": 0.78, + "latex": "\\gamma _ { 1 } , \\gamma _ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1540.0, + 644.0, + 1540.0, + 644.0, + 1579.0, + 295.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1004.0, + 731.0, + 1004.0, + 731.0, + 1051.0, + 295.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1079.0, + 615.0, + 1079.0, + 615.0, + 1119.0, + 294.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1458.0, + 1402.0, + 1458.0, + 1402.0, + 1485.0, + 1378.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1458.0, + 1402.0, + 1458.0, + 1402.0, + 1486.0, + 1378.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 860.0, + 470.0, + 860.0, + 470.0, + 904.0, + 291.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 860.0, + 1408.0, + 860.0, + 1408.0, + 904.0, + 1088.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 893.0, + 443.0, + 893.0, + 443.0, + 932.0, + 294.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 893.0, + 672.0, + 893.0, + 672.0, + 932.0, + 621.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 893.0, + 1405.0, + 893.0, + 1405.0, + 932.0, + 859.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 926.0, + 1407.0, + 926.0, + 1407.0, + 961.0, + 295.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1603.0, + 925.0, + 1603.0, + 925.0, + 1638.0, + 295.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1603.0, + 1404.0, + 1603.0, + 1404.0, + 1638.0, + 996.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1634.0, + 472.0, + 1634.0, + 472.0, + 1665.0, + 295.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1634.0, + 978.0, + 1634.0, + 978.0, + 1665.0, + 517.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1634.0, + 1404.0, + 1634.0, + 1404.0, + 1665.0, + 1023.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1666.0, + 917.0, + 1666.0, + 917.0, + 1697.0, + 295.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1134.0, + 470.0, + 1134.0, + 470.0, + 1179.0, + 295.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1134.0, + 685.0, + 1134.0, + 685.0, + 1179.0, + 600.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1134.0, + 1039.0, + 1134.0, + 1039.0, + 1179.0, + 777.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1134.0, + 1303.0, + 1134.0, + 1303.0, + 1179.0, + 1060.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1134.0, + 1406.0, + 1134.0, + 1406.0, + 1179.0, + 1323.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1157.0, + 441.0, + 1157.0, + 441.0, + 1226.0, + 287.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1157.0, + 805.0, + 1157.0, + 805.0, + 1226.0, + 600.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1896.0, + 1407.0, + 1896.0, + 1407.0, + 1932.0, + 293.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1924.0, + 493.0, + 1924.0, + 493.0, + 1964.0, + 295.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1924.0, + 651.0, + 1924.0, + 651.0, + 1964.0, + 639.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 573.0, + 1407.0, + 573.0, + 1407.0, + 615.0, + 294.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 608.0, + 515.0, + 608.0, + 515.0, + 644.0, + 296.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 608.0, + 652.0, + 608.0, + 652.0, + 644.0, + 643.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 608.0, + 839.0, + 608.0, + 839.0, + 644.0, + 783.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 608.0, + 1123.0, + 608.0, + 1123.0, + 644.0, + 897.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 371.0, + 366.0, + 371.0, + 366.0, + 412.0, + 293.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 371.0, + 880.0, + 371.0, + 880.0, + 412.0, + 503.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 371.0, + 1154.0, + 371.0, + 1154.0, + 412.0, + 1004.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 371.0, + 1407.0, + 371.0, + 1407.0, + 412.0, + 1211.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 403.0, + 599.0, + 403.0, + 599.0, + 442.0, + 297.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 716.0, + 374.0, + 716.0, + 374.0, + 756.0, + 294.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 716.0, + 557.0, + 716.0, + 557.0, + 756.0, + 543.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 716.0, + 1406.0, + 716.0, + 1406.0, + 756.0, + 747.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 744.0, + 298.0, + 744.0, + 298.0, + 787.0, + 295.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 744.0, + 1025.0, + 744.0, + 1025.0, + 787.0, + 421.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1731.0, + 388.0, + 1731.0, + 388.0, + 1774.0, + 295.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1731.0, + 726.0, + 1731.0, + 726.0, + 1774.0, + 467.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1731.0, + 1126.0, + 1731.0, + 1126.0, + 1774.0, + 916.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1731.0, + 1408.0, + 1731.0, + 1408.0, + 1774.0, + 1398.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1766.0, + 928.0, + 1766.0, + 928.0, + 1801.0, + 293.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1319.0, + 561.0, + 1319.0, + 561.0, + 1360.0, + 294.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1319.0, + 964.0, + 1319.0, + 964.0, + 1360.0, + 594.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1319.0, + 1325.0, + 1319.0, + 1325.0, + 1360.0, + 980.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1452.0, + 365.0, + 1452.0, + 365.0, + 1488.0, + 297.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1452.0, + 534.0, + 1452.0, + 534.0, + 1488.0, + 449.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1452.0, + 896.0, + 1452.0, + 896.0, + 1488.0, + 620.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 811.0, + 227.0, + 811.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 508, + 1403, + 508, + 1403, + 606, + 299, + 606 + ], + "score": 0.976 + }, + { + "category_id": 8, + "poly": [ + 412, + 620, + 1288, + 620, + 1288, + 806, + 412, + 806 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 359, + 1547, + 1339, + 1547, + 1339, + 1775, + 359, + 1775 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 314, + 868, + 1383, + 868, + 1383, + 1315, + 314, + 1315 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 733, + 1836, + 965, + 1836, + 965, + 1907, + 733, + 1907 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 291, + 1328, + 1402, + 1328, + 1402, + 1397, + 291, + 1397 + ], + "score": 0.956 + }, + { + "category_id": 8, + "poly": [ + 706, + 416, + 992, + 416, + 992, + 497, + 706, + 497 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 476, + 1410, + 1220, + 1410, + 1220, + 1487, + 476, + 1487 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 293, + 227, + 1404, + 227, + 1404, + 295, + 293, + 295 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 653, + 307, + 1044, + 307, + 1044, + 351, + 653, + 351 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 1499, + 705, + 1499, + 705, + 1533, + 297, + 1533 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 299, + 1787, + 785, + 1787, + 785, + 1819, + 299, + 1819 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 291, + 1935, + 1376, + 1935, + 1376, + 1973, + 291, + 1973 + ], + "score": 0.929 + }, + { + "category_id": 8, + "poly": [ + 653, + 1985, + 1044, + 1985, + 1044, + 2029, + 653, + 2029 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 296, + 368, + 1350, + 368, + 1350, + 404, + 296, + 404 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 294, + 819, + 1270, + 819, + 1270, + 855, + 294, + 855 + ], + "score": 0.92 + }, + { + "category_id": 9, + "poly": [ + 1351, + 439, + 1400, + 439, + 1400, + 470, + 1351, + 470 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1850, + 1401, + 1850, + 1401, + 1882, + 1351, + 1882 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1263, + 1401, + 1263, + 1401, + 1294, + 1351, + 1294 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1351, + 315, + 1400, + 315, + 1400, + 346, + 1351, + 346 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1994, + 1401, + 1994, + 1401, + 2024, + 1351, + 2024 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1720, + 1401, + 1720, + 1401, + 1750, + 1351, + 1750 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1351, + 755, + 1401, + 755, + 1401, + 787, + 1351, + 787 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1432, + 1401, + 1432, + 1401, + 1464, + 1350, + 1464 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 863, + 2087, + 863, + 2113, + 835, + 2113 + ], + "score": 0.869 + }, + { + "category_id": 14, + "poly": [ + 318, + 870, + 1383, + 870, + 1383, + 1319, + 318, + 1319 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { B = \\| { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } \\beta - \\beta \\| _ { 2 } ^ { 2 } } \\\\ & { \\quad = \\beta ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) \\beta } \\\\ & { \\stackrel { ( \\psi ) } { = } \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( { \\cal { W } } ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma X ^ { \\top } ( V \\Sigma X ^ { \\top } X X ^ { \\top } U \\Sigma \\Sigma { \\cal { W } } ^ { \\top } ) ^ { - 1 } V \\Sigma \\Sigma ^ { \\top } { \\cal { U } } ^ { \\top } X X ^ { \\top } - I _ { d } \\Big ) ^ { \\top } \\Big ( \\cdots ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( { \\cal { W } } \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma ^ { \\top } \\Sigma \\sigma \\Sigma ^ { \\top } \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & { \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( \\Big ( \\Sigma \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\Sigma \\Big ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } { \\cal { U } } ^ { \\top } - { \\cal { W } } \\Sigma \\Big ) ^ { \\top } \\Big ( \\cdots \\Big ) ) } \\\\ & \\quad = \\frac { { r ^ { 2 } } } { d } \\mathrm { t r } ( ( \\Sigma ^ { \\top } X ^ { \\top } \\Sigma ) ^ { - 1 } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 705, + 416, + 991, + 416, + 991, + 495, + 705, + 495 + ], + "score": 0.95, + "latex": "\\Sigma = \\left[ \\Sigma _ { 0 } \\right] , X = \\left[ \\Sigma _ { 1 } \\right] ," + }, + { + "category_id": 14, + "poly": [ + 732, + 1835, + 968, + 1835, + 968, + 1904, + 732, + 1904 + ], + "score": 0.95, + "latex": "B \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } ( 1 - \\gamma _ { 2 } ) } r ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 410, + 617, + 1290, + 617, + 1290, + 811, + 410, + 811 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } X ^ { \\top } \\sigma ^ { 2 } X W \\left( \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } W ^ { \\top } \\right) \\right. } \\\\ & { \\left. = \\sigma ^ { 2 } \\mathrm { t r } \\left( W ^ { \\top } W \\left( W ^ { \\top } X X ^ { \\top } W \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } U ^ { \\top } X X ^ { \\top } U \\Sigma \\right) ^ { - 1 } \\right) \\right. } \\\\ & { \\left. \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( \\Sigma ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\right) = \\sigma ^ { 2 } \\mathrm { t r } \\left( \\left( X _ { 0 } X _ { 0 } ^ { \\top } \\right) ^ { - 1 } \\right) \\right. \\sigma ^ { 2 } \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 357, + 1545, + 1340, + 1545, + 1340, + 1776, + 357, + 1776 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { { \\displaystyle B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) ^ { \\top } \\left( \\Sigma ( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma ) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } \\right) \\right) } } \\\\ { { \\mathrm { } = \\frac { r ^ { 2 } } { d } \\left( \\mathrm { t r } \\left( ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } X _ { 1 } X _ { 0 } ^ { \\top } ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } \\right) + ( d - h ) \\right) } } \\\\ { { \\mathrm { } \\to \\frac { r ^ { 2 } } { d } \\left( \\frac { ( d - h ) h } { n - h - 1 } + d - h \\right) \\to \\frac { \\gamma _ { 1 } - \\gamma _ { 2 } } { \\gamma _ { 1 } \\left( 1 - \\gamma _ { 2 } \\right) } r ^ { 2 } . } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 652, + 307, + 1047, + 307, + 1047, + 351, + 652, + 351 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\hat { \\beta } = W ( W ^ { \\top } X X ^ { \\top } W ) ^ { - 1 } W ^ { \\top } X y . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 770, + 1939, + 941, + 1939, + 941, + 1972, + 770, + 1972 + ], + "score": 0.92, + "latex": "\\gamma _ { 1 } > 1 , \\gamma _ { 2 } > 1" + }, + { + "category_id": 13, + "poly": [ + 814, + 232, + 1007, + 232, + 1007, + 264, + 814, + 264 + ], + "score": 0.92, + "latex": "\\gamma _ { 2 } < 1 , \\gamma _ { 1 } > \\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 385, + 368, + 536, + 368, + 536, + 399, + 385, + 399 + ], + "score": 0.92, + "latex": "W = U \\Sigma V ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 653, + 1986, + 1044, + 1986, + 1044, + 2028, + 653, + 2028 + ], + "score": 0.91, + "latex": "\\hat { \\beta } = W W ^ { \\top } X ( X ^ { \\top } W W ^ { \\top } X ) ^ { - 1 } \\pmb { y } ," + }, + { + "category_id": 13, + "poly": [ + 1052, + 513, + 1138, + 513, + 1138, + 544, + 1052, + 544 + ], + "score": 0.91, + "latex": "X _ { 0 } , X _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 784, + 1361, + 905, + 1361, + 905, + 1397, + 784, + 1397 + ], + "score": 0.91, + "latex": "\\beta ^ { \\top } \\beta = r ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 477, + 1409, + 1222, + 1409, + 1222, + 1487, + 477, + 1487 + ], + "score": 0.91, + "latex": "\\Sigma \\left( \\Sigma ^ { \\top } X X ^ { \\top } \\Sigma \\right) ^ { - 1 } \\Sigma ^ { \\top } X X ^ { \\top } - I _ { d } = \\left[ \\begin{array} { l l } { 0 } & { ( X _ { 0 } X _ { 0 } ^ { \\top } ) ^ { - 1 } X _ { 0 } X _ { 1 } ^ { \\top } } \\\\ { 0 } & { - I _ { d - h } } \\end{array} \\right] ." + }, + { + "category_id": 13, + "poly": [ + 637, + 1788, + 783, + 1788, + 783, + 1819, + 637, + 1819 + ], + "score": 0.9, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 373, + 507, + 862, + 507, + 862, + 545, + 373, + 545 + ], + "score": 0.9, + "latex": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { h \\times h } , X _ { 0 } \\in \\mathbb { R } ^ { h \\times n } , X _ { 1 } \\in \\mathbb { R } ^ { ( d - h ) \\times n }" + }, + { + "category_id": 13, + "poly": [ + 409, + 1936, + 486, + 1936, + 486, + 1967, + 409, + 1967 + ], + "score": 0.88, + "latex": "W ^ { \\top } X" + }, + { + "category_id": 13, + "poly": [ + 401, + 227, + 479, + 227, + 479, + 260, + 401, + 260 + ], + "score": 0.85, + "latex": "W ^ { \\top } X" + }, + { + "category_id": 13, + "poly": [ + 803, + 1331, + 861, + 1331, + 861, + 1360, + 803, + 1360 + ], + "score": 0.81, + "latex": "( \\cdot \\cdot \\cdot )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 500.0, + 372.0, + 500.0, + 372.0, + 551.0, + 290.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 500.0, + 1051.0, + 500.0, + 1051.0, + 551.0, + 863.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 500.0, + 1410.0, + 500.0, + 1410.0, + 551.0, + 1139.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 541.0, + 1407.0, + 541.0, + 1407.0, + 578.0, + 293.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 573.0, + 1181.0, + 573.0, + 1181.0, + 606.0, + 295.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1327.0, + 802.0, + 1327.0, + 802.0, + 1364.0, + 295.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1327.0, + 1405.0, + 1327.0, + 1405.0, + 1364.0, + 862.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1360.0, + 783.0, + 1360.0, + 783.0, + 1401.0, + 293.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1360.0, + 1297.0, + 1360.0, + 1297.0, + 1401.0, + 906.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 224.0, + 400.0, + 224.0, + 400.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 224.0, + 813.0, + 224.0, + 813.0, + 266.0, + 480.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 224.0, + 1406.0, + 224.0, + 1406.0, + 266.0, + 1008.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 605.0, + 260.0, + 605.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1496.0, + 704.0, + 1496.0, + 704.0, + 1536.0, + 295.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1781.0, + 636.0, + 1781.0, + 636.0, + 1824.0, + 295.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1930.0, + 408.0, + 1930.0, + 408.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1930.0, + 769.0, + 1930.0, + 769.0, + 1978.0, + 487.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1930.0, + 1380.0, + 1930.0, + 1380.0, + 1978.0, + 942.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 364.0, + 384.0, + 364.0, + 384.0, + 408.0, + 292.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 364.0, + 1350.0, + 364.0, + 1350.0, + 408.0, + 537.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 818.0, + 1271.0, + 818.0, + 1271.0, + 858.0, + 294.0, + 858.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 317, + 727, + 1462, + 727, + 1462, + 975, + 317, + 975 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 436, + 282, + 1262, + 282, + 1262, + 465, + 436, + 465 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 615, + 1883, + 1081, + 1883, + 1081, + 1961, + 615, + 1961 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 732, + 1229, + 965, + 1229, + 965, + 1307, + 732, + 1307 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 701, + 573, + 995, + 573, + 995, + 654, + 701, + 654 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 295, + 1653, + 1404, + 1653, + 1404, + 1720, + 295, + 1720 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 631, + 1549, + 1068, + 1549, + 1068, + 1619, + 631, + 1619 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 633, + 1742, + 1067, + 1742, + 1067, + 1812, + 633, + 1812 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 295, + 486, + 1406, + 486, + 1406, + 552, + 295, + 552 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 1828, + 685, + 1828, + 685, + 1862, + 297, + 1862 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 293, + 672, + 1405, + 672, + 1405, + 712, + 293, + 712 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 294, + 1980, + 1351, + 1980, + 1351, + 2016, + 294, + 2016 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 299, + 1340, + 774, + 1340, + 774, + 1374, + 299, + 1374 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 583, + 229, + 583, + 263, + 297, + 263 + ], + "score": 0.929 + }, + { + "category_id": 8, + "poly": [ + 306, + 1046, + 1382, + 1046, + 1382, + 1123, + 306, + 1123 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 295, + 1175, + 1035, + 1175, + 1035, + 1210, + 295, + 1210 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 297, + 993, + 945, + 993, + 945, + 1027, + 297, + 1027 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 299, + 1432, + 681, + 1432, + 681, + 1466, + 299, + 1466 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 294, + 1492, + 1388, + 1492, + 1388, + 1529, + 294, + 1529 + ], + "score": 0.921 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1569, + 1400, + 1569, + 1400, + 1601, + 1351, + 1601 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1762, + 1400, + 1762, + 1400, + 1794, + 1351, + 1794 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1901, + 1400, + 1901, + 1400, + 1933, + 1351, + 1933 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1351, + 419, + 1401, + 419, + 1401, + 451, + 1351, + 451 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1250, + 1401, + 1250, + 1401, + 1282, + 1351, + 1282 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 596, + 1401, + 596, + 1401, + 627, + 1351, + 627 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1123, + 1401, + 1123, + 1401, + 1151, + 1351, + 1151 + ], + "score": 0.857 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1341, + 1403, + 1341, + 1403, + 1371, + 1374, + 1371 + ], + "score": 0.828 + }, + { + "category_id": 2, + "poly": [ + 1376, + 1983, + 1402, + 1983, + 1402, + 2009, + 1376, + 2009 + ], + "score": 0.661 + }, + { + "category_id": 9, + "poly": [ + 1353, + 927, + 1401, + 927, + 1401, + 961, + 1353, + 961 + ], + "score": 0.194 + }, + { + "category_id": 14, + "poly": [ + 316, + 727, + 1428, + 727, + 1428, + 979, + 316, + 979 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { V = \\sigma ^ { 2 } \\mathrm { t r } ( W W ^ { \\top } \\Sigma ( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma ) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } ) } \\\\ & \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( [ \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { \\cdots } & { \\cdots } \\\\ { ( \\begin{array} { l l l } { W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 0 } ^ { \\top } } & { W _ { 1 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ( \\Sigma _ { 0 } ^ { \\top } W _ { 0 } W _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 2 } \\Sigma _ { 0 } W _ { 0 } W _ { 1 } ^ { \\top } } \\end{array} ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( \\mathrm { t r } ( \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ) + \\mathrm { t r } ( W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } \\Sigma _ { 0 } ^ { - T } \\Sigma _ { 0 } ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } W _ { 1 } ^ { \\top } ) ) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } ( ( X ^ { \\top } X ) ^ { - 1 } ) + \\sigma ^ { 2 } \\mathrm { t r } ( W _ { 1 } ^ { \\top } W _ { 1 } W _ { 0 } ^ { \\top } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } W _ { 0 } ) \\cdot \\ ( 3 9 ) } \\end{array} \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 434, + 281, + 1264, + 281, + 1264, + 469, + 434, + 469 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { V = \\mathrm { t r } \\left( W W ^ { \\top } X \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } \\sigma ^ { 2 } \\left( X ^ { \\top } W W ^ { \\top } X \\right) ^ { - 1 } X ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } U \\Sigma V ^ { \\top } \\left( V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } U \\Sigma V ^ { \\top } \\right) ^ { - 2 } V \\Sigma ^ { \\top } U ^ { \\top } W W ^ { \\top } \\right) } \\\\ & { \\quad \\sim \\sigma ^ { 2 } \\mathrm { t r } \\left( W W ^ { \\top } \\Sigma \\left( \\Sigma ^ { \\top } W W ^ { \\top } \\Sigma \\right) ^ { - 2 } \\Sigma ^ { \\top } W W ^ { \\top } \\right) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 361, + 1685, + 551, + 1685, + 551, + 1720, + 361, + 1720 + ], + "score": 0.95, + "latex": "W ( t ) = \\bar { \\hat { w } } ( t ) \\mathbf { { a } } ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 732, + 1228, + 968, + 1228, + 968, + 1305, + 732, + 1305 + ], + "score": 0.95, + "latex": "B \\frac { \\gamma _ { 2 } ( \\gamma _ { 1 } - 1 ) } { \\gamma _ { 1 } ( \\gamma _ { 2 } - 1 ) } r ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 632, + 1739, + 1066, + 1739, + 1066, + 1811, + 632, + 1811 + ], + "score": 0.94, + "latex": "\\frac { \\partial \\pmb { \\hat { w } } ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - \\boldsymbol { X } ^ { \\top } \\pmb { \\hat { w } } ( t ) \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } ) ," + }, + { + "category_id": 14, + "poly": [ + 617, + 1881, + 1083, + 1881, + 1083, + 1961, + 617, + 1961 + ], + "score": 0.94, + "latex": "{ \\hat { \\pmb { w } } } ^ { * } = { \\frac { 1 } { \\left\\| \\pmb { a } \\right\\| _ { 2 } ^ { 2 } } } { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } \\Rightarrow { \\hat { \\beta } } = { \\boldsymbol { W } } ^ { * } \\pmb { a } = { \\boldsymbol { X } } ^ { \\dagger } \\pmb { y } ." + }, + { + "category_id": 14, + "poly": [ + 631, + 1547, + 1069, + 1547, + 1069, + 1618, + 631, + 1618 + ], + "score": 0.94, + "latex": "\\frac { \\partial W ( t ) } { \\partial t } = - \\frac { 1 } { n } X ( \\pmb { y } - X ^ { \\top } W ( t ) \\pmb { a } ) \\pmb { a } ^ { \\top } ." + }, + { + "category_id": 14, + "poly": [ + 701, + 572, + 995, + 572, + 995, + 652, + 701, + 652 + ], + "score": 0.93, + "latex": "\\Sigma = \\left[ \\stackrel { \\Sigma _ { 0 } } { 0 } \\right] , W = \\left[ \\stackrel { W _ { 0 } } { W _ { 1 } } \\right] ," + }, + { + "category_id": 13, + "poly": [ + 1184, + 1655, + 1306, + 1655, + 1306, + 1689, + 1184, + 1689 + ], + "score": 0.93, + "latex": "W ( 0 ) = 0" + }, + { + "category_id": 13, + "poly": [ + 634, + 487, + 780, + 487, + 780, + 519, + 634, + 519 + ], + "score": 0.91, + "latex": "X = U \\Sigma V ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 308, + 1046, + 1388, + 1046, + 1388, + 1123, + 308, + 1123 + ], + "score": 0.91, + "latex": "V \\sigma ^ { 2 } \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\sigma ^ { 2 } ( d - n ) \\mathbb { E } _ { W , X } V \\mathrm { t r } ( ( W _ { 0 } W _ { 0 } ^ { \\top } ) ^ { - 1 } ( \\Sigma _ { 0 } ^ { \\top } \\Sigma _ { 0 } ) ^ { - 1 } ) \\sigma ^ { 2 } ( \\frac { 1 } { \\gamma _ { 1 } - 1 } + \\frac { 1 } { \\gamma _ { 2 } - 1 } ) ." + }, + { + "category_id": 13, + "poly": [ + 908, + 676, + 1000, + 676, + 1000, + 709, + 908, + 709 + ], + "score": 0.9, + "latex": "W _ { 0 } , W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 370, + 672, + 851, + 672, + 851, + 709, + 370, + 709 + ], + "score": 0.88, + "latex": "\\Sigma _ { 0 } \\in \\mathbb { R } ^ { n \\times n } , W _ { 0 } \\in \\mathbb { R } ^ { n \\times h } , W _ { 1 } \\in \\mathbb { R } ^ { ( d - n ) \\times h }" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1687, + 1033, + 1687, + 1033, + 1714, + 1007, + 1714 + ], + "score": 0.84, + "latex": "\\hat { w }" + }, + { + "category_id": 13, + "poly": [ + 662, + 1687, + 688, + 1687, + 688, + 1714, + 662, + 1714 + ], + "score": 0.83, + "latex": "\\hat { \\textbf { \\textit { w } } }" + }, + { + "category_id": 13, + "poly": [ + 1347, + 679, + 1370, + 679, + 1370, + 704, + 1347, + 704 + ], + "score": 0.8, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 571, + 522, + 603, + 522, + 603, + 548, + 571, + 548 + ], + "score": 0.8, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 567, + 1657, + 600, + 1657, + 600, + 1683, + 567, + 1683 + ], + "score": 0.79, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 913, + 1496, + 945, + 1496, + 945, + 1523, + 913, + 1523 + ], + "score": 0.77, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1662, + 1098, + 1662, + 1098, + 1683, + 1078, + 1683 + ], + "score": 0.76, + "latex": "^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 1502, + 1234, + 1502, + 1234, + 1523, + 1214, + 1523 + ], + "score": 0.73, + "latex": "^ { a }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1432.0, + 685.0, + 1432.0, + 685.0, + 1469.0, + 296.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2122.0, + 831.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1344.0, + 1404.0, + 1344.0, + 1404.0, + 1375.0, + 1376.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1986.0, + 1403.0, + 1986.0, + 1403.0, + 2013.0, + 1378.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1650.0, + 566.0, + 1650.0, + 566.0, + 1693.0, + 292.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1650.0, + 1077.0, + 1650.0, + 1077.0, + 1693.0, + 601.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1650.0, + 1183.0, + 1650.0, + 1183.0, + 1693.0, + 1099.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1650.0, + 1405.0, + 1650.0, + 1405.0, + 1693.0, + 1307.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1685.0, + 360.0, + 1685.0, + 360.0, + 1722.0, + 295.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1685.0, + 661.0, + 1685.0, + 661.0, + 1722.0, + 552.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 1685.0, + 1006.0, + 1685.0, + 1006.0, + 1722.0, + 689.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1685.0, + 1064.0, + 1685.0, + 1064.0, + 1722.0, + 1034.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 484.0, + 633.0, + 484.0, + 633.0, + 527.0, + 293.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 484.0, + 1406.0, + 484.0, + 1406.0, + 527.0, + 781.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 519.0, + 570.0, + 519.0, + 570.0, + 553.0, + 294.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 519.0, + 616.0, + 519.0, + 616.0, + 553.0, + 604.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1829.0, + 685.0, + 1829.0, + 685.0, + 1863.0, + 297.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 664.0, + 369.0, + 664.0, + 369.0, + 717.0, + 291.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 664.0, + 907.0, + 664.0, + 907.0, + 717.0, + 852.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 664.0, + 1346.0, + 664.0, + 1346.0, + 717.0, + 1001.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 664.0, + 1407.0, + 664.0, + 1407.0, + 717.0, + 1371.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1977.0, + 1352.0, + 1977.0, + 1352.0, + 2018.0, + 292.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1334.0, + 778.0, + 1334.0, + 778.0, + 1380.0, + 294.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 584.0, + 229.0, + 584.0, + 266.0, + 297.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1175.0, + 1035.0, + 1175.0, + 1035.0, + 1212.0, + 296.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 992.0, + 945.0, + 992.0, + 945.0, + 1029.0, + 296.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1491.0, + 912.0, + 1491.0, + 912.0, + 1530.0, + 296.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1491.0, + 1213.0, + 1491.0, + 1213.0, + 1530.0, + 946.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 1491.0, + 1389.0, + 1491.0, + 1389.0, + 1530.0, + 1235.0, + 1530.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1350, + 1406, + 1350, + 1406, + 1452, + 297, + 1452 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 748, + 1406, + 748, + 1406, + 874, + 297, + 874 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 496, + 325, + 1201, + 325, + 1201, + 572, + 496, + 572 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 399, + 1459, + 1292, + 1459, + 1292, + 1880, + 399, + 1880 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 296, + 956, + 1405, + 956, + 1405, + 1055, + 296, + 1055 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 293, + 887, + 1401, + 887, + 1401, + 951, + 293, + 951 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 341, + 1271, + 1348, + 1271, + 1348, + 1342, + 341, + 1342 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 296, + 683, + 1019, + 683, + 1019, + 737, + 296, + 737 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 656, + 619, + 1043, + 619, + 1043, + 675, + 656, + 675 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 296, + 579, + 1084, + 579, + 1084, + 614, + 296, + 614 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 292, + 1165, + 1344, + 1165, + 1344, + 1202, + 292, + 1202 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 295, + 1229, + 853, + 1229, + 853, + 1263, + 295, + 1263 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 296, + 1885, + 1403, + 1885, + 1403, + 1924, + 296, + 1924 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 291, + 285, + 1367, + 285, + 1367, + 320, + 291, + 320 + ], + "score": 0.909 + }, + { + "category_id": 8, + "poly": [ + 576, + 1063, + 1119, + 1063, + 1119, + 1105, + 576, + 1105 + ], + "score": 0.9 + }, + { + "category_id": 1, + "poly": [ + 297, + 1111, + 967, + 1111, + 967, + 1147, + 297, + 1147 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1351, + 632, + 1401, + 632, + 1401, + 663, + 1351, + 663 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1836, + 1401, + 1836, + 1401, + 1868, + 1351, + 1868 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1351, + 528, + 1401, + 528, + 1401, + 559, + 1351, + 559 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.871 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1969, + 1401, + 1969, + 1401, + 2001, + 1350, + 2001 + ], + "score": 0.864 + }, + { + "category_id": 0, + "poly": [ + 298, + 228, + 642, + 228, + 642, + 262, + 298, + 262 + ], + "score": 0.826 + }, + { + "category_id": 8, + "poly": [ + 559, + 1930, + 1134, + 1930, + 1134, + 2045, + 559, + 2045 + ], + "score": 0.731 + }, + { + "category_id": 8, + "poly": [ + 559, + 1930, + 1135, + 1930, + 1135, + 2045, + 559, + 2045 + ], + "score": 0.699 + }, + { + "category_id": 14, + "poly": [ + 394, + 1460, + 1294, + 1460, + 1294, + 1876, + 394, + 1876 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { \\mathrm { t r } \\left( h ^ { - 1 } \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) - h ^ { - 1 } I \\right) } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } \\left\\| \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } - \\left( \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right) ^ { - 1 } \\right\\| _ { F } } \\\\ & { \\quad \\cdot \\left\\| \\frac { n } { h } \\frac { K _ { W } } { 1 + h ^ { - 1 } \\mathrm { t r } \\big ( h \\big ( \\Phi \\Phi ^ { \\top } - \\xi I \\big ) ^ { - 1 } K _ { W } \\big ) } \\right\\| _ { 2 } } \\\\ & { \\le \\displaystyle \\frac { 1 } { h } O ( n ^ { - 1 / 2 + \\varepsilon } ) O ( n ^ { 1 / 2 } ) \\to 0 . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 493, + 324, + 1201, + 324, + 1201, + 576, + 493, + 576 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { V = \\mathbb { E } _ { \\pmb { x } , \\pmb { \\varepsilon } } \\Big [ \\| \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) - \\mathbb { E } _ { \\pmb { \\varepsilon } } \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\Big \\| \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\phi ( W ^ { \\top } \\pmb { x } ) \\Big \\| _ { 2 } ^ { 2 } \\Big ] } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] \\right) } \\\\ & { ~ = \\mathrm { t r } \\left( \\big [ \\phi ( X ^ { \\top } W ) \\big ] ^ { \\dagger } \\big [ \\phi ( W ^ { \\top } \\pmb { X } ) \\big ] ^ { \\dagger } K _ { W } \\right) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 600, + 1410, + 775, + 1410, + 775, + 1456, + 600, + 1456 + ], + "score": 0.94, + "latex": "\\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 }" + }, + { + "category_id": 14, + "poly": [ + 560, + 1928, + 1137, + 1928, + 1137, + 2045, + 560, + 2045 + ], + "score": 0.94, + "latex": "\\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } \\frac { n } { h } \\frac { \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } { 1 + \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) } = 1 ." + }, + { + "category_id": 13, + "poly": [ + 1235, + 1350, + 1399, + 1350, + 1399, + 1384, + 1235, + 1384 + ], + "score": 0.93, + "latex": "\\gamma _ { 2 } = h / n \\neq 1" + }, + { + "category_id": 13, + "poly": [ + 678, + 1887, + 942, + 1887, + 942, + 1923, + 678, + 1923 + ], + "score": 0.92, + "latex": "\\operatorname { t r } \\left( A B \\right) \\leq \\left\\| A \\right\\| _ { F } \\left\\| B \\right\\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1112, + 460, + 1112, + 460, + 1146, + 371, + 1146 + ], + "score": 0.92, + "latex": "\\mu ^ { + } ( d \\lambda )" + }, + { + "category_id": 13, + "poly": [ + 432, + 987, + 516, + 987, + 516, + 1020, + 432, + 1020 + ], + "score": 0.92, + "latex": "O ( { \\sqrt { n } } )" + }, + { + "category_id": 13, + "poly": [ + 777, + 1350, + 862, + 1350, + 862, + 1383, + 777, + 1383 + ], + "score": 0.92, + "latex": "\\xi = 0 ^ { - }" + }, + { + "category_id": 13, + "poly": [ + 502, + 1021, + 618, + 1021, + 618, + 1054, + 502, + 1054 + ], + "score": 0.92, + "latex": "d / n \\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 901, + 578, + 1050, + 578, + 1050, + 613, + 901, + 613 + ], + "score": 0.92, + "latex": "K _ { W } \\in \\mathbb { R } ^ { h \\times h }" + }, + { + "category_id": 13, + "poly": [ + 1249, + 1890, + 1322, + 1890, + 1322, + 1921, + 1249, + 1921 + ], + "score": 0.92, + "latex": "\\xi 0" + }, + { + "category_id": 13, + "poly": [ + 1176, + 989, + 1322, + 989, + 1322, + 1024, + 1176, + 1024 + ], + "score": 0.91, + "latex": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )" + }, + { + "category_id": 14, + "poly": [ + 352, + 1267, + 1343, + 1267, + 1343, + 1344, + 352, + 1344 + ], + "score": 0.91, + "latex": "V = \\operatorname { t r } \\left( \\left( \\phi ( W ^ { \\top } X ) \\phi ( X ^ { \\top } W ) \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } \\operatorname { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\operatorname* { l i m } _ { \\xi \\to 0 ^ { - } } V _ { \\xi } ," + }, + { + "category_id": 13, + "poly": [ + 298, + 1021, + 443, + 1021, + 443, + 1053, + 298, + 1053 + ], + "score": 0.91, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1126, + 953, + 1400, + 953, + 1400, + 989, + 1126, + 989 + ], + "score": 0.91, + "latex": "\\Phi = \\phi ( W ^ { \\top } X ) \\in \\mathbb { R } ^ { h \\times n }" + }, + { + "category_id": 13, + "poly": [ + 980, + 1418, + 1125, + 1418, + 1125, + 1451, + 980, + 1451 + ], + "score": 0.9, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 718, + 1351, + 749, + 1351, + 749, + 1384, + 718, + 1384 + ], + "score": 0.89, + "latex": "V _ { \\xi }" + }, + { + "category_id": 14, + "poly": [ + 654, + 618, + 1043, + 618, + 1043, + 677, + 654, + 677 + ], + "score": 0.88, + "latex": "K _ { W } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( W ^ { \\top } \\pmb { x } ) \\phi ( W ^ { \\top } \\pmb { x } ) ^ { \\top } \\Big ] ." + }, + { + "category_id": 13, + "poly": [ + 886, + 1114, + 954, + 1114, + 954, + 1145, + 886, + 1145 + ], + "score": 0.88, + "latex": "\\rho > 0" + }, + { + "category_id": 13, + "poly": [ + 390, + 1232, + 462, + 1232, + 462, + 1260, + 390, + 1260 + ], + "score": 0.87, + "latex": "h < n" + }, + { + "category_id": 13, + "poly": [ + 753, + 1113, + 827, + 1113, + 827, + 1146, + 753, + 1146 + ], + "score": 0.87, + "latex": "\\lbrack \\rho , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 802, + 1020, + 947, + 1020, + 947, + 1058, + 802, + 1058 + ], + "score": 0.87, + "latex": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda )" + }, + { + "category_id": 14, + "poly": [ + 579, + 1063, + 1116, + 1063, + 1116, + 1106, + 579, + 1106 + ], + "score": 0.86, + "latex": "\\mu _ { \\Phi \\Phi ^ { \\top } / n } ( d \\lambda ) [ 1 - \\gamma _ { 2 } ^ { - 1 } ] _ { + } \\delta _ { 0 } ( \\lambda ) d \\lambda + \\mu ^ { + } ( d \\lambda ) ," + }, + { + "category_id": 13, + "poly": [ + 714, + 1169, + 738, + 1169, + 738, + 1195, + 714, + 1195 + ], + "score": 0.85, + "latex": "\\Phi" + }, + { + "category_id": 13, + "poly": [ + 604, + 683, + 1011, + 683, + 1011, + 740, + 604, + 740 + ], + "score": 0.84, + "latex": "( K _ { W } ) _ { [ i , j ] } = \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { j } ^ { \\top } \\pmb { x } ) \\Big ] ." + }, + { + "category_id": 13, + "poly": [ + 1229, + 286, + 1261, + 286, + 1261, + 316, + 1229, + 316 + ], + "score": 0.84, + "latex": "\\cdot \\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 392, + 1382, + 411, + 1382, + 411, + 1413, + 392, + 1413 + ], + "score": 0.83, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 546, + 958, + 566, + 958, + 566, + 988, + 546, + 988 + ], + "score": 0.8, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 671, + 1020, + 790, + 1020, + 790, + 1054, + 671, + 1054 + ], + "score": 0.7, + "latex": "h / n \\gamma _ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 229.0, + 642.0, + 229.0, + 642.0, + 263.0, + 298.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1347.0, + 717.0, + 1347.0, + 717.0, + 1386.0, + 294.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1347.0, + 776.0, + 1347.0, + 776.0, + 1386.0, + 750.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1347.0, + 1234.0, + 1347.0, + 1234.0, + 1386.0, + 863.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1347.0, + 1409.0, + 1347.0, + 1409.0, + 1386.0, + 1400.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1377.0, + 391.0, + 1377.0, + 391.0, + 1415.0, + 292.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1377.0, + 1406.0, + 1377.0, + 1406.0, + 1415.0, + 412.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1412.0, + 599.0, + 1412.0, + 599.0, + 1457.0, + 292.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 1412.0, + 979.0, + 1412.0, + 979.0, + 1457.0, + 776.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1412.0, + 1140.0, + 1412.0, + 1140.0, + 1457.0, + 1126.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 744.0, + 1406.0, + 744.0, + 1406.0, + 785.0, + 292.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 777.0, + 1408.0, + 777.0, + 1408.0, + 815.0, + 291.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 810.0, + 1406.0, + 810.0, + 1406.0, + 843.0, + 295.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 839.0, + 1168.0, + 839.0, + 1168.0, + 875.0, + 294.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 950.0, + 545.0, + 950.0, + 545.0, + 992.0, + 290.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 950.0, + 1125.0, + 950.0, + 1125.0, + 992.0, + 567.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 950.0, + 1406.0, + 950.0, + 1406.0, + 992.0, + 1401.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 983.0, + 431.0, + 983.0, + 431.0, + 1025.0, + 293.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 983.0, + 1175.0, + 983.0, + 1175.0, + 1025.0, + 517.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 983.0, + 1406.0, + 983.0, + 1406.0, + 1025.0, + 1323.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1019.0, + 297.0, + 1019.0, + 297.0, + 1058.0, + 294.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1019.0, + 501.0, + 1019.0, + 501.0, + 1058.0, + 444.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1019.0, + 670.0, + 1019.0, + 670.0, + 1058.0, + 619.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1019.0, + 801.0, + 1019.0, + 801.0, + 1058.0, + 791.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1019.0, + 1182.0, + 1019.0, + 1182.0, + 1058.0, + 948.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 886.0, + 1406.0, + 886.0, + 1406.0, + 922.0, + 295.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 916.0, + 1292.0, + 916.0, + 1292.0, + 952.0, + 295.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 674.0, + 603.0, + 674.0, + 603.0, + 745.0, + 285.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 674.0, + 1023.0, + 674.0, + 1023.0, + 745.0, + 1012.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 577.0, + 900.0, + 577.0, + 900.0, + 616.0, + 294.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 577.0, + 1086.0, + 577.0, + 1086.0, + 616.0, + 1051.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1163.0, + 713.0, + 1163.0, + 713.0, + 1203.0, + 294.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1163.0, + 1345.0, + 1163.0, + 1345.0, + 1203.0, + 739.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1224.0, + 389.0, + 1224.0, + 389.0, + 1269.0, + 294.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1224.0, + 853.0, + 1224.0, + 853.0, + 1269.0, + 463.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 677.0, + 1881.0, + 677.0, + 1929.0, + 293.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1881.0, + 1248.0, + 1881.0, + 1248.0, + 1929.0, + 943.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1881.0, + 1406.0, + 1881.0, + 1406.0, + 1929.0, + 1323.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 281.0, + 1228.0, + 281.0, + 1228.0, + 325.0, + 294.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 281.0, + 1367.0, + 281.0, + 1367.0, + 325.0, + 1262.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1108.0, + 370.0, + 1108.0, + 370.0, + 1151.0, + 294.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1108.0, + 752.0, + 1108.0, + 752.0, + 1151.0, + 461.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1108.0, + 885.0, + 1108.0, + 885.0, + 1151.0, + 828.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1108.0, + 965.0, + 1108.0, + 965.0, + 1151.0, + 955.0, + 1151.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 446, + 1406, + 446, + 1406, + 604, + 296, + 604 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 616, + 1405, + 616, + 1405, + 755, + 297, + 755 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 1160, + 1406, + 1160, + 1406, + 1293, + 296, + 1293 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 295, + 223, + 1405, + 223, + 1405, + 342, + 295, + 342 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 769, + 357, + 928, + 357, + 928, + 421, + 769, + 421 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 534, + 1031, + 1162, + 1031, + 1162, + 1096, + 534, + 1096 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 496, + 896, + 1201, + 896, + 1201, + 968, + 496, + 968 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 297, + 1513, + 1409, + 1513, + 1409, + 1580, + 297, + 1580 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 415, + 1597, + 1282, + 1597, + 1282, + 1686, + 415, + 1686 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 605, + 1308, + 1095, + 1308, + 1095, + 1371, + 605, + 1371 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 299, + 981, + 990, + 981, + 990, + 1015, + 299, + 1015 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 297, + 1704, + 1399, + 1704, + 1399, + 1741, + 297, + 1741 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 297, + 1759, + 1429, + 1759, + 1429, + 1847, + 297, + 1847 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 1382, + 1399, + 1382, + 1399, + 1417, + 297, + 1417 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 296, + 1113, + 1034, + 1113, + 1034, + 1151, + 296, + 1151 + ], + "score": 0.924 + }, + { + "category_id": 8, + "poly": [ + 309, + 1431, + 1327, + 1431, + 1327, + 1498, + 309, + 1498 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 291, + 845, + 1287, + 845, + 1287, + 881, + 291, + 881 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 298, + 1897, + 1367, + 1897, + 1367, + 1943, + 298, + 1943 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1351, + 367, + 1400, + 367, + 1400, + 398, + 1351, + 398 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1351, + 915, + 1401, + 915, + 1401, + 947, + 1351, + 947 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1324, + 1400, + 1324, + 1400, + 1356, + 1351, + 1356 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2113, + 834, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1625, + 1401, + 1625, + 1401, + 1657, + 1351, + 1657 + ], + "score": 0.866 + }, + { + "category_id": 8, + "poly": [ + 462, + 1954, + 1236, + 1954, + 1236, + 2032, + 462, + 2032 + ], + "score": 0.864 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1847, + 1401, + 1847, + 1401, + 1877, + 1351, + 1877 + ], + "score": 0.855 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1974, + 1401, + 1974, + 1401, + 2005, + 1350, + 2005 + ], + "score": 0.827 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1448, + 1400, + 1448, + 1400, + 1482, + 1352, + 1482 + ], + "score": 0.826 + }, + { + "category_id": 1, + "poly": [ + 297, + 788, + 841, + 788, + 841, + 822, + 297, + 822 + ], + "score": 0.541 + }, + { + "category_id": 0, + "poly": [ + 297, + 788, + 841, + 788, + 841, + 822, + 297, + 822 + ], + "score": 0.475 + }, + { + "category_id": 8, + "poly": [ + 462, + 1955, + 1234, + 1955, + 1234, + 2031, + 462, + 2031 + ], + "score": 0.164 + }, + { + "category_id": 13, + "poly": [ + 434, + 1545, + 527, + 1545, + 527, + 1580, + 434, + 1580 + ], + "score": 0.95, + "latex": "\\lambda _ { i } = \\phi _ { i } ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 459, + 1954, + 1238, + 1954, + 1238, + 2031, + 459, + 2031 + ], + "score": 0.94, + "latex": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } V _ { \\xi } = \\operatorname* { l i m } _ { \\xi \\to 0 } \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda } \\mu _ { \\infty } ^ { + } ( d \\lambda ) ." + }, + { + "category_id": 14, + "poly": [ + 415, + 1592, + 1284, + 1592, + 1284, + 1689, + 415, + 1689 + ], + "score": 0.94, + "latex": "V _ { \\xi } \\to \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) = \\gamma _ { 2 } \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } { \\pmb { u } } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\pmb { u } } _ { i } ." + }, + { + "category_id": 14, + "poly": [ + 768, + 355, + 930, + 355, + 930, + 421, + 768, + 421 + ], + "score": 0.93, + "latex": "V \\frac { \\gamma _ { 2 } } { 1 - \\gamma _ { 2 } } ." + }, + { + "category_id": 13, + "poly": [ + 1211, + 310, + 1342, + 310, + 1342, + 342, + 1211, + 342 + ], + "score": 0.93, + "latex": "n , d , h 0" + }, + { + "category_id": 14, + "poly": [ + 535, + 1027, + 1163, + 1027, + 1163, + 1097, + 535, + 1097 + ], + "score": 0.93, + "latex": "V = \\operatorname * { l i m } _ { \\xi 0 } \\frac { 1 } { n } \\Big [ \\mathrm { t r } ( S ( S ^ { \\top } S - \\xi I _ { n } ) ^ { - 2 } S ^ { \\top } K _ { W } ) \\Big ] = \\operatorname * { l i m } _ { \\xi 0 } V _ { \\xi } ." + }, + { + "category_id": 13, + "poly": [ + 484, + 1898, + 876, + 1898, + 876, + 1942, + 484, + 1942 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mu _ { n } ( x ) = \\frac { 1 } { h } \\sum _ { i = 1 } ^ { h } \\delta _ { \\lambda _ { i } } ( x ) \\pmb { u } _ { i } ^ { \\top } \\tilde { K } _ { W } \\pmb { u } _ { i } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 321, + 1429, + 1331, + 1429, + 1331, + 1497, + 321, + 1497 + ], + "score": 0.92, + "latex": "V _ { \\xi } = \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) \\Big ] \\to \\frac { 1 } { n } \\Big [ \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } \\tilde { K } _ { W } \\right) \\Big ] ," + }, + { + "category_id": 14, + "poly": [ + 497, + 892, + 1203, + 892, + 1203, + 970, + 497, + 970 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { V = \\mathrm { t r } \\bigg ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } \\bigg ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 602, + 1306, + 1098, + 1306, + 1098, + 1372, + 602, + 1372 + ], + "score": 0.92, + "latex": "V _ { \\xi } = \\frac { 1 } { n } \\mathrm { t r } \\left( U \\Sigma ( \\Sigma ^ { \\top } \\Sigma + \\xi I _ { n } ) ^ { - 2 } \\Sigma U ^ { \\top } K _ { W } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 1196, + 1704, + 1291, + 1704, + 1291, + 1741, + 1196, + 1741 + ], + "score": 0.92, + "latex": "1 - \\gamma _ { 2 } ^ { - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1104, + 1904, + 1178, + 1904, + 1178, + 1938, + 1104, + 1938 + ], + "score": 0.92, + "latex": "\\mu _ { n } ^ { + } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1086, + 311, + 1160, + 311, + 1160, + 342, + 1086, + 342 + ], + "score": 0.92, + "latex": "\\xi 0" + }, + { + "category_id": 13, + "poly": [ + 843, + 650, + 916, + 650, + 916, + 681, + 843, + 681 + ], + "score": 0.92, + "latex": "\\xi 0" + }, + { + "category_id": 13, + "poly": [ + 831, + 1160, + 1062, + 1160, + 1062, + 1197, + 831, + 1197 + ], + "score": 0.91, + "latex": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n }" + }, + { + "category_id": 13, + "poly": [ + 410, + 224, + 1163, + 224, + 1163, + 281, + 410, + 281 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\xi \\to 0 } \\operatorname* { l i m } _ { n , d , h \\to \\infty } n / h \\cdot \\mathrm { t r } \\left( \\left( \\Phi \\Phi ^ { \\top } - \\xi I \\right) ^ { - 1 } K _ { W } \\right) = \\gamma _ { 2 } / ( 1 - \\gamma _ { 2 } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1386, + 1401, + 1386, + 1401, + 1417, + 1256, + 1417 + ], + "score": 0.91, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1229, + 1136, + 1229, + 1136, + 1258, + 1012, + 1258 + ], + "score": 0.91, + "latex": "V \\in \\mathbb { R } ^ { n \\times n }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1753, + 1427, + 1753, + 1427, + 1850, + 310, + 1850 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\gamma _ { \\xi } \\to \\gamma _ { 2 } \\displaystyle \\frac { 1 } { h } \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\sum _ { i = 1 } ^ { h } \\frac { \\lambda _ { i } } { ( \\lambda _ { i } + \\xi ) ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } = \\gamma _ { 2 } \\displaystyle \\int \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\gamma _ { 2 } \\displaystyle \\int _ { \\lambda > \\rho } \\frac { \\lambda } { ( \\lambda + \\xi ) ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 473, + 1258, + 820, + 1258, + 820, + 1293, + 473, + 1293 + ], + "score": 0.9, + "latex": "S ^ { \\top } S = \\phi ( X ^ { \\top } W ) \\phi ( \\dot { W } ^ { \\top } X ) / n" + }, + { + "category_id": 13, + "poly": [ + 1281, + 233, + 1406, + 233, + 1406, + 269, + 1281, + 269 + ], + "score": 0.9, + "latex": "\\partial V _ { \\xi } / \\partial \\xi =" + }, + { + "category_id": 13, + "poly": [ + 298, + 1192, + 644, + 1192, + 644, + 1229, + 298, + 1229 + ], + "score": 0.9, + "latex": "\\mathrm { d i a g } _ { h \\times n } \\big ( \\mathring { \\phi _ { 1 } } , \\cdot \\cdot \\cdot , \\phi _ { n } \\big ) \\in \\mathbb { R } ^ { h \\times n }" + }, + { + "category_id": 13, + "poly": [ + 296, + 275, + 584, + 275, + 584, + 314, + 296, + 314 + ], + "score": 0.9, + "latex": "\\mathrm { t r } \\left( \\xi ( \\Phi \\Phi ^ { \\top } - \\xi I ) ^ { - 2 } K _ { W } \\right)" + }, + { + "category_id": 13, + "poly": [ + 371, + 1113, + 786, + 1113, + 786, + 1150, + 371, + 1150 + ], + "score": 0.9, + "latex": "S = \\phi ( W ^ { \\top } X ) / \\sqrt { n } = \\Phi / \\sqrt { n } , \\xi \\in \\mathbb { C }" + }, + { + "category_id": 13, + "poly": [ + 586, + 1547, + 840, + 1547, + 840, + 1578, + 586, + 1578 + ], + "score": 0.9, + "latex": "\\phi _ { n + 1 } = \\cdot \\cdot \\cdot = \\phi _ { h } = 0 )" + }, + { + "category_id": 13, + "poly": [ + 400, + 1510, + 450, + 1510, + 450, + 1546, + 400, + 1546 + ], + "score": 0.9, + "latex": "\\tilde { K } _ { W }" + }, + { + "category_id": 13, + "poly": [ + 964, + 1192, + 1269, + 1192, + 1269, + 1227, + 964, + 1227 + ], + "score": 0.89, + "latex": "U \\overset { ^ { \\prime } } { = } [ \\pmb { u } _ { 1 } , \\cdots , \\pmb { u } _ { h } ] \\in \\mathbb { R } ^ { h \\times h }" + }, + { + "category_id": 13, + "poly": [ + 837, + 1116, + 924, + 1116, + 924, + 1149, + 837, + 1149 + ], + "score": 0.89, + "latex": "\\Im \\xi > 0" + }, + { + "category_id": 13, + "poly": [ + 605, + 1226, + 953, + 1226, + 953, + 1262, + 605, + 1262 + ], + "score": 0.89, + "latex": "S \\underline { { S } } ^ { \\top } = \\phi ( \\underline { { W } } ^ { \\top } X ) \\phi ( X ^ { \\top } W ) / n" + }, + { + "category_id": 13, + "poly": [ + 1040, + 280, + 1107, + 280, + 1107, + 310, + 1040, + 310 + ], + "score": 0.89, + "latex": "\\xi = 0" + }, + { + "category_id": 13, + "poly": [ + 1102, + 1161, + 1251, + 1161, + 1251, + 1191, + 1102, + 1191 + ], + "score": 0.88, + "latex": "S = U \\Sigma V ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 710, + 1517, + 760, + 1517, + 760, + 1546, + 710, + 1546 + ], + "score": 0.88, + "latex": "K _ { W }" + }, + { + "category_id": 13, + "poly": [ + 959, + 1117, + 1025, + 1117, + 1025, + 1149, + 959, + 1149 + ], + "score": 0.88, + "latex": "\\xi < 0" + }, + { + "category_id": 13, + "poly": [ + 1079, + 686, + 1110, + 686, + 1110, + 720, + 1079, + 720 + ], + "score": 0.88, + "latex": "V _ { \\xi }" + }, + { + "category_id": 13, + "poly": [ + 652, + 649, + 684, + 649, + 684, + 683, + 652, + 683 + ], + "score": 0.88, + "latex": "V _ { \\xi }" + }, + { + "category_id": 13, + "poly": [ + 400, + 450, + 483, + 450, + 483, + 479, + 400, + 479 + ], + "score": 0.88, + "latex": "h > n" + }, + { + "category_id": 13, + "poly": [ + 1344, + 1163, + 1404, + 1163, + 1404, + 1193, + 1344, + 1193 + ], + "score": 0.85, + "latex": "\\Sigma =" + }, + { + "category_id": 13, + "poly": [ + 515, + 681, + 540, + 681, + 540, + 713, + 515, + 713 + ], + "score": 0.85, + "latex": "\\tilde { A }" + }, + { + "category_id": 13, + "poly": [ + 729, + 1710, + 747, + 1710, + 747, + 1735, + 729, + 1735 + ], + "score": 0.83, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 763, + 791, + 837, + 791, + 837, + 819, + 763, + 819 + ], + "score": 0.81, + "latex": "h > n" + }, + { + "category_id": 13, + "poly": [ + 539, + 717, + 562, + 717, + 562, + 749, + 539, + 749 + ], + "score": 0.8, + "latex": "\\tilde { A }" + }, + { + "category_id": 13, + "poly": [ + 1258, + 855, + 1277, + 855, + 1277, + 875, + 1258, + 875 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 871, + 619, + 896, + 619, + 896, + 646, + 871, + 646 + ], + "score": 0.74, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 680, + 791, + 706, + 791, + 706, + 818, + 680, + 818 + ], + "score": 0.67, + "latex": "V" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2124.0, + 828.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 788.0, + 679.0, + 788.0, + 679.0, + 825.0, + 296.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 788.0, + 762.0, + 788.0, + 762.0, + 825.0, + 707.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 788.0, + 842.0, + 788.0, + 842.0, + 825.0, + 838.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 444.0, + 399.0, + 444.0, + 399.0, + 483.0, + 294.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 444.0, + 1408.0, + 444.0, + 1408.0, + 483.0, + 484.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 479.0, + 1404.0, + 479.0, + 1404.0, + 513.0, + 295.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 506.0, + 1406.0, + 506.0, + 1406.0, + 545.0, + 293.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 541.0, + 1404.0, + 541.0, + 1404.0, + 575.0, + 296.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 571.0, + 1048.0, + 571.0, + 1048.0, + 604.0, + 295.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 615.0, + 870.0, + 615.0, + 870.0, + 654.0, + 292.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 615.0, + 1408.0, + 615.0, + 1408.0, + 654.0, + 897.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 648.0, + 651.0, + 648.0, + 651.0, + 683.0, + 295.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 648.0, + 842.0, + 648.0, + 842.0, + 683.0, + 685.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 648.0, + 1403.0, + 648.0, + 1403.0, + 683.0, + 917.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 682.0, + 514.0, + 682.0, + 514.0, + 722.0, + 292.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 682.0, + 1078.0, + 682.0, + 1078.0, + 722.0, + 541.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 682.0, + 1405.0, + 682.0, + 1405.0, + 722.0, + 1111.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 719.0, + 538.0, + 719.0, + 538.0, + 754.0, + 297.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 719.0, + 740.0, + 719.0, + 740.0, + 754.0, + 563.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1157.0, + 830.0, + 1157.0, + 830.0, + 1197.0, + 293.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1157.0, + 1101.0, + 1157.0, + 1101.0, + 1197.0, + 1063.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1157.0, + 1343.0, + 1157.0, + 1343.0, + 1197.0, + 1252.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1186.0, + 297.0, + 1186.0, + 297.0, + 1230.0, + 291.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1186.0, + 963.0, + 1186.0, + 963.0, + 1230.0, + 645.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1186.0, + 1409.0, + 1186.0, + 1409.0, + 1230.0, + 1270.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1224.0, + 604.0, + 1224.0, + 604.0, + 1265.0, + 294.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1224.0, + 1011.0, + 1224.0, + 1011.0, + 1265.0, + 954.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1224.0, + 1408.0, + 1224.0, + 1408.0, + 1265.0, + 1137.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1257.0, + 472.0, + 1257.0, + 472.0, + 1294.0, + 295.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1257.0, + 1232.0, + 1257.0, + 1232.0, + 1294.0, + 821.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 224.0, + 409.0, + 224.0, + 409.0, + 276.0, + 292.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 224.0, + 1280.0, + 224.0, + 1280.0, + 276.0, + 1164.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 270.0, + 1039.0, + 270.0, + 1039.0, + 315.0, + 585.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 270.0, + 1406.0, + 270.0, + 1406.0, + 315.0, + 1108.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 307.0, + 1085.0, + 307.0, + 1085.0, + 345.0, + 295.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 307.0, + 1210.0, + 307.0, + 1210.0, + 345.0, + 1161.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 307.0, + 1353.0, + 307.0, + 1353.0, + 345.0, + 1343.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1509.0, + 399.0, + 1509.0, + 399.0, + 1553.0, + 293.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1509.0, + 709.0, + 1509.0, + 709.0, + 1553.0, + 451.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1509.0, + 1405.0, + 1509.0, + 1405.0, + 1553.0, + 761.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1543.0, + 433.0, + 1543.0, + 433.0, + 1583.0, + 295.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 1543.0, + 585.0, + 1543.0, + 585.0, + 1583.0, + 528.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 1543.0, + 955.0, + 1543.0, + 955.0, + 1583.0, + 841.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 980.0, + 993.0, + 980.0, + 993.0, + 1017.0, + 295.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1700.0, + 728.0, + 1700.0, + 728.0, + 1748.0, + 295.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1700.0, + 1195.0, + 1700.0, + 1195.0, + 1748.0, + 748.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1700.0, + 1401.0, + 1700.0, + 1401.0, + 1748.0, + 1292.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1380.0, + 1255.0, + 1380.0, + 1255.0, + 1422.0, + 293.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1110.0, + 370.0, + 1110.0, + 370.0, + 1155.0, + 294.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1110.0, + 836.0, + 1110.0, + 836.0, + 1155.0, + 787.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1110.0, + 958.0, + 1110.0, + 958.0, + 1155.0, + 925.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1110.0, + 1036.0, + 1110.0, + 1036.0, + 1155.0, + 1026.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 843.0, + 1257.0, + 843.0, + 1257.0, + 886.0, + 294.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 843.0, + 1291.0, + 843.0, + 1291.0, + 886.0, + 1278.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1893.0, + 483.0, + 1893.0, + 483.0, + 1946.0, + 291.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 1893.0, + 1103.0, + 1893.0, + 1103.0, + 1946.0, + 877.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1893.0, + 1372.0, + 1893.0, + 1372.0, + 1946.0, + 1179.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 788.0, + 679.0, + 788.0, + 679.0, + 825.0, + 296.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 788.0, + 762.0, + 788.0, + 762.0, + 825.0, + 707.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 788.0, + 842.0, + 788.0, + 842.0, + 825.0, + 838.0, + 825.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1736, + 1405, + 1736, + 1405, + 1838, + 297, + 1838 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 312, + 1073, + 1338, + 1073, + 1338, + 1242, + 312, + 1242 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 322, + 689, + 1374, + 689, + 1374, + 974, + 322, + 974 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 525, + 544, + 1171, + 544, + 1171, + 634, + 525, + 634 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 661, + 1646, + 1038, + 1646, + 1038, + 1724, + 661, + 1724 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 652, + 1299, + 1044, + 1299, + 1044, + 1367, + 652, + 1367 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 473, + 1950, + 1223, + 1950, + 1223, + 2034, + 473, + 2034 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 594, + 275, + 1105, + 275, + 1105, + 352, + 594, + 352 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 290, + 1517, + 1404, + 1517, + 1404, + 1585, + 290, + 1585 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 574, + 417, + 1122, + 417, + 1122, + 481, + 574, + 481 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 757, + 1434, + 941, + 1434, + 941, + 1487, + 757, + 1487 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 297, + 1376, + 1285, + 1376, + 1285, + 1422, + 297, + 1422 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 296, + 646, + 679, + 646, + 679, + 679, + 296, + 679 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 298, + 494, + 919, + 494, + 919, + 530, + 298, + 530 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 296, + 224, + 1108, + 224, + 1108, + 265, + 296, + 265 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 299, + 368, + 784, + 368, + 784, + 403, + 299, + 403 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 295, + 1252, + 890, + 1252, + 890, + 1286, + 295, + 1286 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 297, + 72, + 818, + 72, + 818, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 298, + 1900, + 704, + 1900, + 704, + 1936, + 298, + 1936 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 295, + 1600, + 1220, + 1600, + 1220, + 1636, + 295, + 1636 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 296, + 1019, + 1291, + 1019, + 1291, + 1057, + 296, + 1057 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1318, + 1401, + 1318, + 1401, + 1349, + 1351, + 1349 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 296, + 1844, + 1322, + 1844, + 1322, + 1883, + 296, + 1883 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1350, + 298, + 1401, + 298, + 1401, + 329, + 1350, + 329 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1350, + 572, + 1401, + 572, + 1401, + 604, + 1350, + 604 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1668, + 1400, + 1668, + 1400, + 1700, + 1351, + 1700 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1977, + 1402, + 1977, + 1402, + 2008, + 1351, + 2008 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1350, + 972, + 1402, + 972, + 1402, + 1003, + 1350, + 1003 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1350, + 433, + 1401, + 433, + 1401, + 465, + 1350, + 465 + ], + "score": 0.818 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1185, + 1401, + 1185, + 1401, + 1216, + 1354, + 1216 + ], + "score": 0.813 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1437, + 1400, + 1437, + 1400, + 1469, + 1351, + 1469 + ], + "score": 0.76 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 864, + 2087, + 864, + 2114, + 834, + 2114 + ], + "score": 0.622 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1437, + 1401, + 1437, + 1401, + 1469, + 1351, + 1469 + ], + "score": 0.228 + }, + { + "category_id": 9, + "poly": [ + 1350, + 433, + 1402, + 433, + 1402, + 466, + 1350, + 466 + ], + "score": 0.162 + }, + { + "category_id": 14, + "poly": [ + 304, + 1069, + 1343, + 1069, + 1343, + 1245, + 304, + 1245 + ], + "score": 0.95, + "latex": "\\begin{array} { l } { \\displaystyle \\left. - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\right. _ { x = 0 } = \\displaystyle \\operatorname* { l i m } _ { h , d , n \\to \\infty } \\frac { 1 } { n } \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } { \\boldsymbol u } _ { i } ^ { \\top } \\tilde { K } _ { W } { \\boldsymbol u } _ { i } } \\\\ { \\displaystyle = \\gamma _ { 2 } \\int _ { \\lambda \\geq 0 } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ( d \\lambda ) = \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } + \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 318, + 689, + 1375, + 689, + 1375, + 978, + 318, + 978 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { - \\displaystyle \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\xi ^ { 2 } I _ { h } + S S ^ { \\top } } & { 0 } \\\\ { 0 } & { I _ { n } + S ^ { \\top } S } \\end{array} \\right] ^ { - 1 } \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } } & { 0 } \\\\ { 0 } & { 0 } \\end{array} \\right] \\right) } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad = \\frac { 1 } { n } \\mathrm { t r } \\left( U ( \\Sigma \\Sigma ^ { \\top } + \\xi ^ { 2 } I _ { h } ) ^ { - 1 } U ^ { \\top } \\tilde { K } _ { W } \\right) = \\frac { 1 } { n } \\displaystyle \\sum _ { i = 1 } ^ { h } \\frac { 1 } { \\lambda + \\xi ^ { 2 } } u _ { i } ^ { \\top } \\tilde { K } _ { W } u _ { i } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 526, + 541, + 1173, + 541, + 1173, + 635, + 526, + 635 + ], + "score": 0.94, + "latex": "\\tilde { m } _ { n } ( \\xi , r x , s x , t x ) = \\frac { 1 } { n } \\mathrm { t r } \\left( \\left[ \\begin{array} { c c } { \\tilde { K } _ { W } x - \\xi I _ { h } } & { S } \\\\ { S ^ { \\top } } & { - I _ { n } } \\end{array} \\right] ^ { - 1 } \\right) ," + }, + { + "category_id": 14, + "poly": [ + 653, + 1297, + 1044, + 1297, + 1044, + 1366, + 653, + 1366 + ], + "score": 0.94, + "latex": "q ( \\xi ) = - \\frac { \\partial } { \\partial x } \\tilde { m } ( \\xi , r x , s x , t x ) \\Big | _ { x = 0 } ," + }, + { + "category_id": 14, + "poly": [ + 658, + 1645, + 1039, + 1645, + 1039, + 1723, + 658, + 1723 + ], + "score": 0.94, + "latex": "A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] ," + }, + { + "category_id": 14, + "poly": [ + 591, + 274, + 1108, + 274, + 1108, + 352, + 591, + 352 + ], + "score": 0.93, + "latex": "\\boldsymbol { \\tilde { A } _ { n } } ( \\rho , \\varsigma , \\tau ) = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\varsigma \\mathbf { 1 } _ { h } \\mathbf { 1 } _ { h } ^ { \\top } + \\tau Q } & { S } \\\\ { S ^ { \\top } } & { 0 _ { n } } \\end{array} \\right] ." + }, + { + "category_id": 14, + "poly": [ + 473, + 1947, + 1224, + 1947, + 1224, + 2032, + 473, + 2032 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - A _ { n } ( \\rho , \\tau ) = \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] \\left[ \\begin{array} { c c } { \\xi } & { a _ { 0 } } \\\\ { a _ { 0 } } & { 0 } \\end{array} \\right] \\left[ \\begin{array} { c c } { 1 _ { h } } & { 0 _ { h } } \\\\ { 0 _ { n } } & { 1 _ { n } } \\end{array} \\right] ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 965, + 231, + 1096, + 231, + 1096, + 260, + 965, + 260 + ], + "score": 0.93, + "latex": "N = n + h" + }, + { + "category_id": 13, + "poly": [ + 374, + 1735, + 578, + 1735, + 578, + 1775, + 374, + 1775 + ], + "score": 0.93, + "latex": "S = \\tilde { S } - a _ { 0 } I _ { p \\times n }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1519, + 1351, + 1519, + 1351, + 1553, + 1211, + 1553 + ], + "score": 0.92, + "latex": "\\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau )" + }, + { + "category_id": 13, + "poly": [ + 645, + 1519, + 800, + 1519, + 800, + 1553, + 645, + 1553 + ], + "score": 0.92, + "latex": "m _ { n } ( \\xi , \\rho , \\varsigma , \\tau )" + }, + { + "category_id": 13, + "poly": [ + 648, + 225, + 886, + 225, + 886, + 264, + 648, + 264 + ], + "score": 0.92, + "latex": "\\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) \\in \\mathbb { R } ^ { N \\times N }" + }, + { + "category_id": 13, + "poly": [ + 659, + 493, + 709, + 493, + 709, + 529, + 659, + 529 + ], + "score": 0.92, + "latex": "\\tilde { K } _ { W }" + }, + { + "category_id": 13, + "poly": [ + 430, + 1377, + 956, + 1377, + 956, + 1423, + 430, + 1423 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { q _ { + } ( \\xi ) = q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } = \\gamma _ { 2 } \\int _ { \\lambda > \\rho } \\frac 1 { \\lambda + \\xi ^ { 2 } } \\mu _ { \\infty } ^ { + } ( d \\lambda ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 414, + 1604, + 452, + 1604, + 452, + 1633, + 414, + 1633 + ], + "score": 0.92, + "latex": "A _ { n }" + }, + { + "category_id": 13, + "poly": [ + 713, + 365, + 751, + 365, + 751, + 401, + 713, + 401 + ], + "score": 0.91, + "latex": "{ \\tilde { A } } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 484, + 1021, + 955, + 1021, + 955, + 1056, + 484, + 1056 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\tilde { m } ( \\xi , \\rho , \\varsigma , \\tau ) = \\operatorname* { l i m } _ { n , h , d \\infty } \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1074, + 1739, + 1235, + 1739, + 1235, + 1774, + 1074, + 1774 + ], + "score": 0.91, + "latex": "a _ { 0 } = \\mathbb { E } [ \\phi ( x ) ]" + }, + { + "category_id": 14, + "poly": [ + 576, + 413, + 1123, + 413, + 1123, + 481, + 576, + 481 + ], + "score": 0.91, + "latex": "\\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( \\tilde { A } _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 1171, + 1597, + 1208, + 1597, + 1208, + 1633, + 1171, + 1633 + ], + "score": 0.91, + "latex": "{ \\tilde { A } } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 567, + 1901, + 605, + 1901, + 605, + 1934, + 567, + 1934 + ], + "score": 0.9, + "latex": "{ \\bar { A } } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1550, + 650, + 1550, + 650, + 1585, + 297, + 1585 + ], + "score": 0.9, + "latex": "q ( \\xi ) = - \\tilde { m } _ { x } ^ { \\prime } ( \\xi , r x , s x , t x ) | _ { x = 0 }" + }, + { + "category_id": 13, + "poly": [ + 988, + 1846, + 1316, + 1846, + 1316, + 1880, + 988, + 1880 + ], + "score": 0.9, + "latex": "m _ { n } ( \\xi , \\rho , \\tau ) \\tilde { m } _ { n } ( \\xi , \\rho , \\varsigma , \\tau ) ." + }, + { + "category_id": 14, + "poly": [ + 755, + 1433, + 944, + 1433, + 944, + 1488, + 755, + 1488 + ], + "score": 0.9, + "latex": "V = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) ." + }, + { + "category_id": 13, + "poly": [ + 346, + 1807, + 388, + 1807, + 388, + 1837, + 346, + 1837 + ], + "score": 0.89, + "latex": "\\tilde { m } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 536, + 1520, + 586, + 1520, + 586, + 1552, + 536, + 1552 + ], + "score": 0.89, + "latex": "q ( \\xi )" + }, + { + "category_id": 13, + "poly": [ + 444, + 1776, + 481, + 1776, + 481, + 1806, + 444, + 1806 + ], + "score": 0.89, + "latex": "A _ { n }" + }, + { + "category_id": 13, + "poly": [ + 657, + 1904, + 694, + 1904, + 694, + 1934, + 657, + 1934 + ], + "score": 0.89, + "latex": "A _ { n }" + }, + { + "category_id": 13, + "poly": [ + 509, + 1850, + 600, + 1850, + 600, + 1875, + 509, + 1875 + ], + "score": 0.87, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 640, + 1736, + 1060, + 1736, + 1060, + 1773, + 640, + 1773 + ], + "score": 0.87, + "latex": "S _ { i k } = \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } ) - a _ { 0 } = \\varphi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { k } )" + }, + { + "category_id": 13, + "poly": [ + 439, + 1810, + 481, + 1810, + 481, + 1836, + 439, + 1836 + ], + "score": 0.87, + "latex": "m _ { n }" + }, + { + "category_id": 13, + "poly": [ + 590, + 1773, + 1059, + 1773, + 1059, + 1811, + 590, + 1811 + ], + "score": 0.87, + "latex": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 862, + 1255, + 878, + 1255, + 878, + 1286, + 862, + 1286 + ], + "score": 0.82, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 691, + 1846, + 779, + 1846, + 779, + 1879, + 691, + 1879 + ], + "score": 0.72, + "latex": "\\Im \\xi > 0" + }, + { + "category_id": 13, + "poly": [ + 813, + 1847, + 880, + 1847, + 880, + 1879, + 813, + 1879 + ], + "score": 0.67, + "latex": "\\xi < 0" + }, + { + "category_id": 13, + "poly": [ + 692, + 1846, + 881, + 1846, + 881, + 1879, + 692, + 1879 + ], + "score": 0.38, + "latex": "\\Im \\xi > 0 o r \\xi < 0 " + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1733.0, + 373.0, + 1733.0, + 373.0, + 1778.0, + 291.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1733.0, + 639.0, + 1733.0, + 639.0, + 1778.0, + 579.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1733.0, + 1073.0, + 1733.0, + 1073.0, + 1778.0, + 1061.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1733.0, + 1406.0, + 1733.0, + 1406.0, + 1778.0, + 1236.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1771.0, + 443.0, + 1771.0, + 443.0, + 1810.0, + 294.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1771.0, + 589.0, + 1771.0, + 589.0, + 1810.0, + 482.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1771.0, + 1405.0, + 1771.0, + 1405.0, + 1810.0, + 1060.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 345.0, + 1802.0, + 345.0, + 1839.0, + 294.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1802.0, + 438.0, + 1802.0, + 438.0, + 1839.0, + 389.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1802.0, + 716.0, + 1802.0, + 716.0, + 1839.0, + 482.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1517.0, + 535.0, + 1517.0, + 535.0, + 1554.0, + 294.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1517.0, + 644.0, + 1517.0, + 644.0, + 1554.0, + 587.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1517.0, + 1210.0, + 1517.0, + 1210.0, + 1554.0, + 801.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1517.0, + 1405.0, + 1517.0, + 1405.0, + 1554.0, + 1352.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1547.0, + 1346.0, + 1547.0, + 1346.0, + 1586.0, + 651.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1359.0, + 429.0, + 1359.0, + 429.0, + 1445.0, + 281.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1359.0, + 1302.0, + 1359.0, + 1302.0, + 1445.0, + 957.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 642.0, + 679.0, + 642.0, + 679.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 490.0, + 658.0, + 490.0, + 658.0, + 536.0, + 292.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 490.0, + 922.0, + 490.0, + 922.0, + 536.0, + 710.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 218.0, + 647.0, + 218.0, + 647.0, + 271.0, + 291.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 218.0, + 964.0, + 218.0, + 964.0, + 271.0, + 887.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 218.0, + 1111.0, + 218.0, + 1111.0, + 271.0, + 1097.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 362.0, + 712.0, + 362.0, + 712.0, + 408.0, + 293.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 362.0, + 787.0, + 362.0, + 787.0, + 408.0, + 752.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1245.0, + 861.0, + 1245.0, + 861.0, + 1294.0, + 292.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1245.0, + 892.0, + 1245.0, + 892.0, + 1294.0, + 879.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1896.0, + 566.0, + 1896.0, + 566.0, + 1943.0, + 293.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1896.0, + 656.0, + 1896.0, + 656.0, + 1943.0, + 606.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1896.0, + 708.0, + 1896.0, + 708.0, + 1943.0, + 695.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1597.0, + 413.0, + 1597.0, + 413.0, + 1640.0, + 293.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1597.0, + 1170.0, + 1597.0, + 1170.0, + 1640.0, + 453.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1597.0, + 1219.0, + 1597.0, + 1219.0, + 1640.0, + 1209.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1016.0, + 483.0, + 1016.0, + 483.0, + 1061.0, + 293.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1016.0, + 1294.0, + 1016.0, + 1294.0, + 1061.0, + 956.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1843.0, + 508.0, + 1843.0, + 508.0, + 1884.0, + 293.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1843.0, + 690.0, + 1843.0, + 690.0, + 1884.0, + 601.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1843.0, + 987.0, + 1843.0, + 987.0, + 1884.0, + 882.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1843.0, + 1325.0, + 1843.0, + 1325.0, + 1884.0, + 1317.0, + 1884.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 735, + 1408, + 735, + 1408, + 866, + 296, + 866 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 295, + 1314, + 1408, + 1314, + 1408, + 1440, + 295, + 1440 + ], + "score": 0.976 + }, + { + "category_id": 8, + "poly": [ + 375, + 1155, + 1324, + 1155, + 1324, + 1286, + 375, + 1286 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 451, + 1605, + 1245, + 1605, + 1245, + 2031, + 451, + 2031 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 603, + 1454, + 1094, + 1454, + 1094, + 1539, + 603, + 1539 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 289, + 228, + 1403, + 228, + 1403, + 294, + 289, + 294 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 289, + 389, + 1404, + 389, + 1404, + 456, + 289, + 456 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 513, + 1015, + 1180, + 1015, + 1180, + 1092, + 513, + 1092 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 635, + 310, + 1065, + 310, + 1065, + 371, + 635, + 371 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 498, + 538, + 1077, + 538, + 1077, + 605, + 498, + 605 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 297, + 1103, + 886, + 1103, + 886, + 1137, + 297, + 1137 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 413, + 876, + 1285, + 876, + 1285, + 952, + 413, + 952 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 299, + 966, + 778, + 966, + 778, + 999, + 299, + 999 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 295, + 1557, + 819, + 1557, + 819, + 1590, + 295, + 1590 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 297, + 485, + 1270, + 485, + 1270, + 521, + 297, + 521 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 299, + 687, + 1067, + 687, + 1067, + 725, + 299, + 725 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1467, + 1401, + 1467, + 1401, + 1499, + 1350, + 1499 + ], + "score": 0.899 + }, + { + "category_id": 8, + "poly": [ + 500, + 613, + 1196, + 613, + 1196, + 674, + 500, + 674 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1238, + 1400, + 1238, + 1400, + 1270, + 1351, + 1270 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1351, + 627, + 1401, + 627, + 1401, + 659, + 1351, + 659 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1912, + 1402, + 1912, + 1402, + 1944, + 1350, + 1944 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1351, + 321, + 1400, + 321, + 1400, + 352, + 1351, + 352 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1352, + 555, + 1400, + 555, + 1400, + 586, + 1352, + 586 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1036, + 1401, + 1036, + 1401, + 1068, + 1351, + 1068 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1350, + 897, + 1401, + 897, + 1401, + 929, + 1350, + 929 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2114, + 834, + 2114 + ], + "score": 0.834 + }, + { + "category_id": 14, + "poly": [ + 501, + 534, + 1199, + 534, + 1199, + 676, + 501, + 676 + ], + "score": 0.95, + "latex": "\\begin{array} { l } { { m _ { 1 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { p } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ 1 . . p , 1 . . p ] } ^ { - 1 } \\right) , } } \\\\ { { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) _ { [ p + 1 . . p + n , p + 1 . . p + n ] } ^ { - 1 } \\right) . } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 414, + 874, + 1287, + 874, + 1287, + 952, + 414, + 952 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau Q } & { S _ { * } } \\\\ { S _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] ; \\quad \\pmb { a } ^ { \\top } = [ \\phi ( \\boldsymbol { W } ^ { \\top } \\mathbf { x } _ { n } ) ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] = [ \\boldsymbol { s } ^ { \\top } \\quad \\mathbf { 0 } _ { n - 1 } ^ { \\top } ] . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 631, + 307, + 1063, + 307, + 1063, + 373, + 631, + 373 + ], + "score": 0.94, + "latex": "\\operatorname* { s u p } _ { x } | F ^ { \\tilde { A } _ { n } } ( x ) - F ^ { A _ { n } } ( x ) | \\leq O \\left( n ^ { - 1 } \\right) ," + }, + { + "category_id": 14, + "poly": [ + 373, + 1151, + 1325, + 1151, + 1325, + 1289, + 373, + 1289 + ], + "score": 0.93, + "latex": "\\begin{array} { l } { \\displaystyle m _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\frac { 1 } { n } \\mathrm { t r } \\left( ( A _ { n } ( \\rho , \\varsigma , \\tau ) - \\xi I _ { N } ) _ { [ h + 1 , N ] } ^ { - 1 } \\right) = \\mathbb { E } _ { a } \\left[ ( A _ { n } ( \\rho , \\tau ) - \\xi I _ { N } ) ^ { - 1 } \\right] _ { N N } } \\\\ { \\displaystyle = \\mathbb { E } _ { a } \\bigg [ \\Big ( - \\xi - a ^ { \\top } ( A ^ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ) ^ { - 1 } \\bigg ] . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 602, + 1453, + 1096, + 1453, + 1096, + 1542, + 602, + 1542 + ], + "score": 0.93, + "latex": "\\pmb { w } _ { i } = \\underbrace { \\pmb { w } _ { i } ^ { \\top } \\pmb { x } _ { n } } _ { \\eta _ { i } } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } = \\eta _ { i } \\frac { \\pmb { x } _ { n } } { \\| \\pmb { x } _ { n } \\| } + \\tilde { \\pmb { w } } _ { i } ." + }, + { + "category_id": 13, + "poly": [ + 857, + 390, + 988, + 390, + 988, + 420, + 857, + 420 + ], + "score": 0.92, + "latex": "M \\in \\mathbb { R } ^ { n \\times n }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1377, + 577, + 1377, + 577, + 1412, + 298, + 1412 + ], + "score": 0.92, + "latex": "a _ { 1 } = \\mathbb { E } _ { x \\sim \\mathcal { N } ( 0 , 1 ) } [ x \\varphi ( x ) ]" + }, + { + "category_id": 14, + "poly": [ + 515, + 1012, + 1185, + 1012, + 1185, + 1091, + 515, + 1091 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { ( A - \\xi I _ { N } ) ^ { - 1 } = \\left[ \\begin{array} { l l } { * } & { * } \\\\ { * } & { [ - \\xi - { \\pmb a } ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } { \\pmb a } ] ^ { - 1 } } \\end{array} \\right] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1196, + 784, + 1403, + 784, + 1403, + 819, + 1196, + 819 + ], + "score": 0.91, + "latex": "\\left( N - 1 \\right) \\times \\left( N - 1 \\right)" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1346, + 1309, + 1346, + 1309, + 1380, + 1029, + 1380 + ], + "score": 0.91, + "latex": "\\varphi ( x ) \\ = \\ a _ { 1 } x + \\varphi _ { \\perp } ( x )" + }, + { + "category_id": 13, + "poly": [ + 754, + 767, + 1028, + 767, + 1028, + 839, + 754, + 839 + ], + "score": 0.9, + "latex": "A _ { n } = A = { \\left[ \\begin{array} { l l } { A _ { * } } & { a } \\\\ { \\mathbf { 1 } } & { 0 } \\end{array} \\right] }" + }, + { + "category_id": 13, + "poly": [ + 1226, + 489, + 1263, + 489, + 1263, + 519, + 1226, + 519 + ], + "score": 0.9, + "latex": "A _ { n }" + }, + { + "category_id": 13, + "poly": [ + 371, + 390, + 419, + 390, + 419, + 419, + 371, + 419 + ], + "score": 0.9, + "latex": "F ^ { M }" + }, + { + "category_id": 13, + "poly": [ + 708, + 1376, + 895, + 1376, + 895, + 1410, + 708, + 1410 + ], + "score": 0.9, + "latex": "{ \\pmb w } _ { i } ( \\bar { 1 } \\leq i \\leq \\bar { h } )" + }, + { + "category_id": 13, + "poly": [ + 958, + 1408, + 994, + 1408, + 994, + 1438, + 958, + 1438 + ], + "score": 0.89, + "latex": "\\tilde { \\mathbf { \\pmb { w } } } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 768, + 1322, + 827, + 1322, + 827, + 1350, + 768, + 1350 + ], + "score": 0.89, + "latex": "m _ { 2 , n }" + }, + { + "category_id": 13, + "poly": [ + 695, + 1412, + 731, + 1412, + 731, + 1438, + 695, + 1438 + ], + "score": 0.88, + "latex": "{ \\bf { x } } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 537, + 688, + 1062, + 688, + 1062, + 725, + 537, + 725 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { m _ { n } ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) + m _ { 2 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1196, + 742, + 1264, + 742, + 1264, + 769, + 1196, + 769 + ], + "score": 0.87, + "latex": "\\rho , \\varsigma , \\tau" + }, + { + "category_id": 13, + "poly": [ + 689, + 494, + 730, + 494, + 730, + 519, + 689, + 519 + ], + "score": 0.87, + "latex": "m _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1185, + 1411, + 1222, + 1411, + 1222, + 1438, + 1185, + 1438 + ], + "score": 0.85, + "latex": "{ \\mathbf { \\mathcal { x } } } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 786, + 1148, + 786, + 1148, + 813, + 1113, + 813 + ], + "score": 0.84, + "latex": "A ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 695, + 836, + 718, + 836, + 718, + 861, + 695, + 861 + ], + "score": 0.83, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 769, + 1565, + 788, + 1565, + 788, + 1585, + 769, + 1585 + ], + "score": 0.8, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 495, + 1352, + 517, + 1352, + 517, + 1378, + 495, + 1378 + ], + "score": 0.79, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 1041, + 1802, + 1084, + 1802, + 1084, + 1819, + 1041, + 1819 + ], + "score": 0.78, + "latex": "n - 1" + }, + { + "category_id": 13, + "poly": [ + 872, + 841, + 889, + 841, + 889, + 860, + 872, + 860 + ], + "score": 0.77, + "latex": "\\pmb { s }" + }, + { + "category_id": 13, + "poly": [ + 862, + 744, + 881, + 744, + 881, + 764, + 862, + 764 + ], + "score": 0.76, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 831, + 2001, + 954, + 2001, + 954, + 2028, + 831, + 2028 + ], + "score": 0.63, + "latex": "\\pmb { a } _ { 2 } ^ { \\top } = [ \\pmb { s } _ { 2 } ^ { \\top } , 0 ^ { \\top } ]" + }, + { + "category_id": 14, + "poly": [ + 451, + 1603, + 1170, + 1603, + 1170, + 1821, + 451, + 1821 + ], + "score": 0.54, + "latex": "\\begin{array} { c c c } { \\pmb { a } ^ { \\top } = \\left[ \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| \\pmb { x } _ { n } \\| \\eta _ { 1 } ) } & { \\cdots } & { \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| \\pmb { x } _ { n } \\| \\eta _ { h } ) } & { \\underbrace { 0 \\cdots 0 } _ { n - 1 } \\right] } \\\\ { = \\left[ \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| \\pmb { x } _ { n } \\| \\eta _ { 1 } } & { \\cdots } & { \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| \\pmb { x } _ { n } \\| \\eta _ { h } } & { \\underbrace { 0 \\cdots 0 } _ { n - 1 } \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 775, + 1844, + 898, + 1844, + 898, + 1869, + 775, + 1869 + ], + "score": 0.52, + "latex": "\\pmb { a } _ { 1 } ^ { \\top } = [ \\pmb { s } _ { 1 } ^ { \\top } , 0 ^ { \\top } ]" + }, + { + "category_id": 13, + "poly": [ + 1128, + 1961, + 1171, + 1961, + 1171, + 1977, + 1128, + 1977 + ], + "score": 0.51, + "latex": "n - 1" + }, + { + "category_id": 14, + "poly": [ + 450, + 1605, + 1249, + 1605, + 1249, + 2038, + 450, + 2038 + ], + "score": 0.5, + "latex": "\\begin{array} { r l } { a ^ { \\top } = \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { 1 } ) } & { \\cdots \\cdot \\frac { 1 } { \\sqrt { n } } \\varphi ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { = \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { 1 } } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } a _ { 1 } \\| x _ { n } \\| \\eta _ { h } \\quad \\underbrace { 0 \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] } \\\\ & { + \\underbrace { \\Bigg [ \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { 1 } ) } _ { \\alpha _ { 1 } ^ { \\top } = \\| x _ { n } ^ { \\top } \\| ^ { \\alpha } } \\cdots \\quad \\frac { 1 } { \\sqrt { n } } \\varphi _ { \\bot } ( \\| x _ { n } \\| \\eta _ { h } ) \\quad \\underbrace { 0 \\cdot \\cdots \\cdot 0 } _ { n - 1 } \\Bigg ] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1110, + 1762, + 1125, + 1762, + 1125, + 1783, + 1110, + 1783 + ], + "score": 0.31, + "latex": "0" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 731.0, + 861.0, + 731.0, + 861.0, + 773.0, + 293.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 731.0, + 1195.0, + 731.0, + 1195.0, + 773.0, + 882.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 731.0, + 1405.0, + 731.0, + 1405.0, + 773.0, + 1265.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 782.0, + 753.0, + 782.0, + 753.0, + 819.0, + 293.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 784.0, + 1112.0, + 784.0, + 1112.0, + 821.0, + 1029.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 784.0, + 1195.0, + 784.0, + 1195.0, + 821.0, + 1149.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 832.0, + 694.0, + 832.0, + 694.0, + 867.0, + 294.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 832.0, + 871.0, + 832.0, + 871.0, + 867.0, + 719.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 832.0, + 1132.0, + 832.0, + 1132.0, + 867.0, + 890.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1312.0, + 767.0, + 1312.0, + 767.0, + 1352.0, + 292.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1312.0, + 1407.0, + 1312.0, + 1407.0, + 1352.0, + 828.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1344.0, + 494.0, + 1344.0, + 494.0, + 1384.0, + 292.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1344.0, + 1028.0, + 1344.0, + 1028.0, + 1384.0, + 518.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1344.0, + 1409.0, + 1344.0, + 1409.0, + 1384.0, + 1310.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1375.0, + 297.0, + 1375.0, + 297.0, + 1412.0, + 292.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1375.0, + 707.0, + 1375.0, + 707.0, + 1412.0, + 578.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1375.0, + 1409.0, + 1375.0, + 1409.0, + 1412.0, + 896.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1404.0, + 694.0, + 1404.0, + 694.0, + 1443.0, + 294.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1404.0, + 957.0, + 1404.0, + 957.0, + 1443.0, + 732.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1404.0, + 1184.0, + 1404.0, + 1184.0, + 1443.0, + 995.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1404.0, + 1243.0, + 1404.0, + 1243.0, + 1443.0, + 1223.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1402.0, + 229.0, + 1402.0, + 261.0, + 296.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 260.0, + 1032.0, + 260.0, + 1032.0, + 295.0, + 296.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 383.0, + 370.0, + 383.0, + 370.0, + 426.0, + 293.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 383.0, + 856.0, + 383.0, + 856.0, + 426.0, + 420.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 383.0, + 1407.0, + 383.0, + 1407.0, + 426.0, + 989.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 420.0, + 876.0, + 420.0, + 876.0, + 456.0, + 295.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 426.0, + 1402.0, + 426.0, + 1402.0, + 449.0, + 1378.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1103.0, + 886.0, + 1103.0, + 886.0, + 1140.0, + 295.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 965.0, + 778.0, + 965.0, + 778.0, + 1000.0, + 294.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1556.0, + 768.0, + 1556.0, + 768.0, + 1592.0, + 296.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1556.0, + 822.0, + 1556.0, + 822.0, + 1592.0, + 789.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 484.0, + 688.0, + 484.0, + 688.0, + 524.0, + 294.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 484.0, + 1225.0, + 484.0, + 1225.0, + 524.0, + 731.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 484.0, + 1275.0, + 484.0, + 1275.0, + 524.0, + 1264.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 684.0, + 536.0, + 684.0, + 536.0, + 729.0, + 295.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 684.0, + 1069.0, + 684.0, + 1069.0, + 729.0, + 1063.0, + 729.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 295, + 1019, + 1451, + 1019, + 1451, + 1339, + 295, + 1339 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 312, + 431, + 1471, + 431, + 1471, + 631, + 312, + 631 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 311, + 1692, + 1389, + 1692, + 1389, + 1851, + 311, + 1851 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 292, + 943, + 1405, + 943, + 1405, + 1010, + 292, + 1010 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 290, + 1524, + 1403, + 1524, + 1403, + 1591, + 290, + 1591 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 676, + 1602, + 1021, + 1602, + 1021, + 1640, + 676, + 1640 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 305, + 1426, + 1392, + 1426, + 1392, + 1486, + 305, + 1486 + ], + "score": 0.926 + }, + { + "category_id": 8, + "poly": [ + 595, + 1970, + 1100, + 1970, + 1100, + 2045, + 595, + 2045 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 296, + 1646, + 768, + 1646, + 768, + 1680, + 296, + 1680 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 296, + 389, + 892, + 389, + 892, + 423, + 296, + 423 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 295, + 1381, + 885, + 1381, + 885, + 1416, + 295, + 1416 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 296, + 228, + 740, + 228, + 740, + 265, + 296, + 265 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 299, + 642, + 872, + 642, + 872, + 675, + 299, + 675 + ], + "score": 0.916 + }, + { + "category_id": 8, + "poly": [ + 487, + 272, + 1207, + 272, + 1207, + 378, + 487, + 378 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 291, + 1905, + 1312, + 1905, + 1312, + 1959, + 291, + 1959 + ], + "score": 0.896 + }, + { + "category_id": 8, + "poly": [ + 304, + 680, + 1399, + 680, + 1399, + 935, + 304, + 935 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1481, + 1401, + 1481, + 1401, + 1512, + 1351, + 1512 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1350, + 902, + 1401, + 902, + 1401, + 933, + 1350, + 933 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1604, + 1401, + 1604, + 1401, + 1635, + 1351, + 1635 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1991, + 1401, + 1991, + 1401, + 2022, + 1350, + 2022 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1859, + 1401, + 1859, + 1401, + 1890, + 1351, + 1890 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2114, + 834, + 2114 + ], + "score": 0.871 + }, + { + "category_id": 9, + "poly": [ + 1349, + 299, + 1402, + 299, + 1402, + 332, + 1349, + 332 + ], + "score": 0.871 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1339, + 1402, + 1339, + 1402, + 1370, + 1350, + 1370 + ], + "score": 0.869 + }, + { + "category_id": 8, + "poly": [ + 304, + 680, + 1399, + 680, + 1399, + 935, + 304, + 935 + ], + "score": 0.415 + }, + { + "category_id": 14, + "poly": [ + 313, + 430, + 1428, + 430, + 1428, + 632, + 313, + 632 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { { \\displaystyle { S _ { i k } = \\frac { 1 } { \\sqrt n } \\varphi \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) = \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } a _ { 1 } \\tilde { w } _ { i } ^ { \\top } x _ { k } + \\frac { 1 } { \\sqrt n } \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) } } \\\\ { { \\displaystyle { \\quad = \\frac { 1 } { \\sqrt n } \\varphi ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) + \\frac { 1 } { \\sqrt n } a _ { 1 } \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\frac { 1 } { \\sqrt n } \\left[ \\varphi _ { \\bot } \\left( \\eta _ { i } \\frac { x _ { n } ^ { \\top } x _ { k } } { \\| x _ { n } \\| } + \\tilde { w } _ { i } ^ { \\top } x _ { k } \\right) - \\varphi _ { \\bot } ( \\tilde { w } _ { i } ^ { \\top } x _ { k } ) \\right] } } \\cdot { \\displaystyle ( 7 4 ) } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 310, + 1015, + 1428, + 1015, + 1428, + 1345, + 310, + 1345 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { n _ { 2 , n } ( \\xi , \\rho , \\tau ) = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( A _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } + U C U ^ { \\top } ) ^ { - 1 } a \\bigg ) ^ { - 1 } \\bigg ] } \\\\ & { \\qquad = \\mathbb { E } _ { a } \\bigg [ \\bigg ( - \\xi - \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } } _ { u } + } \\\\ & { \\qquad \\underbrace { a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U } _ { v ^ { \\top } } \\underbrace { ( C ^ { - 1 } + U ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } U ) ^ { - 1 } } _ { S } \\underbrace { U ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a } _ { v } \\bigg ) ^ { - 1 } \\bigg ] } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 598, + 1971, + 1102, + 1971, + 1102, + 2045, + 598, + 2045 + ], + "score": 0.92, + "latex": "v \\mathbb { E } _ { a } v = [ a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\quad 0 ] ." + }, + { + "category_id": 13, + "poly": [ + 531, + 1384, + 609, + 1384, + 609, + 1415, + 531, + 1415 + ], + "score": 0.92, + "latex": "u , v , S" + }, + { + "category_id": 14, + "poly": [ + 488, + 269, + 1209, + 269, + 1209, + 381, + 488, + 381 + ], + "score": 0.92, + "latex": "Q _ { i j } = \\left( \\eta _ { i } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { i } \\right) ^ { \\top } \\left( \\eta _ { j } \\frac { \\pmb { x _ { n } } } { \\lVert \\pmb { x _ { n } } \\rVert } + \\tilde { \\pmb { w } } _ { j } \\right) = \\eta _ { i } \\eta _ { j } + \\underbrace { \\tilde { \\pmb { w } } _ { i } ^ { \\top } \\tilde { \\pmb { w } } _ { j } } _ { \\tilde { Q } _ { i j } } ." + }, + { + "category_id": 14, + "poly": [ + 308, + 1690, + 1386, + 1690, + 1386, + 1864, + 308, + 1864 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } { \\mathbb { E } _ { a } \\boldsymbol { v } ^ { \\top } = \\mathbb { E } _ { a } [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } \\boldsymbol { U } ] = \\mathbb { E } _ { a } [ [ \\boldsymbol { s } ^ { \\top } , \\boldsymbol { 0 } _ { n - 1 } ^ { \\top } ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\boldsymbol { \\eta } } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { \\boldsymbol { u } } \\end{array} ] ] } & { } \\\\ { = [ \\mathbb { E } _ { s } [ \\boldsymbol { s } ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\boldsymbol { \\eta } ] ] } & { 0 ] = [ \\underbrace { \\mathrm { t r } ( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 - h , 1 - h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\boldsymbol { \\eta } \\boldsymbol { s } ^ { \\top } ] ) } _ { \\boldsymbol { v } } } & { 0 ] . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 301, + 680, + 1390, + 680, + 1390, + 936, + 301, + 936 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { A _ { * } = \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { t \\eta \\eta ^ { \\top } } & { a _ { 1 } \\eta u ^ { \\top } } \\\\ { a _ { 1 } u \\eta ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] + \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } \\\\ & { \\quad = \\underbrace { \\left[ \\begin{array} { c c } { \\rho I _ { h } + \\tau \\tilde { Q } } & { \\tilde { S } _ { * } } \\\\ { \\tilde { S } _ { * } ^ { \\top } } & { 0 _ { n - 1 } } \\end{array} \\right] } _ { \\tilde { A } _ { * } } + \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] } _ { U } \\underbrace { \\left[ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} \\right] } _ { C } \\underbrace { \\left[ \\begin{array} { c c } { \\eta } & { \\mathbf { 0 } _ { h } } \\\\ { \\mathbf { 0 } _ { n - 1 } } & { u } \\end{array} \\right] ^ { \\top } } _ { U ^ { \\top } } + \\underbrace { \\left[ \\begin{array} { c c } { E _ { 0 } } & { E _ { 1 } } \\\\ { E _ { 1 } ^ { \\top } } & { \\mathbf { 0 } _ { n - 1 } } \\end{array} \\right] } _ { E } } \\\\ & { \\quad = \\tilde { A } _ { * } + U C U ^ { \\top } + E . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 677, + 1601, + 1023, + 1601, + 1023, + 1639, + 677, + 1639 + ], + "score": 0.89, + "latex": "u \\mathbb { E } _ { \\pmb { a } } u = r \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) ." + }, + { + "category_id": 13, + "poly": [ + 384, + 230, + 732, + 230, + 732, + 263, + 384, + 263 + ], + "score": 0.89, + "latex": "1 \\leq i \\neq j \\leq h , 1 \\leq k \\leq n - 1" + }, + { + "category_id": 13, + "poly": [ + 714, + 1563, + 860, + 1563, + 860, + 1591, + 714, + 1591 + ], + "score": 0.89, + "latex": "n , h , d \\infty" + }, + { + "category_id": 13, + "poly": [ + 1145, + 981, + 1204, + 981, + 1204, + 1010, + 1145, + 1010 + ], + "score": 0.89, + "latex": "m _ { 2 , n }" + }, + { + "category_id": 13, + "poly": [ + 373, + 1527, + 1042, + 1527, + 1042, + 1564, + 373, + 1564 + ], + "score": 0.88, + "latex": "b = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ \\varphi ( \\boldsymbol { x } ) ^ { 2 } ] = \\mathbb { E } _ { \\boldsymbol { x } \\sim \\mathcal { N } ( 0 , 1 ) } [ ( \\phi ( \\boldsymbol { x } ) - \\mathbb { E } \\phi ( \\boldsymbol { x } ) ) ^ { 2 } ] = r _ { ! }" + }, + { + "category_id": 13, + "poly": [ + 744, + 1387, + 803, + 1387, + 803, + 1417, + 744, + 1417 + ], + "score": 0.88, + "latex": "m _ { 2 , n }" + }, + { + "category_id": 14, + "poly": [ + 305, + 1425, + 1386, + 1425, + 1386, + 1486, + 305, + 1486 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\mathbb { E } _ { a } u = \\mathbb { E } _ { a } \\Big [ a ^ { \\top } ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) ^ { - 1 } a \\Big ] = \\mathrm { t r } \\left( \\mathbb { E } _ { s } \\big [ s s ^ { \\top } \\big ] ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . { h } , 1 . { h } ] } ^ { - 1 } \\right) = b \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1099, + 951, + 1191, + 951, + 1191, + 974, + 1099, + 974 + ], + "score": 0.87, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 839, + 644, + 874, + 644, + 874, + 673, + 839, + 673 + ], + "score": 0.87, + "latex": "A _ { * }" + }, + { + "category_id": 13, + "poly": [ + 526, + 1650, + 551, + 1650, + 551, + 1675, + 526, + 1675 + ], + "score": 0.83, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 369, + 1906, + 1152, + 1906, + 1152, + 1961, + 369, + 1961 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { v = \\mathrm { t r } \\left( ( \\tilde { A } _ { * } - \\xi I _ { N - 1 } ) _ { [ 1 . . h , 1 . . h ] } ^ { - 1 } \\mathbb { E } _ { s } [ \\eta s ^ { \\top } ] \\right) = a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } ( \\xi , \\rho , \\tau ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 861, + 392, + 882, + 392, + 882, + 418, + 861, + 418 + ], + "score": 0.81, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 865, + 947, + 891, + 947, + 891, + 974, + 865, + 974 + ], + "score": 0.81, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 862, + 1389, + 881, + 1389, + 881, + 1410, + 862, + 1410 + ], + "score": 0.78, + "latex": "u" + }, + { + "category_id": 13, + "poly": [ + 391, + 1654, + 410, + 1654, + 410, + 1675, + 391, + 1675 + ], + "score": 0.76, + "latex": "\\textbf { { v } }" + }, + { + "category_id": 13, + "poly": [ + 736, + 1654, + 756, + 1654, + 756, + 1675, + 736, + 1675 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 944.0, + 864.0, + 944.0, + 864.0, + 979.0, + 294.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 944.0, + 1098.0, + 944.0, + 1098.0, + 979.0, + 892.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 944.0, + 1404.0, + 944.0, + 1404.0, + 979.0, + 1192.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 975.0, + 1144.0, + 975.0, + 1144.0, + 1012.0, + 295.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 975.0, + 1289.0, + 975.0, + 1289.0, + 1012.0, + 1205.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1524.0, + 372.0, + 1524.0, + 372.0, + 1568.0, + 293.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1524.0, + 1408.0, + 1524.0, + 1408.0, + 1568.0, + 1043.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1560.0, + 713.0, + 1560.0, + 713.0, + 1593.0, + 297.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1640.0, + 390.0, + 1640.0, + 390.0, + 1687.0, + 293.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1640.0, + 525.0, + 1640.0, + 525.0, + 1687.0, + 411.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1640.0, + 735.0, + 1640.0, + 735.0, + 1687.0, + 552.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1640.0, + 770.0, + 1640.0, + 770.0, + 1687.0, + 757.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 387.0, + 860.0, + 387.0, + 860.0, + 426.0, + 295.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 387.0, + 895.0, + 387.0, + 895.0, + 426.0, + 883.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1374.0, + 530.0, + 1374.0, + 530.0, + 1423.0, + 293.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1374.0, + 743.0, + 1374.0, + 743.0, + 1423.0, + 610.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1374.0, + 861.0, + 1374.0, + 861.0, + 1423.0, + 804.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1374.0, + 886.0, + 1374.0, + 886.0, + 1423.0, + 882.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 383.0, + 229.0, + 383.0, + 266.0, + 297.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 229.0, + 741.0, + 229.0, + 741.0, + 266.0, + 733.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 639.0, + 838.0, + 639.0, + 838.0, + 679.0, + 294.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1895.0, + 368.0, + 1895.0, + 368.0, + 1967.0, + 287.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1895.0, + 1323.0, + 1895.0, + 1323.0, + 1967.0, + 1153.0, + 1967.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 296, + 996, + 1424, + 996, + 1424, + 1253, + 296, + 1253 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 299, + 279, + 1610, + 279, + 1610, + 794, + 299, + 794 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 299, + 1321, + 1404, + 1321, + 1404, + 1456, + 299, + 1456 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 299, + 1861, + 1471, + 1861, + 1471, + 2007, + 299, + 2007 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 463, + 855, + 1230, + 855, + 1230, + 932, + 463, + 932 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 293, + 1810, + 1270, + 1810, + 1270, + 1848, + 293, + 1848 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 298, + 1591, + 670, + 1591, + 670, + 1625, + 298, + 1625 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 295, + 1648, + 1110, + 1648, + 1110, + 1684, + 295, + 1684 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 298, + 807, + 607, + 807, + 607, + 840, + 298, + 840 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 295, + 945, + 1274, + 945, + 1274, + 980, + 295, + 980 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 297, + 1503, + 1204, + 1503, + 1204, + 1539, + 297, + 1539 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 297, + 1267, + 779, + 1267, + 779, + 1302, + 297, + 1302 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 297, + 228, + 687, + 228, + 687, + 264, + 297, + 264 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1758, + 1401, + 1758, + 1401, + 1789, + 1351, + 1789 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1707, + 1401, + 1707, + 1401, + 1739, + 1351, + 1739 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1350, + 878, + 1402, + 878, + 1402, + 909, + 1350, + 909 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1455, + 1401, + 1455, + 1401, + 1485, + 1351, + 1485 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1351, + 2004, + 1402, + 2004, + 1402, + 2034, + 1351, + 2034 + ], + "score": 0.839 + }, + { + "category_id": 2, + "poly": [ + 1373, + 1505, + 1403, + 1505, + 1403, + 1534, + 1373, + 1534 + ], + "score": 0.789 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1189, + 1402, + 1189, + 1402, + 1220, + 1350, + 1220 + ], + "score": 0.786 + }, + { + "category_id": 9, + "poly": [ + 1351, + 739, + 1402, + 739, + 1402, + 770, + 1351, + 770 + ], + "score": 0.725 + }, + { + "category_id": 8, + "poly": [ + 627, + 1700, + 1067, + 1700, + 1067, + 1795, + 627, + 1795 + ], + "score": 0.548 + }, + { + "category_id": 8, + "poly": [ + 630, + 1750, + 1065, + 1750, + 1065, + 1795, + 630, + 1795 + ], + "score": 0.519 + }, + { + "category_id": 8, + "poly": [ + 630, + 1701, + 919, + 1701, + 919, + 1743, + 630, + 1743 + ], + "score": 0.455 + }, + { + "category_id": 13, + "poly": [ + 328, + 232, + 470, + 232, + 470, + 263, + 328, + 263 + ], + "score": 0.93, + "latex": "n , d , p \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 598, + 1268, + 745, + 1268, + 745, + 1303, + 598, + 1303 + ], + "score": 0.93, + "latex": "m _ { 1 , n } ( \\xi , \\rho , \\tau )" + }, + { + "category_id": 14, + "poly": [ + 310, + 995, + 1422, + 995, + 1422, + 1255, + 310, + 1255 + ], + "score": 0.92, + "latex": "\\begin{array} { r l r } & { } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) \\to \\mathbb { E } _ { a } \\Big [ \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } \\Big ] \\to \\Big ( - \\xi - u + v ^ { \\top } S v \\Big ) ^ { - 1 } } \\\\ & { \\to \\Big ( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\Big ( a _ { 1 } \\sqrt { \\gamma _ { 2 } ^ { 2 } / \\gamma _ { 1 } } m _ { 1 , n } \\Big ) ^ { 2 } \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] _ { [ 1 , 1 ] } ^ { - 1 } \\Big ) ^ { - 1 } } \\\\ & { } & { = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 , n } + \\frac { \\gamma _ { 2 } a _ { 1 } ^ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) } { m _ { 1 , n } \\left( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau \\right) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 310, + 277, + 1428, + 277, + 1428, + 798, + 310, + 798 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } { \\iota _ { \\alpha } S ^ { - 1 } = \\mathbb { E } _ { \\alpha } [ C ^ { - 1 } + U ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } U ] } \\\\ { = } & { [ \\begin{array} { c c } { \\tau } & { a _ { 1 } } \\\\ { a _ { 1 } } & { 0 } \\end{array} ] ^ { - 1 } + \\mathbb { E } _ { \\alpha } [ [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) ^ { - 1 } [ \\begin{array} { c c } { \\eta } & { 0 _ { h } } \\\\ { 0 _ { h - 1 } } & { u } \\end{array} ] ] } \\\\ { = } & { [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + \\mathbb { E } _ { \\alpha } [ \\begin{array} { c c } { \\eta ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\eta } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } ] } \\\\ { 0 } & { u ^ { \\top } ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } \\frac { 1 } { | \\mu _ { 1 } + 1 , h - w - 1 | ^ { u } } } \\end{array} ] } \\\\ { = } & [ \\begin{array} { c c } { 0 } & { 1 / \\rho _ { 2 } } \\\\ { 1 / \\mu _ { 1 } } & { - \\tau / a _ { 1 } ^ { 2 } } \\end{array} ] + [ \\begin{array} { c c } { \\mathbb { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } { ( 1 - A ) ! } \\mathbb { E } _ { \\alpha } | \\eta \\eta ^ { \\top } ) } & \\mathrm { t r } ( ( \\mathring { A } _ { \\star } - \\xi I _ { N - 1 } ) \\frac { 1 } | \\end{array} \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 424, + 1811, + 889, + 1811, + 889, + 1847, + 424, + 1847 + ], + "score": 0.92, + "latex": "m ( \\xi , \\rho , \\tau ) = \\gamma _ { 2 } m _ { 1 } ( \\xi , \\rho , \\tau ) + m _ { 2 } ( \\xi , \\rho , \\tau )" + }, + { + "category_id": 13, + "poly": [ + 453, + 808, + 598, + 808, + 598, + 839, + 453, + 839 + ], + "score": 0.92, + "latex": "n , d , h \\infty" + }, + { + "category_id": 14, + "poly": [ + 628, + 1697, + 1070, + 1697, + 1070, + 1797, + 628, + 1797 + ], + "score": 0.91, + "latex": "\\begin{array} { l } { { m _ { 2 } = \\left( - \\xi - r \\gamma _ { 2 } m _ { 1 } \\right) ^ { - 1 } , } } \\\\ { { \\nonumber } } \\\\ { { m _ { 1 } = \\left( - \\xi - \\rho - \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 } - r m _ { 2 } \\right) ^ { - 1 } . } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 465, + 853, + 1234, + 853, + 1234, + 932, + 465, + 932 + ], + "score": 0.91, + "latex": "S ^ { - 1 } \\to \\mathbb { E } _ { a } S ^ { - 1 } = \\left[ \\begin{array} { c c } { \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ( \\xi , \\rho , \\tau ) } & { 1 / a _ { 1 } } \\\\ { 1 / a _ { 1 } } & { m _ { 2 , n } ( \\xi , \\rho , \\tau ) - \\tau / a ^ { 2 } } \\end{array} \\right] ." + }, + { + "category_id": 14, + "poly": [ + 310, + 1860, + 1428, + 1860, + 1428, + 2014, + 310, + 2014 + ], + "score": 0.91, + "latex": "\\begin{array} { l } { \\displaystyle - q ( \\xi ) = \\frac { \\partial } { \\partial x } m ( \\xi , r x , t x ) \\Big | _ { x = 0 } = \\left. r \\frac { \\partial } { \\partial \\rho } m ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } } \\\\ { \\displaystyle \\quad = \\left. r \\gamma _ { 2 } \\frac { \\partial } { \\partial \\rho } m _ { 1 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. r \\frac { \\partial } { \\partial \\rho } m _ { 2 } ( \\xi , \\rho , 0 ) \\right| _ { \\rho = 0 } + \\left. t \\gamma _ { 2 } \\frac { \\partial } { \\partial \\tau } m _ { 1 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } + \\left. t \\frac { \\partial } { \\partial \\tau } m _ { 2 } ( \\xi , 0 , \\tau ) \\right| _ { \\tau = 0 } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 291, + 1320, + 1391, + 1320, + 1391, + 1454, + 291, + 1454 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { m _ { 1 , n } ( \\xi , \\rho , \\tau ) \\to } \\\\ & { \\left( - \\xi - \\rho - \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } \\tau ^ { 2 } m _ { 1 , n } - r m _ { 2 , n } + \\frac { \\tau ^ { 2 } \\gamma _ { 1 } ^ { - 1 } \\gamma _ { 2 } m _ { 1 , n } ^ { 2 } ( a _ { 1 } ^ { 2 } m _ { 2 } - \\tau ) - 2 \\tau a _ { 1 } ^ { 2 } m _ { 1 , n } m _ { 2 , n } + a _ { 1 } ^ { 4 } m _ { 1 , n } m _ { 2 , n } ^ { 2 } } { m _ { 1 , n } ( a _ { 1 } ^ { 2 } m _ { 2 , n } - \\tau ) - \\gamma _ { 1 } \\gamma _ { 2 } ^ { - 1 } } \\right) ^ { - 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 651, + 1655, + 752, + 1655, + 752, + 1683, + 651, + 1683 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1204, + 954, + 1262, + 954, + 1262, + 982, + 1204, + 982 + ], + "score": 0.89, + "latex": "m _ { 2 , n }" + }, + { + "category_id": 13, + "poly": [ + 655, + 232, + 676, + 232, + 676, + 258, + 655, + 258 + ], + "score": 0.78, + "latex": "S" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1589.0, + 673.0, + 1589.0, + 673.0, + 1629.0, + 296.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1509.0, + 1402.0, + 1509.0, + 1402.0, + 1536.0, + 1378.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1807.0, + 423.0, + 1807.0, + 423.0, + 1850.0, + 292.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1807.0, + 1272.0, + 1807.0, + 1272.0, + 1850.0, + 890.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1643.0, + 650.0, + 1643.0, + 650.0, + 1689.0, + 293.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1643.0, + 1109.0, + 1643.0, + 1109.0, + 1689.0, + 753.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 803.0, + 452.0, + 803.0, + 452.0, + 844.0, + 294.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 803.0, + 610.0, + 803.0, + 610.0, + 844.0, + 599.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 935.0, + 1203.0, + 935.0, + 1203.0, + 992.0, + 290.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 935.0, + 1279.0, + 935.0, + 1279.0, + 992.0, + 1263.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1503.0, + 1206.0, + 1503.0, + 1206.0, + 1539.0, + 296.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1261.0, + 597.0, + 1261.0, + 597.0, + 1307.0, + 294.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1261.0, + 783.0, + 1261.0, + 783.0, + 1307.0, + 746.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 327.0, + 227.0, + 327.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 227.0, + 654.0, + 227.0, + 654.0, + 267.0, + 471.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 227.0, + 688.0, + 227.0, + 688.0, + 267.0, + 677.0, + 267.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 297, + 1040, + 1484, + 1040, + 1484, + 1182, + 297, + 1182 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 349, + 1280, + 1306, + 1280, + 1306, + 1429, + 349, + 1429 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 400, + 308, + 1294, + 308, + 1294, + 440, + 400, + 440 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 292, + 228, + 1403, + 228, + 1403, + 295, + 292, + 295 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 291, + 1443, + 1406, + 1443, + 1406, + 1509, + 291, + 1509 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 639, + 1404, + 639, + 1404, + 707, + 297, + 707 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 296, + 961, + 1409, + 961, + 1409, + 1027, + 296, + 1027 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 510, + 504, + 1186, + 504, + 1186, + 578, + 510, + 578 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 297, + 1198, + 1408, + 1198, + 1408, + 1269, + 297, + 1269 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 301, + 1732, + 1116, + 1732, + 1116, + 1768, + 301, + 1768 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 444, + 900, + 1252, + 900, + 1252, + 946, + 444, + 946 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 1599, + 899, + 1599, + 899, + 1634, + 297, + 1634 + ], + "score": 0.927 + }, + { + "category_id": 8, + "poly": [ + 345, + 1522, + 1304, + 1522, + 1304, + 1588, + 345, + 1588 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 297, + 1925, + 509, + 1925, + 509, + 1957, + 297, + 1957 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 295, + 590, + 1072, + 590, + 1072, + 624, + 295, + 624 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 298, + 739, + 671, + 739, + 671, + 773, + 298, + 773 + ], + "score": 0.923 + }, + { + "category_id": 8, + "poly": [ + 294, + 1968, + 1440, + 1968, + 1440, + 2042, + 294, + 2042 + ], + "score": 0.923 + }, + { + "category_id": 8, + "poly": [ + 327, + 1782, + 1329, + 1782, + 1329, + 1844, + 327, + 1844 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 296, + 846, + 1401, + 846, + 1401, + 888, + 296, + 888 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 299, + 1680, + 746, + 1680, + 746, + 1714, + 299, + 1714 + ], + "score": 0.906 + }, + { + "category_id": 9, + "poly": [ + 1351, + 523, + 1401, + 523, + 1401, + 555, + 1351, + 555 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 367, + 1401, + 367, + 1401, + 398, + 1351, + 398 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1868, + 1402, + 1868, + 1402, + 1899, + 1350, + 1899 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1540, + 1401, + 1540, + 1401, + 1572, + 1351, + 1572 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1351, + 909, + 1400, + 909, + 1400, + 939, + 1351, + 939 + ], + "score": 0.866 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1379, + 1401, + 1379, + 1401, + 1409, + 1351, + 1409 + ], + "score": 0.862 + }, + { + "category_id": 0, + "poly": [ + 300, + 797, + 776, + 797, + 776, + 831, + 300, + 831 + ], + "score": 0.861 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1802, + 1401, + 1802, + 1401, + 1831, + 1352, + 1831 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 1374, + 591, + 1402, + 591, + 1402, + 620, + 1374, + 620 + ], + "score": 0.797 + }, + { + "category_id": 8, + "poly": [ + 323, + 1851, + 885, + 1851, + 885, + 1911, + 323, + 1911 + ], + "score": 0.792 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1601, + 1402, + 1601, + 1402, + 1629, + 1374, + 1629 + ], + "score": 0.788 + }, + { + "category_id": 1, + "poly": [ + 298, + 454, + 852, + 454, + 852, + 488, + 298, + 488 + ], + "score": 0.535 + }, + { + "category_id": 14, + "poly": [ + 348, + 1279, + 1305, + 1279, + 1305, + 1433, + 348, + 1433 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { d \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } \\mathrm { t r } ( Q _ { 1 } ) \\geq \\frac { 1 } { n } \\mathrm { t r } ( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X ) \\sim \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot W ^ { \\top } W ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\frac { 1 } { n } \\mathrm { t r } ( ( S ^ { \\top } S ) ^ { - 1 } \\cdot ( I + Q ) ) = \\operatorname* { l i m } _ { \\xi 0 } - \\frac { \\partial } { \\partial x } \\tilde { m } _ { n } ( \\xi , x , 0 , x ) \\infty , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 512, + 501, + 1186, + 501, + 1186, + 578, + 512, + 578 + ], + "score": 0.94, + "latex": "V _ { ( \\gamma _ { 1 } \\to \\infty ) } = \\operatorname* { l i m } _ { \\xi \\to 0 } q _ { + } ( \\xi ) = \\operatorname* { l i m } _ { \\xi \\to 0 } \\left( q ( \\xi ) - \\frac { \\gamma _ { 2 } - 1 } { \\xi ^ { 2 } } \\right) = \\frac { 1 } { \\gamma _ { 2 } - 1 } ." + }, + { + "category_id": 14, + "poly": [ + 401, + 306, + 1298, + 306, + 1298, + 444, + 401, + 444 + ], + "score": 0.94, + "latex": "q ( \\xi ) = \\frac { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) - \\xi ^ { 2 } \\right) \\left( \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } + r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) } { 2 \\xi ^ { 2 } \\sqrt { \\left( r \\left( \\gamma _ { 2 } - 1 \\right) + \\xi ^ { 2 } \\right) ^ { 2 } - 4 r \\gamma _ { 2 } \\xi ^ { 2 } } } ." + }, + { + "category_id": 13, + "poly": [ + 1022, + 846, + 1321, + 846, + 1321, + 888, + 1022, + 888 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { B = \\frac { r ^ { 2 } } { d } \\mathrm { t r } \\left( Q _ { 1 } + Q _ { 2 } + I _ { d } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 465, + 1199, + 543, + 1199, + 543, + 1237, + 465, + 1237 + ], + "score": 0.93, + "latex": "X / { \\sqrt { d } }" + }, + { + "category_id": 13, + "poly": [ + 907, + 1473, + 1035, + 1473, + 1035, + 1508, + 907, + 1508 + ], + "score": 0.93, + "latex": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 1 } \\right)" + }, + { + "category_id": 14, + "poly": [ + 324, + 1779, + 1326, + 1779, + 1326, + 1915, + 324, + 1915 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { Q _ { 1 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } X ^ { \\top } , } \\\\ & { Q _ { 2 } = X \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) W ^ { \\top } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1234, + 416, + 1234, + 416, + 1269, + 298, + 1269 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( 0 , 1 / d )" + }, + { + "category_id": 13, + "poly": [ + 574, + 1474, + 795, + 1474, + 795, + 1509, + 574, + 1509 + ], + "score": 0.92, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) = O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1249, + 1475, + 1376, + 1475, + 1376, + 1508, + 1249, + 1508 + ], + "score": 0.92, + "latex": "d ^ { - 1 } \\mathrm { t r } \\left( Q _ { 2 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 951, + 961, + 1066, + 961, + 1066, + 996, + 951, + 996 + ], + "score": 0.92, + "latex": "\\phi ( W ^ { \\top } X )" + }, + { + "category_id": 13, + "poly": [ + 598, + 592, + 679, + 592, + 679, + 622, + 598, + 622 + ], + "score": 0.92, + "latex": "\\gamma _ { 2 } < 1" + }, + { + "category_id": 13, + "poly": [ + 607, + 1602, + 694, + 1602, + 694, + 1632, + 607, + 1632 + ], + "score": 0.91, + "latex": "\\gamma _ { 2 } 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 673, + 431, + 673, + 431, + 706, + 298, + 706 + ], + "score": 0.91, + "latex": "\\log ( 1 + e ^ { x } )" + }, + { + "category_id": 13, + "poly": [ + 763, + 263, + 883, + 263, + 883, + 294, + 763, + 294 + ], + "score": 0.91, + "latex": "\\rho = \\tau = 0" + }, + { + "category_id": 13, + "poly": [ + 597, + 1200, + 778, + 1200, + 778, + 1232, + 597, + 1232 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { d \\times n } ~ = ~ \\mathbb { R } ^ { d \\times p }" + }, + { + "category_id": 14, + "poly": [ + 346, + 1520, + 1305, + 1520, + 1305, + 1588, + 346, + 1588 + ], + "score": 0.91, + "latex": "\\frac { 1 } { d } \\mathrm { t r } ( Q _ { 2 } ) = \\frac { 1 } { d } \\mathrm { t r } \\left( W ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) \\leq \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) ^ { - 1 } ) \\mathrm { t r } \\left( W ^ { \\top } X \\right) = O ( 1 ) ." + }, + { + "category_id": 14, + "poly": [ + 310, + 1038, + 1428, + 1038, + 1428, + 1186, + 310, + 1186 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\overset { ! } { \\operatorname { t r } } ( Q _ { 1 } ) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } \\right) = \\displaystyle \\frac { 1 } { d } \\mathrm { t r } \\left( K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\\\ { \\geq \\displaystyle \\frac { 1 } { d } \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) \\mathrm { t r } \\left( [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) = \\frac { \\lambda _ { \\operatorname* { m i n } } ( K _ { W } ) } { n } \\mathrm { t r } \\left( ( S S ^ { \\top } ) ^ { - 1 } \\cdot \\frac { 1 } { d } X ^ { \\top } X [ \\phi ( W ^ { \\top } X ) ] ^ { - 1 } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 795, + 1601, + 890, + 1601, + 890, + 1628, + 795, + 1628 + ], + "score": 0.9, + "latex": "B \\infty" + }, + { + "category_id": 13, + "poly": [ + 366, + 1736, + 438, + 1736, + 438, + 1763, + 366, + 1763 + ], + "score": 0.9, + "latex": "h > n" + }, + { + "category_id": 13, + "poly": [ + 298, + 992, + 632, + 992, + 632, + 1028, + 298, + 1028 + ], + "score": 0.9, + "latex": "[ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } = [ \\phi ( X ^ { \\top } W ) ] ^ { - 1 }" + }, + { + "category_id": 13, + "poly": [ + 376, + 965, + 468, + 965, + 468, + 991, + 376, + 991 + ], + "score": 0.89, + "latex": "\\textit { h } = \\textit { n }" + }, + { + "category_id": 13, + "poly": [ + 915, + 674, + 1020, + 674, + 1020, + 706, + 915, + 706 + ], + "score": 0.89, + "latex": "c _ { 2 } = 1 / 4" + }, + { + "category_id": 14, + "poly": [ + 310, + 1964, + 1428, + 1964, + 1428, + 2044, + 310, + 2044 + ], + "score": 0.88, + "latex": "\\operatorname { I } ^ { \\mathrm { { T } } } ( Q _ { 1 } ) = 2 \\mathrm { { t r } } \\left( { \\frac { X ^ { \\top } X } { d } } { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) { \\Big ( } \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) { \\Big ) } ^ { - 1 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1019, + 1450, + 1113, + 1450, + 1113, + 1472, + 1019, + 1472 + ], + "score": 0.88, + "latex": "n \\infty" + }, + { + "category_id": 13, + "poly": [ + 460, + 641, + 673, + 641, + 673, + 676, + 460, + 676 + ], + "score": 0.88, + "latex": "\\phi ( x ) = \\mathrm { R e L U } ( x )" + }, + { + "category_id": 13, + "poly": [ + 466, + 1605, + 557, + 1605, + 557, + 1629, + 466, + 1629 + ], + "score": 0.87, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1121, + 640, + 1404, + 640, + 1404, + 677, + 1121, + 677 + ], + "score": 0.87, + "latex": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) =" + }, + { + "category_id": 13, + "poly": [ + 1017, + 997, + 1053, + 997, + 1053, + 1026, + 1017, + 1026 + ], + "score": 0.87, + "latex": "Q _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 446, + 900, + 1253, + 900, + 1253, + 944, + 446, + 944 + ], + "score": 0.87, + "latex": "Q _ { 1 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } K _ { W } [ \\phi ( X ^ { \\top } W ) ] ^ { \\dagger } X ^ { \\top } ; \\quad Q _ { 2 } = X [ \\phi ( W ^ { \\top } X ) ] ^ { \\dagger } W ^ { \\top } ." + }, + { + "category_id": 13, + "poly": [ + 466, + 267, + 510, + 267, + 510, + 294, + 466, + 294 + ], + "score": 0.86, + "latex": "\\tau , \\rho" + }, + { + "category_id": 13, + "poly": [ + 856, + 854, + 880, + 854, + 880, + 880, + 856, + 880 + ], + "score": 0.86, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 804, + 965, + 824, + 965, + 824, + 995, + 804, + 995 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 665, + 1683, + 747, + 1683, + 747, + 1714, + 665, + 1714 + ], + "score": 0.83, + "latex": "\\gamma _ { 2 } > 1" + }, + { + "category_id": 13, + "poly": [ + 760, + 676, + 901, + 676, + 901, + 704, + 760, + 704 + ], + "score": 0.83, + "latex": "c _ { 1 } \\approx 0 . 2 7 1 5" + }, + { + "category_id": 13, + "poly": [ + 371, + 1203, + 405, + 1203, + 405, + 1231, + 371, + 1231 + ], + "score": 0.8, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 938, + 642, + 1057, + 642, + 1057, + 675, + 938, + 675 + ], + "score": 0.79, + "latex": "c _ { 2 } = 1 / 4" + }, + { + "category_id": 13, + "poly": [ + 689, + 642, + 925, + 642, + 925, + 675, + 689, + 675 + ], + "score": 0.77, + "latex": "c _ { 1 } = 1 / 2 - 1 / ( 2 \\pi )" + }, + { + "category_id": 13, + "poly": [ + 700, + 800, + 774, + 800, + 774, + 828, + 700, + 828 + ], + "score": 0.73, + "latex": "h = n" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 737.0, + 675.0, + 737.0, + 675.0, + 777.0, + 295.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1676.0, + 664.0, + 1676.0, + 664.0, + 1718.0, + 294.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1676.0, + 751.0, + 1676.0, + 751.0, + 1718.0, + 748.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 793.0, + 699.0, + 793.0, + 699.0, + 835.0, + 294.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 793.0, + 780.0, + 793.0, + 780.0, + 835.0, + 775.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 595.0, + 1402.0, + 595.0, + 1402.0, + 623.0, + 1378.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1605.0, + 1402.0, + 1605.0, + 1402.0, + 1632.0, + 1378.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 228.0, + 1404.0, + 228.0, + 1404.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 259.0, + 465.0, + 259.0, + 465.0, + 298.0, + 294.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 259.0, + 762.0, + 259.0, + 762.0, + 298.0, + 511.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 259.0, + 952.0, + 259.0, + 952.0, + 298.0, + 884.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1442.0, + 1018.0, + 1442.0, + 1018.0, + 1479.0, + 295.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1442.0, + 1404.0, + 1442.0, + 1404.0, + 1479.0, + 1114.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1471.0, + 573.0, + 1471.0, + 573.0, + 1511.0, + 294.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 1471.0, + 906.0, + 1471.0, + 906.0, + 1511.0, + 796.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1471.0, + 1248.0, + 1471.0, + 1248.0, + 1511.0, + 1036.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1471.0, + 1391.0, + 1471.0, + 1391.0, + 1511.0, + 1377.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 638.0, + 459.0, + 638.0, + 459.0, + 678.0, + 292.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 638.0, + 688.0, + 638.0, + 688.0, + 678.0, + 674.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 638.0, + 937.0, + 638.0, + 937.0, + 678.0, + 926.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 638.0, + 1120.0, + 638.0, + 1120.0, + 678.0, + 1058.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 669.0, + 297.0, + 669.0, + 297.0, + 708.0, + 291.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 669.0, + 759.0, + 669.0, + 759.0, + 708.0, + 432.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 669.0, + 914.0, + 669.0, + 914.0, + 708.0, + 902.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 669.0, + 1035.0, + 669.0, + 1035.0, + 708.0, + 1021.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 956.0, + 375.0, + 956.0, + 375.0, + 1001.0, + 292.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 956.0, + 803.0, + 956.0, + 803.0, + 1001.0, + 469.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 956.0, + 950.0, + 956.0, + 950.0, + 1001.0, + 825.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 956.0, + 1406.0, + 956.0, + 1406.0, + 1001.0, + 1067.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 988.0, + 1016.0, + 988.0, + 1016.0, + 1032.0, + 633.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 988.0, + 1058.0, + 988.0, + 1058.0, + 1032.0, + 1054.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1192.0, + 370.0, + 1192.0, + 370.0, + 1240.0, + 291.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1192.0, + 464.0, + 1192.0, + 464.0, + 1240.0, + 406.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1192.0, + 596.0, + 1192.0, + 596.0, + 1240.0, + 544.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1192.0, + 1411.0, + 1192.0, + 1411.0, + 1240.0, + 779.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1227.0, + 297.0, + 1227.0, + 297.0, + 1274.0, + 289.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1227.0, + 433.0, + 1227.0, + 433.0, + 1274.0, + 417.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1731.0, + 365.0, + 1731.0, + 365.0, + 1770.0, + 295.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1731.0, + 1117.0, + 1731.0, + 1117.0, + 1770.0, + 439.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1598.0, + 465.0, + 1598.0, + 465.0, + 1638.0, + 295.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1598.0, + 606.0, + 1598.0, + 606.0, + 1638.0, + 558.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1598.0, + 794.0, + 1598.0, + 794.0, + 1638.0, + 695.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1598.0, + 901.0, + 1598.0, + 901.0, + 1638.0, + 891.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1921.0, + 510.0, + 1921.0, + 510.0, + 1960.0, + 294.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 588.0, + 597.0, + 588.0, + 597.0, + 628.0, + 298.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 588.0, + 1072.0, + 588.0, + 1072.0, + 628.0, + 680.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 841.0, + 855.0, + 841.0, + 855.0, + 893.0, + 289.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 841.0, + 1021.0, + 841.0, + 1021.0, + 893.0, + 881.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 841.0, + 1408.0, + 841.0, + 1408.0, + 893.0, + 1322.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 452.0, + 854.0, + 452.0, + 854.0, + 492.0, + 295.0, + 492.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1141, + 1406, + 1141, + 1406, + 1248, + 297, + 1248 + ], + "score": 0.977 + }, + { + "category_id": 8, + "poly": [ + 349, + 595, + 1348, + 595, + 1348, + 828, + 349, + 828 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 400, + 219, + 1552, + 219, + 1552, + 502, + 400, + 502 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 294, + 1608, + 1405, + 1608, + 1405, + 1674, + 294, + 1674 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 293, + 931, + 1405, + 931, + 1405, + 996, + 293, + 996 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 298, + 1967, + 1402, + 1967, + 1402, + 2036, + 298, + 2036 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 298, + 1880, + 1401, + 1880, + 1401, + 1951, + 298, + 1951 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 844, + 1400, + 844, + 1400, + 913, + 296, + 913 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 446, + 1264, + 1252, + 1264, + 1252, + 1336, + 446, + 1336 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 295, + 513, + 1407, + 513, + 1407, + 580, + 295, + 580 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 303, + 1403, + 1396, + 1403, + 1396, + 1487, + 303, + 1487 + ], + "score": 0.939 + }, + { + "category_id": 0, + "poly": [ + 299, + 1552, + 741, + 1552, + 741, + 1586, + 299, + 1586 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 300, + 1352, + 751, + 1352, + 751, + 1386, + 300, + 1386 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 298, + 1083, + 640, + 1083, + 640, + 1115, + 298, + 1115 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.917 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1808, + 1400, + 1808, + 1400, + 1840, + 1338, + 1840 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1485, + 1401, + 1485, + 1401, + 1515, + 1351, + 1515 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1350, + 1282, + 1401, + 1282, + 1401, + 1313, + 1350, + 1313 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1351, + 774, + 1400, + 774, + 1400, + 804, + 1351, + 804 + ], + "score": 0.865 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1722, + 1400, + 1722, + 1400, + 1754, + 1352, + 1754 + ], + "score": 0.848 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2114, + 834, + 2114 + ], + "score": 0.829 + }, + { + "category_id": 8, + "poly": [ + 328, + 1696, + 1318, + 1696, + 1318, + 1779, + 328, + 1779 + ], + "score": 0.81 + }, + { + "category_id": 9, + "poly": [ + 1350, + 462, + 1402, + 462, + 1402, + 494, + 1350, + 494 + ], + "score": 0.792 + }, + { + "category_id": 8, + "poly": [ + 345, + 1788, + 932, + 1788, + 932, + 1864, + 345, + 1864 + ], + "score": 0.357 + }, + { + "category_id": 8, + "poly": [ + 330, + 1694, + 1321, + 1694, + 1321, + 1864, + 330, + 1864 + ], + "score": 0.142 + }, + { + "category_id": 14, + "poly": [ + 348, + 594, + 1347, + 594, + 1347, + 829, + 348, + 829 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { \\frac { 2 } { d } \\mathrm { t r } \\left( Q _ { 2 } \\right) = 2 \\mathrm { t r } \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\cdot \\frac { 1 } { d } W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq \\mathrm { t r } \\left( \\left( \\left( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\right) ^ { - 1 } \\phi ( X ^ { \\top } W ) \\right) ( \\ldots ) ^ { \\top } \\right) + \\mathrm { t r } \\left( d ^ { - 2 } X ^ { \\top } W W ^ { \\top } X \\right) } \\\\ & { \\qquad \\leq n \\cdot \\sigma _ { \\operatorname* { m i n } } \\left( \\phi ( X ^ { \\top } W ) \\right) ^ { - 2 } + \\frac { 1 } { d } \\lambda _ { \\operatorname* { m a x } } \\left( \\frac { 1 } { d } X X ^ { \\top } \\right) \\mathrm { t r } \\left( W W ^ { \\top } \\right) = O ( 1 ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 400, + 220, + 1502, + 220, + 1502, + 503, + 400, + 503 + ], + "score": 0.95, + "latex": "\\begin{array} { r l r } { { \\le 2 \\lambda _ { \\operatorname* { m a x } } ( \\frac { X ^ { \\top } X } { d } ) \\operatorname { t r } ( \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 1 } \\phi ( X ^ { \\top } W ) K _ { W } \\phi ( W ^ { \\top } X ) \\Big ) } } \\\\ & { = O ( 1 ) \\cdot \\operatorname { t r } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) K _ { W } ) } \\\\ & { \\le O ( 1 ) \\cdot \\lambda _ { \\operatorname* { m a x } } ( \\phi ( W ^ { \\top } X ) \\Big ( \\phi ( X ^ { \\top } W ) \\phi ( W ^ { \\top } X ) \\Big ) ^ { - 2 } \\phi ( X ^ { \\top } W ) ) \\cdot \\operatorname { t r } ( K _ { W } ) } \\\\ & { = O ( 1 ) \\cdot \\sigma _ { \\operatorname* { m i n } } ^ { - 2 } ( \\phi ( X ^ { \\top } W ) ) \\operatorname { t r } ( K _ { W } ) = O ( 1 ) \\cdot O ( n ^ { - 1 } ) \\cdot O ( n ) = O ( 1 ) , } & { \\quad \\mathrm { ~ \\displaystyle ( 9 5 ) ~ } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 329, + 1690, + 1324, + 1690, + 1324, + 1866, + 329, + 1866 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { \\displaystyle \\frac { \\partial W _ { + } ^ { O } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { + } ^ { O \\top } \\mathbf { x } _ { i } ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\boldsymbol { \\phi } ( W _ { - } ^ { O \\top } \\mathbf { x } _ { i } ) \\Big ) \\mathbf { x } _ { i } \\boldsymbol { \\phi } ^ { \\prime } ( { \\mathbf { x } } _ { i } ^ { \\top } W _ { + } ^ { O } ) \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\left[ X ( \\boldsymbol { y } - \\boldsymbol { y } ^ { O } ( t ) ) \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\circ \\boldsymbol { \\phi } ^ { \\prime } ( X W _ { + } ^ { O } ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 942, + 1176, + 1075, + 1176, + 1075, + 1211, + 942, + 1211 + ], + "score": 0.93, + "latex": "a _ { i } = 1 / \\sqrt { h }" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1176, + 1324, + 1176, + 1324, + 1212, + 1129, + 1212 + ], + "score": 0.93, + "latex": "a _ { i + h _ { 0 } } = - 1 / \\sqrt { h }" + }, + { + "category_id": 13, + "poly": [ + 996, + 1214, + 1182, + 1214, + 1182, + 1247, + 996, + 1247 + ], + "score": 0.92, + "latex": "W = [ W _ { + } , W _ { - } ]" + }, + { + "category_id": 13, + "poly": [ + 675, + 516, + 977, + 516, + 977, + 550, + 675, + 550 + ], + "score": 0.92, + "latex": "\\operatorname { t r } \\left( A B \\right) \\leq \\lambda _ { \\operatorname* { m a x } } ( A ) \\operatorname { t r } \\left( B \\right)" + }, + { + "category_id": 13, + "poly": [ + 495, + 845, + 936, + 845, + 936, + 886, + 495, + 886 + ], + "score": 0.92, + "latex": "\\mathrm { t r } \\left( A B \\right) \\leq ( \\mathrm { t r } \\left( A ^ { \\top } A \\right) + \\mathrm { t r } \\left( B ^ { \\top } B \\right) ) / 2" + }, + { + "category_id": 13, + "poly": [ + 762, + 1881, + 1053, + 1881, + 1053, + 1918, + 762, + 1918 + ], + "score": 0.92, + "latex": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 1246, + 2002, + 1374, + 2002, + 1374, + 2038, + 1246, + 2038 + ], + "score": 0.92, + "latex": "[ W _ { + } ^ { D } , W _ { - } ^ { D } ]" + }, + { + "category_id": 14, + "poly": [ + 444, + 1262, + 1253, + 1262, + 1253, + 1337, + 444, + 1337 + ], + "score": 0.92, + "latex": "f ( \\pmb { x } ; W _ { - } , W _ { + } ) = \\pmb { a } ^ { \\top } \\phi ( W ^ { \\top } \\pmb { x } ) = \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } \\pmb { x } ) - \\frac { 1 } { \\sqrt { h } } \\pmb { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } \\pmb { x } ) ." + }, + { + "category_id": 14, + "poly": [ + 310, + 1403, + 1399, + 1403, + 1399, + 1488, + 310, + 1488 + ], + "score": 0.91, + "latex": "L ( X ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } L ( x ; W _ { + } , W _ { - } ) = \\frac { 1 } { n } \\sum _ { \\substack { x \\in X } } \\left[ y - \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { + } ^ { \\top } x ) + \\frac { 1 } { \\sqrt { h } } \\mathbf { 1 } ^ { \\top } \\phi ( W _ { - } ^ { \\top } x ) \\right] ^ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 594, + 1145, + 669, + 1145, + 669, + 1175, + 594, + 1175 + ], + "score": 0.91, + "latex": "n , d , h" + }, + { + "category_id": 13, + "poly": [ + 296, + 1213, + 424, + 1213, + 424, + 1245, + 296, + 1245 + ], + "score": 0.91, + "latex": "1 \\leq i \\leq h _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1167, + 1143, + 1280, + 1143, + 1280, + 1176, + 1167, + 1176 + ], + "score": 0.91, + "latex": "h _ { 0 } = h / 2" + }, + { + "category_id": 13, + "poly": [ + 487, + 883, + 605, + 883, + 605, + 912, + 487, + 912 + ], + "score": 0.9, + "latex": "h , n \\infty" + }, + { + "category_id": 13, + "poly": [ + 537, + 1175, + 854, + 1175, + 854, + 1211, + 537, + 1211 + ], + "score": 0.9, + "latex": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}" + }, + { + "category_id": 13, + "poly": [ + 652, + 1211, + 944, + 1211, + 944, + 1247, + 652, + 1247 + ], + "score": 0.9, + "latex": "\\pmb { a } ^ { \\top } = h ^ { - 1 / 2 } [ \\mathbf { 1 } _ { h _ { 0 } } , - \\mathbf { 1 } _ { h _ { 0 } } ] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 363, + 882, + 436, + 882, + 436, + 909, + 363, + 909 + ], + "score": 0.9, + "latex": "h > n" + }, + { + "category_id": 13, + "poly": [ + 1270, + 519, + 1326, + 519, + 1326, + 548, + 1270, + 548 + ], + "score": 0.9, + "latex": "A , B" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1966, + 1403, + 1966, + 1403, + 2002, + 1260, + 2002 + ], + "score": 0.9, + "latex": "{ \\pmb w } _ { i } ^ { D } ( 0 ) = { \\bf 0 }" + }, + { + "category_id": 13, + "poly": [ + 447, + 1915, + 498, + 1915, + 498, + 1952, + 447, + 1952 + ], + "score": 0.88, + "latex": "W _ { - } ^ { O }" + }, + { + "category_id": 13, + "poly": [ + 879, + 1143, + 988, + 1143, + 988, + 1176, + 879, + 1176 + ], + "score": 0.88, + "latex": "d _ { 0 } = d / 2" + }, + { + "category_id": 13, + "poly": [ + 1000, + 1144, + 1114, + 1144, + 1114, + 1176, + 1000, + 1176 + ], + "score": 0.87, + "latex": "n _ { 0 } = n / 2" + }, + { + "category_id": 13, + "poly": [ + 1250, + 852, + 1275, + 852, + 1275, + 877, + 1250, + 877 + ], + "score": 0.85, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 296, + 1641, + 337, + 1641, + 337, + 1669, + 296, + 1669 + ], + "score": 0.71, + "latex": "\\pmb { G F }" + }, + { + "category_id": 13, + "poly": [ + 381, + 2003, + 422, + 2003, + 422, + 2031, + 381, + 2031 + ], + "score": 0.69, + "latex": "\\pmb { G F }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1551.0, + 743.0, + 1551.0, + 743.0, + 1589.0, + 295.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1081.0, + 643.0, + 1081.0, + 643.0, + 1118.0, + 296.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2123.0, + 828.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1139.0, + 593.0, + 1139.0, + 593.0, + 1179.0, + 294.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1139.0, + 878.0, + 1139.0, + 878.0, + 1179.0, + 670.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1139.0, + 999.0, + 1139.0, + 999.0, + 1179.0, + 989.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1139.0, + 1166.0, + 1139.0, + 1166.0, + 1179.0, + 1115.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1139.0, + 1406.0, + 1139.0, + 1406.0, + 1179.0, + 1281.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1174.0, + 536.0, + 1174.0, + 536.0, + 1216.0, + 291.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1174.0, + 941.0, + 1174.0, + 941.0, + 1216.0, + 855.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1174.0, + 1128.0, + 1174.0, + 1128.0, + 1216.0, + 1076.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1174.0, + 1407.0, + 1174.0, + 1407.0, + 1216.0, + 1325.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1205.0, + 295.0, + 1205.0, + 295.0, + 1251.0, + 290.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1205.0, + 651.0, + 1205.0, + 651.0, + 1251.0, + 425.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1205.0, + 995.0, + 1205.0, + 995.0, + 1251.0, + 945.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1205.0, + 1198.0, + 1205.0, + 1198.0, + 1251.0, + 1183.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1607.0, + 1408.0, + 1607.0, + 1408.0, + 1644.0, + 294.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1640.0, + 862.0, + 1640.0, + 862.0, + 1676.0, + 338.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 927.0, + 1405.0, + 927.0, + 1405.0, + 967.0, + 292.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 963.0, + 896.0, + 963.0, + 896.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1965.0, + 1259.0, + 1965.0, + 1259.0, + 2005.0, + 294.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1965.0, + 1407.0, + 1965.0, + 1407.0, + 2005.0, + 1404.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1996.0, + 380.0, + 1996.0, + 380.0, + 2041.0, + 292.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1996.0, + 1245.0, + 1996.0, + 1245.0, + 2041.0, + 423.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1996.0, + 1408.0, + 1996.0, + 1408.0, + 2041.0, + 1375.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 761.0, + 1879.0, + 761.0, + 1921.0, + 294.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1921.0, + 1054.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1913.0, + 446.0, + 1913.0, + 446.0, + 1954.0, + 293.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1913.0, + 783.0, + 1913.0, + 783.0, + 1954.0, + 499.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 843.0, + 494.0, + 843.0, + 494.0, + 886.0, + 292.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 843.0, + 1249.0, + 843.0, + 1249.0, + 886.0, + 937.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 843.0, + 1406.0, + 843.0, + 1406.0, + 886.0, + 1276.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 878.0, + 362.0, + 878.0, + 362.0, + 914.0, + 295.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 878.0, + 486.0, + 878.0, + 486.0, + 914.0, + 437.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 878.0, + 616.0, + 878.0, + 616.0, + 914.0, + 606.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 885.0, + 1401.0, + 885.0, + 1401.0, + 905.0, + 1380.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 510.0, + 674.0, + 510.0, + 674.0, + 553.0, + 290.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 510.0, + 1269.0, + 510.0, + 1269.0, + 553.0, + 978.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 510.0, + 1409.0, + 510.0, + 1409.0, + 553.0, + 1327.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 539.0, + 359.0, + 539.0, + 359.0, + 586.0, + 290.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1351.0, + 752.0, + 1351.0, + 752.0, + 1388.0, + 297.0, + 1388.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 25, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1081, + 1406, + 1081, + 1406, + 1176, + 298, + 1176 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 586, + 1939, + 1109, + 1939, + 1109, + 2031, + 586, + 2031 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 299, + 625, + 1406, + 625, + 1406, + 691, + 299, + 691 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 295, + 226, + 1397, + 226, + 1397, + 299, + 295, + 299 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 289, + 776, + 1402, + 776, + 1402, + 841, + 289, + 841 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 481, + 991, + 1217, + 991, + 1217, + 1068, + 481, + 1068 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 480, + 855, + 1219, + 855, + 1219, + 932, + 480, + 932 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 1434, + 1403, + 1434, + 1403, + 1499, + 296, + 1499 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 433, + 1189, + 1261, + 1189, + 1261, + 1268, + 433, + 1268 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 476, + 1375, + 1220, + 1375, + 1220, + 1422, + 476, + 1422 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 293, + 1513, + 1465, + 1513, + 1465, + 1859, + 293, + 1859 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 327, + 320, + 1370, + 320, + 1370, + 408, + 327, + 408 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 321, + 517, + 1314, + 517, + 1314, + 607, + 321, + 607 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 295, + 1294, + 1405, + 1294, + 1405, + 1361, + 295, + 1361 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 946, + 589, + 946, + 589, + 978, + 297, + 978 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 297, + 1895, + 744, + 1895, + 744, + 1926, + 297, + 1926 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 295, + 461, + 1109, + 461, + 1109, + 498, + 295, + 498 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 299, + 723, + 802, + 723, + 802, + 756, + 299, + 756 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.924 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1013, + 1401, + 1013, + 1401, + 1045, + 1338, + 1045 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1337, + 877, + 1401, + 877, + 1401, + 909, + 1337, + 909 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1337, + 408, + 1401, + 408, + 1401, + 439, + 1337, + 439 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1384, + 1400, + 1384, + 1400, + 1416, + 1337, + 1416 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1336, + 1965, + 1402, + 1965, + 1402, + 1998, + 1336, + 1998 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1828, + 1402, + 1828, + 1402, + 1860, + 1337, + 1860 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1339, + 542, + 1401, + 542, + 1401, + 578, + 1339, + 578 + ], + "score": 0.865 + }, + { + "category_id": 8, + "poly": [ + 294, + 1805, + 1062, + 1805, + 1062, + 1879, + 294, + 1879 + ], + "score": 0.09 + }, + { + "category_id": 13, + "poly": [ + 574, + 624, + 719, + 624, + 719, + 662, + 574, + 662 + ], + "score": 0.94, + "latex": "{ \\pmb w } _ { \\pm } ^ { D } ( 0 ) = { \\bf 0 }" + }, + { + "category_id": 14, + "poly": [ + 326, + 317, + 1372, + 317, + 1372, + 412, + 326, + 412 + ], + "score": 0.94, + "latex": "\\frac { \\partial w _ { + } ^ { D } } { \\partial t } = g _ { + } ^ { D } ( \\boldsymbol { w } _ { + } ^ { D } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) + \\sqrt { h } \\phi ( \\boldsymbol { w } _ { - } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\boldsymbol { w } _ { + } ^ { D \\top } \\boldsymbol { x } _ { i } ) \\boldsymbol { x } _ { i } \\right] ." + }, + { + "category_id": 14, + "poly": [ + 481, + 990, + 1219, + 990, + 1219, + 1068, + 481, + 1068 + ], + "score": 0.94, + "latex": "\\pmb { w } _ { + } ^ { S } ( t ) = - \\pmb { w } _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } X \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X ^ { \\top } X t } \\right) ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } ." + }, + { + "category_id": 14, + "poly": [ + 292, + 1510, + 1391, + 1510, + 1391, + 1885, + 292, + 1885 + ], + "score": 0.94, + "latex": "\\begin{array} { r l r } { { \\| { \\boldsymbol w } _ { + } ^ { D } ( T ) - { \\boldsymbol w } _ { + } ^ { S } ( T ) } \\| _ { 2 } = \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) d t \\| _ { 2 } } \\\\ & { \\le \\| \\int _ { 0 } ^ { T } g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { S } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } + \\| \\int _ { 0 } ^ { T } g _ { + } ^ { D } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) - g _ { + } ^ { S } ( { \\boldsymbol w } _ { + } ^ { D } ( t ) ) ~ \\mathrm { d } t \\| _ { 2 } } \\\\ & { = \\| \\int _ { 0 } ^ { T } \\frac { 1 } { 2 n _ { 0 } } \\frac { 2 n _ { 0 } } { i = 1 } [ ( - \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } + \\phi ^ { \\prime } ( 0 ) ( { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) ) ^ { \\top } { \\boldsymbol x } _ { i } ) \\phi ^ { \\prime } ( 0 ) { \\boldsymbol x } _ { i } ] \\mathrm { d } t \\| _ { 2 } } & \\\\ & { = O ( 1 ) \\int _ { 0 } ^ { T } ( \\| { \\boldsymbol w } _ { + } ^ { D } ( t ) - { \\boldsymbol w } _ { + } ^ { S } ( t ) \\| _ { 2 } + \\| { \\boldsymbol w } _ { - } ^ { D } ( t ) - { \\boldsymbol w } _ { - } ^ { S } ( t ) \\| _ { 2 } ) \\mathrm { d } t + { \\boldsymbol E } _ { + } , } & { ( 1 0 6 ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 322, + 513, + 1319, + 513, + 1319, + 609, + 322, + 609 + ], + "score": 0.94, + "latex": "\\frac { \\partial w _ { + } ^ { S } } { \\partial t } = g _ { + } ^ { S } ( w _ { + } ^ { S } ) = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\frac { 1 } { \\sqrt { h } } \\Big ( y _ { i } - \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { + } ^ { S \\top } x _ { i } + \\sqrt { h } \\phi ^ { \\prime } ( 0 ) w _ { - } ^ { 1 \\top } x _ { i } \\Big ) \\phi ^ { \\prime } ( 0 ) x _ { i } \\Big ] ." + }, + { + "category_id": 14, + "poly": [ + 586, + 1936, + 1112, + 1936, + 1112, + 2030, + 586, + 2030 + ], + "score": 0.93, + "latex": "E _ { + } = \\left\\| \\int _ { 0 } ^ { T } \\pmb { g } _ { + } ^ { D } ( \\pmb { w } _ { + } ^ { D } ( s ) ) - \\pmb { g } _ { + } ^ { S } ( \\pmb { w } _ { + } ^ { D } ( s ) ) ~ \\mathrm { d } t \\right\\| _ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 371, + 262, + 449, + 262, + 449, + 302, + 371, + 302 + ], + "score": 0.93, + "latex": "{ \\pmb w } _ { \\pm } ^ { D } ( t )" + }, + { + "category_id": 14, + "poly": [ + 481, + 855, + 1217, + 855, + 1217, + 932, + 481, + 932 + ], + "score": 0.92, + "latex": "w _ { + } ^ { S } ( t ) = - w _ { - } ^ { S } ( t ) = \\frac { 1 } { 2 \\sqrt { h } } \\left( I - e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } ," + }, + { + "category_id": 13, + "poly": [ + 937, + 461, + 1095, + 461, + 1095, + 499, + 937, + 499 + ], + "score": 0.92, + "latex": "{ \\pmb w } _ { \\pm } = { \\pmb w } _ { \\pm } ^ { S } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1170, + 227, + 1400, + 227, + 1400, + 267, + 1170, + 267 + ], + "score": 0.91, + "latex": "W _ { \\pm } ^ { D } ( t ) = { \\pmb w } _ { \\pm } ^ { D } ( t ) { \\bf 1 } ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 434, + 1189, + 1265, + 1189, + 1265, + 1268, + 434, + 1268 + ], + "score": 0.91, + "latex": "\\mathrm { C o n d i t i o n \\ A : } \\ \\| w ( t ) \\| _ { 2 } = O \\left( { \\frac { 1 } { \\sqrt { d } } } \\right) ; \\left\\| X ^ { \\top } w ( t ) \\right\\| _ { \\infty } = O \\left( { \\frac { \\mathrm { p o l y } \\log d } { \\sqrt { d } } } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 365, + 1084, + 437, + 1084, + 437, + 1111, + 365, + 1111 + ], + "score": 0.9, + "latex": "d > n" + }, + { + "category_id": 13, + "poly": [ + 364, + 948, + 434, + 948, + 434, + 975, + 364, + 975 + ], + "score": 0.9, + "latex": "d < n" + }, + { + "category_id": 13, + "poly": [ + 977, + 1083, + 1048, + 1083, + 1048, + 1111, + 977, + 1111 + ], + "score": 0.89, + "latex": "d < n" + }, + { + "category_id": 14, + "poly": [ + 478, + 1375, + 1222, + 1375, + 1222, + 1422, + 478, + 1422 + ], + "score": 0.88, + "latex": "f ( \\pmb { x } ; \\pmb { w } _ { \\pm } ) = f ( \\pmb { x } ; \\pmb { w } _ { + } \\pmb { 1 } ^ { \\top } , \\pmb { w } _ { - } \\pmb { 1 } ^ { \\top } ) = \\sqrt { h } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } ) - \\sqrt { h } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } ) ." + }, + { + "category_id": 13, + "poly": [ + 1285, + 631, + 1304, + 631, + 1304, + 660, + 1285, + 660 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 451, + 1469, + 494, + 1469, + 494, + 1498, + 451, + 1498 + ], + "score": 0.84, + "latex": "{ \\pmb w } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1467, + 1350, + 1467, + 1350, + 1493, + 1327, + 1493 + ], + "score": 0.81, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 740, + 1146, + 754, + 1146, + 754, + 1171, + 740, + 1171 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 533, + 464, + 574, + 464, + 574, + 492, + 533, + 492 + ], + "score": 0.74, + "latex": "\\pmb { G F }" + }, + { + "category_id": 13, + "poly": [ + 508, + 1469, + 553, + 1469, + 553, + 1498, + 508, + 1498 + ], + "score": 0.69, + "latex": "\\mathbf { \\nabla } w _ { - }" + }, + { + "category_id": 13, + "poly": [ + 488, + 812, + 500, + 812, + 500, + 836, + 488, + 836 + ], + "score": 0.67, + "latex": "t" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 720.0, + 805.0, + 720.0, + 805.0, + 760.0, + 295.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2125.0, + 830.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1081.0, + 364.0, + 1081.0, + 364.0, + 1115.0, + 295.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1081.0, + 976.0, + 1081.0, + 976.0, + 1115.0, + 438.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 1081.0, + 1404.0, + 1081.0, + 1404.0, + 1115.0, + 1049.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1114.0, + 1403.0, + 1114.0, + 1403.0, + 1144.0, + 296.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1144.0, + 739.0, + 1144.0, + 739.0, + 1175.0, + 296.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1144.0, + 917.0, + 1144.0, + 917.0, + 1175.0, + 755.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 624.0, + 573.0, + 624.0, + 573.0, + 664.0, + 294.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 624.0, + 1284.0, + 624.0, + 1284.0, + 664.0, + 720.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 624.0, + 1406.0, + 624.0, + 1406.0, + 664.0, + 1305.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 657.0, + 758.0, + 657.0, + 758.0, + 692.0, + 296.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 226.0, + 1169.0, + 226.0, + 1169.0, + 266.0, + 290.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 370.0, + 261.0, + 370.0, + 303.0, + 294.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 261.0, + 810.0, + 261.0, + 810.0, + 303.0, + 450.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 778.0, + 1404.0, + 778.0, + 1404.0, + 814.0, + 296.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 810.0, + 487.0, + 810.0, + 487.0, + 842.0, + 296.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 810.0, + 992.0, + 810.0, + 992.0, + 842.0, + 501.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1432.0, + 1405.0, + 1432.0, + 1405.0, + 1472.0, + 294.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1465.0, + 450.0, + 1465.0, + 450.0, + 1501.0, + 295.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 1465.0, + 507.0, + 1465.0, + 507.0, + 1501.0, + 495.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1465.0, + 1326.0, + 1465.0, + 1326.0, + 1501.0, + 554.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1296.0, + 1406.0, + 1296.0, + 1406.0, + 1333.0, + 295.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1325.0, + 1253.0, + 1325.0, + 1253.0, + 1363.0, + 294.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 942.0, + 363.0, + 942.0, + 363.0, + 981.0, + 294.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 942.0, + 590.0, + 942.0, + 590.0, + 981.0, + 435.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1893.0, + 747.0, + 1893.0, + 747.0, + 1930.0, + 294.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 458.0, + 532.0, + 458.0, + 532.0, + 502.0, + 293.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 458.0, + 936.0, + 458.0, + 936.0, + 502.0, + 575.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 458.0, + 1114.0, + 458.0, + 1114.0, + 502.0, + 1096.0, + 502.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 26, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1140, + 1406, + 1140, + 1406, + 1328, + 296, + 1328 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 295, + 1357, + 1402, + 1357, + 1402, + 1424, + 295, + 1424 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 294, + 228, + 1403, + 228, + 1403, + 293, + 294, + 293 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 470, + 1057, + 1228, + 1057, + 1228, + 1129, + 470, + 1129 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 981, + 1409, + 981, + 1409, + 1047, + 299, + 1047 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 296, + 1972, + 1403, + 1972, + 1403, + 2036, + 296, + 2036 + ], + "score": 0.935 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 817, + 74, + 817, + 105, + 298, + 105 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1909, + 1400, + 1909, + 1400, + 1940, + 1338, + 1940 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1077, + 1400, + 1077, + 1400, + 1108, + 1338, + 1108 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1338, + 938, + 1401, + 938, + 1401, + 968, + 1338, + 968 + ], + "score": 0.866 + }, + { + "category_id": 8, + "poly": [ + 410, + 1436, + 1301, + 1436, + 1301, + 1966, + 410, + 1966 + ], + "score": 0.82 + }, + { + "category_id": 8, + "poly": [ + 299, + 301, + 1488, + 301, + 1488, + 946, + 299, + 946 + ], + "score": 0.751 + }, + { + "category_id": 8, + "poly": [ + 304, + 590, + 970, + 590, + 970, + 776, + 304, + 776 + ], + "score": 0.565 + }, + { + "category_id": 8, + "poly": [ + 306, + 302, + 1490, + 302, + 1490, + 586, + 306, + 586 + ], + "score": 0.489 + }, + { + "category_id": 8, + "poly": [ + 419, + 1434, + 1301, + 1434, + 1301, + 1680, + 419, + 1680 + ], + "score": 0.214 + }, + { + "category_id": 8, + "poly": [ + 299, + 782, + 1420, + 782, + 1420, + 945, + 299, + 945 + ], + "score": 0.164 + }, + { + "category_id": 14, + "poly": [ + 310, + 306, + 1428, + 306, + 1428, + 950, + 310, + 950 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { E _ { + } = \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { \\Phi } { u } \\int _ { 0 } ^ { u } [ u , \\frac { \\Phi } { u } ] ( \\boldsymbol { \\cdot } } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } \\\\ & { \\leq \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ \\frac { 1 } { \\sqrt { \\delta } } ( \\boldsymbol { \\cdot } - \\sqrt { \\delta } \\dot { u } ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } ) - \\boldsymbol { \\cdot } u _ { + } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } ( \\boldsymbol { \\cdot } , \\boldsymbol { \\cdot } ) ) ( \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } u ( \\boldsymbol { \\cdot } ^ { \\Phi } , \\boldsymbol { \\cdot } ) u _ { - } ^ { \\Phi } \\rangle ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] \\Bigg \\| _ { 0 } ^ { 2 } , } \\\\ & \\quad + \\Bigg \\| \\int _ { 0 } ^ { T } \\frac { 1 } { u _ { + } ^ { T } } \\sum _ { \\mathrm { i } = 1 } ^ { N } \\Bigg [ ( - \\mathcal { \\cdot } \\langle u ( u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) - \\mathcal { \\cdot } ( u _ { + } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } u _ { - } ^ { \\Phi } ) ) \\boldsymbol { \\cdot } u _ { - } ^ { \\Phi } \\Bigg ] | \\int _ { 0 } ^ { T } \\ \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 472, + 1054, + 1227, + 1054, + 1227, + 1132, + 472, + 1132 + ], + "score": 0.94, + "latex": "\\left\\| w _ { + } ^ { D } ( T ) - w _ { + } ^ { S } ( T ) \\right\\| _ { 2 } \\leq C _ { 1 } \\cdot \\frac { \\log ^ { c } h } { h } e ^ { C _ { 2 } T } = O \\left( \\frac { \\mathrm { p o l y l o g } h } { h } \\right) \\to 0 ." + }, + { + "category_id": 13, + "poly": [ + 580, + 1201, + 829, + 1201, + 829, + 1237, + 580, + 1237 + ], + "score": 0.93, + "latex": "{ \\pmb w } ^ { S } ( 0 ) = { \\pmb w } ^ { D } ( 0 ) = 0" + }, + { + "category_id": 13, + "poly": [ + 883, + 1388, + 929, + 1388, + 929, + 1425, + 883, + 1425 + ], + "score": 0.92, + "latex": "\\pmb { w } _ { \\pm } ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 363, + 262, + 483, + 262, + 483, + 293, + 363, + 293 + ], + "score": 0.91, + "latex": "0 \\leq t \\leq T" + }, + { + "category_id": 13, + "poly": [ + 340, + 1142, + 547, + 1142, + 547, + 1176, + 340, + 1176 + ], + "score": 0.91, + "latex": "T \\in O ( \\log \\log h )" + }, + { + "category_id": 13, + "poly": [ + 786, + 1388, + 832, + 1388, + 832, + 1425, + 786, + 1425 + ], + "score": 0.91, + "latex": "\\pmb { w } _ { \\pm } ^ { S }" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1204, + 1398, + 1204, + 1398, + 1232, + 1322, + 1232 + ], + "score": 0.9, + "latex": "T > 0" + }, + { + "category_id": 13, + "poly": [ + 389, + 1388, + 432, + 1388, + 432, + 1417, + 389, + 1417 + ], + "score": 0.9, + "latex": "R ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 753, + 232, + 819, + 232, + 819, + 259, + 753, + 259 + ], + "score": 0.9, + "latex": "t = 0" + }, + { + "category_id": 13, + "poly": [ + 1290, + 1171, + 1335, + 1171, + 1335, + 1200, + 1290, + 1200 + ], + "score": 0.89, + "latex": "\\mathbf { \\Delta } w ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 1198, + 1201, + 1240, + 1201, + 1240, + 1232, + 1198, + 1232 + ], + "score": 0.89, + "latex": "\\pmb { w } ^ { S }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1388, + 336, + 1388, + 336, + 1417, + 298, + 1417 + ], + "score": 0.89, + "latex": "R ^ { S }" + }, + { + "category_id": 13, + "poly": [ + 1289, + 1262, + 1334, + 1262, + 1334, + 1292, + 1289, + 1292 + ], + "score": 0.88, + "latex": "\\mathbf { \\Delta } w ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 922, + 986, + 942, + 986, + 942, + 1015, + 922, + 1015 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 14, + "poly": [ + 399, + 1433, + 1293, + 1433, + 1293, + 1969, + 399, + 1969 + ], + "score": 0.8, + "latex": "\\begin{array} { r l } & { \\quad | ( { \\mathcal R } ^ { 3 } - { \\mathcal R } ^ { D } ) | = | \\mathbb { E } _ { { \\mathbf z } } ( x ^ { \\top } ) - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) | ^ { 2 } - \\mathbb { E } _ { { \\mathbf z } _ { \\mathbf z } } ( x ^ { \\top } \\beta - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { D } ) ) ^ { 2 } | } \\\\ & { \\stackrel { ( i ) } { \\le } \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ f ( { \\mathbf x } ; { \\mathbf x } _ { \\mathbf z } ^ { \\top } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } \\mathbb { E } _ { { \\mathbf z } } [ | { \\mathcal R } ^ { 7 } \\cdot f ( { \\mathbf x } ) - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) + \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) ] ^ { 2 } } } \\\\ & \\le \\sqrt { \\mathbb { H } _ { { \\mathbf z } } ^ { 5 } [ | { \\mathcal R } | _ { { \\mathbf z } } [ \\langle \\delta | ^ { \\mathcal { R } } \\rangle - \\phi ; \\langle { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ] ^ { 2 } \\mathbb { E } _ { | { \\mathbf z } } ] + | ( { \\mathcal R } \\langle \\mathbf x ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\rangle ) [ ^ { 2 } } \\\\ & { \\quad \\cdot \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } ^ { 7 } - f ( x ; { \\mathcal R } _ { \\mathbf z } ^ { \\theta } ) ] + | \\beta ^ { \\top } x - f ( x ; { \\mathbf x } _ { \\mathbf z } ^ { D } ) | ] ^ { 2 } } } \\\\ & \\stackrel { ( i i ) } { \\le } 2 \\sqrt { \\mathbb { H } _ { { \\mathbf z } } [ | { \\mathcal R } | _ { { \\mathbf z } } \\mathbb { R } ^ { 5 } ] - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } ] ^ { 2 } + | \\beta \\langle \\mathbf w _ { \\mathbf z } ^ { \\top } \\mathbf x \\rangle - \\phi ( { \\mathcal R } _ { \\mathbf z } ^ { D } \\mathbb { I } _ { { \\mathbf z } } \\rangle | ^ { 2 } } \\\\ & \\quad \\cdot \\sqrt \\mathbb { E } _ { \\mathbf z } [ | { \\mathcal R } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 889, + 1143, + 912, + 1143, + 912, + 1171, + 889, + 1171 + ], + "score": 0.26, + "latex": "\\mathrm { T } ," + }, + { + "category_id": 14, + "poly": [ + 441, + 1434, + 1274, + 1434, + 1274, + 1494, + 441, + 1494 + ], + "score": 0.25, + "latex": "\\left| 2 ( R ^ { S } - R ^ { D } ) \\right| = \\left| \\mathbb { E } _ { \\pmb { x } } \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - f ( \\pmb { x } ; \\pmb { w } _ { \\pmb { \\pm \\pm } } ^ { S } ) \\right) ^ { 2 } - \\mathbb { E } _ { \\pmb { x } } \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - f ( \\pmb { x } ; \\pmb { w } _ { \\pmb { \\pm } } ^ { D } ) \\right) ^ { 2 } \\right|" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1139.0, + 339.0, + 1139.0, + 339.0, + 1175.0, + 294.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1139.0, + 888.0, + 1139.0, + 888.0, + 1175.0, + 548.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1139.0, + 1406.0, + 1139.0, + 1406.0, + 1175.0, + 913.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1167.0, + 1289.0, + 1167.0, + 1289.0, + 1208.0, + 291.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1167.0, + 1407.0, + 1167.0, + 1407.0, + 1208.0, + 1336.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1199.0, + 579.0, + 1199.0, + 579.0, + 1238.0, + 290.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1199.0, + 1197.0, + 1199.0, + 1197.0, + 1238.0, + 830.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1199.0, + 1321.0, + 1199.0, + 1321.0, + 1238.0, + 1241.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1199.0, + 1411.0, + 1199.0, + 1411.0, + 1238.0, + 1399.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1230.0, + 1407.0, + 1230.0, + 1407.0, + 1269.0, + 293.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1261.0, + 1288.0, + 1261.0, + 1288.0, + 1297.0, + 293.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1261.0, + 1406.0, + 1261.0, + 1406.0, + 1297.0, + 1335.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1292.0, + 1356.0, + 1292.0, + 1356.0, + 1329.0, + 295.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1355.0, + 1405.0, + 1355.0, + 1405.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1381.0, + 297.0, + 1381.0, + 297.0, + 1431.0, + 292.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1381.0, + 388.0, + 1381.0, + 388.0, + 1431.0, + 337.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1381.0, + 785.0, + 1381.0, + 785.0, + 1431.0, + 433.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1381.0, + 882.0, + 1381.0, + 882.0, + 1431.0, + 833.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1381.0, + 942.0, + 1381.0, + 942.0, + 1431.0, + 930.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 752.0, + 228.0, + 752.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 264.0, + 820.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 362.0, + 261.0, + 362.0, + 295.0, + 293.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 261.0, + 1029.0, + 261.0, + 1029.0, + 295.0, + 484.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 978.0, + 921.0, + 978.0, + 921.0, + 1021.0, + 293.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 978.0, + 1405.0, + 978.0, + 1405.0, + 1021.0, + 943.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1014.0, + 770.0, + 1014.0, + 770.0, + 1050.0, + 296.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2005.0, + 298.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2003.0, + 1405.0, + 2003.0, + 1405.0, + 2039.0, + 296.0, + 2039.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 27, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 889, + 1407, + 889, + 1407, + 1046, + 297, + 1046 + ], + "score": 0.976 + }, + { + "category_id": 8, + "poly": [ + 340, + 1057, + 1354, + 1057, + 1354, + 1446, + 340, + 1446 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 363, + 1593, + 1335, + 1593, + 1335, + 1736, + 363, + 1736 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 480, + 641, + 1215, + 641, + 1215, + 765, + 480, + 765 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 299, + 400, + 1402, + 400, + 1402, + 560, + 299, + 560 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 593, + 1789, + 1103, + 1789, + 1103, + 1864, + 593, + 1864 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 292, + 227, + 1403, + 227, + 1403, + 296, + 292, + 296 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 301, + 772, + 1105, + 772, + 1105, + 806, + 301, + 806 + ], + "score": 0.938 + }, + { + "category_id": 8, + "poly": [ + 450, + 1986, + 1248, + 1986, + 1248, + 2030, + 450, + 2030 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 298, + 597, + 1179, + 597, + 1179, + 632, + 298, + 632 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 300, + 1455, + 809, + 1455, + 809, + 1487, + 300, + 1487 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 885, + 1942, + 885, + 1978, + 299, + 1978 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 817, + 74, + 817, + 106, + 298, + 106 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 302, + 837, + 831, + 837, + 831, + 871, + 302, + 871 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 295, + 1745, + 1092, + 1745, + 1092, + 1779, + 295, + 1779 + ], + "score": 0.915 + }, + { + "category_id": 8, + "poly": [ + 441, + 1497, + 1254, + 1497, + 1254, + 1546, + 441, + 1546 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 298, + 1553, + 1029, + 1553, + 1029, + 1585, + 298, + 1585 + ], + "score": 0.912 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1810, + 1401, + 1810, + 1401, + 1842, + 1337, + 1842 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1993, + 1402, + 1993, + 1402, + 2025, + 1337, + 2025 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1698, + 1401, + 1698, + 1401, + 1730, + 1338, + 1730 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 298, + 1888, + 734, + 1888, + 734, + 1922, + 298, + 1922 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1337, + 709, + 1401, + 709, + 1401, + 741, + 1337, + 741 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1505, + 1401, + 1505, + 1401, + 1537, + 1337, + 1537 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1392, + 1401, + 1392, + 1401, + 1424, + 1337, + 1424 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1337, + 552, + 1401, + 552, + 1401, + 584, + 1337, + 584 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2113, + 834, + 2113 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 296, + 324, + 1400, + 324, + 1400, + 390, + 296, + 390 + ], + "score": 0.864 + }, + { + "category_id": 14, + "poly": [ + 342, + 1055, + 1358, + 1055, + 1358, + 1448, + 342, + 1448 + ], + "score": 0.96, + "latex": "\\begin{array} { l } { \\displaystyle \\left\\| W ^ { O } ( T ) - W ^ { D } ( T ) \\right\\| _ { F } ^ { 2 } } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) - W ^ { D } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\frac { \\partial \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } } { \\partial t } \\mathrm { d } t } \\\\ { \\displaystyle = \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) ) \\right\\| _ { F } \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } + \\left\\| \\frac { \\partial L \\left( X ; W ^ { O } ( t ) \\right) } { \\partial W } - \\frac { \\partial L \\left( X ; W ^ { D } ( t ) \\right) } { \\partial W } \\right\\| _ { F } ^ { 2 } \\mathrm { d } t } \\\\ { \\displaystyle \\leq \\left\\| W ^ { O } ( 0 ) \\right\\| _ { F } ^ { 2 } + \\left( 1 + L _ { f } \\right) \\int _ { 0 } ^ { T } \\left\\| W ^ { O } ( t ) - W ^ { D } ( t ) \\right\\| _ { F } ^ { 2 } \\mathrm { d } t , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 594, + 1788, + 1106, + 1788, + 1106, + 1865, + 594, + 1865 + ], + "score": 0.94, + "latex": "| R ^ { O } ( T ) - R ^ { D } ( T ) | = O \\left( \\frac { \\mathrm { p o l y l o g } h } { d ^ { \\epsilon / 2 } } \\right) \\to 0 ." + }, + { + "category_id": 14, + "poly": [ + 360, + 1592, + 1336, + 1592, + 1336, + 1740, + 360, + 1740 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\quad \\left\\| f ( \\pmb { x } , W ^ { D } ( T ) ) - f ( \\pmb { x } , W ^ { O } ( T ) ) \\right\\| _ { 2 } = \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) \\pmb { a } - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\pmb { a } \\right\\| _ { 2 } } \\\\ & { \\leq \\left\\| \\phi ( \\pmb { x } ^ { \\top } W ^ { D } ( T ) ) - \\phi ( \\pmb { x } ^ { \\top } W ^ { O } ( T ) ) \\right\\| _ { 2 } \\left\\| \\pmb { a } \\right\\| _ { 2 } \\leq L _ { \\phi } \\left\\| \\pmb { x } \\right\\| _ { 2 } \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } \\left\\| \\pmb { a } \\right\\| _ { 2 } } \\\\ & { = O ( \\sqrt { d } ) \\cdot \\left\\| W ^ { D } ( T ) - W ^ { O } ( T ) \\right\\| _ { F } = O ( d ^ { - \\epsilon / 2 } ) e ^ { C T } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1264, + 921, + 1398, + 921, + 1398, + 957, + 1264, + 957 + ], + "score": 0.93, + "latex": "{ \\pmb w } _ { i } ^ { D } ( 0 ) \\dot { = } { \\bf 0 }" + }, + { + "category_id": 14, + "poly": [ + 480, + 639, + 1218, + 639, + 1218, + 769, + 480, + 769 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { S } ( T ) - R ^ { D } ( T ) | + | R ^ { S } ( T ) - R ^ { S } ( \\infty ) | } \\\\ { = O ( \\frac { \\mathrm { p o l y l o g } h } { \\sqrt { h } } ) + O ( \\frac { 1 } { \\mathrm { p o l y l o g } h } ) 0 . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 380, + 1746, + 536, + 1746, + 536, + 1778, + 380, + 1778 + ], + "score": 0.92, + "latex": "T = \\log \\log h" + }, + { + "category_id": 13, + "poly": [ + 796, + 261, + 993, + 261, + 993, + 295, + 796, + 295 + ], + "score": 0.92, + "latex": "T = O ( \\log \\log h )" + }, + { + "category_id": 14, + "poly": [ + 310, + 398, + 1404, + 398, + 1404, + 560, + 310, + 560 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\left| R ^ { S } ( t ) - R ^ { S } ( \\infty ) \\right| \\leq C \\sqrt { h } \\cdot \\left\\| w _ { + } ^ { S } ( t ) - w _ { + } ^ { S } ( \\infty ) \\right\\| = C \\sqrt { h } \\left\\| \\frac { 1 } { 2 \\sqrt { h } } e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X y \\right\\| _ { 2 } } \\\\ & { = C \\left\\| e ^ { - \\frac { \\phi ^ { \\prime } ( 0 ) ^ { 2 } } { n _ { 0 } } X X ^ { \\top } t } ( X X ^ { \\top } ) ^ { - 1 } X X ^ { \\top } \\beta \\right\\| _ { 2 } \\leq C \\exp \\left( - \\phi ^ { \\prime } ( 0 ) ^ { 2 } \\left\\| \\frac { 1 } { n } X X ^ { \\top } \\right\\| _ { 2 } t \\right) \\| \\beta \\| _ { 2 } = C _ { 3 } e ^ { - C _ { 4 } t } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 447, + 600, + 577, + 600, + 577, + 631, + 447, + 631 + ], + "score": 0.92, + "latex": "C _ { 3 } , C _ { 4 } > 0" + }, + { + "category_id": 13, + "poly": [ + 698, + 231, + 784, + 231, + 784, + 263, + 698, + 263 + ], + "score": 0.91, + "latex": "\\gamma _ { 1 } \\neq 1" + }, + { + "category_id": 13, + "poly": [ + 719, + 921, + 1010, + 921, + 1010, + 956, + 719, + 956 + ], + "score": 0.91, + "latex": "{ \\pmb w } _ { i } ^ { O } ( 0 ) \\sim \\mathcal { N } ( { \\bf 0 } , I / d h ^ { 1 + \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 606, + 259, + 649, + 259, + 649, + 289, + 606, + 289 + ], + "score": 0.91, + "latex": "R ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 719, + 1945, + 875, + 1945, + 875, + 1977, + 719, + 1977 + ], + "score": 0.9, + "latex": "T = \\log \\log h" + }, + { + "category_id": 13, + "poly": [ + 1015, + 600, + 1170, + 600, + 1170, + 632, + 1015, + 632 + ], + "score": 0.9, + "latex": "T = \\log \\log h" + }, + { + "category_id": 13, + "poly": [ + 604, + 775, + 675, + 775, + 675, + 802, + 604, + 802 + ], + "score": 0.9, + "latex": "d > n" + }, + { + "category_id": 14, + "poly": [ + 441, + 1497, + 1258, + 1497, + 1258, + 1546, + 441, + 1546 + ], + "score": 0.9, + "latex": "\\left\\| { W ^ { O } ( T ) - W ^ { D } ( T ) } \\right\\| _ { F } \\leq \\left\\| { W ^ { O } ( 0 ) } \\right\\| _ { F } e ^ { ( 1 + L _ { f } ) T / 2 } = O ( d ^ { - ( 1 + \\epsilon ) / 2 } ) e ^ { C T } ." + }, + { + "category_id": 13, + "poly": [ + 406, + 228, + 446, + 228, + 446, + 258, + 406, + 258 + ], + "score": 0.89, + "latex": "R ^ { S }" + }, + { + "category_id": 13, + "poly": [ + 500, + 228, + 543, + 228, + 543, + 258, + 500, + 258 + ], + "score": 0.89, + "latex": "R ^ { D }" + }, + { + "category_id": 14, + "poly": [ + 453, + 1986, + 1244, + 1986, + 1244, + 2029, + 453, + 2029 + ], + "score": 0.88, + "latex": "| R ^ { O } ( T ) - R ^ { S } ( \\infty ) | \\leq | R ^ { O } ( T ) - R ^ { D } ( T ) | + | R ^ { D } ( T ) - R ^ { S } ( \\infty ) | \\to 0 ." + }, + { + "category_id": 13, + "poly": [ + 587, + 954, + 863, + 954, + 863, + 987, + 587, + 987 + ], + "score": 0.88, + "latex": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } = { \\cal O } ( { \\sqrt { n } } )" + }, + { + "category_id": 13, + "poly": [ + 953, + 359, + 1030, + 359, + 1030, + 384, + 953, + 384 + ], + "score": 0.88, + "latex": "t = \\infty" + }, + { + "category_id": 13, + "poly": [ + 515, + 259, + 555, + 259, + 555, + 289, + 515, + 289 + ], + "score": 0.88, + "latex": "R ^ { S }" + }, + { + "category_id": 13, + "poly": [ + 626, + 360, + 640, + 360, + 640, + 384, + 626, + 384 + ], + "score": 0.77, + "latex": "t" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 836.0, + 835.0, + 836.0, + 835.0, + 874.0, + 295.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1890.0, + 737.0, + 1890.0, + 737.0, + 1923.0, + 297.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2084.0, + 872.0, + 2084.0, + 872.0, + 2125.0, + 828.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 890.0, + 1407.0, + 890.0, + 1407.0, + 927.0, + 294.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 915.0, + 718.0, + 915.0, + 718.0, + 959.0, + 292.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 915.0, + 1263.0, + 915.0, + 1263.0, + 959.0, + 1011.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 915.0, + 1410.0, + 915.0, + 1410.0, + 959.0, + 1399.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 951.0, + 586.0, + 951.0, + 586.0, + 989.0, + 294.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 951.0, + 1407.0, + 951.0, + 1407.0, + 989.0, + 864.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 981.0, + 1405.0, + 981.0, + 1405.0, + 1017.0, + 294.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1013.0, + 494.0, + 1013.0, + 494.0, + 1048.0, + 292.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 223.0, + 405.0, + 223.0, + 405.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 223.0, + 499.0, + 223.0, + 499.0, + 267.0, + 447.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 223.0, + 697.0, + 223.0, + 697.0, + 267.0, + 544.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 223.0, + 1406.0, + 223.0, + 1406.0, + 267.0, + 785.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 251.0, + 514.0, + 251.0, + 514.0, + 298.0, + 291.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 251.0, + 605.0, + 251.0, + 605.0, + 298.0, + 556.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 251.0, + 795.0, + 251.0, + 795.0, + 298.0, + 650.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 251.0, + 1008.0, + 251.0, + 1008.0, + 298.0, + 994.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 770.0, + 603.0, + 770.0, + 603.0, + 809.0, + 296.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 770.0, + 1108.0, + 770.0, + 1108.0, + 809.0, + 676.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 593.0, + 446.0, + 593.0, + 446.0, + 638.0, + 292.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 593.0, + 1014.0, + 593.0, + 1014.0, + 638.0, + 578.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 593.0, + 1182.0, + 593.0, + 1182.0, + 638.0, + 1171.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1455.0, + 810.0, + 1455.0, + 810.0, + 1491.0, + 296.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 718.0, + 1941.0, + 718.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1941.0, + 886.0, + 1941.0, + 886.0, + 1981.0, + 876.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1742.0, + 379.0, + 1742.0, + 379.0, + 1782.0, + 295.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1742.0, + 1095.0, + 1742.0, + 1095.0, + 1782.0, + 537.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1549.0, + 1028.0, + 1549.0, + 1028.0, + 1588.0, + 295.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 322.0, + 1406.0, + 322.0, + 1406.0, + 364.0, + 292.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 353.0, + 625.0, + 353.0, + 625.0, + 393.0, + 294.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 353.0, + 952.0, + 353.0, + 952.0, + 393.0, + 641.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 353.0, + 1042.0, + 353.0, + 1042.0, + 393.0, + 1031.0, + 393.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 28, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1333, + 1406, + 1333, + 1406, + 1527, + 296, + 1527 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 226, + 1406, + 226, + 1406, + 360, + 296, + 360 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 605, + 1406, + 605, + 1406, + 699, + 297, + 699 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 296, + 1024, + 1407, + 1024, + 1407, + 1151, + 296, + 1151 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 296, + 1593, + 1407, + 1593, + 1407, + 1713, + 296, + 1713 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 324, + 1803, + 1314, + 1803, + 1314, + 1954, + 324, + 1954 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 297, + 1163, + 1400, + 1163, + 1400, + 1257, + 297, + 1257 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 633, + 515, + 1066, + 515, + 1066, + 594, + 633, + 594 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 634, + 832, + 1065, + 832, + 1065, + 896, + 634, + 896 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 296, + 1966, + 1406, + 1966, + 1406, + 2038, + 296, + 2038 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 298, + 1727, + 1403, + 1727, + 1403, + 1792, + 298, + 1792 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 587, + 1539, + 1109, + 1539, + 1109, + 1583, + 587, + 1583 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 298, + 905, + 728, + 905, + 728, + 937, + 298, + 937 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 296, + 781, + 886, + 781, + 886, + 818, + 296, + 818 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 297, + 463, + 1088, + 463, + 1088, + 502, + 297, + 502 + ], + "score": 0.929 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 817, + 74, + 817, + 105, + 298, + 105 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 298, + 408, + 641, + 408, + 641, + 441, + 298, + 441 + ], + "score": 0.912 + }, + { + "category_id": 9, + "poly": [ + 1338, + 848, + 1400, + 848, + 1400, + 880, + 1338, + 880 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1338, + 968, + 1400, + 968, + 1400, + 1000, + 1338, + 1000 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 298, + 729, + 749, + 729, + 749, + 763, + 298, + 763 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1338, + 539, + 1400, + 539, + 1400, + 572, + 1338, + 572 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1544, + 1400, + 1544, + 1400, + 1575, + 1338, + 1575 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1907, + 1400, + 1907, + 1400, + 1939, + 1338, + 1939 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 866, + 2087, + 866, + 2113, + 835, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 8, + "poly": [ + 614, + 952, + 1084, + 952, + 1084, + 1015, + 614, + 1015 + ], + "score": 0.831 + }, + { + "category_id": 8, + "poly": [ + 648, + 1269, + 1049, + 1269, + 1049, + 1308, + 648, + 1308 + ], + "score": 0.783 + }, + { + "category_id": 8, + "poly": [ + 612, + 952, + 1086, + 952, + 1086, + 1015, + 612, + 1015 + ], + "score": 0.304 + }, + { + "category_id": 1, + "poly": [ + 648, + 1269, + 1049, + 1269, + 1049, + 1308, + 648, + 1308 + ], + "score": 0.103 + }, + { + "category_id": 13, + "poly": [ + 1164, + 1627, + 1403, + 1627, + 1403, + 1685, + 1164, + 1685 + ], + "score": 0.95, + "latex": "\\left\\| \\frac { \\partial { \\cal L } ( X ; W ( T ) ) } { \\partial W ( T ) } \\right\\| _ { 2 } \\to 0" + }, + { + "category_id": 14, + "poly": [ + 632, + 828, + 1065, + 828, + 1065, + 896, + 632, + 896 + ], + "score": 0.94, + "latex": "\\mathrm { d } { \\pmb y } _ { N N } ( t ) = \\frac { 1 } { n } K ( t ) ( { \\pmb y } - { \\pmb y } _ { N N } ( t ) ) \\ \\mathrm { d } t ," + }, + { + "category_id": 14, + "poly": [ + 614, + 949, + 1083, + 949, + 1083, + 1016, + 614, + 1016 + ], + "score": 0.94, + "latex": "\\mathrm { d } { \\pmb y } _ { N T K } ( t ) = \\frac { 1 } { n } K ( 0 ) ( { \\pmb y } - { \\pmb y } _ { N T K } ( t ) ) \\ \\mathrm { d } t ," + }, + { + "category_id": 14, + "poly": [ + 632, + 512, + 1064, + 512, + 1064, + 595, + 632, + 595 + ], + "score": 0.94, + "latex": "K ( t ) = \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ," + }, + { + "category_id": 13, + "poly": [ + 346, + 289, + 645, + 289, + 645, + 329, + 346, + 329 + ], + "score": 0.94, + "latex": "\\left\\| \\mathbf { \\dot { W } } ^ { O } ( t ) - W ^ { S } ( t ) \\right\\| _ { F } \\to 0" + }, + { + "category_id": 14, + "poly": [ + 323, + 1802, + 1313, + 1802, + 1313, + 1958, + 323, + 1958 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\left\\| \\frac { \\partial L ( X ; \\boldsymbol { w } _ { i } ( t ) ) } { \\partial \\boldsymbol { w } _ { i } ( t ) } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { n } X \\left[ \\frac { 1 } { \\sqrt { h } } ( \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) ) \\circ \\phi ^ { \\prime } ( X ^ { \\top } \\boldsymbol { w } _ { i } ( t ) ) \\right] \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i ) } { \\leq } O ( 1 ) \\frac { 1 } { d ^ { 1 . 5 } } \\left\\| X \\right\\| _ { 2 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( t ) \\right\\| _ { 2 } \\leq O ( 1 ) d ^ { - 1 } \\left\\| \\boldsymbol { y } - \\boldsymbol { y } _ { N N } ( 0 ) \\right\\| _ { 2 } e ^ { - t } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 824, + 228, + 914, + 228, + 914, + 264, + 824, + 264 + ], + "score": 0.93, + "latex": "R ^ { S } ( \\infty )" + }, + { + "category_id": 13, + "poly": [ + 367, + 783, + 686, + 783, + 686, + 818, + 367, + 818 + ], + "score": 0.93, + "latex": "\\pmb { y } _ { N N } ( t ) = f _ { N N } ( \\boldsymbol { X } , t ) \\in \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 463, + 2001, + 760, + 2001, + 760, + 2038, + 463, + 2038 + ], + "score": 0.92, + "latex": "\\| W ( t ) - W ( 0 ) \\| _ { 2 } = O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 369, + 1594, + 681, + 1594, + 681, + 1631, + 369, + 1631 + ], + "score": 0.92, + "latex": "\\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } = O ( \\sqrt { n } )" + }, + { + "category_id": 13, + "poly": [ + 956, + 1594, + 1118, + 1594, + 1118, + 1629, + 956, + 1629 + ], + "score": 0.92, + "latex": "T = O ( \\log d )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1460, + 409, + 1460, + 409, + 1497, + 298, + 1497 + ], + "score": 0.92, + "latex": "O ( d ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 931, + 1726, + 1042, + 1726, + 1042, + 1763, + 931, + 1763 + ], + "score": 0.92, + "latex": "O ( d ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 385, + 467, + 747, + 467, + 747, + 501, + 385, + 501 + ], + "score": 0.92, + "latex": "\\omega = \\mathrm { v e c } ( W ) = \\mathrm { v e c } ( [ W _ { + } , W _ { - } ] )" + }, + { + "category_id": 13, + "poly": [ + 1055, + 289, + 1397, + 289, + 1397, + 329, + 1055, + 329 + ], + "score": 0.92, + "latex": "\\left\\| \\partial L ( X ; W ^ { \\mathcal { O } } ( t ) ) / \\partial W \\right\\| _ { F } \\to 0" + }, + { + "category_id": 13, + "poly": [ + 339, + 1087, + 485, + 1087, + 485, + 1119, + 339, + 1119 + ], + "score": 0.92, + "latex": "h = \\mathrm { p o l y } ( n )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1999, + 409, + 1999, + 409, + 2037, + 298, + 2037 + ], + "score": 0.92, + "latex": "O ( d ^ { - 1 / 2 } )" + }, + { + "category_id": 14, + "poly": [ + 586, + 1537, + 1113, + 1537, + 1113, + 1581, + 586, + 1581 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\| { \\pmb y } _ { N N } ( t ) - { \\pmb y } \\| _ { 2 } \\le C _ { 1 } \\| { \\pmb y } _ { N N } ( 0 ) - { \\pmb y } \\| _ { 2 } e ^ { - C _ { 2 } t } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 638, + 1396, + 978, + 1396, + 978, + 1432, + 638, + 1432 + ], + "score": 0.91, + "latex": "\\lVert W ( t ) - W ( 0 ) \\rVert _ { 2 } = O ( d ^ { 1 - \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 350, + 1634, + 622, + 1634, + 622, + 1676, + 350, + 1676 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\frac { 1 } { n } \\| { \\pmb y } _ { N N } ( T ) - { \\pmb y } \\| _ { 2 } ^ { 2 } 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 919, + 262, + 1075, + 262, + 1075, + 292, + 919, + 292 + ], + "score": 0.91, + "latex": "T = \\log \\log h" + }, + { + "category_id": 13, + "poly": [ + 807, + 466, + 997, + 466, + 997, + 501, + 807, + 501 + ], + "score": 0.91, + "latex": "\\omega _ { 0 } = \\mathrm { v e c } ( W ^ { \\mathrm { i n i t } } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1366, + 538, + 1366, + 538, + 1401, + 298, + 1401 + ], + "score": 0.91, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( K ( 0 ) ) \\stackrel { } { = } O ( d )" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1166, + 1258, + 1166, + 1258, + 1197, + 1203, + 1197 + ], + "score": 0.91, + "latex": "f _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 1162, + 1966, + 1405, + 1966, + 1405, + 2002, + 1162, + 2002 + ], + "score": 0.91, + "latex": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =" + }, + { + "category_id": 13, + "poly": [ + 821, + 326, + 872, + 326, + 872, + 360, + 821, + 360 + ], + "score": 0.9, + "latex": "o ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1132, + 1397, + 1365, + 1397, + 1365, + 1430, + 1132, + 1430 + ], + "score": 0.9, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( K ( t ) ) \\stackrel { } { = } O ( d )" + }, + { + "category_id": 13, + "poly": [ + 1163, + 1430, + 1405, + 1430, + 1405, + 1463, + 1163, + 1463 + ], + "score": 0.89, + "latex": "\\| { \\pmb w } _ { i } ( t ) - { \\pmb w } _ { i } ( 0 ) \\| _ { 2 } =" + }, + { + "category_id": 13, + "poly": [ + 775, + 1468, + 835, + 1468, + 835, + 1497, + 775, + 1497 + ], + "score": 0.88, + "latex": "{ \\pmb y } _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 811, + 1271, + 1041, + 1271, + 1041, + 1306, + 811, + 1306 + ], + "score": 0.88, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( K ( t ) ) = O ( d )" + }, + { + "category_id": 13, + "poly": [ + 725, + 1735, + 760, + 1735, + 760, + 1760, + 725, + 1760 + ], + "score": 0.87, + "latex": "{ \\pmb w } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 435, + 1765, + 470, + 1765, + 470, + 1791, + 435, + 1791 + ], + "score": 0.87, + "latex": "{ \\pmb w } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 780, + 1968, + 806, + 1968, + 806, + 2000, + 780, + 2000 + ], + "score": 0.86, + "latex": "\\phi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1643, + 746, + 1643, + 746, + 1667, + 655, + 1667 + ], + "score": 0.84, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 928, + 1432, + 952, + 1432, + 952, + 1456, + 928, + 1456 + ], + "score": 0.81, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1094, + 2003, + 1127, + 2003, + 1127, + 2031, + 1094, + 2031 + ], + "score": 0.77, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 821, + 1680, + 870, + 1680, + 870, + 1711, + 821, + 1711 + ], + "score": 0.33, + "latex": "\\mathsf { o } ( 1 )" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 406.0, + 639.0, + 406.0, + 639.0, + 443.0, + 294.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 728.0, + 752.0, + 728.0, + 752.0, + 766.0, + 295.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1333.0, + 1406.0, + 1333.0, + 1406.0, + 1371.0, + 293.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1363.0, + 297.0, + 1363.0, + 297.0, + 1404.0, + 293.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1363.0, + 1406.0, + 1363.0, + 1406.0, + 1404.0, + 539.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1392.0, + 637.0, + 1392.0, + 637.0, + 1437.0, + 290.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1392.0, + 1131.0, + 1392.0, + 1131.0, + 1437.0, + 979.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 1392.0, + 1411.0, + 1392.0, + 1411.0, + 1437.0, + 1366.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1423.0, + 927.0, + 1423.0, + 927.0, + 1467.0, + 291.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1423.0, + 1162.0, + 1423.0, + 1162.0, + 1467.0, + 953.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1456.0, + 297.0, + 1456.0, + 297.0, + 1504.0, + 291.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1456.0, + 774.0, + 1456.0, + 774.0, + 1504.0, + 410.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1456.0, + 1408.0, + 1456.0, + 1408.0, + 1504.0, + 836.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1493.0, + 724.0, + 1493.0, + 724.0, + 1530.0, + 294.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 226.0, + 823.0, + 226.0, + 823.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 226.0, + 1404.0, + 226.0, + 1404.0, + 265.0, + 915.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 918.0, + 260.0, + 918.0, + 294.0, + 295.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 260.0, + 1406.0, + 260.0, + 1406.0, + 294.0, + 1076.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 286.0, + 345.0, + 286.0, + 345.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 286.0, + 1054.0, + 286.0, + 1054.0, + 330.0, + 646.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 286.0, + 1411.0, + 286.0, + 1411.0, + 330.0, + 1398.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 325.0, + 820.0, + 325.0, + 820.0, + 362.0, + 293.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 325.0, + 1181.0, + 325.0, + 1181.0, + 362.0, + 873.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 326.0, + 1407.0, + 326.0, + 1407.0, + 356.0, + 1373.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 605.0, + 1404.0, + 605.0, + 1404.0, + 639.0, + 295.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 635.0, + 1406.0, + 635.0, + 1406.0, + 669.0, + 295.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 664.0, + 1404.0, + 664.0, + 1404.0, + 700.0, + 292.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1022.0, + 1410.0, + 1022.0, + 1410.0, + 1059.0, + 294.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1056.0, + 1408.0, + 1056.0, + 1408.0, + 1090.0, + 293.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1086.0, + 338.0, + 1086.0, + 338.0, + 1119.0, + 294.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1086.0, + 1405.0, + 1086.0, + 1405.0, + 1119.0, + 486.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1117.0, + 1408.0, + 1117.0, + 1408.0, + 1153.0, + 293.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1592.0, + 368.0, + 1592.0, + 368.0, + 1631.0, + 296.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1592.0, + 955.0, + 1592.0, + 955.0, + 1631.0, + 682.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1592.0, + 1407.0, + 1592.0, + 1407.0, + 1631.0, + 1119.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1634.0, + 349.0, + 1634.0, + 349.0, + 1672.0, + 295.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1634.0, + 654.0, + 1634.0, + 654.0, + 1672.0, + 623.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1634.0, + 1163.0, + 1634.0, + 1163.0, + 1672.0, + 747.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1627.0, + 1409.0, + 1627.0, + 1409.0, + 1689.0, + 1404.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1676.0, + 820.0, + 1676.0, + 820.0, + 1714.0, + 295.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1676.0, + 1176.0, + 1676.0, + 1176.0, + 1714.0, + 871.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1161.0, + 1202.0, + 1161.0, + 1202.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1161.0, + 1405.0, + 1161.0, + 1405.0, + 1199.0, + 1259.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1193.0, + 1404.0, + 1193.0, + 1404.0, + 1228.0, + 293.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1226.0, + 662.0, + 1226.0, + 662.0, + 1260.0, + 296.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1960.0, + 779.0, + 1960.0, + 779.0, + 2007.0, + 293.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1960.0, + 1161.0, + 1960.0, + 1161.0, + 2007.0, + 807.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1997.0, + 462.0, + 1997.0, + 462.0, + 2038.0, + 410.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1997.0, + 1093.0, + 1997.0, + 1093.0, + 2038.0, + 761.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1997.0, + 1404.0, + 1997.0, + 1404.0, + 2038.0, + 1128.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1726.0, + 724.0, + 1726.0, + 724.0, + 1765.0, + 292.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1726.0, + 930.0, + 1726.0, + 930.0, + 1765.0, + 761.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1726.0, + 1406.0, + 1726.0, + 1406.0, + 1765.0, + 1043.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1761.0, + 434.0, + 1761.0, + 434.0, + 1794.0, + 293.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1761.0, + 693.0, + 1761.0, + 693.0, + 1794.0, + 471.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 903.0, + 730.0, + 903.0, + 730.0, + 943.0, + 294.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 779.0, + 366.0, + 779.0, + 366.0, + 822.0, + 294.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 779.0, + 888.0, + 779.0, + 888.0, + 822.0, + 687.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 462.0, + 384.0, + 462.0, + 384.0, + 506.0, + 295.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 462.0, + 806.0, + 462.0, + 806.0, + 506.0, + 748.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 462.0, + 1091.0, + 462.0, + 1091.0, + 506.0, + 998.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1265.0, + 810.0, + 1265.0, + 810.0, + 1312.0, + 648.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1265.0, + 1051.0, + 1265.0, + 1051.0, + 1312.0, + 1042.0, + 1312.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 29, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1696, + 1404, + 1696, + 1404, + 1822, + 296, + 1822 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 296, + 308, + 1409, + 308, + 1409, + 435, + 296, + 435 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 418, + 623, + 1277, + 623, + 1277, + 1101, + 418, + 1101 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 292, + 228, + 1402, + 228, + 1402, + 295, + 292, + 295 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 614, + 1973, + 1082, + 1973, + 1082, + 2046, + 614, + 2046 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 300, + 1391, + 1403, + 1391, + 1403, + 1481, + 300, + 1481 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 1899, + 1399, + 1899, + 1399, + 1968, + 299, + 1968 + ], + "score": 0.94 + }, + { + "category_id": 0, + "poly": [ + 301, + 1850, + 770, + 1850, + 770, + 1883, + 301, + 1883 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 817, + 74, + 817, + 105, + 299, + 105 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 296, + 1119, + 548, + 1119, + 548, + 1151, + 296, + 1151 + ], + "score": 0.915 + }, + { + "category_id": 8, + "poly": [ + 356, + 1634, + 1279, + 1634, + 1279, + 1693, + 356, + 1693 + ], + "score": 0.911 + }, + { + "category_id": 8, + "poly": [ + 300, + 1483, + 1394, + 1483, + 1394, + 1559, + 300, + 1559 + ], + "score": 0.905 + }, + { + "category_id": 8, + "poly": [ + 297, + 1152, + 1409, + 1152, + 1409, + 1387, + 297, + 1387 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1047, + 1401, + 1047, + 1401, + 1079, + 1337, + 1079 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1997, + 1401, + 1997, + 1401, + 2029, + 1337, + 2029 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1334, + 1400, + 1334, + 1400, + 1365, + 1337, + 1365 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 296, + 440, + 1408, + 440, + 1408, + 620, + 296, + 620 + ], + "score": 0.886 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1557, + 1400, + 1557, + 1400, + 1587, + 1338, + 1587 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1648, + 1400, + 1648, + 1400, + 1680, + 1338, + 1680 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 293, + 1593, + 1404, + 1593, + 1404, + 1629, + 293, + 1629 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.804 + }, + { + "category_id": 8, + "poly": [ + 297, + 1310, + 1159, + 1310, + 1159, + 1386, + 297, + 1386 + ], + "score": 0.119 + }, + { + "category_id": 8, + "poly": [ + 300, + 440, + 1402, + 440, + 1402, + 505, + 300, + 505 + ], + "score": 0.09 + }, + { + "category_id": 14, + "poly": [ + 418, + 618, + 1279, + 618, + 1279, + 1106, + 418, + 1106 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\quad | \\int _ { \\mathbf { N } ^ { \\mathrm { N } } } ( \\hat { x } , t ) - \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) | = | \\int _ { 0 } ^ { T } \\left[ \\frac { \\mathrm { d } } { \\mathrm { d } t } \\int _ { \\mathbf { N } \\cap \\mathbf { K } } ( \\hat { x } , t ) - \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\right] \\mathrm { d } t | } \\\\ & { = \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left[ \\mathbf { a } _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N N } ( t ) ) - u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) ^ { \\top } ( y - y _ { N T } ( \\hat { x } ) ) \\right] \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } | \\int _ { 0 } ^ { T } u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) ^ { \\top } ( y _ { N N } ( t ) - y _ { N \\wedge \\mathbf { K } } ( t ) ) \\mathrm { d } t | } \\\\ & { \\quad + \\frac { 1 } { n } | \\int _ { 0 } ^ { T } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) ^ { \\top } ( y - y _ { N N } ( t ) ) \\mathrm { d } t | } \\\\ & { \\leq \\frac { 1 } { n } \\| u _ { N \\cap \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y _ { N N } ( t ) - y _ { N T \\mathbf { K } } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\| u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { \\quad + \\frac { 1 } { n } \\operatorname* { m a x } \\left( u _ { N \\wedge \\mathbf { N } } ( \\hat { x } , t ) - u _ { N T \\mathbf { K } } ( \\hat { x } , t ) \\right) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| y - y _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 908, + 520, + 1047, + 520, + 1047, + 556, + 908, + 556 + ], + "score": 0.94, + "latex": "{ \\pmb u } _ { N T K } ( { \\pmb x } , t )" + }, + { + "category_id": 14, + "poly": [ + 616, + 1969, + 1081, + 1969, + 1081, + 2045, + 616, + 2045 + ], + "score": 0.94, + "latex": "\\pmb { y } = \\pmb { f } ( X ; \\omega ) = \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } ( \\pmb { \\omega } - \\pmb { \\omega } _ { 0 } ) ," + }, + { + "category_id": 14, + "poly": [ + 363, + 1630, + 1274, + 1630, + 1274, + 1695, + 363, + 1695 + ], + "score": 0.93, + "latex": "| f _ { N N } ( \\pmb { \\hat { x } } , t ) - f _ { N T K } ( \\pmb { \\hat { x } } , t ) | \\leq \\frac { 1 } { n } \\| \\pmb { u } _ { N T K } ( \\pmb { \\hat { x } } , t ) \\| _ { 2 } O ( d ^ { - \\epsilon ^ { \\prime } } ) + O ( d ^ { - \\epsilon ^ { \\prime } } ) \\overset { ( i ) } { = } O ( d ^ { - \\epsilon ^ { \\prime } } ) ," + }, + { + "category_id": 13, + "poly": [ + 373, + 510, + 854, + 510, + 854, + 563, + 373, + 563 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { { \\pmb u } _ { N N } ( { \\pmb x } , t ) = \\frac { \\partial { \\pmb f } ( X ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } ^ { \\top } \\frac { \\partial { \\pmb f } ( { \\pmb x } ; \\omega ( t ) ) } { \\partial { \\pmb \\omega } ( t ) } \\in \\mathbb { R } ^ { n } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 763, + 307, + 945, + 307, + 945, + 347, + 763, + 347 + ], + "score": 0.93, + "latex": "\\| \\hat { \\pmb { x } } \\| _ { 2 } = O ( \\sqrt { d } )" + }, + { + "category_id": 13, + "poly": [ + 689, + 1900, + 865, + 1900, + 865, + 1937, + 689, + 1937 + ], + "score": 0.92, + "latex": "\\pmb { y } = \\pmb { \\beta } ^ { \\top } \\pmb { X } + \\pmb { \\varepsilon }" + }, + { + "category_id": 13, + "poly": [ + 440, + 1902, + 571, + 1902, + 571, + 1933, + 440, + 1933 + ], + "score": 0.92, + "latex": "X \\in \\mathbb { R } ^ { d \\times n }" + }, + { + "category_id": 14, + "poly": [ + 302, + 1479, + 1390, + 1479, + 1390, + 1560, + 302, + 1560 + ], + "score": 0.92, + "latex": "\\frac { 1 } { n \\hbar \\epsilon \\mathcal { T } } \\left. u _ { N N } ( \\hat { x } , t ) - u _ { N T K } ( \\hat { x } , t ) \\right. _ { 2 } \\int _ { 0 } ^ { T } \\left. y - y _ { N N } ( t ) \\right. _ { 2 } \\mathrm { d } t \\overset { ( i ) } { \\leq } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) = O ( d ^ { - \\epsilon ^ { \\prime } } ) ," + }, + { + "category_id": 13, + "poly": [ + 626, + 1756, + 814, + 1756, + 814, + 1791, + 626, + 1791 + ], + "score": 0.92, + "latex": "\\bar { | | \\mathbf { y } | | _ { 2 } } = \\bar { O ( \\sqrt { n } ) } )" + }, + { + "category_id": 13, + "poly": [ + 328, + 1727, + 384, + 1727, + 384, + 1759, + 328, + 1759 + ], + "score": 0.91, + "latex": "f _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 435, + 1728, + 508, + 1728, + 508, + 1759, + 435, + 1759 + ], + "score": 0.91, + "latex": "f _ { N T K }" + }, + { + "category_id": 14, + "poly": [ + 291, + 1148, + 1390, + 1148, + 1390, + 1390, + 291, + 1390 + ], + "score": 0.91, + "latex": "\\begin{array} { r l r } { { \\| { \\boldsymbol y } _ { N N } ( T ) - { \\boldsymbol y } _ { N T K } ( T ) \\| _ { 2 } \\le \\frac { 1 } { n } \\int _ { 0 } ^ { T } \\| K ( t ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) ) - K ( 0 ) ( { \\boldsymbol y } - { \\boldsymbol y } _ { N T K } ( t ) ) \\| _ { 2 } \\mathrm { d } t } } \\\\ & { } & { \\le \\frac { 1 } { n 0 < t < T } \\| K ( t ) - K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } - { \\boldsymbol y } _ { N N } ( t ) \\| _ { 2 } \\mathrm { d } t + \\frac { 1 } { n } \\| K ( 0 ) \\| _ { 2 } \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t } \\\\ & { } & { \\overset { ( i ) } { \\le } \\frac { 1 } { n } O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) O ( \\sqrt { d } ) + O ( 1 ) \\int _ { 0 } ^ { T } \\| { \\boldsymbol y } _ { N N } ( t ) - { \\boldsymbol y } _ { N T K } ( t ) \\| _ { 2 } \\mathrm { d } t \\overset { ( i i ) } { \\le } O ( d ^ { - \\epsilon ^ { \\prime } } ) , \\quad \\quad \\quad \\quad ( 1 2 4 ) \\mathrm { d } { \\boldsymbol z } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 869, + 262, + 926, + 262, + 926, + 294, + 869, + 294 + ], + "score": 0.91, + "latex": "f _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 917, + 1599, + 978, + 1599, + 978, + 1628, + 917, + 1628 + ], + "score": 0.9, + "latex": "{ \\pmb y } _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 882, + 1392, + 955, + 1392, + 955, + 1420, + 882, + 1420 + ], + "score": 0.9, + "latex": "\\epsilon ^ { \\prime } > 0" + }, + { + "category_id": 13, + "poly": [ + 969, + 263, + 1033, + 263, + 1033, + 290, + 969, + 290 + ], + "score": 0.89, + "latex": "t > 0" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1397, + 1346, + 1397, + 1346, + 1424, + 1287, + 1424 + ], + "score": 0.88, + "latex": "{ \\bf { \\it { \\mathbf { y } } } } _ { N N }" + }, + { + "category_id": 13, + "poly": [ + 449, + 560, + 504, + 560, + 504, + 593, + 449, + 593 + ], + "score": 0.86, + "latex": "\\omega ( 0 )" + }, + { + "category_id": 14, + "poly": [ + 310, + 438, + 1407, + 438, + 1407, + 504, + 310, + 504 + ], + "score": 0.85, + "latex": "\\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N N } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N N } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N N } ( t ) } ) ; \\frac { \\mathrm { d } } { \\mathrm { d } t } f _ { N T K } ( \\hat { x } , t ) = \\frac { 1 } { n } { u _ { N T K } ( \\hat { x } , t ) } ^ { \\top } ( { y - y _ { N T K } ( t ) } ) ," + }, + { + "category_id": 13, + "poly": [ + 977, + 1423, + 1001, + 1423, + 1001, + 1448, + 977, + 1448 + ], + "score": 0.85, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 676, + 314, + 698, + 314, + 698, + 340, + 676, + 340 + ], + "score": 0.83, + "latex": "\\hat { \\pmb x }" + }, + { + "category_id": 13, + "poly": [ + 1127, + 561, + 1149, + 561, + 1149, + 587, + 1127, + 587 + ], + "score": 0.8, + "latex": "\\hat { \\pmb x }" + }, + { + "category_id": 13, + "poly": [ + 744, + 1791, + 816, + 1791, + 816, + 1820, + 744, + 1820 + ], + "score": 0.8, + "latex": "f _ { N T K }" + }, + { + "category_id": 13, + "poly": [ + 713, + 1942, + 736, + 1942, + 736, + 1962, + 713, + 1962 + ], + "score": 0.77, + "latex": "\\omega" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1849.0, + 774.0, + 1849.0, + 774.0, + 1885.0, + 296.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2123.0, + 830.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1694.0, + 1406.0, + 1694.0, + 1406.0, + 1731.0, + 295.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1722.0, + 327.0, + 1722.0, + 327.0, + 1764.0, + 293.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1722.0, + 434.0, + 1722.0, + 434.0, + 1764.0, + 385.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1722.0, + 1407.0, + 1722.0, + 1407.0, + 1764.0, + 509.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1757.0, + 625.0, + 1757.0, + 625.0, + 1793.0, + 295.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1757.0, + 1405.0, + 1757.0, + 1405.0, + 1793.0, + 815.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1781.0, + 743.0, + 1781.0, + 743.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1781.0, + 829.0, + 1781.0, + 829.0, + 1826.0, + 817.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 309.0, + 675.0, + 309.0, + 675.0, + 349.0, + 291.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 309.0, + 762.0, + 309.0, + 762.0, + 349.0, + 699.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 309.0, + 1408.0, + 309.0, + 1408.0, + 349.0, + 946.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 342.0, + 1406.0, + 342.0, + 1406.0, + 375.0, + 295.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 371.0, + 1408.0, + 371.0, + 1408.0, + 408.0, + 294.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 401.0, + 490.0, + 401.0, + 490.0, + 438.0, + 294.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 868.0, + 262.0, + 868.0, + 294.0, + 296.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 262.0, + 968.0, + 262.0, + 968.0, + 294.0, + 927.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 262.0, + 1042.0, + 262.0, + 1042.0, + 294.0, + 1034.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1386.0, + 881.0, + 1386.0, + 881.0, + 1429.0, + 291.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1386.0, + 1286.0, + 1386.0, + 1286.0, + 1429.0, + 956.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1386.0, + 1408.0, + 1386.0, + 1408.0, + 1429.0, + 1347.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1422.0, + 976.0, + 1422.0, + 976.0, + 1455.0, + 295.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1422.0, + 1405.0, + 1422.0, + 1405.0, + 1455.0, + 1002.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1453.0, + 453.0, + 1453.0, + 453.0, + 1483.0, + 295.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 439.0, + 1896.0, + 439.0, + 1941.0, + 294.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1896.0, + 688.0, + 1896.0, + 688.0, + 1941.0, + 572.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1896.0, + 1404.0, + 1896.0, + 1404.0, + 1941.0, + 866.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1932.0, + 712.0, + 1932.0, + 712.0, + 1972.0, + 294.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1932.0, + 749.0, + 1932.0, + 749.0, + 1972.0, + 737.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1115.0, + 550.0, + 1115.0, + 550.0, + 1154.0, + 293.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 439.0, + 309.0, + 439.0, + 309.0, + 506.0, + 302.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 505.0, + 372.0, + 505.0, + 372.0, + 568.0, + 292.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 505.0, + 907.0, + 505.0, + 907.0, + 568.0, + 855.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 505.0, + 1411.0, + 505.0, + 1411.0, + 568.0, + 1048.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 557.0, + 448.0, + 557.0, + 448.0, + 592.0, + 294.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 557.0, + 1126.0, + 557.0, + 1126.0, + 592.0, + 505.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 557.0, + 1406.0, + 557.0, + 1406.0, + 592.0, + 1150.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 594.0, + 329.0, + 594.0, + 329.0, + 624.0, + 293.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1588.0, + 916.0, + 1588.0, + 916.0, + 1632.0, + 293.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1588.0, + 1404.0, + 1588.0, + 1404.0, + 1632.0, + 979.0, + 1632.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 30, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1145, + 1404, + 1145, + 1404, + 1241, + 297, + 1241 + ], + "score": 0.977 + }, + { + "category_id": 8, + "poly": [ + 442, + 1791, + 1261, + 1791, + 1261, + 2033, + 442, + 2033 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 297, + 506, + 1399, + 506, + 1399, + 916, + 297, + 916 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 575, + 1451, + 1126, + 1451, + 1126, + 1543, + 575, + 1543 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 452, + 313, + 1243, + 313, + 1243, + 407, + 452, + 407 + ], + "score": 0.956 + }, + { + "category_id": 8, + "poly": [ + 585, + 1618, + 1113, + 1618, + 1113, + 1709, + 585, + 1709 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 424, + 1403, + 424, + 1403, + 490, + 297, + 490 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 289, + 227, + 1406, + 227, + 1406, + 295, + 289, + 295 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 470, + 985, + 1227, + 985, + 1227, + 1064, + 470, + 1064 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 295, + 1359, + 1402, + 1359, + 1402, + 1432, + 295, + 1432 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 728, + 1257, + 971, + 1257, + 971, + 1298, + 728, + 1298 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 933, + 1185, + 933, + 1185, + 969, + 297, + 969 + ], + "score": 0.928 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 289, + 1312, + 1307, + 1312, + 1307, + 1347, + 289, + 1347 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 296, + 1564, + 671, + 1564, + 671, + 1599, + 296, + 1599 + ], + "score": 0.92 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1260, + 1400, + 1260, + 1400, + 1292, + 1338, + 1292 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1479, + 1401, + 1479, + 1401, + 1512, + 1337, + 1512 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1646, + 1401, + 1646, + 1401, + 1678, + 1337, + 1678 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1011, + 1401, + 1011, + 1401, + 1044, + 1337, + 1044 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1336, + 346, + 1401, + 346, + 1401, + 379, + 1336, + 379 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1983, + 1402, + 1983, + 1402, + 2015, + 1337, + 2015 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1337, + 827, + 1402, + 827, + 1402, + 860, + 1337, + 860 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 297, + 1091, + 864, + 1091, + 864, + 1125, + 297, + 1125 + ], + "score": 0.83 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 866, + 2087, + 866, + 2114, + 835, + 2114 + ], + "score": 0.791 + }, + { + "category_id": 1, + "poly": [ + 294, + 1740, + 864, + 1740, + 864, + 1776, + 294, + 1776 + ], + "score": 0.61 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2114, + 834, + 2114 + ], + "score": 0.171 + }, + { + "category_id": 1, + "poly": [ + 297, + 1091, + 864, + 1091, + 864, + 1125, + 297, + 1125 + ], + "score": 0.105 + }, + { + "category_id": 14, + "poly": [ + 440, + 1789, + 1259, + 1789, + 1259, + 2039, + 440, + 2039 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\quad \\displaystyle \\frac 1 d [ \\hat { K } _ { X } ] _ { i j } = \\frac 1 d \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } } \\\\ & { = \\displaystyle \\frac 1 d { \\sum _ { k = 1 } ^ { h } } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial { \\pmb w } _ { k } } = \\frac 1 { d h } { \\displaystyle \\sum _ { k = 1 } ^ { h } } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ^ { \\top } { \\pmb x } _ { j } ) } \\\\ & { \\to \\displaystyle \\frac 1 d { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\mathbb { E } _ { \\pmb w } \\Big [ \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } ^ { \\top } { \\pmb x } _ { j } ) \\Big ] = \\frac 1 d H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 471, + 983, + 1227, + 983, + 1227, + 1062, + 471, + 1062 + ], + "score": 0.94, + "latex": "\\hat { \\pmb { u } } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( { \\pmb { x } } ; \\omega _ { 0 } ) } { \\partial \\omega } , \\quad \\hat { K } _ { X } = \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial { \\pmb { f } } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ." + }, + { + "category_id": 14, + "poly": [ + 290, + 503, + 1390, + 503, + 1390, + 924, + 290, + 924 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { 2 R = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - f ( x ; \\omega _ { 1 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } ( \\omega _ { 1 } - \\omega _ { 0 } ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha , \\epsilon } [ ( x ^ { \\top } \\beta - \\frac { \\partial f ( x ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial f ( X ; \\omega _ { 0 } ) ^ { \\top } } { \\partial \\omega } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) ) ^ { 2 } ] } \\\\ & { \\quad = \\mathbb { E } _ { \\alpha } [ ( x ^ { \\top } \\beta - \\frac { \\partial \\tilde { f } ( \\kappa ^ { - 1 } X ^ { \\top } \\beta ) } { \\partial \\beta } \\frac { \\partial \\tilde { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ( \\frac { \\partial \\tilde { f } ( \\tilde { X } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } \\tilde { \\alpha } ^ { - 1 } ) } { 2 V } \\frac { \\partial ^ { 2 } } { \\partial \\omega } , \\qquad ( 1 2 9 ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1165, + 1177, + 1341, + 1177, + 1341, + 1211, + 1165, + 1211 + ], + "score": 0.93, + "latex": "b _ { 0 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1206, + 533, + 1206, + 533, + 1242, + 298, + 1242 + ], + "score": 0.93, + "latex": "b _ { 1 } ^ { 2 } = \\mathbb { E } [ \\phi ^ { \\prime } ( G ) ^ { 2 } ] - b _ { 0 } ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 454, + 309, + 1245, + 309, + 1245, + 409, + 454, + 409 + ], + "score": 0.93, + "latex": "\\omega _ { 1 } = \\omega _ { 0 } + \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\left( \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } ^ { \\top } \\frac { \\partial { f ( X ; \\omega _ { 0 } ) } } { \\partial { \\omega } } \\right) ^ { - 1 } ( X ^ { \\top } \\beta + \\varepsilon ) ." + }, + { + "category_id": 14, + "poly": [ + 728, + 1256, + 971, + 1256, + 971, + 1296, + 728, + 1296 + ], + "score": 0.93, + "latex": "\\phi ^ { \\prime } ( x ) = b _ { 0 } + \\phi _ { \\perp } ^ { \\prime } ( x ) ," + }, + { + "category_id": 14, + "poly": [ + 582, + 1616, + 1115, + 1616, + 1115, + 1709, + 582, + 1709 + ], + "score": 0.93, + "latex": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } - ( b _ { 0 } ^ { 2 } + b _ { 1 } ^ { 2 } ) \\right| < \\varepsilon ." + }, + { + "category_id": 13, + "poly": [ + 553, + 455, + 680, + 455, + 680, + 490, + 553, + 490 + ], + "score": 0.93, + "latex": "f ^ { \\mathrm { i n i t } } ( \\cdot ) = 0 ." + }, + { + "category_id": 14, + "poly": [ + 570, + 1449, + 1123, + 1449, + 1123, + 1544, + 570, + 1544 + ], + "score": 0.92, + "latex": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 } ," + }, + { + "category_id": 13, + "poly": [ + 919, + 230, + 1099, + 230, + 1099, + 264, + 919, + 264 + ], + "score": 0.92, + "latex": "\\partial f ( \\pmb { x } _ { i } ; \\pmb { \\omega } _ { 0 } ) / \\partial \\omega" + }, + { + "category_id": 13, + "poly": [ + 927, + 1176, + 1050, + 1176, + 1050, + 1210, + 927, + 1210 + ], + "score": 0.92, + "latex": "L ^ { 2 } ( \\mathbb { R } , \\mu _ { G } )" + }, + { + "category_id": 13, + "poly": [ + 824, + 1177, + 888, + 1177, + 888, + 1211, + 824, + 1211 + ], + "score": 0.92, + "latex": "\\phi ^ { \\prime } ( x )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1312, + 537, + 1312, + 537, + 1348, + 371, + 1348 + ], + "score": 0.92, + "latex": "\\mathbb { E } [ \\phi _ { \\perp } ^ { \\prime } ( G ) ] = 0" + }, + { + "category_id": 13, + "poly": [ + 1277, + 1363, + 1344, + 1363, + 1344, + 1395, + 1277, + 1395 + ], + "score": 0.92, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 655, + 1357, + 744, + 1357, + 744, + 1398, + 655, + 1398 + ], + "score": 0.91, + "latex": "( \\hat { K } _ { X } ) _ { i j } ." + }, + { + "category_id": 13, + "poly": [ + 372, + 230, + 553, + 230, + 553, + 263, + 372, + 263 + ], + "score": 0.91, + "latex": "\\partial f ( X ; \\omega _ { 0 } ) / \\partial \\omega" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1362, + 1119, + 1362, + 1119, + 1394, + 1018, + 1394 + ], + "score": 0.91, + "latex": "c , c ^ { \\prime } > 0" + }, + { + "category_id": 13, + "poly": [ + 602, + 232, + 686, + 232, + 686, + 260, + 602, + 260 + ], + "score": 0.9, + "latex": "d h \\times n" + }, + { + "category_id": 13, + "poly": [ + 458, + 1743, + 520, + 1743, + 520, + 1775, + 458, + 1775 + ], + "score": 0.9, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 530, + 1559, + 662, + 1559, + 662, + 1597, + 530, + 1597 + ], + "score": 0.9, + "latex": "1 - e ^ { - c ^ { \\prime } n \\varepsilon ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 427, + 1395, + 552, + 1395, + 552, + 1430, + 427, + 1430 + ], + "score": 0.89, + "latex": "1 - e ^ { - c n \\varepsilon ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 264, + 475, + 264, + 475, + 295, + 298, + 295 + ], + "score": 0.88, + "latex": "\\gamma _ { 1 } , \\gamma _ { 2 } \\in ( 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 1263, + 235, + 1353, + 235, + 1353, + 259, + 1263, + 259 + ], + "score": 0.86, + "latex": "n \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1210, + 1043, + 1210, + 1043, + 1240, + 1015, + 1240 + ], + "score": 0.83, + "latex": "\\phi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 486, + 264, + 573, + 264, + 573, + 291, + 486, + 291 + ], + "score": 0.82, + "latex": "d h > n" + }, + { + "category_id": 13, + "poly": [ + 731, + 1153, + 748, + 1153, + 748, + 1174, + 731, + 1174 + ], + "score": 0.75, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1089.0, + 864.0, + 1089.0, + 864.0, + 1127.0, + 296.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 872.0, + 2084.0, + 872.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1143.0, + 730.0, + 1143.0, + 730.0, + 1181.0, + 294.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1143.0, + 1406.0, + 1143.0, + 1406.0, + 1181.0, + 749.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1173.0, + 823.0, + 1173.0, + 823.0, + 1211.0, + 292.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1173.0, + 926.0, + 1173.0, + 926.0, + 1211.0, + 889.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1173.0, + 1164.0, + 1173.0, + 1164.0, + 1211.0, + 1051.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1173.0, + 1404.0, + 1173.0, + 1404.0, + 1211.0, + 1342.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1205.0, + 1014.0, + 1205.0, + 1014.0, + 1243.0, + 534.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 423.0, + 1407.0, + 423.0, + 1407.0, + 460.0, + 295.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 449.0, + 552.0, + 449.0, + 552.0, + 493.0, + 291.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 449.0, + 701.0, + 449.0, + 701.0, + 493.0, + 681.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 228.0, + 371.0, + 228.0, + 371.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 228.0, + 601.0, + 228.0, + 601.0, + 265.0, + 554.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 228.0, + 918.0, + 228.0, + 918.0, + 265.0, + 687.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 228.0, + 1262.0, + 228.0, + 1262.0, + 265.0, + 1100.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 265.0, + 1354.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 257.0, + 297.0, + 257.0, + 297.0, + 300.0, + 292.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 257.0, + 485.0, + 257.0, + 485.0, + 300.0, + 476.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 257.0, + 1355.0, + 257.0, + 1355.0, + 300.0, + 574.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1358.0, + 654.0, + 1358.0, + 654.0, + 1400.0, + 294.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1358.0, + 1017.0, + 1358.0, + 1017.0, + 1400.0, + 745.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1358.0, + 1276.0, + 1358.0, + 1276.0, + 1400.0, + 1120.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1358.0, + 1405.0, + 1358.0, + 1405.0, + 1400.0, + 1345.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1392.0, + 426.0, + 1392.0, + 426.0, + 1439.0, + 289.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1392.0, + 660.0, + 1392.0, + 660.0, + 1439.0, + 553.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 929.0, + 1186.0, + 929.0, + 1186.0, + 974.0, + 292.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1311.0, + 370.0, + 1311.0, + 370.0, + 1350.0, + 293.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1311.0, + 1310.0, + 1311.0, + 1310.0, + 1350.0, + 538.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1558.5, + 673.0, + 1558.5, + 673.0, + 1605.5, + 291.0, + 1605.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1739.0, + 457.0, + 1739.0, + 457.0, + 1779.0, + 296.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1739.0, + 864.0, + 1739.0, + 864.0, + 1779.0, + 521.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1089.0, + 864.0, + 1089.0, + 864.0, + 1127.0, + 296.0, + 1127.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 31, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1485, + 1407, + 1485, + 1407, + 1613, + 296, + 1613 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 295, + 458, + 1404, + 458, + 1404, + 564, + 295, + 564 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 295, + 223, + 1406, + 223, + 1406, + 338, + 295, + 338 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 367, + 1874, + 1274, + 1874, + 1274, + 2029, + 367, + 2029 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 578, + 705, + 1121, + 705, + 1121, + 796, + 578, + 796 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 492, + 922, + 1206, + 922, + 1206, + 1015, + 492, + 1015 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 622, + 1743, + 1073, + 1743, + 1073, + 1817, + 622, + 1817 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 445, + 1199, + 1256, + 1199, + 1256, + 1292, + 445, + 1292 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 593, + 577, + 1101, + 577, + 1101, + 642, + 593, + 642 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 297, + 1419, + 1405, + 1419, + 1405, + 1483, + 297, + 1483 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 419, + 351, + 1279, + 351, + 1279, + 441, + 419, + 441 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 297, + 655, + 1039, + 655, + 1039, + 691, + 297, + 691 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 297, + 1697, + 1018, + 1697, + 1018, + 1731, + 297, + 1731 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 298, + 1643, + 592, + 1643, + 592, + 1675, + 298, + 1675 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 813, + 1213, + 813, + 1213, + 850, + 297, + 850 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 297, + 1028, + 523, + 1028, + 523, + 1064, + 297, + 1064 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 297, + 1366, + 803, + 1366, + 803, + 1401, + 297, + 1401 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 295, + 1830, + 844, + 1830, + 844, + 1864, + 295, + 1864 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 818, + 73, + 818, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 298, + 1305, + 644, + 1305, + 644, + 1345, + 298, + 1345 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 298, + 1090, + 1168, + 1090, + 1168, + 1127, + 298, + 1127 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 298, + 873, + 1337, + 873, + 1337, + 909, + 298, + 909 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 300, + 1147, + 1096, + 1147, + 1096, + 1187, + 300, + 1187 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1337, + 733, + 1401, + 733, + 1401, + 767, + 1337, + 767 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1337, + 594, + 1401, + 594, + 1401, + 626, + 1337, + 626 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1337, + 379, + 1401, + 379, + 1401, + 412, + 1337, + 412 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1337, + 951, + 1401, + 951, + 1401, + 983, + 1337, + 983 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1764, + 1401, + 1764, + 1401, + 1796, + 1337, + 1796 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1227, + 1401, + 1227, + 1401, + 1260, + 1337, + 1260 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1968, + 1401, + 1968, + 1401, + 2001, + 1337, + 2001 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.868 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1093, + 1404, + 1093, + 1404, + 1123, + 1374, + 1123 + ], + "score": 0.842 + }, + { + "category_id": 2, + "poly": [ + 1375, + 1368, + 1403, + 1368, + 1403, + 1397, + 1375, + 1397 + ], + "score": 0.829 + }, + { + "category_id": 2, + "poly": [ + 1374, + 817, + 1403, + 817, + 1403, + 846, + 1374, + 846 + ], + "score": 0.818 + }, + { + "category_id": 14, + "poly": [ + 623, + 1742, + 1076, + 1742, + 1076, + 1819, + 623, + 1819 + ], + "score": 0.94, + "latex": "2 B = \\mathbb { E } _ { \\pmb { x } } \\left[ \\left( \\pmb { x } ^ { \\top } \\pmb { \\beta } - \\hat { \\pmb { u } } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\pmb { \\beta } \\right) ^ { 2 } \\right] ." + }, + { + "category_id": 13, + "poly": [ + 788, + 1421, + 1021, + 1421, + 1021, + 1455, + 788, + 1455 + ], + "score": 0.94, + "latex": "{ \\pmb w } _ { i } ( 0 ) \\sim N ( 0 , I _ { d } / d )" + }, + { + "category_id": 14, + "poly": [ + 495, + 919, + 1204, + 919, + 1204, + 1015, + 495, + 1015 + ], + "score": 0.94, + "latex": "\\frac { 1 } { d } \\left\\| \\hat { \\pmb { u } } - \\tilde { \\pmb { u } } \\right\\| _ { 2 } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( \\pmb { x } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial \\pmb { f } ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { \\pmb { u } } \\right\\| _ { 2 } < \\frac { \\log ^ { l } d } { d } ," + }, + { + "category_id": 14, + "poly": [ + 593, + 574, + 1105, + 574, + 1105, + 642, + 593, + 642 + ], + "score": 0.94, + "latex": "\\frac { 1 } { d } H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = b _ { 0 } ^ { 2 } \\frac { 1 } { d } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } + O \\big ( ( \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d ) ^ { 2 } \\big ) ." + }, + { + "category_id": 14, + "poly": [ + 447, + 1197, + 1252, + 1197, + 1252, + 1292, + 447, + 1292 + ], + "score": 0.93, + "latex": "\\frac { 1 } { d } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } = \\left\\| \\frac { 1 } { d } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( X ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } \\tilde { K } _ { X } \\right\\| _ { F } < \\log ^ { l } d ," + }, + { + "category_id": 14, + "poly": [ + 366, + 1874, + 1270, + 1874, + 1270, + 2033, + 366, + 2033 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\left\\| \\hat { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta - \\tilde { \\boldsymbol u } ^ { \\top } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\boldsymbol X ^ { \\top } \\boldsymbol \\beta \\right\\| _ { 2 } \\leq \\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\boldsymbol X \\right\\| _ { 2 } \\left\\| \\boldsymbol \\beta \\right\\| _ { 2 } } \\\\ & { \\quad \\quad \\quad \\quad \\stackrel { ( i ) } { = } O \\left( \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot \\sqrt { d } \\cdot 1 \\right) = O \\left( \\frac { \\log ^ { l } d } { \\sqrt { d } } \\right) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 422, + 224, + 944, + 224, + 944, + 280, + 422, + 280 + ], + "score": 0.93, + "latex": "H ( \\pmb { x } _ { i } , \\pmb { x } _ { j } ) = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } \\mathbb { E } _ { \\pmb { w } } \\Big [ \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } ^ { \\top } \\pmb { x } _ { j } ) \\Big ]" + }, + { + "category_id": 14, + "poly": [ + 420, + 350, + 1279, + 350, + 1279, + 444, + 420, + 444 + ], + "score": 0.93, + "latex": "\\operatorname* { P r } \\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } H ( { \\pmb x } _ { i } , { \\pmb x } _ { j } ) \\right| < \\frac { 1 } { d } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\varepsilon > 1 - e ^ { - c _ { 1 } h \\varepsilon ^ { 2 } } ." + }, + { + "category_id": 13, + "poly": [ + 455, + 654, + 604, + 654, + 604, + 692, + 455, + 692 + ], + "score": 0.93, + "latex": "\\varepsilon = \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d" + }, + { + "category_id": 13, + "poly": [ + 468, + 1088, + 613, + 1088, + 613, + 1127, + 468, + 1127 + ], + "score": 0.93, + "latex": "\\varepsilon = \\log ^ { l } d / d" + }, + { + "category_id": 14, + "poly": [ + 574, + 703, + 1122, + 703, + 1122, + 798, + 574, + 798 + ], + "score": 0.93, + "latex": "\\left| \\frac { 1 } { d } \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega _ { 0 } ) } { \\partial \\omega } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega _ { 0 } ) } { \\partial \\omega } - \\frac { 1 } { d } b _ { 0 } ^ { 2 } { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\right| < \\varepsilon ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1028, + 515, + 1028, + 515, + 1065, + 371, + 1065 + ], + "score": 0.93, + "latex": "\\tilde { \\pmb { u } } = b _ { 0 } ^ { 2 } \\pmb { x } ^ { \\top } \\boldsymbol { X }" + }, + { + "category_id": 13, + "poly": [ + 982, + 457, + 1349, + 457, + 1349, + 498, + 982, + 498 + ], + "score": 0.92, + "latex": "\\mathrm { P r } \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j } / d > \\varepsilon < 1 - e ^ { - c _ { 2 } d \\varepsilon ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 720, + 459, + 789, + 459, + 789, + 498, + 720, + 498 + ], + "score": 0.92, + "latex": "\\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { j }" + }, + { + "category_id": 13, + "poly": [ + 621, + 304, + 684, + 304, + 684, + 338, + 621, + 338 + ], + "score": 0.92, + "latex": "\\phi ^ { \\prime } ( x )" + }, + { + "category_id": 13, + "poly": [ + 481, + 810, + 605, + 810, + 605, + 846, + 481, + 846 + ], + "score": 0.92, + "latex": "1 - e ^ { - c d \\varepsilon ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 1086, + 500, + 1335, + 500, + 1335, + 535, + 1086, + 535 + ], + "score": 0.92, + "latex": "\\phi ^ { \\prime } ( x ) = b _ { 0 } x + \\phi _ { \\perp } ^ { \\prime } ( x )" + }, + { + "category_id": 13, + "poly": [ + 297, + 495, + 710, + 495, + 710, + 534, + 297, + 534 + ], + "score": 0.91, + "latex": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { x } _ { i } / d - 1 | < \\varepsilon > 1 - e ^ { - c _ { 3 } d \\varepsilon ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 848, + 460, + 917, + 460, + 917, + 497, + 848, + 497 + ], + "score": 0.91, + "latex": "\\| \\pmb { x } _ { i } \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1487, + 1409, + 1487, + 1409, + 1521, + 1078, + 1521 + ], + "score": 0.91, + "latex": "\\phi ( x ) = \\mathrm { s i g m o i d } ( x ) = ( 1 +" + }, + { + "category_id": 13, + "poly": [ + 1141, + 818, + 1203, + 818, + 1203, + 849, + 1141, + 849 + ], + "score": 0.91, + "latex": "i = j" + }, + { + "category_id": 13, + "poly": [ + 426, + 1582, + 513, + 1582, + 513, + 1612, + 426, + 1612 + ], + "score": 0.91, + "latex": "\\gamma _ { 1 } 1" + }, + { + "category_id": 13, + "poly": [ + 371, + 1305, + 636, + 1305, + 636, + 1345, + 371, + 1345 + ], + "score": 0.9, + "latex": "\\tilde { K } _ { X } = b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I" + }, + { + "category_id": 13, + "poly": [ + 297, + 1517, + 384, + 1517, + 384, + 1551, + 297, + 1551 + ], + "score": 0.89, + "latex": "e ^ { - x } ) ^ { - 1 }" + }, + { + "category_id": 13, + "poly": [ + 892, + 1521, + 977, + 1521, + 977, + 1549, + 892, + 1549 + ], + "score": 0.89, + "latex": "b _ { 1 } \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 669, + 1148, + 718, + 1148, + 718, + 1185, + 669, + 1185 + ], + "score": 0.89, + "latex": "{ \\hat { K } } _ { X } { \\mathrm { . } }" + }, + { + "category_id": 13, + "poly": [ + 523, + 1551, + 608, + 1551, + 608, + 1580, + 523, + 1580 + ], + "score": 0.89, + "latex": "b _ { 1 } = 0" + }, + { + "category_id": 13, + "poly": [ + 1172, + 868, + 1335, + 868, + 1335, + 906, + 1172, + 906 + ], + "score": 0.89, + "latex": "1 - d e ^ { - c \\log ^ { l } d }" + }, + { + "category_id": 13, + "poly": [ + 1218, + 1550, + 1245, + 1550, + 1245, + 1580, + 1218, + 1580 + ], + "score": 0.88, + "latex": "b _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 916, + 878, + 978, + 878, + 978, + 906, + 916, + 906 + ], + "score": 0.88, + "latex": "l > 0" + }, + { + "category_id": 13, + "poly": [ + 457, + 1486, + 698, + 1486, + 698, + 1520, + 457, + 1520 + ], + "score": 0.86, + "latex": "\\phi ( x ) = \\mathrm { S o f t P l u s } ( x ) ." + }, + { + "category_id": 13, + "poly": [ + 781, + 1550, + 802, + 1550, + 802, + 1580, + 781, + 1580 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 732, + 1833, + 753, + 1833, + 753, + 1859, + 732, + 1859 + ], + "score": 0.85, + "latex": "\\hat { \\textbf { \\textit { u } } }" + }, + { + "category_id": 13, + "poly": [ + 1281, + 1521, + 1301, + 1521, + 1301, + 1550, + 1281, + 1550 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 814, + 1833, + 834, + 1833, + 834, + 1858, + 814, + 1858 + ], + "score": 0.83, + "latex": "\\tilde { \\mathbf { \\pmb { u } } }" + }, + { + "category_id": 13, + "poly": [ + 586, + 1519, + 759, + 1519, + 759, + 1551, + 586, + 1551 + ], + "score": 0.82, + "latex": "b _ { 1 } ^ { 2 } = 0 . 0 0 2 1 4 4" + }, + { + "category_id": 13, + "poly": [ + 930, + 1145, + 1094, + 1145, + 1094, + 1184, + 930, + 1184 + ], + "score": 0.82, + "latex": "1 - d e ^ { - c \\log ^ { l } d }" + }, + { + "category_id": 13, + "poly": [ + 869, + 533, + 897, + 533, + 897, + 559, + 869, + 559 + ], + "score": 0.81, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 669, + 879, + 691, + 879, + 691, + 904, + 669, + 904 + ], + "score": 0.78, + "latex": "\\hat { \\textbf { \\textit { u } } }" + }, + { + "category_id": 13, + "poly": [ + 709, + 1485, + 822, + 1485, + 822, + 1521, + 709, + 1521 + ], + "score": 0.77, + "latex": "b _ { 0 } ^ { 2 } = 1 / 4" + }, + { + "category_id": 13, + "poly": [ + 837, + 1486, + 1010, + 1486, + 1010, + 1520, + 837, + 1520 + ], + "score": 0.76, + "latex": "b _ { 1 } ^ { 2 } = 0 . 0 4 3 3 7 9" + }, + { + "category_id": 13, + "poly": [ + 398, + 1519, + 573, + 1519, + 573, + 1552, + 398, + 1552 + ], + "score": 0.61, + "latex": "b _ { 0 } ^ { 2 } = 0 . 0 4 2 6 9 2" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1638.0, + 595.0, + 1638.0, + 595.0, + 1680.0, + 294.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1098.0, + 1403.0, + 1098.0, + 1403.0, + 1124.0, + 1378.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1373.0, + 1401.0, + 1373.0, + 1401.0, + 1395.0, + 1380.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 821.0, + 1403.0, + 821.0, + 1403.0, + 847.0, + 1378.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1482.0, + 456.0, + 1482.0, + 456.0, + 1523.0, + 293.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1482.0, + 708.0, + 1482.0, + 708.0, + 1523.0, + 699.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1482.0, + 836.0, + 1482.0, + 836.0, + 1523.0, + 823.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 1482.0, + 1077.0, + 1482.0, + 1077.0, + 1523.0, + 1011.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1513.0, + 296.0, + 1513.0, + 296.0, + 1554.0, + 291.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1513.0, + 397.0, + 1513.0, + 397.0, + 1554.0, + 385.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1513.0, + 585.0, + 1513.0, + 585.0, + 1554.0, + 574.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1513.0, + 891.0, + 1513.0, + 891.0, + 1554.0, + 760.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1513.0, + 1280.0, + 1513.0, + 1280.0, + 1554.0, + 978.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1513.0, + 1408.0, + 1513.0, + 1408.0, + 1554.0, + 1302.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1548.0, + 522.0, + 1548.0, + 522.0, + 1585.0, + 295.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1548.0, + 780.0, + 1548.0, + 780.0, + 1585.0, + 609.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 1548.0, + 1217.0, + 1548.0, + 1217.0, + 1585.0, + 803.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1548.0, + 1407.0, + 1548.0, + 1407.0, + 1585.0, + 1246.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 425.0, + 1580.0, + 425.0, + 1613.0, + 294.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1580.0, + 1077.0, + 1580.0, + 1077.0, + 1613.0, + 514.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 453.0, + 719.0, + 453.0, + 719.0, + 499.0, + 289.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 453.0, + 847.0, + 453.0, + 847.0, + 499.0, + 790.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 453.0, + 981.0, + 453.0, + 981.0, + 499.0, + 918.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 453.0, + 1410.0, + 453.0, + 1410.0, + 499.0, + 1350.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 488.0, + 296.0, + 488.0, + 296.0, + 541.0, + 290.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 488.0, + 1085.0, + 488.0, + 1085.0, + 541.0, + 711.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 488.0, + 1409.0, + 488.0, + 1409.0, + 541.0, + 1336.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 529.0, + 868.0, + 529.0, + 868.0, + 565.0, + 293.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 529.0, + 901.0, + 529.0, + 901.0, + 565.0, + 898.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 421.0, + 227.0, + 421.0, + 272.0, + 293.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 227.0, + 1407.0, + 227.0, + 1407.0, + 272.0, + 945.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 267.0, + 1404.0, + 267.0, + 1404.0, + 311.0, + 294.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 302.0, + 620.0, + 302.0, + 620.0, + 340.0, + 294.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 302.0, + 1144.0, + 302.0, + 1144.0, + 340.0, + 685.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1416.0, + 787.0, + 1416.0, + 787.0, + 1459.0, + 291.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1416.0, + 1406.0, + 1416.0, + 1406.0, + 1459.0, + 1022.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1454.0, + 703.0, + 1454.0, + 703.0, + 1486.0, + 296.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 652.0, + 454.0, + 652.0, + 454.0, + 693.0, + 294.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 652.0, + 1041.0, + 652.0, + 1041.0, + 693.0, + 605.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1695.0, + 1018.0, + 1695.0, + 1018.0, + 1735.0, + 295.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 802.0, + 480.0, + 802.0, + 480.0, + 860.0, + 288.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 802.0, + 1140.0, + 802.0, + 1140.0, + 860.0, + 606.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 802.0, + 1222.0, + 802.0, + 1222.0, + 860.0, + 1204.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1025.0, + 370.0, + 1025.0, + 370.0, + 1067.0, + 293.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1025.0, + 527.0, + 1025.0, + 527.0, + 1067.0, + 516.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1366.0, + 805.0, + 1366.0, + 805.0, + 1402.0, + 296.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1827.0, + 731.0, + 1827.0, + 731.0, + 1867.0, + 295.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1827.0, + 813.0, + 1827.0, + 813.0, + 1867.0, + 754.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1827.0, + 845.0, + 1827.0, + 845.0, + 1867.0, + 835.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1303.0, + 370.0, + 1303.0, + 370.0, + 1348.0, + 295.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1303.0, + 645.0, + 1303.0, + 645.0, + 1348.0, + 637.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 467.0, + 1089.0, + 467.0, + 1130.0, + 294.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1089.0, + 1172.0, + 1089.0, + 1172.0, + 1130.0, + 614.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 871.0, + 668.0, + 871.0, + 668.0, + 913.0, + 291.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 871.0, + 915.0, + 871.0, + 915.0, + 913.0, + 692.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 871.0, + 1171.0, + 871.0, + 1171.0, + 913.0, + 979.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 871.0, + 1344.0, + 871.0, + 1344.0, + 913.0, + 1336.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1146.0, + 668.0, + 1146.0, + 668.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1146.0, + 929.0, + 1146.0, + 929.0, + 1190.0, + 719.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1146.0, + 1099.0, + 1146.0, + 1099.0, + 1190.0, + 1095.0, + 1190.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 32, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 416, + 1063, + 1282, + 1063, + 1282, + 1377, + 416, + 1377 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 326, + 527, + 1369, + 527, + 1369, + 682, + 326, + 682 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 329, + 772, + 1316, + 772, + 1316, + 1008, + 329, + 1008 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 511, + 1433, + 1186, + 1433, + 1186, + 1660, + 511, + 1660 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 369, + 319, + 1267, + 319, + 1267, + 471, + 369, + 471 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 220, + 1403, + 220, + 1403, + 309, + 298, + 309 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 452, + 1715, + 1249, + 1715, + 1249, + 1806, + 452, + 1806 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 677, + 1986, + 1022, + 1986, + 1022, + 2046, + 677, + 2046 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 1668, + 1004, + 1668, + 1004, + 1702, + 297, + 1702 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 296, + 1940, + 918, + 1940, + 918, + 1975, + 296, + 1975 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 296, + 1021, + 453, + 1021, + 453, + 1053, + 296, + 1053 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 1389, + 813, + 1389, + 813, + 1423, + 297, + 1423 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 297, + 724, + 1141, + 724, + 1141, + 764, + 297, + 764 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 818, + 73, + 818, + 106, + 297, + 106 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 297, + 1818, + 510, + 1818, + 510, + 1857, + 297, + 1857 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 298, + 1885, + 659, + 1885, + 659, + 1920, + 298, + 1920 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 297, + 482, + 1396, + 482, + 1396, + 517, + 297, + 517 + ], + "score": 0.912 + }, + { + "category_id": 9, + "poly": [ + 1336, + 1744, + 1402, + 1744, + 1402, + 1777, + 1336, + 1777 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1336, + 1612, + 1402, + 1612, + 1402, + 1644, + 1336, + 1644 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1998, + 1402, + 1998, + 1402, + 2030, + 1337, + 2030 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1336, + 1317, + 1402, + 1317, + 1402, + 1350, + 1336, + 1350 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1337, + 676, + 1402, + 676, + 1402, + 706, + 1337, + 706 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1337, + 418, + 1401, + 418, + 1401, + 450, + 1337, + 450 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1338, + 948, + 1402, + 948, + 1402, + 980, + 1338, + 980 + ], + "score": 0.843 + }, + { + "category_id": 14, + "poly": [ + 417, + 1066, + 1277, + 1066, + 1277, + 1381, + 417, + 1381 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { ~ \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| } \\\\ & { = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } ) X ^ { \\top } X ( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } ) \\right) \\right| } \\\\ & { < \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 511, + 1433, + 1187, + 1433, + 1187, + 1663, + 511, + 1663 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { 2 B \\to \\mathbb { E } _ { x } [ ( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\quad \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ( I - b _ { 0 } ^ { 2 } X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } ) ) } \\\\ & { \\quad = \\cfrac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } ( ( I - b _ { 0 } ^ { 2 } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 1 } X ^ { \\top } ) ^ { 2 } ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 453, + 1712, + 1245, + 1712, + 1245, + 1806, + 453, + 1806 + ], + "score": 0.95, + "latex": "B = \\beta ^ { \\top } \\beta \\left( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } \\right) ," + }, + { + "category_id": 14, + "poly": [ + 322, + 772, + 1310, + 772, + 1310, + 1013, + 322, + 1013 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } - X \\tilde { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right| = \\left| \\frac { 1 } { d } \\mathrm { t r } \\left( X ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } ( \\hat { K } _ { X } - \\tilde { K } _ { X } ) \\tilde { K } _ { X } ^ { - 1 } \\right) \\right| } \\\\ & { \\qquad < \\displaystyle \\frac { 1 } { d } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ & { \\qquad = O \\left( d ^ { - 1 } \\cdot d \\cdot d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 677, + 1985, + 1023, + 1985, + 1023, + 2045, + 677, + 2045 + ], + "score": 0.94, + "latex": "2 V = \\mathbb { E } _ { \\pmb { x } } \\left[ \\hat { \\pmb { u } } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { K } _ { \\scriptscriptstyle X } ^ { - 1 } \\hat { \\pmb { u } } ^ { \\top } \\right] \\sigma ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 325, + 526, + 1374, + 526, + 1374, + 683, + 325, + 683 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } { \\mathbb { E } _ { x } \\left[ \\left( x ^ { \\top } \\beta - b _ { 0 } ^ { 2 } x ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta \\right) ^ { 2 } \\right] } & { = \\mathbb { E } _ { x } \\left[ \\beta ^ { \\top } \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) ^ { 2 } \\beta \\right] } \\\\ & { = \\frac { \\beta ^ { \\top } \\beta } { d } \\mathrm { t r } \\left( \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\left( I - b _ { 0 } ^ { 2 } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\right) \\right) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 368, + 318, + 1269, + 318, + 1269, + 475, + 368, + 475 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { 2 B = \\mathbb { E } _ { x } [ ( { \\pmb x } ^ { \\top } \\beta - \\hat { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - \\tilde { \\pmb u } ^ { \\top } \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] } \\\\ & { \\qquad = \\mathbb { E } _ { \\pmb { x } } [ ( { \\pmb x } ^ { \\top } \\beta - b _ { 0 } ^ { 2 } { \\pmb x } ^ { \\top } X \\hat { K } _ { X } ^ { - 1 } X ^ { \\top } \\beta ) ^ { 2 } ] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 651, + 223, + 1034, + 223, + 1034, + 282, + 651, + 282 + ], + "score": 0.94, + "latex": "\\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } = \\lambda _ { \\operatorname* { m i n } } ^ { - 1 } ( \\hat { K } _ { X } ) = O ( 1 / d )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1818, + 500, + 1818, + 500, + 1857, + 371, + 1857 + ], + "score": 0.93, + "latex": "m = { b _ { 0 } } ^ { - 2 } { b _ { 1 } } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1087, + 231, + 1272, + 231, + 1272, + 270, + 1087, + 270 + ], + "score": 0.93, + "latex": "\\| X \\| _ { 2 } = O ( { \\sqrt { d } } )" + }, + { + "category_id": 13, + "poly": [ + 900, + 725, + 946, + 725, + 946, + 761, + 900, + 761 + ], + "score": 0.91, + "latex": "\\tilde { K } _ { X }" + }, + { + "category_id": 13, + "poly": [ + 814, + 725, + 860, + 725, + 860, + 761, + 814, + 761 + ], + "score": 0.91, + "latex": "\\hat { K } _ { X }" + }, + { + "category_id": 13, + "poly": [ + 424, + 276, + 570, + 276, + 570, + 309, + 424, + 309 + ], + "score": 0.9, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 602, + 490, + 623, + 490, + 623, + 511, + 602, + 511 + ], + "score": 0.79, + "latex": "_ { \\textbf { \\em x } }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1885.0, + 661.0, + 1885.0, + 661.0, + 1921.0, + 296.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 217.0, + 650.0, + 217.0, + 650.0, + 283.0, + 284.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 217.0, + 1086.0, + 217.0, + 1086.0, + 283.0, + 1035.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 217.0, + 1414.0, + 217.0, + 1414.0, + 283.0, + 1273.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 275.0, + 423.0, + 275.0, + 423.0, + 309.0, + 296.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1667.0, + 1003.0, + 1667.0, + 1003.0, + 1703.0, + 296.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1935.0, + 917.0, + 1935.0, + 917.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1017.0, + 455.0, + 1017.0, + 455.0, + 1057.0, + 295.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1388.0, + 814.0, + 1388.0, + 814.0, + 1425.0, + 297.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 721.0, + 813.0, + 721.0, + 813.0, + 767.0, + 292.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 721.0, + 899.0, + 721.0, + 899.0, + 767.0, + 861.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 721.0, + 1143.0, + 721.0, + 1143.0, + 767.0, + 947.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1814.0, + 370.0, + 1814.0, + 370.0, + 1862.0, + 293.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1814.0, + 513.0, + 1814.0, + 513.0, + 1862.0, + 501.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 480.0, + 601.0, + 480.0, + 601.0, + 518.0, + 294.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 480.0, + 1398.0, + 480.0, + 1398.0, + 518.0, + 624.0, + 518.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 33, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1811, + 1407, + 1811, + 1407, + 1974, + 295, + 1974 + ], + "score": 0.979 + }, + { + "category_id": 8, + "poly": [ + 474, + 1590, + 1221, + 1590, + 1221, + 1780, + 474, + 1780 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 493, + 1080, + 1205, + 1080, + 1205, + 1392, + 493, + 1392 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 486, + 506, + 1209, + 506, + 1209, + 847, + 486, + 847 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 419, + 280, + 1283, + 280, + 1283, + 436, + 419, + 436 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 546, + 919, + 1150, + 919, + 1150, + 1011, + 546, + 1011 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 296, + 456, + 447, + 456, + 447, + 489, + 296, + 489 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 296, + 1536, + 978, + 1536, + 978, + 1571, + 296, + 1571 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 298, + 865, + 1133, + 865, + 1133, + 900, + 298, + 900 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 297, + 1030, + 676, + 1030, + 676, + 1063, + 297, + 1063 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 823, + 229, + 823, + 264, + 297, + 264 + ], + "score": 0.928 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 296, + 1409, + 888, + 1409, + 888, + 1443, + 296, + 1443 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 299, + 1481, + 742, + 1481, + 742, + 1514, + 299, + 1514 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1715, + 1401, + 1715, + 1401, + 1747, + 1337, + 1747 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1337, + 947, + 1402, + 947, + 1402, + 981, + 1337, + 981 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1337, + 784, + 1401, + 784, + 1401, + 816, + 1337, + 816 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1328, + 1402, + 1328, + 1402, + 1360, + 1337, + 1360 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1337, + 375, + 1402, + 375, + 1402, + 408, + 1337, + 408 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1987, + 1402, + 1987, + 1402, + 2016, + 1374, + 2016 + ], + "score": 0.802 + }, + { + "category_id": 14, + "poly": [ + 486, + 507, + 1211, + 507, + 1211, + 850, + 486, + 850 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { \\quad \\left| \\mathbb { E } _ { \\mathbf { x } } \\left[ \\tilde { u } \\hat { K } _ { X } ^ { - 1 } \\hat { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } - \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\right] \\right| } \\\\ & { = \\mathrm { t r } \\left( \\left( \\hat { K } _ { X } ^ { - 1 } - \\tilde { K } _ { X } ^ { - 1 } \\right) \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) \\mathbb { E } _ { \\alpha \\tilde { u } \\tilde { u } ^ { \\top } } \\right) } \\\\ & { = \\mathrm { t r } \\left( \\hat { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } - \\tilde { K } _ { X } \\right) \\tilde { K } _ { X } ^ { - 1 } \\left( \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right) X ^ { T } X \\right) } \\\\ & { \\leq \\left\\| \\hat { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } - \\tilde { K } _ { X } \\right\\| _ { F } \\left\\| \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { K } _ { X } ^ { - 1 } + \\tilde { K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| X ^ { T } X \\right\\| _ { 2 } } \\\\ & { = O ( d ^ { - 1 } \\cdot d \\log ^ { l } d \\cdot d ^ { - 1 } \\cdot d ^ { - 1 } \\cdot d ) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 491, + 1080, + 1206, + 1080, + 1206, + 1396, + 491, + 1396 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { 2 V \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ \\tilde { u } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { u } ^ { \\top } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\mathbb { E } _ { x } \\Big [ b _ { 0 } ^ { 4 } { \\boldsymbol x } ^ { T } X ( b _ { 0 } ^ { 2 } X ^ { \\top } X + b _ { 1 } ^ { 2 } d I ) ^ { - 2 } X ^ { T } { \\boldsymbol x } \\Big ] } \\\\ & { \\quad = \\sigma ^ { 2 } \\frac { 1 } { d } \\mathrm { t r } ( \\frac { 1 } { d } X ^ { T } X \\cdot ( \\frac { 1 } { d } X ^ { T } X + b _ { 0 } ^ { - 2 } b _ { 1 } ^ { 2 } I ) ^ { - 2 } ) } \\\\ & { \\quad = \\sigma ^ { 2 } ( - \\frac { 1 } { 2 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 2 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 473, + 1588, + 1226, + 1588, + 1226, + 1783, + 473, + 1783 + ], + "score": 0.94, + "latex": "\\begin{array} { c } { { R r ^ { 2 } ( \\frac { \\gamma _ { 1 } - 1 } { 2 \\gamma _ { 1 } } + \\frac { \\gamma _ { 1 } ( \\gamma _ { 1 } + \\gamma _ { 1 } m + m - 2 ) + 1 } { 2 \\gamma _ { 1 } \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) } } \\\\ { { + \\sigma ^ { 2 } ( - \\frac { 1 } { 4 } + \\frac { \\gamma _ { 1 } + \\gamma _ { 1 } m + 1 } { 4 \\sqrt { \\gamma _ { 1 } ( \\gamma _ { 1 } + m ( \\gamma _ { 1 } ( m + 2 ) + 2 ) - 2 ) + 1 } } ) . } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 629, + 1905, + 742, + 1905, + 742, + 1942, + 629, + 1942 + ], + "score": 0.94, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } = X ^ { \\dagger } \\boldsymbol { y }" + }, + { + "category_id": 14, + "poly": [ + 415, + 279, + 1278, + 279, + 1278, + 439, + 415, + 439 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\left| \\hat { \\boldsymbol u } \\hat { K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol u } ^ { \\top } - \\tilde { \\boldsymbol u } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\tilde { \\boldsymbol u } ^ { \\top } \\right| \\leq \\left\\| \\hat { \\boldsymbol u } - \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol u } + \\tilde { \\boldsymbol u } \\right\\| _ { 2 } \\left\\| \\hat { \\boldsymbol K } _ { X } ^ { - 1 } \\right\\| _ { 2 } } \\\\ { = O \\left( \\log ^ { l } d \\cdot \\frac { 1 } { d } \\cdot d \\cdot \\frac { 1 } { d } \\right) = O \\left( \\frac { \\log ^ { l } d } { d } \\right) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1265, + 1908, + 1403, + 1908, + 1403, + 1943, + 1265, + 1943 + ], + "score": 0.93, + "latex": "\\gamma _ { 1 } \\in ( 0 , \\infty )" + }, + { + "category_id": 14, + "poly": [ + 547, + 917, + 1151, + 917, + 1151, + 1012, + 547, + 1012 + ], + "score": 0.92, + "latex": "| 2 V - \\mathbb { E } _ { \\pmb { x } } [ \\tilde { \\pmb { u } } \\tilde { K } _ { X } ^ { - 1 } \\tilde { K } _ { X } ^ { - 1 } \\tilde { \\pmb { u } } ^ { \\top } ] \\sigma ^ { 2 } | = O ( \\frac { \\log ^ { l } d } { d } ) 0 ." + }, + { + "category_id": 13, + "poly": [ + 987, + 867, + 1131, + 867, + 1131, + 899, + 987, + 899 + ], + "score": 0.9, + "latex": "n , d , p \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 414, + 1942, + 491, + 1942, + 491, + 1970, + 414, + 1970 + ], + "score": 0.89, + "latex": "m > 0" + }, + { + "category_id": 13, + "poly": [ + 810, + 1873, + 838, + 1873, + 838, + 1902, + 810, + 1902 + ], + "score": 0.86, + "latex": "r ^ { \\bar { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 884, + 1819, + 914, + 1819, + 914, + 1846, + 884, + 1846 + ], + "score": 0.85, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 544, + 1942, + 565, + 1942, + 565, + 1973, + 544, + 1973 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 371, + 1416, + 399, + 1416, + 399, + 1438, + 371, + 1438 + ], + "score": 0.75, + "latex": "m" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1480.0, + 744.0, + 1480.0, + 744.0, + 1516.0, + 296.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1992.0, + 1401.0, + 1992.0, + 1401.0, + 2015.0, + 1379.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1812.0, + 883.0, + 1812.0, + 883.0, + 1846.0, + 296.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1812.0, + 1403.0, + 1812.0, + 1403.0, + 1846.0, + 915.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1843.0, + 1404.0, + 1843.0, + 1404.0, + 1877.0, + 297.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1875.0, + 809.0, + 1875.0, + 809.0, + 1909.0, + 297.0, + 1909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1875.0, + 1405.0, + 1875.0, + 1405.0, + 1909.0, + 839.0, + 1909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1905.0, + 628.0, + 1905.0, + 628.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1905.0, + 1264.0, + 1905.0, + 1264.0, + 1946.0, + 743.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1905.0, + 1407.0, + 1905.0, + 1407.0, + 1946.0, + 1404.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1940.0, + 413.0, + 1940.0, + 413.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1940.0, + 543.0, + 1940.0, + 543.0, + 1975.0, + 492.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1940.0, + 710.0, + 1940.0, + 710.0, + 1975.0, + 566.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 455.0, + 449.0, + 455.0, + 449.0, + 492.0, + 296.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1537.0, + 977.0, + 1537.0, + 977.0, + 1572.0, + 294.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 862.0, + 986.0, + 862.0, + 986.0, + 905.0, + 293.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 862.0, + 1135.0, + 862.0, + 1135.0, + 905.0, + 1132.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1027.0, + 675.0, + 1027.0, + 675.0, + 1065.0, + 294.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 229.0, + 825.0, + 229.0, + 825.0, + 269.0, + 298.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1407.0, + 370.0, + 1407.0, + 370.0, + 1445.0, + 295.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1407.0, + 889.0, + 1407.0, + 889.0, + 1445.0, + 400.0, + 1445.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 34, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1177, + 1406, + 1177, + 1406, + 1298, + 296, + 1298 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1691, + 1403, + 1691, + 1403, + 1792, + 298, + 1792 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 314, + 1401, + 1321, + 1401, + 1321, + 1681, + 314, + 1681 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 296, + 401, + 1406, + 401, + 1406, + 492, + 296, + 492 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 295, + 838, + 1404, + 838, + 1404, + 931, + 295, + 931 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 498, + 597, + 1199, + 597, + 1199, + 705, + 498, + 705 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 528, + 1806, + 1171, + 1806, + 1171, + 1888, + 528, + 1888 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 295, + 517, + 1401, + 517, + 1401, + 585, + 295, + 585 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 1310, + 1406, + 1310, + 1406, + 1389, + 299, + 1389 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 290, + 977, + 1402, + 977, + 1402, + 1043, + 290, + 1043 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 692, + 343, + 1007, + 343, + 1007, + 387, + 692, + 387 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 295, + 714, + 1230, + 714, + 1230, + 749, + 295, + 749 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 366, + 763, + 1268, + 763, + 1268, + 821, + 366, + 821 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 299, + 1899, + 1129, + 1899, + 1129, + 1938, + 299, + 1938 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 289, + 1130, + 1368, + 1130, + 1368, + 1165, + 289, + 1165 + ], + "score": 0.928 + }, + { + "category_id": 8, + "poly": [ + 541, + 1059, + 1155, + 1059, + 1155, + 1104, + 541, + 1104 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 294, + 762, + 294, + 762, + 330, + 297, + 330 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 818, + 73, + 818, + 106, + 297, + 106 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1337, + 665, + 1401, + 665, + 1401, + 697, + 1337, + 697 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1336, + 1829, + 1401, + 1829, + 1401, + 1861, + 1336, + 1861 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1337, + 775, + 1401, + 775, + 1401, + 809, + 1337, + 809 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1337, + 350, + 1401, + 350, + 1401, + 382, + 1337, + 382 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1339, + 1631, + 1401, + 1631, + 1401, + 1663, + 1339, + 1663 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 866, + 2087, + 866, + 2113, + 835, + 2113 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1950, + 1404, + 1950, + 1404, + 1979, + 1374, + 1979 + ], + "score": 0.836 + }, + { + "category_id": 0, + "poly": [ + 298, + 224, + 612, + 224, + 612, + 263, + 298, + 263 + ], + "score": 0.828 + }, + { + "category_id": 14, + "poly": [ + 314, + 1403, + 1325, + 1403, + 1325, + 1684, + 314, + 1684 + ], + "score": 0.96, + "latex": "\\begin{array} { r l } & { \\| X ^ { \\top } \\hat { \\pmb \\beta } ( t ) \\| _ { \\infty } = \\left\\| X ^ { \\top } \\left( I - \\exp ( - \\frac t n X X ^ { \\top } ) \\right) \\left( X X ^ { \\top } \\right) ^ { - 1 } X \\pmb y \\right\\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } U ^ { \\top } U \\bar { \\Sigma } U ^ { \\top } U \\hat { \\Sigma } ^ { - 2 } U ^ { \\top } U \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\| \\pmb y \\| _ { \\infty } } \\\\ & { \\qquad \\leq \\left\\| V \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma V ^ { \\top } \\right\\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) } \\\\ & { \\qquad \\leq \\| V \\| _ { \\infty } \\left\\| \\Sigma ^ { \\top } \\bar { \\Sigma } \\hat { \\Sigma } ^ { - 2 } \\Sigma \\right\\| _ { \\infty } \\| V ^ { \\top } \\| _ { \\infty } \\left( \\| X ^ { \\top } \\pmb \\beta \\| _ { \\infty } + \\| \\varepsilon \\| _ { \\infty } \\right) \\overset { ( i ) } { \\leq } O _ { P } ( \\mathrm { p o l y } \\log d ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 528, + 1806, + 1170, + 1806, + 1170, + 1885, + 528, + 1885 + ], + "score": 0.94, + "latex": "\\left\\| V \\right\\| _ { \\infty } = \\operatorname* { s u p } _ { z } { \\frac { \\left\\| V z \\right\\| _ { \\infty } } { \\left\\| z \\right\\| _ { \\infty } } } = O \\left( { \\frac { \\log d } { \\sqrt { d } } } \\right) { \\frac { \\left\\| z \\right\\| _ { 2 } } { \\left\\| z \\right\\| _ { \\infty } } } = O ( \\log d ) ," + }, + { + "category_id": 13, + "poly": [ + 399, + 1349, + 765, + 1349, + 765, + 1389, + 399, + 1389 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) = U \\bar { \\Sigma } U ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1111, + 838, + 1286, + 838, + 1286, + 876, + 1111, + 876 + ], + "score": 0.93, + "latex": "\\epsilon = \\log ^ { c } d / \\sqrt { d }" + }, + { + "category_id": 13, + "poly": [ + 668, + 874, + 957, + 874, + 957, + 932, + 668, + 932 + ], + "score": 0.93, + "latex": "\\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d ." + }, + { + "category_id": 13, + "poly": [ + 1296, + 1311, + 1398, + 1311, + 1398, + 1348, + 1296, + 1348 + ], + "score": 0.93, + "latex": "\\hat { \\Sigma } _ { i i } = \\lambda _ { i }" + }, + { + "category_id": 13, + "poly": [ + 580, + 402, + 771, + 402, + 771, + 440, + 580, + 440 + ], + "score": 0.93, + "latex": "Q _ { i \\neq j } \\ = \\ w _ { i } ^ { \\top } w _ { j }" + }, + { + "category_id": 13, + "poly": [ + 843, + 1351, + 1127, + 1351, + 1127, + 1388, + 843, + 1388 + ], + "score": 0.93, + "latex": "\\bar { \\Sigma } _ { i , i } = 1 - \\exp ( - t \\lambda _ { i } ^ { 2 } / n )" + }, + { + "category_id": 13, + "poly": [ + 1104, + 1219, + 1406, + 1219, + 1406, + 1261, + 1104, + 1261 + ], + "score": 0.93, + "latex": "\\begin{array} { r l r } { { \\| { \\cal I } - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\| _ { 2 } = } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 372, + 836, + 750, + 836, + 750, + 879, + 372, + 879 + ], + "score": 0.92, + "latex": "| [ K _ { W } ] _ { i j } - [ \\tilde { K } _ { W } ] _ { i j } | < ( \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } ) ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 370, + 762, + 1265, + 762, + 1265, + 822, + 370, + 822 + ], + "score": 0.92, + "latex": "[ K _ { W } ] _ { i j } = \\mathbb { E } _ { \\boldsymbol { x } } \\Big [ \\phi ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) \\Big ] = \\| \\boldsymbol { w } _ { i } \\| _ { 2 } \\| \\boldsymbol { w } _ { j } \\| _ { 2 } + \\mathbb { E } _ { \\boldsymbol { x } } [ \\phi _ { \\bot } ( \\boldsymbol { w } _ { i } ^ { \\top } \\boldsymbol { x } ) \\phi _ { \\bot } ( \\boldsymbol { w } _ { j } ^ { \\top } \\boldsymbol { x } ) ] ," + }, + { + "category_id": 14, + "poly": [ + 492, + 596, + 1201, + 596, + 1201, + 708, + 492, + 708 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { ~ \\Big | [ K _ { W } ] _ { i i } - [ \\tilde { K } _ { W } ] _ { i i } \\Big | = \\Big | \\mathbb { E } _ { \\pmb { x } } \\Big [ \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\phi ( \\pmb { w } _ { i } ^ { \\top } \\pmb { x } ) \\Big ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\Big | } \\\\ & { = \\big | \\mathbb { E } [ \\phi ( \\| \\pmb { w } _ { i } \\| _ { 2 } G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] \\big | = O ( \\epsilon ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 689, + 342, + 1007, + 342, + 1007, + 386, + 689, + 386 + ], + "score": 0.92, + "latex": "\\tilde { K } _ { W } = r I _ { h } + s { \\bf 1 } _ { h } { \\bf 1 } _ { h } ^ { \\top } + t Q ," + }, + { + "category_id": 13, + "poly": [ + 637, + 1261, + 823, + 1261, + 823, + 1298, + 637, + 1298 + ], + "score": 0.91, + "latex": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 675, + 518, + 1123, + 518, + 1123, + 558, + 675, + 558 + ], + "score": 0.91, + "latex": "\\mathcal { A } _ { \\epsilon } = \\big \\{ | \\| \\pmb { w } _ { i } \\| _ { 2 } - 1 | < \\epsilon , | \\pmb { w } _ { i } ^ { \\top } \\pmb { w } _ { j } | < \\epsilon \\big \\}" + }, + { + "category_id": 13, + "poly": [ + 376, + 402, + 514, + 402, + 514, + 438, + 376, + 438 + ], + "score": 0.91, + "latex": "Q \\in \\mathbb { R } ^ { h \\times h }" + }, + { + "category_id": 13, + "poly": [ + 413, + 1220, + 860, + 1220, + 860, + 1260, + 413, + 1260 + ], + "score": 0.91, + "latex": "\\| \\hat { \\pmb \\beta } ( \\infty ) \\| _ { 2 } = \\left\\| ( X X ^ { \\top } ) ^ { - 1 } X \\pmb y \\right\\| _ { 2 } = O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 712, + 1898, + 1121, + 1898, + 1121, + 1937, + 712, + 1937 + ], + "score": 0.91, + "latex": "\\| \\pmb { y } - X ^ { \\top } \\pmb { \\beta } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d )" + }, + { + "category_id": 13, + "poly": [ + 492, + 1723, + 572, + 1723, + 572, + 1756, + 492, + 1756 + ], + "score": 0.91, + "latex": "S O ( n )" + }, + { + "category_id": 13, + "poly": [ + 832, + 406, + 946, + 406, + 946, + 439, + 832, + 439 + ], + "score": 0.91, + "latex": "Q _ { i , i } ~ = ~ 0" + }, + { + "category_id": 13, + "poly": [ + 467, + 1014, + 548, + 1014, + 548, + 1043, + 467, + 1043 + ], + "score": 0.91, + "latex": "\\gamma _ { 1 } \\neq 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 1012, + 415, + 1012, + 415, + 1042, + 298, + 1042 + ], + "score": 0.91, + "latex": "n , d \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 346, + 1756, + 623, + 1756, + 623, + 1793, + 346, + 1793 + ], + "score": 0.91, + "latex": "\\| V z \\| _ { \\infty } = O ( \\log d / \\sqrt { d } )" + }, + { + "category_id": 13, + "poly": [ + 488, + 975, + 541, + 975, + 541, + 1013, + 488, + 1013 + ], + "score": 0.9, + "latex": "{ \\hat { \\boldsymbol { \\beta } } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 529, + 1134, + 600, + 1134, + 600, + 1161, + 529, + 1161 + ], + "score": 0.9, + "latex": "d < n" + }, + { + "category_id": 13, + "poly": [ + 421, + 446, + 597, + 446, + 597, + 482, + 421, + 482 + ], + "score": 0.9, + "latex": "t = \\mathbb { E } [ G \\phi ( G ) ] ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1263, + 357, + 1263, + 357, + 1297, + 298, + 1297 + ], + "score": 0.9, + "latex": "O ( 1 )" + }, + { + "category_id": 14, + "poly": [ + 542, + 1056, + 1156, + 1056, + 1156, + 1101, + 542, + 1101 + ], + "score": 0.9, + "latex": "\\| \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { 2 } = O _ { P } ( 1 ) ; \\quad \\| \\boldsymbol { X } ^ { \\top } \\hat { \\boldsymbol { \\beta } } ( t ) \\| _ { \\infty } = O _ { P } ( \\mathrm { p o l y l o g } d ) ." + }, + { + "category_id": 13, + "poly": [ + 902, + 1224, + 982, + 1224, + 982, + 1256, + 902, + 1256 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } < 1" + }, + { + "category_id": 13, + "poly": [ + 517, + 297, + 568, + 297, + 568, + 328, + 517, + 328 + ], + "score": 0.89, + "latex": "K _ { W }" + }, + { + "category_id": 13, + "poly": [ + 750, + 1312, + 897, + 1312, + 897, + 1345, + 750, + 1345 + ], + "score": 0.89, + "latex": "\\boldsymbol { X } = \\boldsymbol { U \\Sigma V } ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 401, + 1264, + 465, + 1264, + 465, + 1294, + 401, + 1294 + ], + "score": 0.89, + "latex": "t \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1354, + 1220, + 1354, + 1220, + 1384, + 1156, + 1384 + ], + "score": 0.88, + "latex": "i \\leq d" + }, + { + "category_id": 13, + "poly": [ + 905, + 843, + 940, + 843, + 940, + 874, + 905, + 874 + ], + "score": 0.88, + "latex": "\\mathcal { A } _ { \\epsilon }" + }, + { + "category_id": 13, + "poly": [ + 1278, + 522, + 1313, + 522, + 1313, + 552, + 1278, + 552 + ], + "score": 0.88, + "latex": "\\mathcal { A } _ { \\epsilon }" + }, + { + "category_id": 13, + "poly": [ + 386, + 886, + 451, + 886, + 451, + 914, + 386, + 914 + ], + "score": 0.87, + "latex": "c > 0" + }, + { + "category_id": 13, + "poly": [ + 1014, + 403, + 1400, + 403, + 1400, + 439, + 1014, + 439 + ], + "score": 0.86, + "latex": "r = \\mathbb { E } [ \\phi ( G ) ^ { 2 } ] - \\mathbb { E } [ \\phi ( G ) ] ^ { 2 } , s =" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1694, + 1336, + 1694, + 1336, + 1720, + 1312, + 1720 + ], + "score": 0.83, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 424, + 1178, + 932, + 1178, + 932, + 1219, + 424, + 1219 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { \\hat { \\pmb { \\beta } } ( t ) = \\left( I - \\exp ( - \\frac { t } { n } X X ^ { \\top } ) \\right) ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 986, + 1312, + 1112, + 1312, + 1112, + 1351, + 986, + 1351 + ], + "score": 0.82, + "latex": "\\Sigma = [ \\hat { \\Sigma } ; 0 ]" + }, + { + "category_id": 13, + "poly": [ + 297, + 446, + 406, + 446, + 406, + 482, + 297, + 482 + ], + "score": 0.82, + "latex": "\\mathbb { E } [ \\phi ( G ) ] ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1126, + 1311, + 1245, + 1311, + 1245, + 1347, + 1126, + 1347 + ], + "score": 0.78, + "latex": "\\hat { \\Sigma } \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 13, + "poly": [ + 895, + 1728, + 914, + 1728, + 914, + 1750, + 895, + 1750 + ], + "score": 0.77, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 705, + 447, + 848, + 447, + 848, + 480, + 705, + 480 + ], + "score": 0.74, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 1124, + 1723, + 1166, + 1723, + 1166, + 1751, + 1124, + 1751 + ], + "score": 0.73, + "latex": "V z" + }, + { + "category_id": 13, + "poly": [ + 1088, + 1724, + 1113, + 1724, + 1113, + 1750, + 1088, + 1750 + ], + "score": 0.69, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 767, + 1181, + 931, + 1181, + 931, + 1218, + 767, + 1218 + ], + "score": 0.66, + "latex": "( X X ^ { \\top } ) ^ { - 1 } X y" + }, + { + "category_id": 13, + "poly": [ + 735, + 438, + 1187, + 438, + 1187, + 495, + 735, + 495 + ], + "score": 0.62, + "latex": "d , h \\to \\infty , \\left\\| K _ { W } - \\tilde { K } _ { W } \\right\\| _ { F } \\leq \\log ^ { c } d a . s ." + }, + { + "category_id": 13, + "poly": [ + 1348, + 406, + 1403, + 406, + 1403, + 438, + 1348, + 438 + ], + "score": 0.26, + "latex": "s =" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1955.0, + 1402.0, + 1955.0, + 1402.0, + 1978.0, + 1380.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 221.0, + 614.0, + 221.0, + 614.0, + 269.0, + 293.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 423.0, + 1175.0, + 423.0, + 1222.0, + 293.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1175.0, + 1407.0, + 1175.0, + 1407.0, + 1222.0, + 933.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1215.0, + 412.0, + 1215.0, + 412.0, + 1262.0, + 293.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1215.0, + 901.0, + 1215.0, + 901.0, + 1262.0, + 861.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1215.0, + 1103.0, + 1215.0, + 1103.0, + 1262.0, + 983.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1258.0, + 400.0, + 1258.0, + 400.0, + 1300.0, + 358.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1258.0, + 636.0, + 1258.0, + 636.0, + 1300.0, + 466.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1258.0, + 835.0, + 1258.0, + 835.0, + 1300.0, + 824.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1691.0, + 1311.0, + 1691.0, + 1311.0, + 1726.0, + 296.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1691.0, + 1405.0, + 1691.0, + 1405.0, + 1726.0, + 1337.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1720.0, + 491.0, + 1720.0, + 491.0, + 1758.0, + 294.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1720.0, + 894.0, + 1720.0, + 894.0, + 1758.0, + 573.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1720.0, + 1087.0, + 1720.0, + 1087.0, + 1758.0, + 915.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1720.0, + 1123.0, + 1720.0, + 1123.0, + 1758.0, + 1114.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 1720.0, + 1403.0, + 1720.0, + 1403.0, + 1758.0, + 1167.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1755.0, + 345.0, + 1755.0, + 345.0, + 1794.0, + 293.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1755.0, + 859.0, + 1755.0, + 859.0, + 1794.0, + 624.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 395.0, + 375.0, + 395.0, + 375.0, + 445.0, + 291.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 395.0, + 579.0, + 395.0, + 579.0, + 445.0, + 515.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 395.0, + 831.0, + 395.0, + 831.0, + 445.0, + 772.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 395.0, + 1013.0, + 395.0, + 1013.0, + 445.0, + 947.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 395.0, + 1407.0, + 395.0, + 1407.0, + 445.0, + 1404.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 434.0, + 296.0, + 434.0, + 296.0, + 494.0, + 289.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 434.0, + 420.0, + 434.0, + 420.0, + 494.0, + 407.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 434.0, + 704.0, + 434.0, + 704.0, + 494.0, + 598.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 434.0, + 1208.0, + 434.0, + 1208.0, + 494.0, + 1188.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 835.0, + 371.0, + 835.0, + 371.0, + 881.0, + 294.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 835.0, + 904.0, + 835.0, + 904.0, + 881.0, + 751.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 835.0, + 1110.0, + 835.0, + 1110.0, + 881.0, + 941.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 835.0, + 1405.0, + 835.0, + 1405.0, + 881.0, + 1287.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 877.0, + 385.0, + 877.0, + 385.0, + 930.0, + 292.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 877.0, + 667.0, + 877.0, + 667.0, + 930.0, + 452.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 877.0, + 969.0, + 877.0, + 969.0, + 930.0, + 958.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 514.0, + 674.0, + 514.0, + 674.0, + 559.0, + 291.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 514.0, + 1277.0, + 514.0, + 1277.0, + 559.0, + 1124.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 514.0, + 1405.0, + 514.0, + 1405.0, + 559.0, + 1314.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 550.0, + 783.0, + 550.0, + 783.0, + 588.0, + 295.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1305.0, + 749.0, + 1305.0, + 749.0, + 1355.0, + 292.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1305.0, + 985.0, + 1305.0, + 985.0, + 1355.0, + 898.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1305.0, + 1125.0, + 1305.0, + 1125.0, + 1355.0, + 1113.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1305.0, + 1295.0, + 1305.0, + 1295.0, + 1355.0, + 1246.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1305.0, + 1411.0, + 1305.0, + 1411.0, + 1355.0, + 1399.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1347.0, + 398.0, + 1347.0, + 398.0, + 1390.0, + 294.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1347.0, + 842.0, + 1347.0, + 842.0, + 1390.0, + 766.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1347.0, + 1155.0, + 1347.0, + 1155.0, + 1390.0, + 1128.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1347.0, + 1411.0, + 1347.0, + 1411.0, + 1390.0, + 1221.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 975.0, + 487.0, + 975.0, + 487.0, + 1018.0, + 292.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 975.0, + 1407.0, + 975.0, + 1407.0, + 1018.0, + 542.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1009.0, + 297.0, + 1009.0, + 297.0, + 1045.0, + 293.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1009.0, + 466.0, + 1009.0, + 466.0, + 1045.0, + 416.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1009.0, + 785.0, + 1009.0, + 785.0, + 1045.0, + 549.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 714.0, + 1231.0, + 714.0, + 1231.0, + 753.0, + 294.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1897.0, + 711.0, + 1897.0, + 711.0, + 1941.0, + 293.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1897.0, + 1130.0, + 1897.0, + 1130.0, + 1941.0, + 1122.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1131.0, + 528.0, + 1131.0, + 528.0, + 1166.0, + 295.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1131.0, + 1373.0, + 1131.0, + 1373.0, + 1166.0, + 601.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 516.0, + 290.0, + 516.0, + 334.0, + 294.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 290.0, + 763.0, + 290.0, + 763.0, + 334.0, + 569.0, + 334.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 35, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 226, + 1406, + 226, + 1406, + 329, + 295, + 329 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1932, + 1406, + 1932, + 1406, + 2039, + 297, + 2039 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1554, + 1403, + 1554, + 1403, + 1658, + 297, + 1658 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 418, + 515, + 1279, + 515, + 1279, + 895, + 418, + 895 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 537, + 1228, + 1161, + 1228, + 1161, + 1386, + 537, + 1386 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 577, + 1861, + 1120, + 1861, + 1120, + 1924, + 577, + 1924 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 334, + 963, + 1303, + 963, + 1303, + 1151, + 334, + 1151 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 573, + 338, + 1152, + 338, + 1152, + 418, + 573, + 418 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 295, + 440, + 1404, + 440, + 1404, + 509, + 295, + 509 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 1391, + 1402, + 1391, + 1402, + 1455, + 298, + 1455 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 338, + 1719, + 1296, + 1719, + 1296, + 1809, + 338, + 1809 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 298, + 1156, + 1409, + 1156, + 1409, + 1221, + 298, + 1221 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 296, + 1678, + 755, + 1678, + 755, + 1711, + 296, + 1711 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 1818, + 650, + 1818, + 650, + 1850, + 297, + 1850 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 299, + 918, + 767, + 918, + 767, + 951, + 299, + 951 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.921 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1876, + 1401, + 1876, + 1401, + 1909, + 1337, + 1909 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1348, + 1400, + 1348, + 1400, + 1380, + 1337, + 1380 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1337, + 359, + 1401, + 359, + 1401, + 392, + 1337, + 392 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1337, + 839, + 1401, + 839, + 1401, + 871, + 1337, + 871 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1745, + 1401, + 1745, + 1401, + 1778, + 1337, + 1778 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1113, + 1401, + 1113, + 1401, + 1145, + 1338, + 1145 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1470, + 1402, + 1470, + 1402, + 1500, + 1374, + 1500 + ], + "score": 0.796 + }, + { + "category_id": 14, + "poly": [ + 416, + 515, + 1279, + 515, + 1279, + 899, + 416, + 899 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\quad \\displaystyle \\left\\| \\frac { \\partial L ( W _ { 1 } ) } { \\partial W _ { 1 } } - \\frac { \\partial L ( W _ { 2 } ) } { \\partial W _ { 2 } } \\right\\| _ { F } } \\\\ & { = \\displaystyle \\left\\| \\frac { 1 } { n } X \\left[ ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right] - \\frac { 1 } { n } X \\left[ ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right] \\right\\| _ { F } } \\\\ & { \\leq \\displaystyle \\frac { 1 } { n } \\| X \\| _ { 2 } \\left\\| ( y - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - ( y - y _ { 2 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\leq \\ O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y _ { 2 } - y _ { 1 } ) a ^ { \\top } \\circ \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) \\right\\| _ { F } } \\\\ & { \\quad + O \\left( \\frac { 1 } { \\sqrt { d } } \\right) \\left\\| ( y - y _ { 2 } ) a ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } W _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } W _ { 2 } ) ) \\right\\| _ { F } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 333, + 959, + 1305, + 959, + 1305, + 1149, + 333, + 1149 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\left\\| ( y _ { 2 } - y _ { 1 } ) { \\boldsymbol a } ^ { \\top } \\circ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) \\right\\| _ { 2 } \\overset { ( i ) } { \\leq } \\operatorname* { m a x } \\{ \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) _ { i j } \\} \\left\\| y _ { 2 } - y _ { 1 } \\right\\| _ { 2 } \\left\\| { \\boldsymbol a } \\right\\| _ { 2 } } \\\\ & { \\qquad \\overset { ( i i ) } { \\leq } O ( 1 ) \\left\\| \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 1 } ) { \\boldsymbol a } - \\phi ^ { \\prime } ( \\boldsymbol { X } ^ { \\top } \\boldsymbol { W } _ { 2 } ) { \\boldsymbol a } \\right\\| _ { F } } \\\\ & { \\qquad \\overset { ( i i i ) } { \\leq } O ( 1 ) \\left\\| \\boldsymbol { X } \\right\\| _ { 2 } \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\boldsymbol { W } _ { 1 } - \\boldsymbol { W } _ { 2 } \\right\\| _ { F } , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 340, + 1715, + 1296, + 1715, + 1296, + 1811, + 340, + 1811 + ], + "score": 0.94, + "latex": "K _ { i j } ( t ) = \\frac { \\partial f ( { \\pmb x } _ { i } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } ^ { \\top } \\frac { \\partial f ( { \\pmb x } _ { j } ; \\omega ( t ) ) } { \\partial \\omega ( t ) } = { \\pmb x } _ { i } ^ { \\top } { \\pmb x } _ { j } \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { i } ) \\phi ^ { \\prime } ( { \\pmb w } _ { k } ( t ) ^ { \\top } { \\pmb x } _ { j } ) ," + }, + { + "category_id": 14, + "poly": [ + 576, + 1858, + 1120, + 1858, + 1120, + 1924, + 576, + 1924 + ], + "score": 0.94, + "latex": "K ( t ) = X ^ { \\top } X \\circ \\frac { 1 } { h } [ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) ] ." + }, + { + "category_id": 14, + "poly": [ + 536, + 1227, + 1163, + 1227, + 1163, + 1388, + 536, + 1388 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\qquad \\left\\| ( \\pmb { y } - \\pmb { y } _ { 2 } ) \\pmb { a } ^ { \\top } \\circ ( \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) ) \\right\\| _ { F } } \\\\ & { \\leq \\operatorname* { m a x } \\{ \\left| a _ { i } \\right| \\} \\left\\| \\pmb { y } - \\pmb { y } _ { 2 } \\right\\| _ { 2 } \\left\\| \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 1 } ) - \\phi ^ { \\prime } ( X ^ { \\top } \\pmb { W } _ { 2 } ) \\right\\| _ { F } } \\\\ & { \\overset { ( i ) } { \\leq } O ( 1 ) \\left\\| X \\right\\| _ { 2 } \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } = O ( \\sqrt { d } ) \\left\\| \\pmb { W } _ { 1 } - \\pmb { W } _ { 2 } \\right\\| _ { F } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 478, + 442, + 693, + 442, + 693, + 478, + 478, + 478 + ], + "score": 0.93, + "latex": "\\pmb { y } _ { 1 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 1 } ) \\pmb { a }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1934, + 682, + 1934, + 682, + 1967, + 500, + 1967 + ], + "score": 0.92, + "latex": "\\mathbf { { x } } _ { i } ~ \\sim ~ N ( 0 , I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 750, + 441, + 965, + 441, + 965, + 478, + 750, + 478 + ], + "score": 0.92, + "latex": "\\pmb { y } _ { 2 } = \\phi ( \\pmb { X } ^ { \\top } \\pmb { W } _ { 2 } ) \\pmb { a }" + }, + { + "category_id": 13, + "poly": [ + 906, + 229, + 1199, + 229, + 1199, + 265, + 906, + 265 + ], + "score": 0.92, + "latex": "\\| { \\pmb y } - { \\pmb f } ( { \\pmb X } ) \\| _ { 2 } \\ = \\ { \\cal O } ( { \\sqrt { n } } )" + }, + { + "category_id": 14, + "poly": [ + 570, + 337, + 1153, + 337, + 1153, + 417, + 570, + 417 + ], + "score": 0.91, + "latex": "\\left\\| \\frac { \\partial L ( X ; W ) } { \\partial W } - \\frac { \\partial L ( X ; W ^ { \\prime } ) } { \\partial W } \\right\\| _ { F } \\leq L \\left\\| W - W ^ { \\prime } \\right\\| _ { F } ." + }, + { + "category_id": 13, + "poly": [ + 480, + 2002, + 867, + 2002, + 867, + 2038, + 480, + 2038 + ], + "score": 0.91, + "latex": "\\lVert \\pmb { w } _ { k } ( t ) - \\pmb { w } _ { k } ( 0 ) \\rVert _ { 2 } ~ = ~ O ( d ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 828, + 1588, + 1194, + 1588, + 1194, + 1627, + 828, + 1627 + ], + "score": 0.91, + "latex": "\\| K ( t ) - K ( 0 ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )" + }, + { + "category_id": 13, + "poly": [ + 1013, + 444, + 1101, + 444, + 1101, + 477, + 1013, + 477 + ], + "score": 0.91, + "latex": "W _ { 1 } , W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 266, + 411, + 266, + 411, + 300, + 298, + 300 + ], + "score": 0.91, + "latex": "\\phi ( X W ) \\mathbf { a }" + }, + { + "category_id": 13, + "poly": [ + 1216, + 2003, + 1281, + 2003, + 1281, + 2035, + 1216, + 2035 + ], + "score": 0.9, + "latex": "i , j , k" + }, + { + "category_id": 13, + "poly": [ + 651, + 1158, + 1012, + 1158, + 1012, + 1193, + 651, + 1193 + ], + "score": 0.9, + "latex": "\\left\\| A \\circ B \\right\\| _ { F } \\leq \\operatorname* { m a x } \\{ | A _ { i j } | \\} \\left\\| B \\right\\| _ { F }" + }, + { + "category_id": 13, + "poly": [ + 1182, + 1970, + 1280, + 1970, + 1280, + 2000, + 1182, + 2000 + ], + "score": 0.9, + "latex": "\\epsilon _ { 1 } ~ > ~ 0" + }, + { + "category_id": 13, + "poly": [ + 297, + 1589, + 551, + 1589, + 551, + 1625, + 297, + 1625 + ], + "score": 0.9, + "latex": "{ \\pmb w } _ { i } ( 0 ) \\| _ { 2 } = O ( d ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1625, + 409, + 1625, + 409, + 1657, + 298, + 1657 + ], + "score": 0.9, + "latex": "\\epsilon ^ { \\prime } \\in \\Theta ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 527, + 262, + 840, + 262, + 840, + 301, + 527, + 301 + ], + "score": 0.89, + "latex": "a _ { i } \\sim \\mathrm { U n i f } \\{ - 1 / \\sqrt { h } , 1 / \\sqrt { h } \\}" + }, + { + "category_id": 13, + "poly": [ + 1292, + 1556, + 1405, + 1556, + 1405, + 1591, + 1292, + 1591 + ], + "score": 0.89, + "latex": "\\parallel { \\pmb w } _ { i } ( t ) -" + }, + { + "category_id": 13, + "poly": [ + 1296, + 230, + 1402, + 230, + 1402, + 265, + 1296, + 265 + ], + "score": 0.89, + "latex": "f ( X ) \\ =" + }, + { + "category_id": 13, + "poly": [ + 1128, + 1393, + 1155, + 1393, + 1155, + 1424, + 1128, + 1424 + ], + "score": 0.88, + "latex": "\\phi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1261, + 1159, + 1286, + 1159, + 1286, + 1190, + 1261, + 1190 + ], + "score": 0.88, + "latex": "\\phi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 625, + 1968, + 1057, + 1968, + 1057, + 2001, + 625, + 2001 + ], + "score": 0.88, + "latex": "\\operatorname* { P r } | \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } | < \\log d \\ \\leq \\ O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )" + }, + { + "category_id": 13, + "poly": [ + 746, + 1934, + 998, + 1934, + 998, + 1967, + 746, + 1967 + ], + "score": 0.88, + "latex": "{ \\pmb w } _ { k } ( 0 ) ~ \\sim ~ N ( 0 , d ^ { \\epsilon } I _ { d } )" + }, + { + "category_id": 13, + "poly": [ + 698, + 230, + 777, + 230, + 777, + 260, + 698, + 260 + ], + "score": 0.87, + "latex": "W , W ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 467, + 1191, + 487, + 1191, + 487, + 1220, + 467, + 1220 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1206, + 1939, + 1236, + 1939, + 1236, + 1965, + 1206, + 1965 + ], + "score": 0.84, + "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }" + }, + { + "category_id": 13, + "poly": [ + 965, + 2005, + 983, + 2005, + 983, + 2031, + 965, + 2031 + ], + "score": 0.78, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 638, + 1595, + 651, + 1595, + 651, + 1620, + 638, + 1620 + ], + "score": 0.37, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 657, + 300, + 690, + 300, + 690, + 326, + 657, + 326 + ], + "score": 0.33, + "latex": "W" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1476.0, + 1401.0, + 1476.0, + 1401.0, + 1498.0, + 1379.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 225.0, + 697.0, + 225.0, + 697.0, + 267.0, + 292.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 225.0, + 905.0, + 225.0, + 905.0, + 267.0, + 778.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 225.0, + 1295.0, + 225.0, + 1295.0, + 267.0, + 1200.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 266.0, + 526.0, + 266.0, + 526.0, + 302.0, + 412.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 266.0, + 1404.0, + 266.0, + 1404.0, + 302.0, + 841.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 297.0, + 656.0, + 297.0, + 656.0, + 332.0, + 295.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 297.0, + 993.0, + 297.0, + 993.0, + 332.0, + 691.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1929.0, + 499.0, + 1929.0, + 499.0, + 1970.0, + 292.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1929.0, + 745.0, + 1929.0, + 745.0, + 1970.0, + 683.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1929.0, + 1205.0, + 1929.0, + 1205.0, + 1970.0, + 999.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1929.0, + 1406.0, + 1929.0, + 1406.0, + 1970.0, + 1237.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1962.0, + 624.0, + 1962.0, + 624.0, + 2005.0, + 291.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1962.0, + 1181.0, + 1962.0, + 1181.0, + 2005.0, + 1058.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1962.0, + 1411.0, + 1962.0, + 1411.0, + 2005.0, + 1281.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1998.0, + 479.0, + 1998.0, + 479.0, + 2039.0, + 294.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1998.0, + 964.0, + 1998.0, + 964.0, + 2039.0, + 868.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1998.0, + 1215.0, + 1998.0, + 1215.0, + 2039.0, + 984.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1998.0, + 1406.0, + 1998.0, + 1406.0, + 2039.0, + 1282.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1551.0, + 1291.0, + 1551.0, + 1291.0, + 1593.0, + 292.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1584.0, + 296.0, + 1584.0, + 296.0, + 1632.0, + 291.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1584.0, + 637.0, + 1584.0, + 637.0, + 1632.0, + 552.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1584.0, + 827.0, + 1584.0, + 827.0, + 1632.0, + 652.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1584.0, + 1407.0, + 1584.0, + 1407.0, + 1632.0, + 1195.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1614.0, + 297.0, + 1614.0, + 297.0, + 1664.0, + 292.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1614.0, + 425.0, + 1614.0, + 425.0, + 1664.0, + 410.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 440.0, + 477.0, + 440.0, + 477.0, + 481.0, + 293.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 440.0, + 749.0, + 440.0, + 749.0, + 481.0, + 694.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 440.0, + 1012.0, + 440.0, + 1012.0, + 481.0, + 966.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 440.0, + 1405.0, + 440.0, + 1405.0, + 481.0, + 1102.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 472.0, + 1236.0, + 472.0, + 1236.0, + 512.0, + 294.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1390.0, + 1127.0, + 1390.0, + 1127.0, + 1429.0, + 294.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1390.0, + 1404.0, + 1390.0, + 1404.0, + 1429.0, + 1156.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1423.0, + 687.0, + 1423.0, + 687.0, + 1455.0, + 294.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1155.0, + 650.0, + 1155.0, + 650.0, + 1195.0, + 292.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1155.0, + 1260.0, + 1155.0, + 1260.0, + 1195.0, + 1013.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1155.0, + 1406.0, + 1155.0, + 1406.0, + 1195.0, + 1287.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1188.0, + 466.0, + 1188.0, + 466.0, + 1225.0, + 295.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1188.0, + 913.0, + 1188.0, + 913.0, + 1225.0, + 488.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1677.0, + 757.0, + 1677.0, + 757.0, + 1713.0, + 295.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1817.0, + 650.0, + 1817.0, + 650.0, + 1853.0, + 294.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 914.0, + 770.0, + 914.0, + 770.0, + 958.0, + 293.0, + 958.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 36, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1879, + 1406, + 1879, + 1406, + 2036, + 297, + 2036 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 318, + 1406, + 318, + 1406, + 419, + 298, + 419 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 372, + 966, + 1324, + 966, + 1324, + 1218, + 372, + 1218 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 538, + 435, + 1157, + 435, + 1157, + 529, + 538, + 529 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 606, + 619, + 1093, + 619, + 1093, + 708, + 606, + 708 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 298, + 224, + 1403, + 224, + 1403, + 303, + 298, + 303 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 322, + 779, + 1308, + 779, + 1308, + 872, + 322, + 872 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 382, + 1772, + 1257, + 1772, + 1257, + 1848, + 382, + 1848 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 299, + 1251, + 1403, + 1251, + 1403, + 1318, + 299, + 1318 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 291, + 887, + 1404, + 887, + 1404, + 954, + 291, + 954 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 297, + 726, + 1299, + 726, + 1299, + 762, + 297, + 762 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 545, + 1069, + 545, + 1069, + 598, + 297, + 598 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 817, + 73, + 817, + 106, + 298, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 298, + 1552, + 676, + 1552, + 676, + 1585, + 298, + 1585 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 298, + 1426, + 679, + 1426, + 679, + 1463, + 298, + 1463 + ], + "score": 0.92 + }, + { + "category_id": 9, + "poly": [ + 1337, + 465, + 1401, + 465, + 1401, + 497, + 1337, + 497 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1337, + 647, + 1401, + 647, + 1401, + 679, + 1337, + 679 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1338, + 808, + 1401, + 808, + 1401, + 841, + 1338, + 841 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1170, + 1401, + 1170, + 1401, + 1202, + 1337, + 1202 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 296, + 1692, + 1400, + 1692, + 1400, + 1758, + 296, + 1758 + ], + "score": 0.889 + }, + { + "category_id": 8, + "poly": [ + 479, + 1599, + 1219, + 1599, + 1219, + 1640, + 479, + 1640 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1792, + 1400, + 1792, + 1400, + 1823, + 1337, + 1823 + ], + "score": 0.875 + }, + { + "category_id": 1, + "poly": [ + 292, + 1651, + 1386, + 1651, + 1386, + 1686, + 292, + 1686 + ], + "score": 0.843 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1332, + 1403, + 1332, + 1403, + 1362, + 1374, + 1362 + ], + "score": 0.787 + }, + { + "category_id": 1, + "poly": [ + 297, + 1493, + 993, + 1493, + 993, + 1529, + 297, + 1529 + ], + "score": 0.744 + }, + { + "category_id": 0, + "poly": [ + 297, + 1493, + 993, + 1493, + 993, + 1529, + 297, + 1529 + ], + "score": 0.232 + }, + { + "category_id": 14, + "poly": [ + 382, + 1770, + 1259, + 1770, + 1259, + 1847, + 382, + 1847 + ], + "score": 0.94, + "latex": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } ( \\hat { f } ) \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } ( \\hat { f } ) = ( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } ) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 371, + 963, + 1330, + 963, + 1330, + 1221, + 371, + 1221 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { ~ \\| K ( t ) - K ( 0 ) \\| _ { 2 } } \\\\ & { = \\left\\| X ^ { \\top } X \\circ \\frac { 1 } { h } \\left[ \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right] \\right\\| } \\\\ & { \\leq \\frac { 1 } { h } \\left\\| X ^ { \\top } X \\right\\| _ { 2 } \\operatorname* { m a x } \\left\\{ \\left| \\phi ^ { \\prime } ( X ^ { \\top } W ( t ) ) \\phi ^ { \\prime } ( W ( t ) ^ { \\top } X ) - \\phi ^ { \\prime } ( X ^ { \\top } W ( 0 ) ) \\phi ^ { \\prime } ( W ( 0 ) ^ { \\top } X ) \\right| _ { i j } \\right\\} } \\\\ & { \\leq \\frac { 1 } { h } O ( d ) O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 383, + 545, + 528, + 545, + 528, + 601, + 383, + 601 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\varepsilon = \\sqrt { \\frac { c \\log h } { h ^ { 1 + \\epsilon _ { 2 } } } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 326, + 776, + 1308, + 776, + 1308, + 873, + 326, + 873 + ], + "score": 0.94, + "latex": "\\left| \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( t ) ^ { \\top } \\pmb { x } _ { j } ) - \\sum _ { k = 1 } ^ { h } \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { i } ) \\phi ^ { \\prime } ( \\pmb { w } _ { k } ( 0 ) ^ { \\top } \\pmb { x } _ { j } ) \\right| = O ( h ^ { 1 / 2 - \\epsilon _ { 4 } } ) ," + }, + { + "category_id": 13, + "poly": [ + 1102, + 727, + 1197, + 727, + 1197, + 758, + 1102, + 758 + ], + "score": 0.94, + "latex": "1 - h ^ { - 3 }" + }, + { + "category_id": 14, + "poly": [ + 539, + 433, + 1160, + 433, + 1160, + 530, + 539, + 530 + ], + "score": 0.93, + "latex": "\\operatorname* { P r } \\left| \\frac { 1 } { h } \\sum _ { k = 1 } ^ { h } y _ { k } - \\mathbb { E } [ y _ { k } ] \\right| > \\varepsilon \\leq 2 \\exp \\left( - \\frac { h \\varepsilon ^ { 2 } } { 2 \\sigma ^ { 2 } + 2 \\varepsilon / 3 } \\right) ." + }, + { + "category_id": 14, + "poly": [ + 607, + 614, + 1095, + 614, + 1095, + 711, + 607, + 711 + ], + "score": 0.93, + "latex": "{ \\frac { 1 } { h } } \\sum _ { k = 1 } ^ { h } y _ { k } \\leq \\varepsilon + \\mathbb { E } [ y _ { k } ] = O \\left( { \\frac { \\mathrm { p o l y l o g } h } { h ^ { 1 / 2 + \\epsilon _ { 3 } } } } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 991, + 1692, + 1235, + 1692, + 1235, + 1728, + 991, + 1728 + ], + "score": 0.93, + "latex": "\\phi ^ { \\prime } ( { \\pmb x } ) + \\phi ^ { \\prime } ( - { \\pmb x } ) = C" + }, + { + "category_id": 13, + "poly": [ + 1059, + 227, + 1405, + 227, + 1405, + 268, + 1059, + 268 + ], + "score": 0.92, + "latex": "| \\phi ^ { \\prime } ( \\pmb { x } _ { i } ^ { \\top } \\pmb { w } _ { k } ( t ) ) \\phi ^ { \\prime } ( \\pmb { x } _ { j } ^ { \\top } \\pmb { w } _ { k } ( t ) ) -" + }, + { + "category_id": 13, + "poly": [ + 550, + 318, + 842, + 318, + 842, + 355, + 550, + 355 + ], + "score": 0.92, + "latex": "y _ { k } = \\mathbf { 1 } \\{ | x _ { i } ^ { \\top } w _ { k } | < \\log d \\}" + }, + { + "category_id": 13, + "poly": [ + 596, + 355, + 852, + 355, + 852, + 390, + 596, + 390 + ], + "score": 0.91, + "latex": "\\mathbb { E } [ y _ { k } ] = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )" + }, + { + "category_id": 13, + "poly": [ + 384, + 1726, + 530, + 1726, + 530, + 1757, + 384, + 1757 + ], + "score": 0.91, + "latex": "n , d , h \\infty" + }, + { + "category_id": 13, + "poly": [ + 965, + 558, + 1058, + 558, + 1058, + 588, + 965, + 588 + ], + "score": 0.91, + "latex": "1 - h ^ { - c }" + }, + { + "category_id": 13, + "poly": [ + 910, + 353, + 1397, + 353, + 1397, + 391, + 910, + 391 + ], + "score": 0.91, + "latex": "\\mathrm { V a r } [ y _ { k } ] = \\mathbb { E } [ y _ { k } ^ { 2 } ] - \\mathbb { E } [ y _ { k } ] ^ { 2 } = O ( 1 / d ^ { 1 / 2 + \\epsilon _ { 1 } } )" + }, + { + "category_id": 13, + "poly": [ + 893, + 1249, + 1351, + 1249, + 1351, + 1290, + 893, + 1290 + ], + "score": 0.91, + "latex": "\\| { \\pmb u } _ { N N } ( { \\hat { \\pmb x } } ) - { \\pmb u } _ { N T K } ( { \\hat { \\pmb x } } ) \\| _ { 2 } = O ( d ^ { 1 / 2 - \\epsilon ^ { \\prime } } )" + }, + { + "category_id": 13, + "poly": [ + 618, + 227, + 884, + 227, + 884, + 265, + 618, + 265 + ], + "score": 0.9, + "latex": "| x _ { j } ^ { \\top } w _ { k } ( 0 ) | > O ( \\log d )" + }, + { + "category_id": 13, + "poly": [ + 297, + 265, + 744, + 265, + 744, + 306, + 297, + 306 + ], + "score": 0.9, + "latex": "\\phi ^ { \\prime } ( { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) \\phi ^ { \\prime } ( { \\pmb x } _ { j } ^ { \\top } { \\pmb w } _ { k } ( 0 ) ) | = = O ( d ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 227, + 565, + 227, + 565, + 264, + 298, + 264 + ], + "score": 0.89, + "latex": "| { \\pmb x } _ { i } ^ { \\top } { \\pmb w } _ { k } ( 0 ) | > O ( \\log d )" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1605, + 1207, + 1605, + 1207, + 1638, + 1063, + 1638 + ], + "score": 0.88, + "latex": "\\pmb { x } _ { i } + \\pmb { x } _ { j } = 0" + }, + { + "category_id": 13, + "poly": [ + 764, + 890, + 790, + 890, + 790, + 922, + 764, + 922 + ], + "score": 0.88, + "latex": "\\phi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1098, + 888, + 1127, + 888, + 1127, + 917, + 1098, + 917 + ], + "score": 0.87, + "latex": "d ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 537, + 1917, + 567, + 1917, + 567, + 1944, + 537, + 1944 + ], + "score": 0.86, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 568, + 733, + 600, + 733, + 600, + 762, + 568, + 762 + ], + "score": 0.86, + "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { j } }" + }, + { + "category_id": 13, + "poly": [ + 487, + 734, + 517, + 734, + 517, + 759, + 487, + 759 + ], + "score": 0.85, + "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }" + }, + { + "category_id": 13, + "poly": [ + 433, + 325, + 463, + 325, + 463, + 352, + 433, + 352 + ], + "score": 0.85, + "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }" + }, + { + "category_id": 13, + "poly": [ + 1209, + 323, + 1227, + 323, + 1227, + 349, + 1209, + 349 + ], + "score": 0.84, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1317, + 1946, + 1345, + 1946, + 1345, + 1975, + 1317, + 1975 + ], + "score": 0.83, + "latex": "\\gamma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 802, + 733, + 828, + 733, + 828, + 759, + 802, + 759 + ], + "score": 0.83, + "latex": "c _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1379, + 1699, + 1398, + 1699, + 1398, + 1722, + 1379, + 1722 + ], + "score": 0.72, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 752, + 1603, + 872, + 1603, + 872, + 1637, + 752, + 1637 + ], + "score": 0.66, + "latex": "\\forall i \\in [ 1 , n ]" + }, + { + "category_id": 13, + "poly": [ + 751, + 1602, + 1014, + 1602, + 1014, + 1638, + 751, + 1638 + ], + "score": 0.62, + "latex": "\\forall i \\in [ 1 , n ] , \\exists ! j \\in [ 1 , n ]" + }, + { + "category_id": 13, + "poly": [ + 903, + 1603, + 1013, + 1603, + 1013, + 1637, + 903, + 1637 + ], + "score": 0.56, + "latex": "\\lfloor j \\in [ 1 , n ]" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1423.0, + 683.0, + 1423.0, + 683.0, + 1469.0, + 292.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1335.0, + 1404.0, + 1335.0, + 1404.0, + 1366.0, + 1376.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1494.0, + 995.0, + 1494.0, + 995.0, + 1531.0, + 295.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1881.0, + 1404.0, + 1881.0, + 1404.0, + 1915.0, + 296.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 536.0, + 1911.0, + 536.0, + 1947.0, + 294.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1911.0, + 1406.0, + 1911.0, + 1406.0, + 1947.0, + 568.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1941.0, + 1316.0, + 1941.0, + 1316.0, + 1978.0, + 292.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1978.0, + 1346.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1970.0, + 1408.0, + 1970.0, + 1408.0, + 2010.0, + 292.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1076.0, + 2002.0, + 1076.0, + 2035.0, + 295.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 317.0, + 432.0, + 317.0, + 432.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 317.0, + 549.0, + 317.0, + 549.0, + 356.0, + 464.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 317.0, + 1208.0, + 317.0, + 1208.0, + 356.0, + 843.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 317.0, + 1406.0, + 317.0, + 1406.0, + 356.0, + 1228.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 347.0, + 595.0, + 347.0, + 595.0, + 394.0, + 290.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 347.0, + 909.0, + 347.0, + 909.0, + 394.0, + 853.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 347.0, + 1409.0, + 347.0, + 1409.0, + 394.0, + 1398.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 386.0, + 584.0, + 386.0, + 584.0, + 423.0, + 296.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 222.0, + 297.0, + 222.0, + 297.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 222.0, + 617.0, + 222.0, + 617.0, + 269.0, + 566.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 222.0, + 1058.0, + 222.0, + 1058.0, + 269.0, + 885.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 262.0, + 296.0, + 262.0, + 296.0, + 304.0, + 293.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 262.0, + 755.0, + 262.0, + 755.0, + 304.0, + 745.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1249.0, + 892.0, + 1249.0, + 892.0, + 1291.0, + 292.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1249.0, + 1406.0, + 1249.0, + 1406.0, + 1291.0, + 1352.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1285.0, + 574.0, + 1285.0, + 574.0, + 1320.0, + 293.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 885.0, + 763.0, + 885.0, + 763.0, + 926.0, + 293.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 885.0, + 1097.0, + 885.0, + 1097.0, + 926.0, + 791.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 885.0, + 1405.0, + 885.0, + 1405.0, + 926.0, + 1128.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 918.0, + 534.0, + 918.0, + 534.0, + 954.0, + 293.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 725.0, + 486.0, + 725.0, + 486.0, + 764.0, + 293.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 725.0, + 567.0, + 725.0, + 567.0, + 764.0, + 518.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 725.0, + 801.0, + 725.0, + 801.0, + 764.0, + 601.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 725.0, + 1101.0, + 725.0, + 1101.0, + 764.0, + 829.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 725.0, + 1301.0, + 725.0, + 1301.0, + 764.0, + 1198.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 547.0, + 964.0, + 547.0, + 964.0, + 600.0, + 529.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 547.0, + 1071.0, + 547.0, + 1071.0, + 600.0, + 1059.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.25, + 556.5, + 461.25, + 556.5, + 461.25, + 596.5, + 294.25, + 596.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1552.0, + 678.0, + 1552.0, + 678.0, + 1588.0, + 296.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1689.0, + 990.0, + 1689.0, + 990.0, + 1732.0, + 292.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1689.0, + 1378.0, + 1689.0, + 1378.0, + 1732.0, + 1236.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1689.0, + 1406.0, + 1689.0, + 1406.0, + 1732.0, + 1399.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1721.0, + 383.0, + 1721.0, + 383.0, + 1761.0, + 294.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1650.0, + 1389.0, + 1650.0, + 1389.0, + 1690.0, + 295.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1494.0, + 995.0, + 1494.0, + 995.0, + 1531.0, + 295.0, + 1531.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 37, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1891, + 1404, + 1891, + 1404, + 2018, + 298, + 2018 + ], + "score": 0.978 + }, + { + "category_id": 8, + "poly": [ + 525, + 1585, + 1172, + 1585, + 1172, + 1748, + 525, + 1748 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 326, + 224, + 1368, + 224, + 1368, + 622, + 326, + 622 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 362, + 1231, + 1338, + 1231, + 1338, + 1504, + 362, + 1504 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 296, + 645, + 1408, + 645, + 1408, + 787, + 296, + 787 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 459, + 886, + 1240, + 886, + 1240, + 973, + 459, + 973 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 982, + 1405, + 982, + 1405, + 1047, + 298, + 1047 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 449, + 1796, + 1250, + 1796, + 1250, + 1872, + 449, + 1872 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 298, + 1513, + 1402, + 1513, + 1402, + 1576, + 298, + 1576 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 297, + 1754, + 666, + 1754, + 666, + 1786, + 297, + 1786 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 299, + 838, + 1227, + 838, + 1227, + 875, + 299, + 875 + ], + "score": 0.926 + }, + { + "category_id": 8, + "poly": [ + 299, + 1059, + 1432, + 1059, + 1432, + 1145, + 299, + 1145 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1649, + 1401, + 1649, + 1401, + 1681, + 1337, + 1681 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1338, + 912, + 1400, + 912, + 1400, + 945, + 1338, + 945 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1815, + 1400, + 1815, + 1400, + 1847, + 1338, + 1847 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1444, + 1400, + 1444, + 1400, + 1476, + 1337, + 1476 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1146, + 1402, + 1146, + 1402, + 1177, + 1337, + 1177 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 300, + 1187, + 755, + 1187, + 755, + 1220, + 300, + 1220 + ], + "score": 0.865 + }, + { + "category_id": 14, + "poly": [ + 359, + 1230, + 1339, + 1230, + 1339, + 1511, + 359, + 1511 + ], + "score": 0.96, + "latex": "\\begin{array} { l } { \\displaystyle \\frac { \\partial w _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h \\phi \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + h _ { 0 } \\phi \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\Big ) \\big ( \\phi ^ { \\prime } \\big ( w _ { + } ^ { \\top } x _ { i } \\big ) + \\phi ^ { \\prime } \\big ( - w _ { + } ^ { \\top } x _ { i } \\big ) \\big ) x _ { i } \\Big ] } \\\\ { \\displaystyle \\qquad = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { n _ { 0 } } \\Big [ \\Big ( y _ { i } - h _ { 0 } w _ { + } ^ { \\top } x _ { i } \\Big ) x _ { i } \\Big ] = \\frac { 1 } { 2 n _ { 0 } } X _ { 0 } y _ { 0 } - \\frac { 1 } { 2 n _ { 0 } } h _ { 0 } X _ { 0 } X _ { 0 } ^ { \\top } w _ { + } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 525, + 1580, + 1174, + 1580, + 1174, + 1750, + 525, + 1750 + ], + "score": 0.95, + "latex": "\\pmb { w } _ { + } ^ { ( t = \\infty ) } = - \\pmb { w } _ { - } ^ { ( t = \\infty ) } = \\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { 1 } { h _ { 0 } } ( X X ^ { \\top } ) ^ { - 1 } X \\pmb { y } , } & { \\gamma _ { 1 } < 0 . 5 , } \\\\ { \\displaystyle } \\\\ { \\displaystyle \\frac { 1 } { h _ { 0 } } X ( X ^ { \\top } X ) ^ { - 1 } \\pmb { y } , } & { \\gamma _ { 1 } > 0 . 5 . } \\end{array} \\right." + }, + { + "category_id": 14, + "poly": [ + 460, + 882, + 1238, + 882, + 1238, + 977, + 460, + 977 + ], + "score": 0.93, + "latex": "\\frac { \\partial \\pmb { w } _ { + } } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) + h _ { 0 } \\phi ( \\pmb { w } _ { - } ^ { \\top } \\pmb { x } _ { i } ) \\Big ) \\phi ^ { \\prime } ( \\pmb { w } _ { + } ^ { \\top } \\pmb { x } _ { i } ) \\pmb { x } _ { i } \\right] ," + }, + { + "category_id": 13, + "poly": [ + 746, + 1013, + 964, + 1013, + 964, + 1047, + 746, + 1047 + ], + "score": 0.93, + "latex": "\\phi ( x ) - \\phi ( - x ) = x" + }, + { + "category_id": 13, + "poly": [ + 768, + 841, + 952, + 841, + 952, + 875, + 768, + 875 + ], + "score": 0.93, + "latex": "X = [ X _ { 0 } , - X _ { 0 } ]" + }, + { + "category_id": 14, + "poly": [ + 449, + 1794, + 1251, + 1794, + 1251, + 1871, + 449, + 1871 + ], + "score": 0.92, + "latex": "R _ { ( \\gamma _ { 1 } < 0 . 5 ) } \\to \\frac { 2 \\gamma _ { 1 } } { 1 - 2 \\gamma _ { 1 } } \\sigma ^ { 2 } ; \\quad R _ { ( \\gamma _ { 1 } \\geq 0 . 5 ) } = \\left( 1 - \\frac { 1 } { 2 \\gamma _ { 1 } } \\right) r ^ { 2 } + \\frac { 1 } { 2 \\gamma _ { 1 } - 1 } \\sigma ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 310, + 1055, + 1428, + 1055, + 1428, + 1150, + 310, + 1150 + ], + "score": 0.91, + "latex": "\\frac { \\partial ( { \\pmb w } _ { + } ) } { \\partial t } + \\frac { \\partial ( { \\pmb w } _ { - } ) } { \\partial t } = \\frac { 1 } { 2 n _ { 0 } } \\sum _ { i = 1 } ^ { 2 n _ { 0 } } \\left[ \\Big ( y _ { i } - h _ { 0 } \\phi ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) + h _ { 0 } \\phi ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) \\Big ) ( \\phi ^ { \\prime } ( { \\pmb w } _ { + } ^ { \\top } { \\pmb x } _ { i } ) - \\phi ^ { \\prime } ( { \\pmb w } _ { - } ^ { \\top } { \\pmb x } _ { i } ) ) { \\pmb x } _ { i } \\right] = 0" + }, + { + "category_id": 13, + "poly": [ + 683, + 1194, + 726, + 1194, + 726, + 1221, + 683, + 1221 + ], + "score": 0.87, + "latex": "{ \\pmb w } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 497, + 987, + 542, + 987, + 542, + 1014, + 497, + 1014 + ], + "score": 0.86, + "latex": "{ \\pmb w } _ { - }" + }, + { + "category_id": 13, + "poly": [ + 437, + 1518, + 481, + 1518, + 481, + 1546, + 437, + 1546 + ], + "score": 0.85, + "latex": "{ \\pmb w } _ { - }" + }, + { + "category_id": 13, + "poly": [ + 755, + 678, + 783, + 678, + 783, + 704, + 755, + 704 + ], + "score": 0.84, + "latex": "\\gamma _ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 249.0, + 394.0, + 249.0, + 394.0, + 285.0, + 346.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 248.0, + 930.0, + 248.0, + 930.0, + 285.0, + 881.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 318.0, + 394.0, + 318.0, + 394.0, + 352.0, + 346.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 303.0, + 929.0, + 303.0, + 929.0, + 339.0, + 881.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 347.0, + 409.0, + 347.0, + 409.0, + 434.0, + 319.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 357.0, + 934.0, + 357.0, + 934.0, + 416.0, + 846.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 413.0, + 929.0, + 413.0, + 929.0, + 449.0, + 881.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 453.0, + 394.0, + 453.0, + 394.0, + 489.0, + 346.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 471.0, + 554.0, + 471.0, + 554.0, + 480.0, + 545.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 460.0, + 746.0, + 460.0, + 746.0, + 490.0, + 570.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 467.0, + 928.0, + 467.0, + 928.0, + 503.0, + 881.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 461.0, + 1254.0, + 461.0, + 1254.0, + 489.0, + 1115.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 491.0, + 557.0, + 491.0, + 557.0, + 506.0, + 539.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 484.0, + 803.0, + 484.0, + 803.0, + 514.0, + 571.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 494.0, + 1102.0, + 494.0, + 1102.0, + 504.0, + 1090.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 484.0, + 1337.0, + 484.0, + 1337.0, + 514.0, + 1116.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 508.0, + 809.0, + 508.0, + 809.0, + 538.0, + 545.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 516.0, + 949.0, + 516.0, + 949.0, + 526.0, + 939.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 507.0, + 1343.0, + 507.0, + 1343.0, + 537.0, + 1080.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 522.0, + 392.0, + 522.0, + 392.0, + 552.0, + 345.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 523.0, + 938.0, + 523.0, + 938.0, + 553.0, + 879.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 543.0, + 471.0, + 543.0, + 471.0, + 574.0, + 427.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 543.0, + 546.0, + 543.0, + 546.0, + 574.0, + 500.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 542.0, + 619.0, + 542.0, + 619.0, + 573.0, + 572.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 540.0, + 694.0, + 540.0, + 694.0, + 575.0, + 644.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 540.0, + 767.0, + 540.0, + 767.0, + 575.0, + 719.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 542.0, + 839.0, + 542.0, + 839.0, + 573.0, + 792.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 542.0, + 1008.0, + 542.0, + 1008.0, + 576.0, + 960.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 543.0, + 1080.0, + 543.0, + 1080.0, + 574.0, + 1034.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 539.0, + 1156.0, + 539.0, + 1156.0, + 574.0, + 1105.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 540.0, + 1230.0, + 540.0, + 1230.0, + 575.0, + 1179.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 540.0, + 1303.0, + 540.0, + 1303.0, + 575.0, + 1251.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 542.0, + 1371.0, + 542.0, + 1371.0, + 571.0, + 1326.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 564.0, + 647.0, + 564.0, + 647.0, + 593.0, + 555.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 564.0, + 1182.0, + 564.0, + 1182.0, + 593.0, + 1090.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 589.0, + 761.0, + 589.0, + 761.0, + 627.0, + 408.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 589.0, + 1318.0, + 589.0, + 1318.0, + 625.0, + 918.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.25, + 417.0, + 504.25, + 417.0, + 504.25, + 427.0, + 476.25, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 645.0, + 1409.0, + 645.0, + 1409.0, + 677.0, + 297.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 671.0, + 754.0, + 671.0, + 754.0, + 708.0, + 293.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 671.0, + 1406.0, + 671.0, + 1406.0, + 708.0, + 784.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 698.0, + 1409.0, + 698.0, + 1409.0, + 736.0, + 293.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 726.0, + 1406.0, + 726.0, + 1406.0, + 764.0, + 293.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 758.0, + 896.0, + 758.0, + 896.0, + 789.0, + 295.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1891.0, + 1404.0, + 1891.0, + 1404.0, + 1926.0, + 293.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1924.0, + 1405.0, + 1924.0, + 1405.0, + 1957.0, + 294.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1952.0, + 1406.0, + 1952.0, + 1406.0, + 1988.0, + 293.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1985.0, + 950.0, + 1985.0, + 950.0, + 2018.0, + 294.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1985.0, + 1405.0, + 1985.0, + 1405.0, + 2015.0, + 1374.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 980.0, + 496.0, + 980.0, + 496.0, + 1017.0, + 293.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 980.0, + 1405.0, + 980.0, + 1405.0, + 1017.0, + 543.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1011.0, + 745.0, + 1011.0, + 745.0, + 1051.0, + 293.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1011.0, + 1304.0, + 1011.0, + 1304.0, + 1051.0, + 965.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1510.0, + 436.0, + 1510.0, + 436.0, + 1552.0, + 292.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1510.0, + 1406.0, + 1510.0, + 1406.0, + 1552.0, + 482.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1543.0, + 543.0, + 1543.0, + 543.0, + 1579.0, + 294.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1754.0, + 665.0, + 1754.0, + 665.0, + 1790.0, + 297.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 835.0, + 767.0, + 835.0, + 767.0, + 880.0, + 292.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 835.0, + 1229.0, + 835.0, + 1229.0, + 880.0, + 953.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1184.0, + 682.0, + 1184.0, + 682.0, + 1225.0, + 294.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1184.0, + 757.0, + 1184.0, + 757.0, + 1225.0, + 727.0, + 1225.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 38, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 456, + 1405, + 456, + 1405, + 591, + 297, + 591 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 295, + 1404, + 295, + 1404, + 424, + 298, + 424 + ], + "score": 0.974 + }, + { + "category_id": 0, + "poly": [ + 300, + 225, + 645, + 225, + 645, + 262, + 300, + 262 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 817, + 74, + 817, + 105, + 299, + 105 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2112, + 834, + 2112 + ], + "score": 0.846 + }, + { + "category_id": 13, + "poly": [ + 1109, + 527, + 1270, + 527, + 1270, + 563, + 1109, + 563 + ], + "score": 0.94, + "latex": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }" + }, + { + "category_id": 13, + "poly": [ + 581, + 357, + 745, + 357, + 745, + 395, + 581, + 395 + ], + "score": 0.93, + "latex": "\\beta = - \\mathbf { 1 } _ { d } / \\sqrt { d }" + }, + { + "category_id": 13, + "poly": [ + 865, + 361, + 953, + 361, + 953, + 395, + 865, + 395 + ], + "score": 0.93, + "latex": "( \\gamma _ { 1 } , \\gamma _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 358, + 457, + 358, + 457, + 395, + 298, + 395 + ], + "score": 0.92, + "latex": "F ( { \\pmb x } ) = { \\pmb x } ^ { \\top } \\beta" + }, + { + "category_id": 13, + "poly": [ + 1018, + 487, + 1312, + 487, + 1312, + 527, + 1018, + 527 + ], + "score": 0.92, + "latex": "\\| \\nabla _ { W } f ( X , W ) \\| _ { F } ^ { 2 } < 1 0 ^ { - 6 }" + }, + { + "category_id": 13, + "poly": [ + 856, + 332, + 925, + 332, + 925, + 359, + 856, + 359 + ], + "score": 0.89, + "latex": "\\gamma _ { 1 } , \\gamma _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 530, + 396, + 530, + 396, + 559, + 297, + 559 + ], + "score": 0.88, + "latex": "n = 3 2 0" + }, + { + "category_id": 13, + "poly": [ + 620, + 328, + 742, + 328, + 742, + 355, + 620, + 355 + ], + "score": 0.87, + "latex": "n = 1 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 303, + 493, + 401, + 493, + 401, + 525, + 303, + 525 + ], + "score": 0.85, + "latex": "( \\eta = 0 . 1 )" + }, + { + "category_id": 13, + "poly": [ + 500, + 534, + 569, + 534, + 569, + 562, + 500, + 562 + ], + "score": 0.81, + "latex": "\\gamma _ { 1 } , \\gamma _ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 224.0, + 647.0, + 224.0, + 647.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2120.0, + 829.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 455.0, + 1406.0, + 455.0, + 1406.0, + 493.0, + 296.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 489.0, + 302.0, + 489.0, + 302.0, + 528.0, + 291.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 489.0, + 1017.0, + 489.0, + 1017.0, + 528.0, + 402.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 489.0, + 1407.0, + 489.0, + 1407.0, + 528.0, + 1313.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 527.0, + 296.0, + 527.0, + 296.0, + 563.0, + 291.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 527.0, + 499.0, + 527.0, + 499.0, + 563.0, + 397.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 527.0, + 1108.0, + 527.0, + 1108.0, + 563.0, + 570.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 527.0, + 1405.0, + 527.0, + 1405.0, + 563.0, + 1271.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 559.0, + 1012.0, + 559.0, + 1012.0, + 593.0, + 295.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 293.0, + 1405.0, + 293.0, + 1405.0, + 332.0, + 293.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 325.0, + 619.0, + 325.0, + 619.0, + 362.0, + 293.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 325.0, + 855.0, + 325.0, + 855.0, + 362.0, + 743.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 325.0, + 1405.0, + 325.0, + 1405.0, + 362.0, + 926.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 356.0, + 297.0, + 356.0, + 297.0, + 399.0, + 292.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 356.0, + 580.0, + 356.0, + 580.0, + 399.0, + 458.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 356.0, + 864.0, + 356.0, + 864.0, + 399.0, + 746.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 356.0, + 1409.0, + 356.0, + 1409.0, + 399.0, + 954.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 386.0, + 360.0, + 386.0, + 360.0, + 427.0, + 293.0, + 427.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 39, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_layout.pdf b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..056aeb93462dfd6e9c61b326413068ad477a639a --- /dev/null +++ b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3e7514d984b6f24083ed3ca5a156c94f41d2122db26c14ef7c1ba2aae6f94b7 +size 680929 diff --git a/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_origin.pdf b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..65202fafa493dc5c3bbeee4b8a610bd773d10ef2 --- /dev/null +++ b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bcde428a6a3fa19a43ff5a17fb5b0a35671af300ad8d78273244136c538cd77 +size 535578 diff --git a/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_span.pdf b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..04a8aa36321f31b0e016bba347e1f9aa01d1383e --- /dev/null +++ b/parse/train/H4e7mBnC9f0/H4e7mBnC9f0_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:656008ac2c255758cf7bf4e6fbaa119f82a5eb8b78b8fa84992015215a46d580 +size 690282 diff --git a/parse/train/HJTzHtqee/HJTzHtqee_layout.pdf b/parse/train/HJTzHtqee/HJTzHtqee_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8f40f9f1d9b94137d4a355d34e5fc49f6cb0d002 --- /dev/null +++ b/parse/train/HJTzHtqee/HJTzHtqee_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89df5339954000c0cc7083e3bcb1fbd523239541b31271d57b57f88cc6cbfffb +size 472146 diff --git a/parse/train/HJTzHtqee/HJTzHtqee_origin.pdf b/parse/train/HJTzHtqee/HJTzHtqee_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bd0eeb2d1ca33f2e45b3c40fc1d26516f42700f2 --- /dev/null +++ b/parse/train/HJTzHtqee/HJTzHtqee_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2205e7b613a006c28d7eda2eb5141c1b4ab2dcc9b5ebbfe266fdf56a8707fb63 +size 356543 diff --git a/parse/train/HJTzHtqee/HJTzHtqee_span.pdf b/parse/train/HJTzHtqee/HJTzHtqee_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..839f6a3697e4c1ddd7d0b9a97c137d89d0cab584 --- /dev/null +++ b/parse/train/HJTzHtqee/HJTzHtqee_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f836518f6996898a105322a0c7571674a965546ffe292e59dcfa8cbab8164d97 +size 476102 diff --git a/parse/train/HJYoqzbC-/HJYoqzbC-_layout.pdf b/parse/train/HJYoqzbC-/HJYoqzbC-_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6b0be4052727885eed8b4914be56bfa8ff0601f8 --- /dev/null +++ b/parse/train/HJYoqzbC-/HJYoqzbC-_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e78ae3d60e0aeaea669b10fccb7bd6613c91fa934319bf6330f79e85308fb065 +size 3054319 diff --git a/parse/train/HJYoqzbC-/HJYoqzbC-_origin.pdf b/parse/train/HJYoqzbC-/HJYoqzbC-_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..63975a8b50997657ada288e1c1bad32d093206e6 --- /dev/null +++ b/parse/train/HJYoqzbC-/HJYoqzbC-_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2138cf2730735f3a7649c66f0ec2879035543243e98934636ab4592ce1dac2eb +size 2972748 diff --git a/parse/train/HJYoqzbC-/HJYoqzbC-_span.pdf b/parse/train/HJYoqzbC-/HJYoqzbC-_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..98866f6f760bd98c77e961d1e16543d3a4fba80b --- /dev/null +++ b/parse/train/HJYoqzbC-/HJYoqzbC-_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9edc755f7d4af924da3a8565b283256c9676da144265a3e4e0d60cda9639943 +size 3051410 diff --git a/parse/train/HJeq43AqF7/HJeq43AqF7_layout.pdf b/parse/train/HJeq43AqF7/HJeq43AqF7_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3978f4137e0f6e570f4d746082b79b67c7258bf8 --- /dev/null +++ b/parse/train/HJeq43AqF7/HJeq43AqF7_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19f5cb9e5293290babb7692975469edd0fe65beb092e00c1ae211f64637eb3a6 +size 407651 diff --git a/parse/train/HJeq43AqF7/HJeq43AqF7_origin.pdf b/parse/train/HJeq43AqF7/HJeq43AqF7_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..95f558858b716dd426e48cd958c0b3799e975fe5 --- /dev/null +++ b/parse/train/HJeq43AqF7/HJeq43AqF7_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39d94256726f71070aaa7cc6f7f6d8bb04fdb90501ed2958bbaacef03396f17d +size 256695 diff --git a/parse/train/HJeq43AqF7/HJeq43AqF7_span.pdf b/parse/train/HJeq43AqF7/HJeq43AqF7_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..909cb735c4cacf20887316e2d421cc3e6a2b48f3 --- /dev/null +++ b/parse/train/HJeq43AqF7/HJeq43AqF7_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b03e1065acb4f8477f67c978906d8a013e78ef2a524dfc3a4bee6fdaa9a1ba51 +size 405140 diff --git a/parse/train/HJgExaVtwr/HJgExaVtwr_layout.pdf b/parse/train/HJgExaVtwr/HJgExaVtwr_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..cc84adb6abaabd5d03ddece03a8baaae4087a30d --- /dev/null +++ b/parse/train/HJgExaVtwr/HJgExaVtwr_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42d163fda8ced87b786838ea54f1deb7ac9f883c226bed1038edc7c67fe4b7d8 +size 899969 diff --git a/parse/train/HJgExaVtwr/HJgExaVtwr_origin.pdf b/parse/train/HJgExaVtwr/HJgExaVtwr_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4ff932b48fbe88c987c24cc57ab7ef6e7896f787 --- /dev/null +++ b/parse/train/HJgExaVtwr/HJgExaVtwr_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3919260df9c90ac673ec45c983337d33965e55e3b2960cfc5f3fb8ded10810ec +size 709685 diff --git a/parse/train/HJgExaVtwr/HJgExaVtwr_span.pdf b/parse/train/HJgExaVtwr/HJgExaVtwr_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..822bffde79b8c43cf72239f30f616f6ee8228c70 --- /dev/null +++ b/parse/train/HJgExaVtwr/HJgExaVtwr_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f5075ef264f68c75d30cc5640297972f1f528e9ecfd8f97f3cb21c057e568f9 +size 898929 diff --git a/parse/train/HOFxeCutxZR/HOFxeCutxZR_layout.pdf b/parse/train/HOFxeCutxZR/HOFxeCutxZR_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..812251d6c6034d1a764851b5fa7ba2e4d9f7de05 --- /dev/null +++ b/parse/train/HOFxeCutxZR/HOFxeCutxZR_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d33c1208fd123de794cf0f36978881d25b6e887f5ba45a0515dc6cd4cbd3766 +size 38094844 diff --git a/parse/train/HOFxeCutxZR/HOFxeCutxZR_origin.pdf b/parse/train/HOFxeCutxZR/HOFxeCutxZR_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..08b7b45b7d08391fc4bbc2eefba31eff42329828 --- /dev/null +++ b/parse/train/HOFxeCutxZR/HOFxeCutxZR_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c502e0053552b7b5179f2f56506431b47a0e0a5d9c15119375cc53b396f1d9a1 +size 37959368 diff --git a/parse/train/HOFxeCutxZR/HOFxeCutxZR_span.pdf b/parse/train/HOFxeCutxZR/HOFxeCutxZR_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..92abfeddf2e3036dc2dfb54e1dc3ee4ea407f7d7 --- /dev/null +++ b/parse/train/HOFxeCutxZR/HOFxeCutxZR_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bca7c8a1bbfaca5538e09603d59ee0b7a3eafe6a8c4ace4c0eaa0846d2302e3 +size 38098834 diff --git a/parse/train/Hkekl0NFPr/Hkekl0NFPr_layout.pdf b/parse/train/Hkekl0NFPr/Hkekl0NFPr_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..93c88de7ad94ebeb7e5b901a6b1f279e04018bb9 --- /dev/null +++ b/parse/train/Hkekl0NFPr/Hkekl0NFPr_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4071cc9cbff3b1b072bb9a65482bd551ec8365c9fa679593cc4ae21a7453df69 +size 1149624 diff --git a/parse/train/Hkekl0NFPr/Hkekl0NFPr_origin.pdf b/parse/train/Hkekl0NFPr/Hkekl0NFPr_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..dd33de484a41b72b8c49b9d2c57e016ab1d8ab0f --- /dev/null +++ b/parse/train/Hkekl0NFPr/Hkekl0NFPr_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fb732e7659df2bc27499523a2d1eb11c0cf4e776807a9f9edbd9af4df1440ad +size 833448 diff --git a/parse/train/Hkekl0NFPr/Hkekl0NFPr_span.pdf b/parse/train/Hkekl0NFPr/Hkekl0NFPr_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5b878094fb0dc11ede1e04c91142ec353d2ea506 --- /dev/null +++ b/parse/train/Hkekl0NFPr/Hkekl0NFPr_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9753a792ba9e4b4711ab5dc57d37282c23c3c28eabe981f336d2324dc0a67d1 +size 1153136 diff --git a/parse/train/Hkex2a4FPr/Hkex2a4FPr_layout.pdf b/parse/train/Hkex2a4FPr/Hkex2a4FPr_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fa9cae2ce6d317f2f8bd58eede46b7964ec3bc5b --- /dev/null +++ b/parse/train/Hkex2a4FPr/Hkex2a4FPr_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d1f47b16a84176dc388fa450efe7fc36d86cc0c5b0f740690d64421426dd048 +size 779101 diff --git a/parse/train/Hkex2a4FPr/Hkex2a4FPr_origin.pdf b/parse/train/Hkex2a4FPr/Hkex2a4FPr_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6b62660a2bae5d1f164c73d1b72945d5f3a1271a --- /dev/null +++ b/parse/train/Hkex2a4FPr/Hkex2a4FPr_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6740c91f09310a8c5f7000d530f7e48c2cd66fa9308fd2dcf02dd328036e7866 +size 561825 diff --git a/parse/train/Hkex2a4FPr/Hkex2a4FPr_span.pdf b/parse/train/Hkex2a4FPr/Hkex2a4FPr_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d41155d3fa0c960f14f82ab658d6700eabb5812d --- /dev/null +++ b/parse/train/Hkex2a4FPr/Hkex2a4FPr_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0da27fb97cfc5aee74fcbddbe978f7ee2bdcd4e6cd6728beb0c8bdae6a4cd37a +size 780230 diff --git a/parse/train/Hy6b4Pqee/Hy6b4Pqee_layout.pdf b/parse/train/Hy6b4Pqee/Hy6b4Pqee_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4122d42e4b8131bf0f45b6dfeb3865a1a585d3c9 --- /dev/null +++ b/parse/train/Hy6b4Pqee/Hy6b4Pqee_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:427cb7c6815903ba757036db7eb60b67f351e9e24657b9701703c0109f4a2b63 +size 610540 diff --git a/parse/train/Hy6b4Pqee/Hy6b4Pqee_origin.pdf b/parse/train/Hy6b4Pqee/Hy6b4Pqee_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..26d3c67f3b7cc14fc4ce8d5e86ac66d54f2c94dc --- /dev/null +++ b/parse/train/Hy6b4Pqee/Hy6b4Pqee_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3bbd6ed9603ddf70e1b10f0862ed79280a15dd1ba73f49ac9b7586f6e1403fbf +size 339494 diff --git a/parse/train/Hy6b4Pqee/Hy6b4Pqee_span.pdf b/parse/train/Hy6b4Pqee/Hy6b4Pqee_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..094feb97b6c5d55a39bc4b1f64644f2d71573057 --- /dev/null +++ b/parse/train/Hy6b4Pqee/Hy6b4Pqee_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:156d26d9a9564090edff5edea6583e89018fbc8beed4beba232085560cd21178 +size 610141 diff --git a/parse/train/HygrAR4tPS/HygrAR4tPS_layout.pdf b/parse/train/HygrAR4tPS/HygrAR4tPS_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..73bd7bfd975c2b9f83f0f73a4a3f161efc6273d8 --- /dev/null +++ b/parse/train/HygrAR4tPS/HygrAR4tPS_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dbed59e1f22b299fb94927b83912546d7e61db45cdc808bff15d342e2d4c848b +size 937279 diff --git a/parse/train/HygrAR4tPS/HygrAR4tPS_origin.pdf b/parse/train/HygrAR4tPS/HygrAR4tPS_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..81958eefc24faa1d66209dc9f872c37ec973848b --- /dev/null +++ b/parse/train/HygrAR4tPS/HygrAR4tPS_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f13126db5a9af57728f3de199257d372387fe37174cfb850d12a98576ea834f4 +size 571304 diff --git a/parse/train/HygrAR4tPS/HygrAR4tPS_span.pdf b/parse/train/HygrAR4tPS/HygrAR4tPS_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e2c2fdd748e392da8532504dc866e68704355920 --- /dev/null +++ b/parse/train/HygrAR4tPS/HygrAR4tPS_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70ff68a642f48747f480fd36b52d553dc027f0d75d2bf88b6ea2446dc168d994 +size 934439 diff --git a/parse/train/Ig53hpHxS4/Ig53hpHxS4.md b/parse/train/Ig53hpHxS4/Ig53hpHxS4.md new file mode 100644 index 0000000000000000000000000000000000000000..1382553c1711d583243cb99fa3be143bf03fad8f --- /dev/null +++ b/parse/train/Ig53hpHxS4/Ig53hpHxS4.md @@ -0,0 +1,374 @@ +# FLOWTRON: AN AUTOREGRESSIVE FLOW-BASED GENERATIVE NETWORK FOR TEXT-TO-SPEECH SYNTHESIS + +Rafael Valle, Kevin J. Shih, Ryan Prenger & Bryan Catanzaro +NVIDIA +rafaelvalle@nvidia.com + +# ABSTRACT + +In this paper we propose Flowtron: an autoregressive flow-based generative network for text-to-speech synthesis with style transfer and speech variation. Flowtron borrows insights from Autoregressive Flows and revamps Tacotron 2 in order to provide high-quality and expressive mel-spectrogram synthesis. Flowtron is optimized by maximizing the likelihood of the training data, which makes training simple and stable. Flowtron learns an invertible mapping of data to a latent space that can be used to modulate many aspects of speech synthesis (timbre, expressivity, accent). Our mean opinion scores (MOS) show that Flowtron matches state-ofthe-art TTS models in terms of speech quality. We provide results on speech variation, interpolation over time between samples and style transfer between seen and unseen speakers. Code and pre-trained models are publicly available at https://github.com/NVIDIA/flowtron. + +# 1 INTRODUCTION + +Current speech synthesis methods do not give the user enough control over how speech actually sounds. Automatically converting text to audio that successfully communicates the text was achieved a long time ago (Umeda et al., 1968; Badham et al., 1983). However, communicating only the text information leaves out the acoustic properties of the voice that convey much of the meaning and human expressiveness. In spite of this, the typical speech synthesis problem is formulated as a text to speech (TTS) problem in which the user inputs only text since the 1960s. This work proposes a normalizing flow model (Kingma & Dhariwal, 2018; Huang et al., 2018) that learns an unsupervised mapping from non-textual information to manipulable latent Gaussian distributions. + +Taming the non-textual information in speech is difficult because the non-textual is unlabeled. A voice actor may speak the same text with different emphasis or emotion based on context, but it is unclear how to label a particular reading. Without labels for the non-textual information, recent approaches (Shen et al., 2017; Arik et al., 2017a;b; Ping et al., 2017) have formulated speech synthesis as a TTS problem wherein the non-textual information is implicitly learned. Despite their success in recreating non-textual information in the training set, the user has limited insight and control over it. + +It is possible to formulate an unsupervised learning problem in such a way that the user can exploit the unlabeled characteristics of a data set. One way is to formulate the problem such that the data is assumed to have a representation in some latent space, and have the model learn that representation. This latent space can then be investigated and manipulated to give the user more control over the generative model’s output. Such approaches have been popular in image generation, allowing users to interpolate smoothly between images and to identify portions of the latent space that correlate with various features (Radford et al., 2015; Kingma & Dhariwal, 2018; Izmailov et al., 2019). + +Recent deep learning approaches to expressive speech synthesis have combined text and learned latent embeddings for non-textual information (Wang et al., 2018; Skerry-Ryan et al., 2018; Hsu et al., 2018; Habib et al., 2019; Sun et al., 2020). These approaches impose an undesirable paradox: they require making assumptions before hand about the dimensionality of the embeddings when the correct dimensionality can only be determined after the model is trained. Even then, these embeddings are not guaranteed to contain all the non-textual information it takes to reconstruct speech, often + +![](images/0c05d3d0dbb87076db69f42f212b7790e8ecfb5fe50a23eb3336e9f0a2277c75.jpg) + +![](images/55fc3e5bacc9b8755ed7ca12eb90609a6ff9ac878d555efc308b5f8adac3fd79.jpg) + +(a) Time-averaged $_ { z }$ -values from multiple samples from 3 speakers with different timbres. $^ +$ is the centroid computed over samples from the same speaker. + +(b) Time-averaged $_ { z }$ -values from multiple samples from 123 LibriTTS speakers. Each color represents a speaker. $^ +$ is Male (lower $F _ { 0 }$ , third quadrant). is Female (higher $F _ { 0 }$ , first quadrant). + +Figure 1: T-SNE plot showing Flowtron partitioning the $_ z$ -space according to acoustic characteristics. + +resulting in models with dummy or uninterpretable latent dimensions and not enough capacity, as the appendices in Wang et al. (2018); Skerry-Ryan et al. (2018); Hsu et al. (2018) confirm. + +Furthermore, most models are not able to manipulate speech characteristics over time due to fixedlength embeddings. Their assumption is that variable-length embeddings are not robust to text and speaker perturbations (Skerry-Ryan et al., 2018), which we show not to be the case. Finally, although VAEs and GANs (Sun et al., 2020; Habib et al., 2019; Hsu et al., 2018; Binkowski et al., 2019; ´ Akuzawa et al., 2018) provide a latent embedding that can be manipulated, they may be difficult to train, are limited to approximate latent variable prediction, and rely on an implicit generative model or ELBO estimate to perform MLE in the latent space (Kingma & Dhariwal, 2018; Lucic et al., 2018; Kingma et al., 2016). + +In this paper we propose Flowtron: an autoregressive flow-based generative network for melspectrogram synthesis with style transfer over time and speech variation. Flowtron learns an invertible function that maps a distribution over mel-spectrograms to a latent $_ z$ -space parameterized by a spherical Gaussian. Figure 1 shows that acoustic characteristics like timbre and $F _ { 0 }$ correlate with portions of the $_ z$ -space of Flowtron models trained without speaker embeddings. + +With our formalization, we can generate samples containing specific speech characteristics manifested in mel-space by finding and sampling the corresponding region in $_ z$ -space (Gambardella et al., 2019). Our formulation also allows us to impose a structure to the $_ z$ -space and to parametrize it with a Gaussian mixture, similar to Hsu et al. (2018). In our simplest setup, we generate samples with a zero mean spherical Gaussian prior and control the amount of variation by adjusting its variance. + +Compared to VAEs and GANs and their disadvantages enumerated in Kingma & Dhariwal (2018), manipulating a latent prior in Flowtron comes at no cost in speech quality nor optimization challenges. Flowtron is able to generalize and produce sharp mel-spectrograms, even at high $\sigma ^ { 2 }$ values, by simply maximizing the likelihood of the data while not requiring any additional Prenet or Postnet layer (Wang et al., 2017), nor compound loss functions required by most SOTA models (Shen et al., 2017; Ping et al., 2017; Skerry-Ryan et al., 2018; Wang et al., 2018; Binkowski et al., 2019). ´ + +In summary, Flowtron is optimized by maximizing the exact likelihood of the training data, which makes training simple and stable. Using normalizing flows, it learns an invertible mapping from data to latent space that can be manipulated to modulate many aspects of speech synthesis. Concurrent with this work are Glow-TTS (Kim et al., 2020) and Flow-TTS (Miao et al., 2020), both of which incorporate normalizing flows into the TTS task. Our work differs from these two in that Flowtron is an autoregressive architecture where we explore the use of flow to modulate speech and style variation. In contrast, Glow-TTS and Flow-TTS are parallel architectures that focus on inference speed. Our mean opinion scores (MOS) show that Flowtron matches SOTA TTS models in terms of speech quality. Further, we provide results on speech variation, interpolation between samples and interpolation between styles over time, and style transfer between seen and unseen speakers with equal or different sentences. We hope this work, the first to show evidence that normalizing flows can be used for expressive text-to-speech synthesis and style transfer, will further stimulate developments in normalizing flows. + +# 2 FLOWTRON + +Flowtron is an autoregressive flow that generates a sequence of mel-spectrogram frames. A normalizing flow generates samples by first sampling a latent variable from a known distribution $p ( z )$ , and applying a series of invertible transformations to produce a sample from the target distribution $p ( { \pmb x } )$ . These invertible transformations $f$ are known as steps of flow: + +$$ +\pmb { x } = \pmb { f } _ { 1 } \circ \pmb { f } _ { 2 } \circ . . . \pmb { f } _ { k } ( z ) +$$ + +Because each transformation is invertible, we can directly evaluate the exact log-likelihood of the target distribution $p ( { \pmb x } )$ using the change of variables: + +$$ +\begin{array} { l } { \log p _ { \theta } ( \pmb { x } ) = \log p _ { \theta } ( z ) + \displaystyle \sum _ { i = 1 } ^ { k } \log \vert \operatorname* { d e t } ( \pmb { J } ( \pmb { f } _ { i } ^ { - 1 } ( \pmb { x } ) ) ) \vert } \\ { z = \pmb { f } _ { k } ^ { - 1 } \circ \pmb { f } _ { k - 1 } ^ { - 1 } \circ . . . \pmb { f } _ { 1 } ^ { - 1 } ( \pmb { x } ) } \end{array} +$$ + +Where $\textbf { { J } }$ is the Jacobian of the inverse transform $f _ { i } ^ { - 1 } ( { \pmb x } )$ . By cleverly choosing the latent distribution $p ( z )$ and the invertible transformations, the exact log-likelihood becomes simple and tractable. + +# 2.1 LATENT DISTRIBUTIONS + +We consider two simple distributions for the latent distribution $_ z$ : a zero-mean spherical Gaussian and a mixture of spherical Gaussians with fixed or learnable parameters. + +$$ +z \sim \mathcal { N } ( z ; 0 , I ) \quad \mathrm { o r } \quad z \sim \sum _ { k } \hat { \phi } _ { k } \mathcal { N } ( z ; \hat { \mu } _ { k } , \hat { \Sigma } _ { k } ) +$$ + +The zero-mean spherical Gaussian has a simple log-likelihood. The mixture of the spherical Gaussians, has inherent clusters that might result in interesting aspects of the audio information. + +# 2.2 INVERTIBLE TRANSFORMATIONS + +Normalizing flows are typically constructed using coupling layers (Dinh et al., 2014; 2016; Kingma & Dhariwal, 2018). In our case, we use an autoregressive affine coupling layer (Dinh et al., 2016). The latent variable $_ z$ has the same number of dimensions and frames as the resulting mel-spectrogram sample. The previous frames $z _ { 1 : t - 1 }$ produce scale and bias terms, $\mathbf { } _ { s _ { t } }$ and $\mathbf { } _ { b _ { t } }$ respectively, that affine-transform the succeeding time step ${ \boldsymbol { z } } _ { t }$ : + +$$ +\begin{array} { c } { { ( \log s _ { t } , b _ { t } ) = N N ( z _ { 1 : t - 1 } , t e x t , s p e a k e r ) } } \\ { { f ( z _ { t } ) = ( z _ { t } - b _ { t } ) \div s _ { t } } } \\ { { f ^ { - 1 } ( z _ { t } ) = s _ { t } \odot z _ { t } + b _ { t } } } \end{array} +$$ + +Here, $N N ( )$ can be any autoregressive causal transformation (Shumway & Stoffer, 2017). The affine coupling layer is a reversible transformation, even though $N N ( )$ itself need not be invertible. We use a 0-vector for obtaining the scaling and bias terms what will affine transform $z _ { 1 }$ . This 0-vector constant also guarantees that the first $_ z$ is always known. + +![](images/a0d01da5631791f09e9056942672608e597522f1883e26e701c1ed9de4deae52.jpg) +Figure 2: Mapping data $\mathbf { \rho } ( \mathbf { x } )$ to the latent dimension ${ \bf \Pi } ( { \bf z } )$ for K steps of flow. The right side shows a unrolled view of the $N N ( )$ architecture. Text and speaker embeddings are channel-wise concatenated to produce context matrix $c$ . At every time step $t$ , a recurrent attention mechanism computes a weighting distribution over the context matrix $c$ to produce a weighted-sum reduction over $c$ , which is then passed through an LSTM-Conv decoder architecture to generate affine parameters for transforming $x _ { t + 1 } \to x _ { t + 1 } ^ { \prime }$ . As predicted parameters are always for the next time step, the first iteration is conditioned on a pre-defined 0-vector. + +With an affine coupling layer, only the $\mathbf { \delta } _ { s _ { t } }$ term changes the volume of the mapping and adds a change of variables term to the loss. This term also penalizes the model for non-invertible affine mappings. + +$$ +\log | \operatorname* { d e t } ( J ( f _ { c o u p l i n g } ^ { - 1 } ( \pmb { x } ) ) ) | = \log | s | +$$ + +To evaluate the likelihood, we take the mel-spectrograms and pass them through the inverse steps of flow conditioned on the text and optional speaker ids, adding the corresponding $\log | s |$ penalties, and evaluate the result based on the Gaussian likelihoods. + +With this setup, it is also possible to reverse the ordering of the mel-spectrogram frames in time without loss of generality. We reverse the order of frames on even steps of flow, defining a step of flow as a full pass over the input sequence. This allows the model to learn dependencies both forward and backwards in time while remaining causal and invertible. + +# 2.3 MODEL ARCHITECTURE + +Our text encoder modifies the text encoder in Tacotron 2 by replacing batch-norm with instance-norm. Our decoder and $N N$ architecture, depicted in Figure 2, removes the Prenet and Postnet layers from Tacotron previously thought to be essential (Shen et al., 2017). Please compare Figure 2 describing our architecture and Figure 8 in A.4.4 describing Tacotron’s architecture. We also provide model summary views in A.6 + +We use the content-based tanh attention described in Vinyals et al. (2015), which can be easily modified to become also location sensitive. We use the Mel Encoder described in Hsu et al. (2018) to predict the parameters of the Gaussian Mixture. Following (Valle et al., 2019b), we use speakerembeddings channel-wise concatenated with the encoder outputs at every token. We use a single shared embedding for models not conditioned on speaker id. + +The step of flow closest to the latent variable $_ { z }$ has a gating mechanism that prunes extra frames from the $_ z$ -values provided to the model during inference. The length of $_ z$ -values remains fixed on the next steps of flow. + +# 2.4 INFERENCE + +Inference, given a trained model, is simply a matter of sampling $_ z$ values from a spherical Gaussian, or Gaussian Mixture, and running them through the network in the forward direction $f$ , e.g. Eq. 1. The parameters of the Gaussian mixture are either fixed or predicted by Flowtron. Training was conducted with $\sigma ^ { 2 } = 1$ , but we explore the effects of different values for $\sigma ^ { \bar { 2 } }$ in section 3.3. In general, we found that sampling $_ z$ from a Gaussian with lower standard deviation than used during training resulted in better sounding mel-spectrograms, as similarly concluded in Kingma & Dhariwal (2018) and (Parmar et al., 2018). Our inference results use $\sigma ^ { \mathrm { 2 } } = 0 . 5$ while sampling the prior and the posterior variance while sampling the posterior. + +# 2.5 POSTERIOR INFERENCE + +Figure 1 shows that several speech characteristics present in mel-spectrograms are clustered into regions of the $_ z$ -space. Knowing this, we can treat the latent distribution as a prior $q ( z ) = \mathcal { N } ( 0 , I )$ and obtain a posterior over the latent space of the flow model $q ( z | \zeta _ { 1 : m } )$ conditioned on the evidence $\zeta _ { 1 : m }$ , which are $m$ data observations $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { i } }$ mapped to the latent space using $\zeta _ { i } = f ^ { - 1 } ( { \pmb x } _ { i } )$ . We can use a Gaussian likelihood function with covariance matrix $\Sigma$ to compute the posterior above analytically, $q ( z | \boldsymbol { \zeta } _ { 1 : m } ) = \mathcal { N } ( \mu _ { p } , \boldsymbol { \Sigma } _ { p } )$ . Following the approach in Gambardella et al. (2019), defining $\bar { \zeta }$ as the mean of $\zeta _ { i }$ and using $\lambda$ as a hyperparameter, we define the parameters of the posterior below. Please see A.2, Algorithm 1 and Gambardella et al. (2019) for implementation details and a full derivation. + +$$ +\pmb { \mu } _ { p } = \frac { \frac { m } { \lambda } \bar { \zeta } } { \frac { m } { \lambda } + 1 } \quad \pmb { \Sigma } _ { p } = \frac { 1 } { \frac { m } { \lambda } + 1 } \pmb { I } +$$ + +# 3 EXPERIMENTS + +This section describes our training setup and provides quantitative and qualitative results. Our quantitative results show that Flowtron has mean opinion scores that are comparable to the state of the art. Our qualitative results demonstrate many features that are either impossible or inefficient to achieve using Tacotron, Tacotron 2 GST and Tacotron GM-VAE. These features include variation control in speech, interpolation between samples, and style transfer over time. + +We decode all mel-spectrograms into waveforms with a WaveGlow (Prenger et al., 2019) model available on github (Valle et al., 2019a). This suggests that WaveGlow can be used as an universal decoder. In addition to our illustrated and quantitative results, we ask that the readers listen to Flowtron samples in our supplementary materials corresponding to our qualitative experiments. + +# 3.1 TRAINING SETUP + +We train Flowtron, Tacotron 2 and Tacotron 2 GST models using a dataset (LSH) that combines the LJSpeech dataset (Ito et al., 2017) with two proprietary single speaker datasets with 20 and 10 hours each (Sally and Helen). We also train a Flowtron model on the train-clean-100 subset of LibriTTS (Zen et al., 2019) with 123 speakers and 25 minutes on average per speaker. Speakers with less than 5 minutes of data and files that are larger than 10 seconds are filtered out. For each dataset, we use at least 180 samples for the validation set, and the remainder for the training set. + +The models are trained on uniformly sampled normalized text and ARPAbet encodings obtained from the CMU Pronouncing Dictionary (Weide, 1998). We do not perform any data augmentation. We adapt public Tacotron 2 and Tacotron 2 GST repos to include speaker embeddings as described in Section 2. We use the same mel-spectrogram representation used in WaveGlow (Prenger et al., 2019). We train Flowtron with a pre-trained text encoder, progressively adding steps of flow once the last step of flow has learned to attend to text. Flowtron models used in our experiments have 2 steps of flow. We forward readers to A.3 and A.4 for details on our training setup and ablation studies. + +# 3.2 MEAN OPINION SCORE (MOS) COMPARISON + +We use the LJS voice as a reference and compare MOS between real samples, samples from Flowtron with 2 steps of flow, and samples from Tacotron 2. Following guidelines in (Prenger et al., 2019), we crowd-sourced MOS tests on Amazon Mechanical Turk using 30 volume normalized utterances disjoint from the training set for evaluation, and randomly chose the utterances for each subject. The scores provided in (Prenger et al., 2019) are used for real samples. + +The mean opinion scores are shown in Table 1 with $9 5 \%$ confidence intervals computed over approximately 250 scores per source. The results roughly match our subjective qualitative assessment. The larger advantage of Flowtron is in the control over the amount of speech variation and the manipulation of the latent space. + +Table 1: Mean Opinion Scores + +
SourceFlowsMOS
Real14.27 ± 0.13
Flowtron23.66 ± 0.16
Tacotron 213.52 ± 0.17
+ +# 3.3 SAMPLING THE PRIOR + +The simplest approach to generating samples with Flowtron is to sample from a prior distribution $z \sim \mathcal { N } ( 0 , \sigma ^ { 2 } )$ and adjust $\sigma ^ { 2 }$ to control the amount of variation. Whereas $\sigma ^ { 2 } = { \overset { \cdot } { 0 } }$ completely removes variation and produces outputs based on the model bias, increasing $\sigma ^ { 2 }$ will increase the amount of variation in speech. + +# 3.3.1 SPEECH VARIATION + +We illustrate the relationship between $\sigma ^ { 2 }$ and control over variability by synthesizing Flowtron samples with $\sigma ^ { 2 } \in \{ 0 . 0 , 0 . 5 , 1 . 0 \}$ . All samples are generated conditioned on the speaker Sally and the text “How much variation is there?". Despite the variability added by increasing $\sigma ^ { 2 }$ , all Flowtron-synthesized samples produce high quality speech. + +Figure 3 shows that contrary to commonly held wisdom (Shen et al., 2017; Arik et al., 2017a;b; Ping et al., 2017; Skerry-Ryan et al., 2018; Wang et al., 2018; Binkowski et al., 2019), Flowtron generates ´ sharp harmonics and well resolved formants without a compound loss nor Prenet or Postnet layers. + +![](images/e7c8b410365c954bd6e9c021768b7bcaf98aa8346bf4d800c4fd74a32b969a41.jpg) +Figure 3: Flowtron Mel-spectrograms illustrate increasing variability by using different $\sigma ^ { 2 }$ and that Flowtron is able to produce sharp harmonics with high $\sigma ^ { \tilde { 2 } }$ and without Prenet or Postnet layers. + +Now we show that adjusting $\sigma ^ { 2 }$ is a simple and valuable approach that provides more variation and control thereof than Tacotron, without sacrificing speech quality and despite of having a similar but simpler architecture. For this, we synthesize 10 samples with Tacotron 2 using different values for the Prenet dropout probability $p \in \{ 0 . 4 5 , 0 . 5 , 0 . 5 5 \}$ , scaling outputs accordingly. Samples computed on values of $p \in [ 0 . 3 , 0 . 8 ]$ are not included because they sound unintelligible. + +Figure 4 provides plots of $F _ { 0 }$ contours extracted with the YIN algorithm (De Cheveigné & Kawahara, 2002), with minimum $F _ { 0 }$ , maximum $F _ { 0 }$ , and harmonicity threshold equal to $8 0 \mathrm { H z }$ , $4 0 0 \mathrm { H z }$ and 0.3 respectively. Our results are similar to the previous sample duration analysis. As expected, $\sigma ^ { 2 } = 0$ provides no variation in $F _ { 0 }$ contour1, while increasing $\sigma ^ { \bar { 2 } }$ will increase variation in $F _ { 0 }$ contours. + +Our results in Figure 4 also show that Flowtron samples are considerably less monotonous than the samples produced with Tacotron 2 at no cost and with a similar but simpler architecture. Whereas increasing $\sigma ^ { 2 }$ considerably increases variation in $F _ { 0 }$ , modifying $p$ barely produces any variation. This is valuable because expressive speech is associated with non-monotonic $F _ { 0 }$ contours. In A.1 we show similar results with respect to sentence duration. + +# 3.3.2 INTERPOLATION BETWEEN SAMPLES + +With Flowtron, we can perform interpolation in $_ { z }$ -space to achieve interpolation in mel-spectrogram space. This experiment evaluates Flowtron models with and without speaker embeddings. For the experiment with speaker embeddings, we choose the speaker Sally and the phrase $^ { 6 6 } I t$ is well known that deep generative models have a rich latent space.". We generate mel-spectrograms by sampling $z \sim \mathcal { N } ( 0 , 0 . 8 )$ twice and interpolating between them over 100 timesteps. For the experiment without speaker embeddings we interpolate between Sally and Helen using the phrase “We are testing this model.". + +First, we perform inference by sampling $z \sim \mathcal { N } ( 0 , 0 . 5 )$ until we find $_ z$ values, $_ { z _ { h } }$ and $z _ { s }$ , that produce mel-spectrograms with Helen’s and Sally’s voice respectively. We then generate samples by performing inference while linearly interpolating between $z _ { h }$ and $z _ { s }$ . Our same speaker interpolation samples show that Flowtron is able to interpolate between multiple samples while producing correct alignment maps. Our different speaker interpolation samples show that Flowtron is able to gradually and smoothly morph one voice into another. + +![](images/f30f9cb2befac46a0eb35ffb6087c92348c0aedb6f0ca808d2b43b9d9b9d1998.jpg) +Figure 4: $F _ { 0 }$ contours obtained from samples generated by Flowtron and Tacotron 2 with different values for $\sigma ^ { 2 }$ and $p$ . Flowtron provides more variability and expressivity than Tacotron 2. + +# 3.4 SAMPLING THE POSTERIOR (STYLE TRANSFER) + +We generate samples with Flowtron by sampling a posterior distribution conditioned on the evidence containing speech characteristics of interest, as described in 2.5 and Gambardella et al. (2019). Tacotron 2 GST Wang et al. (2018) has an equivalent posterior sampling approach. During inference, the model is conditioned on a weighted sum of global style tokens (posterior) queried through an embedding of existing audio samples (evidence). We evaluate Tacotron 2 GST using a single sample to query a style token, or multiple samples to compute an average style token. For complete results, please refer to audio samples in the supplemental material corresponding to the following sections. + +# 3.4.1 SEEN SPEAKER + +In this section we run two style transfer experiments: the first one (Expressive) uses samples with high variance in pitch, which we use as a proxy for comparing expressivity in speech; the second (High Pitch), uses samples with high average pitch. In these experiments, we provide comparisons between Pitch Mean and Pitch Standard Deviation from the Reference samples providing the style, a Flowtron Baseline and after style transfer using Flowtron Posterior and Tacotron 2 GST. + +Our experiments show that by sampling from the posterior or interpolating between the posterior and the Gaussian prior over time, Flowtron makes a monotonic speaker gradually sound more expressive. Architectures similar to Tacotron 2 GST with fixed-latent embeddings are not able to perform gradual changes in style over time. Table 2 provides pitch summary statistics computed over 5 phrases and 10 takes each and shows that Flowtron is overall closer to the reference providing the style than Tacotron 2 GST. Our supplemental materials also show that Tacotron 2 GST sentences are repetitive and contain vocal-fry like distortions. + +
Pitch MeanPitch Standard Deviation
ModelStyle ExpressiveHigh PitchSurprisedExpressiveHigh PitchSurprised
Reference53.655.258.24.52.31.0
FTA Posterior53.453.355.52.52.33.0
FTA Baseline53.152.652.82.21.91.9
Tacotron 2 GST51.753.651.62.02.41.7
+ +Table 2: Values closer to the Reference are better. Comparison between pitch (MIDI number) summary statistics from Reference providing the style, Flowtron with standard Gaussian prior (FTA Baseline) and samples after style transfer with Flowtron (FTA Posterior) and Tacotron 2 GST. Our results show that FTA Posterior is overall more effective than Tacotron 2 GST in emulating the Reference by better matching its pitch summary statistics. + +# 3.4.2 SEEN SPEAKER WITH UNSEEN STYLE + +We compare samples generated with Flowtron and Tacotron 2 GST to evaluate their ability to emulate a speaking style unseen during training of a speaker seen during training. While Sally’s data used during training consists of news article readings, the evaluation samples contain Sally’s interpretation of the somber and vampiresque novel, Born of Darkness (BOD). + +Our samples show that while Tacotron 2 GST fails to emulate the somber timbre in Born of Darkness, Flowtron succeeds in transferring not only the somber timbre, but also the low $F _ { 0 }$ and the long pauses associated with the narrative style. + +# 3.4.3 UNSEEN SPEAKER + +In this experiment we compare Flowtron and Tacotron 2 GST samples to evaluate their ability to emulate the speaking style of a speaker not seen during training. The styles comes from speaker ID 24 and her “surprised" samples in RAVDESS (Livingstone & Russo, 2018), a dataset with emotion labels. Table 2 shows that while the samples generated with Tacotron 2 GST are not able to emulate the high-pitched style from RAVDESS, Flowtron is able to make Sally sound high-pitched as in the “surprised" style. + +# 3.5 INTERPOLATION BETWEEN STYLES (PRIOR AND POSTERIOR) + +In this experiment we illustrate how to control the speaking style at inference time by adjusting the parameter $\lambda$ in Equation 9 to interpolate between a baseline style (prior) and a target style (posterior). We use a model trained on LibriTTS and use a single sample from Sally’s (unseen speaker) Born of Darkness dataset as evidence providing the target style. We synthesize posterior samples generated with Flowtron with $\lambda \in \{ 0 . 1 , \bar { 0 } . 6 6 6 , \bar { 1 . 0 } , 2 . 0 \}$ . Figure 5 reflects the interpolation in style as interpolation in spectral profiles. Our supplemental materials aurally reflect a similar interpolation in other non-textual characteristics. + +![](images/3aff932d2f12339fcd2f2fce3f3076915313ae7641a2cba1b3937ff1976cb0df.jpg) +Figure 5: Spectral profiles from the target style, from the Flowtron baseline generated using the prior, and from Flowtron samples generated using the posterior with different values for $\lambda$ . These images show that by decreasing the value of $\lambda$ we gradually move the spectral profile from the baseline style (prior) to the target style (posterior). + +# 3.6 SAMPLING THE GAUSSIAN MIXTURE + +In this last section we provide samples from Flowtron Gaussian Mixture (GM) and visualizations. We replicate the experiments in Tacotron GM-VAE (Hsu et al., 2018) to visualize how speakers are assigned to mixture components and provide samples in which we modulate speech characteristics by translating one of the dimensions of an individual mixture component. + +For these experiments, Flowtron GM-LibriTTS is trained on LibriTTS without speaker embeddings and a Gaussian mixture with 8 component with predicted mean, covariance and component assignment probabilities; Flowtron GM-LSH is trained on LSH with speaker embeddings and a Gaussian Mixture with 8 components, fixed mean and covariances and predicted component assignment probabilities. + +# 3.6.1 VISUALIZING ASSIGNMENTS + +We evaluate model interpretability on a subset of LibriTTS with 123 speakers and 1410 utterances, 180 of which come from the validation set. Following Hsu et al. (2018), each utterance is assigned to the component with the highest posterior probability a $\cdot \mathrm { g } \operatorname* { m a x } _ { k } p ( \hat { \phi } _ { k } \mid \mathbf { x } )$ . We obtain posterior probabilities per utterance by using the Mel Encoder described in Section 2.3 and averaging the predicted component assignment probabilities over time. Figure 6 suggests that information in each component of Flowtron GM-LibriTTS is gender dependent. + +e quant al. (201 the association between gender and mixture components with the The assignment consistency with respect to gender is defined as $\begin{array} { r } { \frac { 1 } { M } \sum _ { i = 1 } ^ { N } \sum _ { j = 1 } ^ { N _ { i } } \mathbb { 1 } y _ { i j } = } \end{array}$ $\hat { y } _ { i }$ $M$ is the component assignment of utterance $j$ from speaker $i$ , and ugge $\hat { y } _ { i }$ is the mode of ng that the co $\{ y _ { i j } \} _ { j = 1 } ^ { N _ { i } }$ . The assignment consistency in Flowtron GM-LibriTTS is group utterances by speaker and group speakers by gend $8 2 . 4 \%$ provide visualizations in Figure 6. + +![](images/e736b3bc02eb0a38f6cdff0b56f42799339468a72f5b49b7e13d7cb79dfaa76d.jpg) +Figure 6: Component assignments suggest that information in each component is gender dependent. + +# 3.6.2 TRANSLATING DIMENSIONS + +We use the model Flowtron GM-LSH and focus on translating one of the dimensions of a single mixture component by adding an offset. The samples in our supplementary material show that we are able to modulate specific speech characteristics like pitch and word duration. Although the samples generated by translating one the dimensions associated with pitch height have different pitch contours, they have the same duration. Similarly, our samples show that translating the dimension associated with length of the first word does not modulate the pitch of the first word. We provide visualizations in Figure 9 in A.5. + +# 4 CONCLUSION + +We propose a new text to mel-spectrogram synthesis model based on autoregressive flows that is optimized by maximizing the likelihood and allows for speech variation and style transfer. Our results show that samples generated with Flowtron achieve mean opinion scores similar to SOTA TTS models. We demonstrate that our model learns a latent space that stores non-textual information, supervised using only MLE. Flowtron is able to produce high quality speech with high variability by adjusting $\sigma ^ { 2 }$ . + +Our results show that the latent space over non-textual features that can be investigated and manipulated to give the user more control over the generative model’s output. We provide many examples that showcase this, including transferring the style from speakers seen and unseen during training to another speaker using sentences with similar or different text, and making a monotonic speaker sound more expressive. For future work, we are interested in using normalizing flows for few-shot speech synthesis, speech compression and in semi-supervised settings to exploit datasets with limited labels. + +# REFERENCES + +Kei Akuzawa, Yusuke Iwasawa, and Yutaka Matsuo. Expressive speech synthesis via modeling expressions with variational autoencoder. arXiv preprint arXiv:1804.02135, 2018. + +Sercan Arik, Mike Chrzanowski, Adam Coates, Gregory Diamos, Andrew Gibiansky, Yongguo Kang, Xian Li, John Miller, Andrew Ng, Jonathan Raiman, et al. Deep voice: Real-time neural text-to-speech. arXiv preprint arXiv:1702.07825, 2017a. + +Sercan Arik, Gregory Diamos, Andrew Gibiansky, John Miller, Kainan Peng, Wei Ping, Jonathan Raiman, and Yanqi Zhou. Deep voice 2: Multi-speaker neural text-to-speech. arXiv preprint arXiv:1705.08947, 2017b. + +John Badham, Lawrence Lasker, Walter F Parkes, Arthur B Rubinstein, Matthew Broderick, Dabney Coleman, and John Wood. Wargames, 1983. + +Mikołaj Binkowski, Jeff Donahue, Sander Dieleman, Aidan Clark, Erich Elsen, Norman Casagrande, ´ Luis C Cobo, and Karen Simonyan. High fidelity speech synthesis with adversarial networks. arXiv preprint arXiv:1909.11646, 2019. + +Alain De Cheveigné and Hideki Kawahara. Yin, a fundamental frequency estimator for speech and music. The Journal of the Acoustical Society of America, 111(4):1917–1930, 2002. + +Laurent Dinh, David Krueger, and Yoshua Bengio. Nice: Non-linear independent components estimation. arXiv preprint arXiv:1410.8516, 2014. + +Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real nvp. arXiv preprint arXiv:1605.08803, 2016. + +Andrew Gambardella, Atılım Güne¸s Baydin, and Philip HS Torr. Transflow learning: Repurposing flow models without retraining. arXiv preprint arXiv:1911.13270, 2019. + +Raza Habib, Soroosh Mariooryad, Matt Shannon, Eric Battenberg, RJ Skerry-Ryan, Daisy Stanton, David Kao, and Tom Bagby. Semi-supervised generative modeling for controllable speech synthesis. arXiv preprint arXiv:1910.01709, 2019. + +Wei-Ning Hsu, Yu Zhang, Ron J Weiss, Heiga Zen, Yonghui Wu, Yuxuan Wang, Yuan Cao, Ye Jia, Zhifeng Chen, Jonathan Shen, et al. Hierarchical generative modeling for controllable speech synthesis. arXiv preprint arXiv:1810.07217, 2018. + +Chin-Wei Huang, David Krueger, Alexandre Lacoste, and Aaron Courville. Neural autoregressive flows. arXiv preprint arXiv:1804.00779, 2018. + +Keith Ito et al. The LJ speech dataset, 2017. + +Pavel Izmailov, Polina Kirichenko, Marc Finzi, and Andrew Gordon Wilson. Semi-supervised learning with normalizing flows. arXiv preprint arXiv:1912.13025, 2019. + +Jaehyeon Kim, Sungwon Kim, Jungil Kong, and Sungroh Yoon. Glow-tts: A generative flow for text-to-speech via monotonic alignment search. arXiv preprint arXiv:2005.11129, 2020. + +Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. + +Diederik P Kingma and Prafulla Dhariwal. Glow: Generative flow with invertible 1x1 convolutions. arXiv preprint arXiv:1807.03039, 2018. + +Diederik P Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen, Ilya Sutskever, and Max Welling. Improved variational inference with inverse autoregressive flow. In Advances in Neural Information Processing Systems, pp. 4743–4751, 2016. + +Steven R Livingstone and Frank A Russo. The ryerson audio-visual database of emotional speech and song (ravdess): A dynamic, multimodal set of facial and vocal expressions in north american english. PloS one, 13(5), 2018. + +Mario Lucic, Karol Kurach, Marcin Michalski, Sylvain Gelly, and Olivier Bousquet. Are gans created equal? a large-scale study. In Advances in neural information processing systems, pp. 700–709, 2018. + +C. Miao, S. Liang, M. Chen, J. Ma, S. Wang, and J. Xiao. Flow-tts: A non-autoregressive network for text to speech based on flow. In ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 7209–7213, 2020. + +Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Łukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran. Image transformer. arXiv preprint arXiv:1802.05751, 2018. + +Wei Ping, Kainan Peng, Andrew Gibiansky, Sercan Arik, Ajay Kannan, Sharan Narang, Jonathan Raiman, and John Miller. Deep voice 3: 2000-speaker neural text-to-speech. arXiv preprint arXiv:1710.07654, 2017. + +Ryan Prenger, Rafael Valle, and Bryan Catanzaro. Waveglow: A flow-based generative network for speech synthesis. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 3617–3621. IEEE, 2019. + +Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. + +Jonathan Shen, Ruoming Pang, Ron J Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, RJ Skerry-Ryan, et al. Natural tts synthesis by conditioning wavenet on mel spectrogram predictions. arXiv preprint arXiv:1712.05884, 2017. + +Robert H Shumway and David S Stoffer. Time series analysis and its applications: with R examples. Springer, 2017. + +RJ Skerry-Ryan, Eric Battenberg, Ying Xiao, Yuxuan Wang, Daisy Stanton, Joel Shor, Ron J Weiss, Rob Clark, and Rif A Saurous. Towards end-to-end prosody transfer for expressive speech synthesis with tacotron. arXiv preprint arXiv:1803.09047, 2018. + +Guangzhi Sun, Yu Zhang, Ron J Weiss, Yuan Cao, Heiga Zen, and Yonghui Wu. Fully-hierarchical fine-grained prosody modeling for interpretable speech synthesis. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6264–6268. IEEE, 2020. + +Noriko Umeda, E Matsui, Torazo Suzuki, and Hiroshi Omura. Synthesis of fairy tales using an analog vocal tract. In Proceedings of 6th International Congress on Acoustics, pp. B159–162, 1968. + +Rafael Valle, Jason Li, Ryan Prenger, and Bryan Catanzaro. Mellotron github repo, 2019a. URL https://github.com/NVIDIA/mellotron. + +Rafael Valle, Jason Li, Ryan Prenger, and Bryan Catanzaro. Mellotron: Multispeaker expressive voice synthesis by conditioning on rhythm, pitch and global style tokens. arXiv preprint arXiv:1910.11997, 2019b. + +Oriol Vinyals, Łukasz Kaiser, Terry Koo, Slav Petrov, Ilya Sutskever, and Geoffrey Hinton. Grammar as a foreign language. In Advances in neural information processing systems, pp. 2773–2781, 2015. + +Yuxuan Wang, RJ Skerry-Ryan, Daisy Stanton, Yonghui Wu, Ron J Weiss, Navdeep Jaitly, Zongheng Yang, Ying Xiao, Zhifeng Chen, Samy Bengio, et al. Tacotron: A fully end-to-end text-to-speech synthesis model. arXiv preprint arXiv:1703.10135, 2017. + +Yuxuan Wang, Daisy Stanton, Yu Zhang, RJ Skerry-Ryan, Eric Battenberg, Joel Shor, Ying Xiao, Fei Ren, Ye Jia, and Rif A Saurous. Style tokens: Unsupervised style modeling, control and transfer in end-to-end speech synthesis. arXiv preprint arXiv:1803.09017, 2018. + +Robert L Weide. The CMU pronouncing dictionary. URL: http://www.speech.cs.cmu.edu/cgibin/cmudict, 1998. + +Heiga Zen, Viet Dang, Rob Clark, Yu Zhang, Ron J Weiss, Ye Jia, Zhifeng Chen, and Yonghui Wu. Libritts: A corpus derived from librispeech for text-to-speech. arXiv preprint arXiv:1904.02882, 2019. + +# A APPENDIX + +# A.1 SPEECH VARIATION + +Figure 7 provides plots from sample durations in seconds. Our results show that larger values of $\sigma ^ { 2 }$ produces samples with more variation in duration, whereas $\sigma ^ { 2 } = 0$ is fully deterministic. These results demonstrate that our latent space is able to model duration, which is a critical non-textual component to expressiveness in speech. + +![](images/a1d117d71899bdbad6fd981f8c746b3fe210e9afcef209fbdbb739e32933e0c9.jpg) +Figure 7: Sample duration given $\sigma ^ { 2 }$ and $p$ show that Flowtron provides more variation in sample duration than Tacotron. + +# A.2 POSTERIOR INFERENCE + +We generate posterior samples with Flowtron by sampling a posterior distribution conditioned on evidence containing speech characteristics of interest, as described in (Gambardella et al., 2019). We collect the evidence by performing a forward pass with Flowtron using with a speaker embedding, $( s \sim \mathcal { N } ( 0 , I ) )$ , the observed mel-spectrogram, and the text from a set of samples with the speech characteristics of interest. We use a specific speaker embedding when we want to factor out information about a specific speaker from $\zeta$ . + +Next, we compute $\bar { \zeta }$ by averaging $\zeta _ { i , k }$ over batch $( i )$ or over batch and time $( i , k )$ and use Equation 9 to compute the parameters of the posterior. When averaging over batch, we repeat the $\mathbf { Z }$ -values over the time dimension until they reach the desired length. We find in our experiments that averaging over batch is more efficient for transfering the style than averaging over batch and time. In all experiments, we select the best performing samples given $\lambda$ values between $m * 0 . 1$ and $m * 4$ , where $m$ is the number of samples in the evidence. While small $\lambda$ values move the mean of the posterior closer to the evidence and decreases its variance, large $\lambda$ values move the mean of the posterior closer to the prior and increase the variance. + +Once the parameters of the posterior distribution are computed, we can sample the posterior distribution and perform inference with the desired text and speaker. Algorithm 1 provides a description of posterior inference with Flowtron. + +# A.3 TRAINING DETAILS + +We use the ADAM (Kingma & Ba, 2014) optimizer with default parameters, 1e-4 learning rate and1e-6 weight decay for Flowtron and 1e-3 learning rate and 1e-5 weight decay for the other models,following Wang et al. (2017). We anneal the learning rate once the generalization error starts toplateau and stop training once the the generalization error stops significantly decreasing or startsincreasing. Flowtron models with 2 steps of flow were trained on the LSH dataset for approximately1000 epochs, then fine-tuned on LibriTTS for 500 epochs. Tacotron 2 and Tacotron 2 GST are trained for approximately 500 epochs. Each model is trained on a single NVIDIA DGX-1 with 8 GPUs. + +Algorithm 1: Flowtron Posterior inference + +Input : Trained Flowtron model $f$ , evidence audio samples ${ \boldsymbol { \mathbf { \mathit { x } } } } _ { 1 : m }$ Output : Posterior sample + +1 For each $m e l _ { i , k } , t e x t _ { i }$ , speakeri ∈ x1:m + +2 if average over batch then + +3 repeat each $\zeta _ { k }$ over the time dimension until target length is achieve +4 $\bar { \zeta _ { k } } \gets \mathrm { C o m p u }$ te $\zeta _ { i , k }$ average over batch $k$ +5 else +6 ¯ζ ← Compute $\zeta _ { i , k }$ average over batch and time $k$ +7 end +8 $\mu _ { p }$ , $\Sigma _ { p } $ Compute posterior parameters using Equation 9 +9 Initialize $\boldsymbol { Z } _ { p } \sim \mathcal { N } ( \pmb { \mu } _ { p } , \pmb { \Sigma } _ { p } )$ +10 Sample $z _ { p }$ from $Z _ { p }$ +11 Perform inference with Flowtron using $z _ { p }$ , text and speaker + +# A.4 ABLATION STUDIES + +# A.4.1 COMPOSING FLOWS + +We evaluated Flowtron models with 2, 3 and 6 steps of flow and found that more steps of flow have better likelihood but no significant qualitative improvement, while increasing inference time significantly. Hence, we chose to report results on Flowtron models with 2 steps of flow. + +# A.4.2 BIDIRECTIONAL PROCESSING + +We compared the bidirectional (reversing the ordering of the mel-spectrogram frames in time on even numbered steps of flows) and unidirectional processing and found that bidirectional processing provides better likelihood and audio quality. Hence, we use bidirectional processing in all our Flowtron models. + +# A.4.3 ADDITIVE VS AFFINE TRANSFORMATIONS + +The Tacotron 2 baseline without the postnet layer can be interpreted as additive single step autoregressive normalizing flow (ASSANF). By comparing Flowtron with Tacotron 2, we’re comparing with a model that is better than an (ASSANF), as Tacotron 2 sans Postnet does not have sharp harmonics. Hence, we prefer affine over additive transformations. + +# A.4.4 COMPARISON WITH TACOTRON 2 + +The Tacotron 2 baseline without the postnet layer can be interpreted as additive single step autoregressive normalizing flow (ASSANF). By comparing Flowtron with Tacotron 2, we’re comparing with a model that is better than an (ASSANF), as Tacotron 2 sans Postnet does not have sharp harmonics. Hence, we prefer affine over additive transformations. + +![](images/cefdc8f2171b60bd2119214eca1bde41063f24f86b63edf5cfab97efff387eb2.jpg) +Figure 8: Visualization of the decoder in Tacotron 2 during training. Unlike Flowtron, Tacotron 2 requires Prenet and Postnet layers to learn attention and produce sharp harmonics. + +![](images/62ef8251c7a9a869a84ca1e2925c939f73e3784e68dce95f76c876c37514717f.jpg) +Figure 9: (a) shows that by translating one of the dimensions of $_ z$ we are able to alter the pitch contour of the sentence while keeping the length fixed. (b) shows that by translation one of the dimensions of $_ z$ we are able to alter the length of the sentence while keeping a similar pitch contour. + +# A.6 FLOWTRON AND TACOTRON SUMMARY VIEW + +Flowtron( (speaker_embedding): Embedding(3, 128) (embedding): Embedding(185, 512) (flows): ModuleList( (0): AR_Step( (conv): Conv1d(1024, 160, kernel_size $=$ (1,), stride $=$ (1,)) (lstm): LSTM(1664, 1024, num_layers $^ { = 2 }$ ) (attention_lstm): LSTM(80, 1024) (attention_layer): Attention( (softmax): Softmax(dim $^ { = 2 }$ ) (query): LinearNorm( (linear_layer): Linear(in_features=1024, out_features $= 6 4 0$ , bias=False) ) (key): LinearNorm( (linear_layer): Linear(in_features $= 6 4 0$ , out_feature $_ { 3 } = 6 4 0$ , bias $\mathbf { \tau } = \dot { }$ False) ) (value): LinearNorm( (linear_layer): Linear(in_features $= 6 4 0$ , out_feature $\mathord { : } = 6 4 0$ , bias=False) ) (v): LinearNorm( (linear_layer): Linear(in_features $= 6 4 0$ , out_features $: = 1$ , bias=False) ) ) (dense_layer): DenseLayer( (layers): ModuleList( (0): LinearNorm( (linear_layer): Linear(in_features=1024, out_features=1024, bias=True) ) (1): LinearNorm( (linear_layer): Linear(in_feature $\mathord { \mathrm { 3 } } = 1 0 2 4$ , out_features $_ { \mathrm { : = 1 0 2 4 } }$ , bias $= "$ True) ) + +(1): AR_Back_Step( (ar_step): AR_Step( (conv): Conv1d(1024, 160, kernel_size $=$ (1,), stride $=$ (1,)) (lstm): LSTM(1664, 1024, num_layers $^ { = 2 }$ ) (attention_lstm): LSTM(80, 1024) (attention_layer): Attention( (softmax): Softmax(dim $^ { = 2 }$ ) (query): LinearNorm( (linear_layer): Linear(in_feature $\mathord { \mathrm { 3 } } = 1 0 2 4$ , out_feature $\mathtt { s } = 6 4 0$ , bias $=$ False) ) (key): LinearNorm( (linear_layer): Linear(in_feature $\mathord { : } = 6 4 0$ , out_features $= 6 4 0$ , bias ${ } = \mathbb { E }$ alse) (value): LinearNorm( (linear_layer): Linear(in_features $_ { : = 6 4 0 }$ , out_features $= 6 4 0$ , bias=False) ) (v): LinearNorm( (linear_layer): Linear(in_features $_ { : = 6 4 0 }$ , out_features $^ { = 1 }$ , bias $=$ False) ) ) (dense_layer): DenseLayer( (layers): ModuleList( (0): LinearNorm( (linear_layer): Linear(in_features $\mathbf { \Psi } = \mathbf { \dot { \Psi } }$ 1024, out_features $\mathbf { \Psi } = \mathbf { \dot { \Psi } }$ 1024, bias ${ } , = { }$ True) ) (1): LinearNorm( (linear_layer): Linear(in_features $= 1$ 1024, out_features $=$ 1024, bias $=$ True) ) (gate_layer): LinearNorm( (linear_layer): Linear(in_features $^ { = 1 }$ 664, out_features $^ { = 1 }$ , bias $= "$ True) ) +) +(encoder): Encoder( (convolutions): ModuleList( (0): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding=(2,)) ) (1): InstanceNorm1d(512, eps=1e-05, momentum=0.1, affine $=$ True, track_running_ ) + +(1): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding=(2,)) ) (1): InstanceNorm1d(512, ep $\mathrm { 3 } { = } 1 \mathrm { e } { - } 0 5$ , momentum $\ i { = } 0 \cdot 1$ , affine $= ^ { \prime }$ True, track_running_stats=False) ) (2): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding=(2,)) ) (1): InstanceNorm1d(512, eps $=$ 1e-05, momentum $\ i { = } 0 \cdot 1$ , affine $=$ True, track_running_stats=False) ) ) (lstm): LSTM(512, 256, batch_first $= "$ True, bidirectional $=$ True) ) ) + +acotron2( +(embedding): Embedding(185, 512) +(encoder): Encoder( (convolutions): ModuleList( (0): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $: =$ (5,), stride $=$ (1,), padding=(2,)) ) (1): BatchNorm1d(512, ep $\mathtt { s } = 1 \mathtt { e } - 0 5$ , momentum $\iota { = } 0 \cdot 1$ , affine $=$ True, track_running_stat $s =$ True) ) (1): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding ${ } = { }$ (2,)) ) (1): BatchNorm1d(512, ep $\mathsf { s } { = } 1 \mathsf { e } { - } 0 5$ , momentum $\iota { = } 0 \cdot 1$ , affine $: =$ True, track_running_stat ${ \mathfrak { s } } =$ True) ) (2): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding=(2,)) ) (1): BatchNorm1d(512, eps $=$ 1e-05, momentum $\iota { = } 0 \cdot 1$ , affine $: =$ True, track_running_stats $; =$ True) ) ) (lstm): LSTM(512, 256, batch_first $=$ True, bidirectional ${ } = { }$ True) +) +(decoder): Decoder( (prenet): Prenet( (layers): ModuleList( (0): LinearNorm( (linear_layer): Linear(in_features $= 8 0$ , out_feature $_ { 5 } = 2 5 6$ , bias $=$ False) ) (1): LinearNorm( (linear_layer): Linear(in_features=256, out_features=256, bias=False) ) ) (attention_rnn): LSTMCell(896, 1024) (attention_layer): Attention( (query_layer): LinearNorm( (linear_layer): Linear(in_features=1024, out_features=128, bias=False) ) (memory_layer): LinearNorm( (linear_layer): Linear(in_feature $\mathord { 5 } = 6 4 0$ , out_features $^ { - 1 2 8 }$ , bias $\fallingdotseq$ False) ) (v): LinearNorm( (linear_layer): Linear(in_feature $\mathord { \left. \mathrm { = } \right.} 1 2 8 $ , out_features $^ { = 1 }$ , bias=False) ) (location_layer): LocationLayer( (location_conv): ConvNorm( (conv): Conv1d(2, 32, kernel_size $=$ (31,), stride $\mathbf { \Omega } : = \left( \mathbb { 1 } , \right)$ , padding $=$ (15,), bia $: =$ False) ) (location_dense): LinearNorm( (linear_layer): Linear(in_features $= 3 2$ , out_features $= 1 2 8$ , bias $=$ False) ) ) (decoder_rnn): LSTMCell(1664, 1024, bias $^ { = 1 }$ ) (linear_projection): LinearNorm( (linear_layer): Linear(in_features $^ { = 1 }$ 664, out_features ${ \mathfrak { s } } = 8 0$ , bias $= "$ True) ) (gate_layer): LinearNorm( (linear_layer): Linear(in_features=1664, out_features $^ { = 1 }$ , bias=True) ) +) +(postnet): Postnet( (convolutions): ModuleList( (0): Sequential( (0): ConvNorm( (conv): Conv1d(80, 512, kernel_size $=$ (5,), stride $=$ (1,), padding $=$ (2,)) ) (1): BatchNorm1d(512, eps=1e-05, momentum $= 0 \cdot 1$ , affine $=$ wTrue, track_running_stats $= ^ { \prime }$ True) ) (1): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $: =$ (5,), stride $=$ (1,), padding $\bf \tilde { \tau } =$ (2,)) ) (1): BatchNorm1d(512, eps $= 1 \mathrm { e } - 0 5$ , momentum $\iota { = } 0 \cdot 1$ , affine $=$ True, track_running_stat $\bar { \mathsf { z } } = \mathsf { ^ { \prime } }$ True) ) (2): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding ${ } = { }$ (2,)) ) (1): BatchNorm1d(512, ep $\mathtt { s } = 1 \mathtt { e } - 0 5$ , momentum $\iota { = } 0 \cdot 1$ , affine $: =$ True, track_running_stat ${ \tt S } =$ True) ) (3): Sequential( (0): ConvNorm( (conv): Conv1d(512, 512, kernel_size $=$ (5,), stride $=$ (1,), padding=(2,)) ) (1): BatchNorm1d(512, eps $=$ 1e-05, momentum $\iota { = } 0 \cdot 1$ , affine $: =$ True, track_running_stat $s =$ True) ) (4): Sequential( (0): ConvNorm( (conv): Conv1d(512, 80, kernel_size $=$ (5,), stride $=$ (1,), padding $=$ (2,)) ) (1): BatchNorm1d(80, eps=1e-05, momentum $\mathord { \mathrm { 1 } } = 0 \cdot 1$ , affine $=$ True, track_running_stats $=$ True) ) +(speaker_embedding): Embedding(3, 128) \ No newline at end of file diff --git a/parse/train/Ig53hpHxS4/Ig53hpHxS4_model.json b/parse/train/Ig53hpHxS4/Ig53hpHxS4_model.json new file mode 100644 index 0000000000000000000000000000000000000000..4478b37ba5bc63485ffcb5b7b87ea78804176741 --- /dev/null +++ b/parse/train/Ig53hpHxS4/Ig53hpHxS4_model.json @@ -0,0 +1,29481 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 671, + 1303, + 671, + 1303, + 1036, + 398, + 1036 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1160, + 1404, + 1160, + 1404, + 1405, + 298, + 1405 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1619, + 1404, + 1619, + 1404, + 1835, + 299, + 1835 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1850, + 1403, + 1850, + 1403, + 2033, + 299, + 2033 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1421, + 1404, + 1421, + 1404, + 1604, + 299, + 1604 + ], + "score": 0.977 + }, + { + "category_id": 0, + "poly": [ + 298, + 219, + 1407, + 219, + 1407, + 378, + 298, + 378 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 302, + 1092, + 573, + 1092, + 573, + 1127, + 302, + 1127 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 813, + 76, + 813, + 104, + 299, + 104 + ], + "score": 0.876 + }, + { + "category_id": 0, + "poly": [ + 773, + 605, + 926, + 605, + 926, + 638, + 773, + 638 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 856, + 2088, + 856, + 2112, + 841, + 2112 + ], + "score": 0.724 + }, + { + "category_id": 1, + "poly": [ + 312, + 428, + 1054, + 428, + 1054, + 522, + 312, + 522 + ], + "score": 0.619 + }, + { + "category_id": 15, + "poly": [ + 293.0, + 218.0, + 1404.0, + 218.0, + 1404.0, + 272.0, + 293.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 273.0, + 1412.0, + 273.0, + 1412.0, + 325.0, + 295.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 335.0, + 446.0, + 335.0, + 446.0, + 379.0, + 295.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 579.0, + 1089.0, + 579.0, + 1136.0, + 294.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 604.0, + 932.0, + 604.0, + 932.0, + 641.0, + 769.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 857.0, + 2088.0, + 857.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 672.0, + 1306.0, + 672.0, + 1306.0, + 707.0, + 394.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 702.0, + 1305.0, + 702.0, + 1305.0, + 737.0, + 394.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 732.0, + 1305.0, + 732.0, + 1305.0, + 767.0, + 393.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 763.0, + 1306.0, + 763.0, + 1306.0, + 798.0, + 394.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 793.0, + 1306.0, + 793.0, + 1306.0, + 831.0, + 393.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 821.0, + 1306.0, + 821.0, + 1306.0, + 860.0, + 392.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 855.0, + 1307.0, + 855.0, + 1307.0, + 889.0, + 393.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 886.0, + 1308.0, + 886.0, + 1308.0, + 917.0, + 393.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 917.0, + 1305.0, + 917.0, + 1305.0, + 949.0, + 395.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 947.0, + 1304.0, + 947.0, + 1304.0, + 979.0, + 394.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 976.0, + 1307.0, + 976.0, + 1307.0, + 1012.0, + 392.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1006.0, + 817.0, + 1006.0, + 817.0, + 1039.0, + 395.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1159.0, + 1405.0, + 1159.0, + 1405.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1193.0, + 1405.0, + 1193.0, + 1405.0, + 1225.0, + 293.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1224.0, + 1406.0, + 1224.0, + 1406.0, + 1257.0, + 294.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1251.0, + 1406.0, + 1251.0, + 1406.0, + 1287.0, + 293.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1285.0, + 1406.0, + 1285.0, + 1406.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1312.0, + 1407.0, + 1312.0, + 1407.0, + 1350.0, + 292.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1346.0, + 1406.0, + 1346.0, + 1406.0, + 1380.0, + 294.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1375.0, + 1233.0, + 1375.0, + 1233.0, + 1408.0, + 294.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1617.0, + 1405.0, + 1617.0, + 1405.0, + 1657.0, + 292.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1652.0, + 1406.0, + 1652.0, + 1406.0, + 1686.0, + 294.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1684.0, + 1405.0, + 1684.0, + 1405.0, + 1715.0, + 295.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1715.0, + 1404.0, + 1715.0, + 1404.0, + 1746.0, + 295.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1742.0, + 1406.0, + 1742.0, + 1406.0, + 1779.0, + 293.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1807.0, + 293.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1804.0, + 1285.0, + 1804.0, + 1285.0, + 1838.0, + 294.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1848.0, + 1407.0, + 1848.0, + 1407.0, + 1887.0, + 294.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 1407.0, + 1881.0, + 1407.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1907.0, + 1407.0, + 1907.0, + 1407.0, + 1949.0, + 293.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1943.0, + 1407.0, + 1943.0, + 1407.0, + 1977.0, + 293.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 1405.0, + 2002.0, + 1405.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1422.0, + 1406.0, + 1422.0, + 1406.0, + 1453.0, + 294.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1452.0, + 1408.0, + 1452.0, + 1408.0, + 1486.0, + 293.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1483.0, + 1405.0, + 1483.0, + 1405.0, + 1515.0, + 295.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1511.0, + 1406.0, + 1511.0, + 1406.0, + 1549.0, + 292.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1545.0, + 1404.0, + 1545.0, + 1404.0, + 1577.0, + 295.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1576.0, + 1408.0, + 1576.0, + 1408.0, + 1607.0, + 295.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 428.0, + 1054.0, + 428.0, + 1054.0, + 466.0, + 311.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 463.0, + 419.0, + 463.0, + 419.0, + 492.0, + 313.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 494.0, + 686.0, + 494.0, + 686.0, + 526.0, + 313.0, + 526.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1023, + 1405, + 1023, + 1405, + 1268, + 298, + 1268 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1619, + 1405, + 1619, + 1405, + 1805, + 298, + 1805 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1819, + 1403, + 1819, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1283, + 1404, + 1283, + 1404, + 1437, + 299, + 1437 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1453, + 1404, + 1453, + 1404, + 1605, + 298, + 1605 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 388, + 260, + 823, + 260, + 823, + 678, + 388, + 678 + ], + "score": 0.96 + }, + { + "category_id": 3, + "poly": [ + 878, + 262, + 1315, + 262, + 1315, + 676, + 878, + 676 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 292, + 945, + 1399, + 945, + 1399, + 1008, + 292, + 1008 + ], + "score": 0.946 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 374, + 688, + 844, + 688, + 844, + 799, + 374, + 799 + ], + "score": 0.871 + }, + { + "category_id": 1, + "poly": [ + 862, + 687, + 1327, + 687, + 1327, + 828, + 862, + 828 + ], + "score": 0.818 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.715 + }, + { + "category_id": 1, + "poly": [ + 291, + 853, + 1400, + 853, + 1400, + 886, + 291, + 886 + ], + "score": 0.705 + }, + { + "category_id": 4, + "poly": [ + 862, + 687, + 1327, + 687, + 1327, + 828, + 862, + 828 + ], + "score": 0.184 + }, + { + "category_id": 4, + "poly": [ + 291, + 853, + 1400, + 853, + 1400, + 886, + 291, + 886 + ], + "score": 0.156 + }, + { + "category_id": 4, + "poly": [ + 374, + 688, + 844, + 688, + 844, + 799, + 374, + 799 + ], + "score": 0.12 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.096 + }, + { + "category_id": 13, + "poly": [ + 1204, + 1375, + 1237, + 1375, + 1237, + 1405, + 1204, + 1405 + ], + "score": 0.89, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1174, + 1680, + 1206, + 1680, + 1206, + 1710, + 1174, + 1710 + ], + "score": 0.86, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1068, + 771, + 1099, + 771, + 1099, + 798, + 1068, + 798 + ], + "score": 0.86, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 800, + 1092, + 800, + 1092, + 826, + 1063, + 826 + ], + "score": 0.85, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 561, + 692, + 579, + 692, + 579, + 713, + 561, + 713 + ], + "score": 0.78, + "latex": "_ { z }" + }, + { + "category_id": 13, + "poly": [ + 1084, + 692, + 1102, + 692, + 1102, + 713, + 1084, + 713 + ], + "score": 0.78, + "latex": "_ { z }" + }, + { + "category_id": 13, + "poly": [ + 983, + 1519, + 1003, + 1519, + 1003, + 1541, + 983, + 1541 + ], + "score": 0.76, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 1079, + 1350, + 1099, + 1350, + 1099, + 1372, + 1079, + 1372 + ], + "score": 0.75, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 1017, + 1488, + 1037, + 1488, + 1037, + 1510, + 1017, + 1510 + ], + "score": 0.75, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 908, + 859, + 927, + 859, + 927, + 881, + 908, + 881 + ], + "score": 0.73, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 466, + 1410, + 486, + 1410, + 486, + 1433, + 466, + 1433 + ], + "score": 0.71, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 861, + 773, + 881, + 773, + 881, + 796, + 861, + 796 + ], + "score": 0.69, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 372, + 747, + 392, + 747, + 392, + 768, + 372, + 768 + ], + "score": 0.68, + "latex": "^ +" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 283.0, + 415.0, + 283.0, + 415.0, + 297.0, + 403.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 355.0, + 413.0, + 355.0, + 413.0, + 365.0, + 404.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 419.0, + 417.0, + 419.0, + 417.0, + 441.0, + 402.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 430.0, + 408.0, + 430.0, + 408.0, + 498.0, + 389.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 492.0, + 414.0, + 492.0, + 414.0, + 508.0, + 403.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 561.0, + 415.0, + 561.0, + 415.0, + 581.0, + 393.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 631.0, + 415.0, + 631.0, + 415.0, + 650.0, + 393.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 655.0, + 435.0, + 655.0, + 435.0, + 671.0, + 416.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 645.0, + 696.0, + 645.0, + 696.0, + 689.0, + 462.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 659.0, + 733.0, + 659.0, + 733.0, + 669.0, + 725.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 659.0, + 794.0, + 659.0, + 794.0, + 668.0, + 787.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 300.0, + 903.0, + 300.0, + 903.0, + 314.0, + 892.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 375.0, + 903.0, + 375.0, + 903.0, + 392.0, + 892.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 392.0, + 931.0, + 392.0, + 931.0, + 399.0, + 923.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 427.0, + 903.0, + 427.0, + 903.0, + 491.0, + 875.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 452.0, + 1295.0, + 452.0, + 1295.0, + 461.0, + 1287.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 468.0, + 1231.0, + 468.0, + 1231.0, + 477.0, + 1221.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 530.0, + 902.0, + 530.0, + 902.0, + 547.0, + 883.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 594.0, + 1185.0, + 594.0, + 1185.0, + 604.0, + 1177.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 609.0, + 900.0, + 609.0, + 900.0, + 621.0, + 888.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 617.0, + 1151.0, + 617.0, + 1151.0, + 622.0, + 1146.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 628.0, + 1113.0, + 628.0, + 1113.0, + 640.0, + 1101.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 657.0, + 919.0, + 657.0, + 919.0, + 673.0, + 902.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 653.0, + 999.0, + 653.0, + 999.0, + 674.0, + 977.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 646.0, + 1157.0, + 646.0, + 1157.0, + 686.0, + 1050.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 657.0, + 1227.0, + 657.0, + 1227.0, + 672.0, + 1216.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 660.0, + 1303.0, + 660.0, + 1303.0, + 668.0, + 1295.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 686.0, + 1083.0, + 686.0, + 1083.0, + 717.0, + 860.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 686.0, + 1329.0, + 686.0, + 1329.0, + 717.0, + 1103.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 713.0, + 1330.0, + 713.0, + 1330.0, + 746.0, + 858.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 741.0, + 1295.0, + 741.0, + 1295.0, + 774.0, + 859.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 768.0, + 860.0, + 768.0, + 860.0, + 802.0, + 857.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 768.0, + 1067.0, + 768.0, + 1067.0, + 802.0, + 882.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 768.0, + 1292.0, + 768.0, + 1292.0, + 802.0, + 1100.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 796.0, + 1062.0, + 796.0, + 1062.0, + 830.0, + 860.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 796.0, + 1255.0, + 796.0, + 1255.0, + 830.0, + 1093.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 847.0, + 907.0, + 847.0, + 907.0, + 890.0, + 294.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 847.0, + 1403.0, + 847.0, + 1403.0, + 890.0, + 928.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 687.0, + 560.0, + 687.0, + 560.0, + 719.0, + 372.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 687.0, + 846.0, + 687.0, + 846.0, + 719.0, + 580.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 716.0, + 831.0, + 716.0, + 831.0, + 745.0, + 373.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 744.0, + 843.0, + 744.0, + 843.0, + 773.0, + 393.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 773.0, + 557.0, + 773.0, + 557.0, + 800.0, + 375.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1022.0, + 1410.0, + 1022.0, + 1410.0, + 1059.0, + 293.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1053.0, + 1405.0, + 1053.0, + 1405.0, + 1087.0, + 295.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1084.0, + 1406.0, + 1084.0, + 1406.0, + 1119.0, + 292.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1114.0, + 1407.0, + 1114.0, + 1407.0, + 1148.0, + 296.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1144.0, + 1405.0, + 1144.0, + 1405.0, + 1180.0, + 293.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 1406.0, + 1175.0, + 1406.0, + 1209.0, + 293.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1205.0, + 1410.0, + 1205.0, + 1410.0, + 1243.0, + 293.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1238.0, + 538.0, + 1238.0, + 538.0, + 1271.0, + 296.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1620.0, + 1407.0, + 1620.0, + 1407.0, + 1656.0, + 296.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1652.0, + 1410.0, + 1652.0, + 1410.0, + 1688.0, + 295.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1681.0, + 1173.0, + 1681.0, + 1173.0, + 1717.0, + 293.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1681.0, + 1405.0, + 1681.0, + 1405.0, + 1717.0, + 1207.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1712.0, + 1407.0, + 1712.0, + 1407.0, + 1748.0, + 293.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1741.0, + 1408.0, + 1741.0, + 1408.0, + 1777.0, + 293.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1773.0, + 1258.0, + 1773.0, + 1258.0, + 1809.0, + 295.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1858.0, + 293.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1404.0, + 1882.0, + 1404.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1913.0, + 1404.0, + 1913.0, + 1404.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2005.0, + 1404.0, + 2005.0, + 1404.0, + 2036.0, + 296.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1282.0, + 1409.0, + 1282.0, + 1409.0, + 1319.0, + 293.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1314.0, + 1406.0, + 1314.0, + 1406.0, + 1348.0, + 293.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 1078.0, + 1344.0, + 1078.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1344.0, + 1406.0, + 1344.0, + 1406.0, + 1381.0, + 1100.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1372.0, + 1203.0, + 1372.0, + 1203.0, + 1410.0, + 293.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1372.0, + 1405.0, + 1372.0, + 1405.0, + 1410.0, + 1238.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1404.0, + 465.0, + 1404.0, + 465.0, + 1445.0, + 294.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1404.0, + 1198.0, + 1404.0, + 1198.0, + 1445.0, + 487.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1452.0, + 1405.0, + 1452.0, + 1405.0, + 1488.0, + 294.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 1016.0, + 1482.0, + 1016.0, + 1519.0, + 294.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1482.0, + 1408.0, + 1482.0, + 1408.0, + 1519.0, + 1038.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 982.0, + 1511.0, + 982.0, + 1551.0, + 293.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1511.0, + 1408.0, + 1511.0, + 1408.0, + 1551.0, + 1004.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1541.0, + 1408.0, + 1541.0, + 1408.0, + 1580.0, + 293.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1572.0, + 1376.0, + 1572.0, + 1376.0, + 1610.0, + 292.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 945.0, + 1402.0, + 945.0, + 1402.0, + 980.0, + 295.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 975.0, + 1265.0, + 975.0, + 1265.0, + 1012.0, + 294.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 687.0, + 560.0, + 687.0, + 560.0, + 719.0, + 372.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 687.0, + 846.0, + 687.0, + 846.0, + 719.0, + 580.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 716.0, + 831.0, + 716.0, + 831.0, + 745.0, + 373.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 744.0, + 843.0, + 744.0, + 843.0, + 773.0, + 393.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 773.0, + 557.0, + 773.0, + 557.0, + 800.0, + 375.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 686.0, + 1083.0, + 686.0, + 1083.0, + 717.0, + 860.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 686.0, + 1329.0, + 686.0, + 1329.0, + 717.0, + 1103.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 713.0, + 1330.0, + 713.0, + 1330.0, + 746.0, + 858.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 741.0, + 1295.0, + 741.0, + 1295.0, + 774.0, + 859.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 768.0, + 860.0, + 768.0, + 860.0, + 802.0, + 857.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 768.0, + 1067.0, + 768.0, + 1067.0, + 802.0, + 882.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 768.0, + 1292.0, + 768.0, + 1292.0, + 802.0, + 1100.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 796.0, + 1062.0, + 796.0, + 1062.0, + 830.0, + 860.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 796.0, + 1255.0, + 796.0, + 1255.0, + 830.0, + 1093.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 847.0, + 907.0, + 847.0, + 907.0, + 890.0, + 294.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 847.0, + 1403.0, + 847.0, + 1403.0, + 890.0, + 928.0, + 890.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1544, + 1406, + 1544, + 1406, + 1701, + 296, + 1701 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 527, + 1406, + 527, + 1406, + 651, + 297, + 651 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1404, + 228, + 1404, + 414, + 298, + 414 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1910, + 1404, + 1910, + 1404, + 2035, + 298, + 2035 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 294, + 1389, + 1404, + 1389, + 1404, + 1453, + 294, + 1453 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 293, + 777, + 1401, + 777, + 1401, + 842, + 293, + 842 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 296, + 1176, + 1401, + 1176, + 1401, + 1240, + 296, + 1240 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 295, + 1017, + 1406, + 1017, + 1406, + 1084, + 295, + 1084 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 595, + 1293, + 1103, + 1293, + 1103, + 1359, + 595, + 1359 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 717, + 714, + 981, + 714, + 981, + 752, + 717, + 752 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 577, + 860, + 1120, + 860, + 1120, + 937, + 577, + 937 + ], + "score": 0.929 + }, + { + "category_id": 0, + "poly": [ + 298, + 1120, + 660, + 1120, + 660, + 1151, + 298, + 1151 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 299, + 457, + 515, + 457, + 515, + 494, + 299, + 494 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 300, + 1488, + 756, + 1488, + 756, + 1521, + 300, + 1521 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.906 + }, + { + "category_id": 8, + "poly": [ + 678, + 943, + 1021, + 943, + 1021, + 987, + 678, + 987 + ], + "score": 0.9 + }, + { + "category_id": 8, + "poly": [ + 709, + 1845, + 990, + 1845, + 990, + 1882, + 709, + 1882 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1366, + 715, + 1400, + 715, + 1400, + 745, + 1366, + 745 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1304, + 1400, + 1304, + 1400, + 1334, + 1366, + 1334 + ], + "score": 0.891 + }, + { + "category_id": 8, + "poly": [ + 715, + 1802, + 984, + 1802, + 984, + 1836, + 715, + 1836 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1366, + 951, + 1400, + 951, + 1400, + 980, + 1366, + 980 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1366, + 882, + 1400, + 882, + 1400, + 912, + 1366, + 912 + ], + "score": 0.875 + }, + { + "category_id": 8, + "poly": [ + 619, + 1764, + 1081, + 1764, + 1081, + 1796, + 619, + 1796 + ], + "score": 0.869 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1764, + 1400, + 1764, + 1400, + 1794, + 1366, + 1794 + ], + "score": 0.866 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1805, + 1400, + 1805, + 1400, + 1833, + 1366, + 1833 + ], + "score": 0.86 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1849, + 1400, + 1849, + 1400, + 1878, + 1366, + 1878 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.664 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.462 + }, + { + "category_id": 14, + "poly": [ + 577, + 854, + 1119, + 854, + 1119, + 994, + 577, + 994 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { \\log p _ { \\theta } ( \\pmb { x } ) = \\log p _ { \\theta } ( z ) + \\displaystyle \\sum _ { i = 1 } ^ { k } \\log \\vert \\operatorname* { d e t } ( \\pmb { J } ( \\pmb { f } _ { i } ^ { - 1 } ( \\pmb { x } ) ) ) \\vert } \\\\ { z = \\pmb { f } _ { k } ^ { - 1 } \\circ \\pmb { f } _ { k - 1 } ^ { - 1 } \\circ . . . \\pmb { f } _ { 1 } ^ { - 1 } ( \\pmb { x } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 830, + 1017, + 920, + 1017, + 920, + 1054, + 830, + 1054 + ], + "score": 0.93, + "latex": "f _ { i } ^ { - 1 } ( { \\pmb x } )" + }, + { + "category_id": 14, + "poly": [ + 595, + 1293, + 1105, + 1293, + 1105, + 1361, + 595, + 1361 + ], + "score": 0.93, + "latex": "z \\sim \\mathcal { N } ( z ; 0 , I ) \\quad \\mathrm { o r } \\quad z \\sim \\sum _ { k } \\hat { \\phi } _ { k } \\mathcal { N } ( z ; \\hat { \\mu } _ { k } , \\hat { \\Sigma } _ { k } )" + }, + { + "category_id": 14, + "poly": [ + 615, + 1759, + 1083, + 1759, + 1083, + 1889, + 615, + 1889 + ], + "score": 0.92, + "latex": "\\begin{array} { c } { { ( \\log s _ { t } , b _ { t } ) = N N ( z _ { 1 : t - 1 } , t e x t , s p e a k e r ) } } \\\\ { { f ( z _ { t } ) = ( z _ { t } - b _ { t } ) \\div s _ { t } } } \\\\ { { f ^ { - 1 } ( z _ { t } ) = s _ { t } \\odot z _ { t } + b _ { t } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 501, + 810, + 557, + 810, + 557, + 843, + 501, + 843 + ], + "score": 0.92, + "latex": "p ( { \\pmb x } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1051, + 352, + 1051, + 352, + 1085, + 298, + 1085 + ], + "score": 0.91, + "latex": "p ( z )" + }, + { + "category_id": 13, + "poly": [ + 1342, + 589, + 1398, + 589, + 1398, + 622, + 1342, + 622 + ], + "score": 0.91, + "latex": "p ( { \\pmb x } )" + }, + { + "category_id": 13, + "poly": [ + 1291, + 558, + 1345, + 558, + 1345, + 591, + 1291, + 591 + ], + "score": 0.91, + "latex": "p ( z )" + }, + { + "category_id": 13, + "poly": [ + 957, + 1943, + 1030, + 1943, + 1030, + 1975, + 957, + 1975 + ], + "score": 0.91, + "latex": "N N ( )" + }, + { + "category_id": 13, + "poly": [ + 647, + 1643, + 723, + 1643, + 723, + 1671, + 647, + 1671 + ], + "score": 0.9, + "latex": "z _ { 1 : t - 1 }" + }, + { + "category_id": 14, + "poly": [ + 718, + 714, + 982, + 714, + 982, + 750, + 718, + 750 + ], + "score": 0.89, + "latex": "\\pmb { x } = \\pmb { f } _ { 1 } \\circ \\pmb { f } _ { 2 } \\circ . . . \\pmb { f } _ { k } ( z )" + }, + { + "category_id": 13, + "poly": [ + 364, + 1911, + 438, + 1911, + 438, + 1945, + 364, + 1945 + ], + "score": 0.89, + "latex": "N N ( )" + }, + { + "category_id": 13, + "poly": [ + 1168, + 1640, + 1197, + 1640, + 1197, + 1669, + 1168, + 1669 + ], + "score": 0.87, + "latex": "\\mathbf { } _ { b _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 662, + 621, + 684, + 621, + 684, + 652, + 662, + 652 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1204, + 1976, + 1236, + 1976, + 1236, + 2003, + 1204, + 2003 + ], + "score": 0.85, + "latex": "z _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1083, + 1643, + 1112, + 1643, + 1112, + 1669, + 1083, + 1669 + ], + "score": 0.84, + "latex": "\\mathbf { } _ { s _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 761, + 1674, + 790, + 1674, + 790, + 1699, + 761, + 1699 + ], + "score": 0.84, + "latex": "{ \\boldsymbol { z } } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 375, + 1022, + 398, + 1022, + 398, + 1048, + 375, + 1048 + ], + "score": 0.83, + "latex": "\\textbf { { J } }" + }, + { + "category_id": 13, + "poly": [ + 504, + 1612, + 525, + 1612, + 525, + 1635, + 504, + 1635 + ], + "score": 0.78, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 713, + 2008, + 732, + 2008, + 732, + 2030, + 713, + 2030 + ], + "score": 0.77, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1184, + 1027, + 1184, + 1027, + 1205, + 1007, + 1205 + ], + "score": 0.7, + "latex": "_ z" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1120.0, + 660.0, + 1120.0, + 660.0, + 1153.0, + 295.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 454.0, + 520.0, + 454.0, + 520.0, + 501.0, + 292.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1488.0, + 761.0, + 1488.0, + 761.0, + 1523.0, + 295.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 860.0, + 2085.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1544.0, + 1407.0, + 1544.0, + 1407.0, + 1582.0, + 293.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1574.0, + 1408.0, + 1574.0, + 1408.0, + 1613.0, + 291.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1606.0, + 503.0, + 1606.0, + 503.0, + 1643.0, + 294.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1606.0, + 1404.0, + 1606.0, + 1404.0, + 1643.0, + 526.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1637.0, + 646.0, + 1637.0, + 646.0, + 1675.0, + 294.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1637.0, + 1082.0, + 1637.0, + 1082.0, + 1675.0, + 724.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1637.0, + 1167.0, + 1637.0, + 1167.0, + 1675.0, + 1113.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1637.0, + 1406.0, + 1637.0, + 1406.0, + 1675.0, + 1198.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1667.0, + 760.0, + 1667.0, + 760.0, + 1706.0, + 294.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1667.0, + 802.0, + 1667.0, + 802.0, + 1706.0, + 791.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 527.0, + 1407.0, + 527.0, + 1407.0, + 563.0, + 294.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 559.0, + 1290.0, + 559.0, + 1290.0, + 591.0, + 294.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 559.0, + 1406.0, + 559.0, + 1406.0, + 591.0, + 1346.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 587.0, + 1341.0, + 587.0, + 1341.0, + 626.0, + 292.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 587.0, + 1411.0, + 587.0, + 1411.0, + 626.0, + 1399.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 618.0, + 661.0, + 618.0, + 661.0, + 653.0, + 294.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 618.0, + 996.0, + 618.0, + 996.0, + 653.0, + 685.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 229.0, + 1407.0, + 229.0, + 1407.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1404.0, + 261.0, + 1404.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 1402.0, + 290.0, + 1402.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 320.0, + 1405.0, + 320.0, + 1405.0, + 358.0, + 292.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 1405.0, + 350.0, + 1405.0, + 387.0, + 293.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 382.0, + 541.0, + 382.0, + 541.0, + 418.0, + 293.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 363.0, + 1910.0, + 363.0, + 1947.0, + 293.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1947.0, + 439.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 956.0, + 1942.0, + 956.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1976.0, + 1031.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1974.0, + 1203.0, + 1974.0, + 1203.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1974.0, + 1405.0, + 1974.0, + 1405.0, + 2006.0, + 1237.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 712.0, + 2003.0, + 712.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 2003.0, + 934.0, + 2003.0, + 934.0, + 2038.0, + 733.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1386.0, + 1409.0, + 1386.0, + 1409.0, + 1427.0, + 293.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1420.0, + 1240.0, + 1420.0, + 1240.0, + 1454.0, + 294.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 777.0, + 1404.0, + 777.0, + 1404.0, + 813.0, + 296.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 811.0, + 500.0, + 811.0, + 500.0, + 843.0, + 297.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 811.0, + 895.0, + 811.0, + 895.0, + 843.0, + 558.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1175.0, + 1006.0, + 1175.0, + 1006.0, + 1212.0, + 294.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1175.0, + 1403.0, + 1175.0, + 1403.0, + 1212.0, + 1028.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1207.0, + 1103.0, + 1207.0, + 1103.0, + 1243.0, + 295.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1017.0, + 374.0, + 1017.0, + 374.0, + 1057.0, + 293.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1017.0, + 829.0, + 1017.0, + 829.0, + 1057.0, + 399.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 1017.0, + 1406.0, + 1017.0, + 1406.0, + 1057.0, + 921.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1049.0, + 297.0, + 1049.0, + 297.0, + 1087.0, + 292.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1049.0, + 1353.0, + 1049.0, + 1353.0, + 1087.0, + 353.0, + 1087.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1787, + 1406, + 1787, + 1406, + 2035, + 298, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1022, + 1404, + 1022, + 1404, + 1146, + 299, + 1146 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1251, + 1404, + 1251, + 1404, + 1405, + 298, + 1405 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1420, + 1404, + 1420, + 1404, + 1575, + 299, + 1575 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1590, + 1403, + 1590, + 1403, + 1682, + 299, + 1682 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 298, + 221, + 1403, + 221, + 1403, + 448, + 298, + 448 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 300, + 914, + 1402, + 914, + 1402, + 1008, + 300, + 1008 + ], + "score": 0.965 + }, + { + "category_id": 4, + "poly": [ + 296, + 466, + 1406, + 466, + 1406, + 682, + 296, + 682 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 293, + 749, + 1401, + 749, + 1401, + 812, + 293, + 812 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 638, + 836, + 1060, + 836, + 1060, + 881, + 638, + 881 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 300, + 1192, + 650, + 1192, + 650, + 1223, + 300, + 1223 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 1728, + 503, + 1728, + 503, + 1760, + 299, + 1760 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 813, + 76, + 813, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1366, + 843, + 1399, + 843, + 1399, + 872, + 1366, + 872 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 857, + 2089, + 857, + 2111, + 840, + 2111 + ], + "score": 0.782 + }, + { + "category_id": 13, + "poly": [ + 908, + 1970, + 1019, + 1970, + 1019, + 2002, + 908, + 2002 + ], + "score": 0.93, + "latex": "\\sigma ^ { \\mathrm { 2 } } = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 489, + 620, + 642, + 620, + 642, + 655, + 489, + 655 + ], + "score": 0.93, + "latex": "x _ { t + 1 } \\to x _ { t + 1 } ^ { \\prime }" + }, + { + "category_id": 14, + "poly": [ + 637, + 833, + 1062, + 833, + 1062, + 881, + 637, + 881 + ], + "score": 0.92, + "latex": "\\log | \\operatorname* { d e t } ( J ( f _ { c o u p l i n g } ^ { - 1 } ( \\pmb { x } ) ) ) | = \\log | s |" + }, + { + "category_id": 13, + "poly": [ + 470, + 1880, + 552, + 1880, + 552, + 1910, + 470, + 1910 + ], + "score": 0.92, + "latex": "\\sigma ^ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 521, + 499, + 595, + 499, + 595, + 532, + 521, + 532 + ], + "score": 0.9, + "latex": "N N ( )" + }, + { + "category_id": 13, + "poly": [ + 1170, + 945, + 1243, + 945, + 1243, + 979, + 1170, + 979 + ], + "score": 0.89, + "latex": "\\log | s |" + }, + { + "category_id": 13, + "poly": [ + 1090, + 1880, + 1122, + 1880, + 1122, + 1909, + 1090, + 1909 + ], + "score": 0.88, + "latex": "\\sigma ^ { \\bar { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 719, + 756, + 747, + 756, + 747, + 781, + 719, + 781 + ], + "score": 0.85, + "latex": "\\mathbf { \\delta } _ { s _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1821, + 1276, + 1821, + 1276, + 1853, + 1256, + 1853 + ], + "score": 0.82, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1625, + 1082, + 1625, + 1082, + 1648, + 1063, + 1648 + ], + "score": 0.77, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 339, + 1625, + 359, + 1625, + 359, + 1648, + 339, + 1648 + ], + "score": 0.76, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 787, + 1596, + 808, + 1596, + 808, + 1618, + 787, + 1618 + ], + "score": 0.75, + "latex": "_ { z }" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1795, + 1028, + 1795, + 1028, + 1818, + 1007, + 1818 + ], + "score": 0.74, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 487, + 1284, + 541, + 1284, + 541, + 1312, + 487, + 1312 + ], + "score": 0.72, + "latex": "N N" + }, + { + "category_id": 13, + "poly": [ + 566, + 1916, + 587, + 1916, + 587, + 1940, + 566, + 1940 + ], + "score": 0.71, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 574, + 470, + 609, + 470, + 609, + 498, + 574, + 498 + ], + "score": 0.71, + "latex": "\\mathbf { \\rho } ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 860, + 566, + 877, + 566, + 877, + 587, + 860, + 587 + ], + "score": 0.7, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 875, + 532, + 890, + 532, + 890, + 557, + 875, + 557 + ], + "score": 0.69, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 879, + 470, + 913, + 470, + 913, + 498, + 879, + 498 + ], + "score": 0.56, + "latex": "{ \\bf \\Pi } ( { \\bf z } )" + }, + { + "category_id": 13, + "poly": [ + 609, + 535, + 625, + 535, + 625, + 556, + 609, + 556 + ], + "score": 0.49, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 1381, + 566, + 1397, + 566, + 1397, + 588, + 1381, + 588 + ], + "score": 0.47, + "latex": "c" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 235.0, + 460.0, + 235.0, + 460.0, + 271.0, + 424.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 236.0, + 691.0, + 236.0, + 691.0, + 268.0, + 562.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 250.0, + 750.0, + 250.0, + 750.0, + 286.0, + 737.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 234.0, + 824.0, + 234.0, + 824.0, + 277.0, + 779.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 243.0, + 908.0, + 243.0, + 908.0, + 263.0, + 888.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 232.0, + 1025.0, + 232.0, + 1025.0, + 276.0, + 976.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 241.0, + 1109.0, + 241.0, + 1109.0, + 268.0, + 1084.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 235.0, + 1276.0, + 235.0, + 1276.0, + 277.0, + 1226.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 245.0, + 1356.0, + 245.0, + 1356.0, + 260.0, + 1340.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 271.0, + 1177.0, + 271.0, + 1177.0, + 283.0, + 1154.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 300.0, + 337.0, + 300.0, + 337.0, + 333.0, + 294.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 300.0, + 644.0, + 300.0, + 644.0, + 336.0, + 430.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 278.0, + 829.0, + 278.0, + 829.0, + 317.0, + 766.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 294.0, + 916.0, + 294.0, + 916.0, + 332.0, + 845.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 278.0, + 1027.0, + 278.0, + 1027.0, + 331.0, + 963.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 292.0, + 1116.0, + 292.0, + 1116.0, + 334.0, + 1044.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 299.0, + 1208.0, + 299.0, + 1208.0, + 315.0, + 1166.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 279.0, + 1278.0, + 279.0, + 1278.0, + 332.0, + 1215.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 310.0, + 1368.0, + 310.0, + 1368.0, + 336.0, + 1320.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 324.0, + 918.0, + 324.0, + 918.0, + 349.0, + 869.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 324.0, + 1118.0, + 324.0, + 1118.0, + 350.0, + 1069.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 326.0, + 1369.0, + 326.0, + 1369.0, + 351.0, + 1320.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 342.0, + 834.0, + 342.0, + 834.0, + 368.0, + 769.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 343.0, + 1033.0, + 343.0, + 1033.0, + 369.0, + 965.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 347.0, + 1177.0, + 347.0, + 1177.0, + 360.0, + 1153.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 342.0, + 1282.0, + 342.0, + 1282.0, + 368.0, + 1216.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 357.0, + 805.0, + 357.0, + 805.0, + 379.0, + 769.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 360.0, + 1001.0, + 360.0, + 1001.0, + 378.0, + 966.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 364.0, + 1212.0, + 364.0, + 1212.0, + 374.0, + 1201.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 360.0, + 1253.0, + 360.0, + 1253.0, + 378.0, + 1217.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 372.0, + 456.0, + 372.0, + 456.0, + 398.0, + 429.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 370.0, + 879.0, + 370.0, + 879.0, + 383.0, + 843.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 368.0, + 1081.0, + 368.0, + 1081.0, + 384.0, + 1036.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 373.0, + 1212.0, + 373.0, + 1212.0, + 391.0, + 1166.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 410.0, + 694.0, + 410.0, + 694.0, + 438.0, + 422.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 407.0, + 776.0, + 407.0, + 776.0, + 425.0, + 761.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 410.0, + 842.0, + 410.0, + 842.0, + 425.0, + 827.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 410.0, + 980.0, + 410.0, + 980.0, + 425.0, + 962.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 410.0, + 1044.0, + 410.0, + 1044.0, + 425.0, + 1027.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 410.0, + 1166.0, + 410.0, + 1166.0, + 425.0, + 1148.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 414.0, + 1291.0, + 414.0, + 1291.0, + 424.0, + 1280.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 414.0, + 1375.0, + 414.0, + 1375.0, + 425.0, + 1363.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.25, + 410.5, + 1230.25, + 410.5, + 1230.25, + 428.5, + 1209.25, + 428.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 468.0, + 573.0, + 468.0, + 573.0, + 502.0, + 295.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 468.0, + 878.0, + 468.0, + 878.0, + 502.0, + 610.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 468.0, + 1404.0, + 468.0, + 1404.0, + 502.0, + 914.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 499.0, + 520.0, + 499.0, + 520.0, + 533.0, + 294.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 499.0, + 1406.0, + 499.0, + 1406.0, + 533.0, + 596.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 530.0, + 608.0, + 530.0, + 608.0, + 564.0, + 294.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 530.0, + 874.0, + 530.0, + 874.0, + 564.0, + 626.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 530.0, + 1406.0, + 530.0, + 1406.0, + 564.0, + 891.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 560.0, + 859.0, + 560.0, + 859.0, + 595.0, + 293.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 560.0, + 1380.0, + 560.0, + 1380.0, + 595.0, + 878.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 560.0, + 1408.0, + 560.0, + 1408.0, + 595.0, + 1398.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 590.0, + 1406.0, + 590.0, + 1406.0, + 625.0, + 294.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 618.0, + 488.0, + 618.0, + 488.0, + 659.0, + 291.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 618.0, + 1408.0, + 618.0, + 1408.0, + 659.0, + 643.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 648.0, + 850.0, + 648.0, + 850.0, + 686.0, + 293.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1192.0, + 653.0, + 1192.0, + 653.0, + 1225.0, + 296.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1725.0, + 508.0, + 1725.0, + 508.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1790.0, + 1006.0, + 1790.0, + 1006.0, + 1824.0, + 295.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1790.0, + 1408.0, + 1790.0, + 1408.0, + 1824.0, + 1029.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1255.0, + 1822.0, + 1255.0, + 1855.0, + 295.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1822.0, + 1409.0, + 1822.0, + 1409.0, + 1855.0, + 1277.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1886.0, + 295.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 469.0, + 1880.0, + 469.0, + 1916.0, + 293.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1880.0, + 1089.0, + 1880.0, + 1089.0, + 1916.0, + 553.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1880.0, + 1408.0, + 1880.0, + 1408.0, + 1916.0, + 1123.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 565.0, + 1909.0, + 565.0, + 1950.0, + 292.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1909.0, + 1407.0, + 1909.0, + 1407.0, + 1950.0, + 588.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 907.0, + 1973.0, + 907.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2006.0, + 1020.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2004.0, + 831.0, + 2004.0, + 831.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1022.0, + 1404.0, + 1022.0, + 1404.0, + 1057.0, + 294.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1055.0, + 1405.0, + 1055.0, + 1405.0, + 1088.0, + 295.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1084.0, + 1405.0, + 1084.0, + 1405.0, + 1120.0, + 294.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1116.0, + 981.0, + 1116.0, + 981.0, + 1148.0, + 295.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1251.0, + 1408.0, + 1251.0, + 1408.0, + 1287.0, + 296.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1280.0, + 486.0, + 1280.0, + 486.0, + 1319.0, + 293.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1280.0, + 1404.0, + 1280.0, + 1404.0, + 1319.0, + 542.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1309.0, + 1405.0, + 1309.0, + 1405.0, + 1353.0, + 292.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1344.0, + 1405.0, + 1344.0, + 1405.0, + 1380.0, + 293.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1377.0, + 555.0, + 1377.0, + 555.0, + 1407.0, + 294.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1422.0, + 1404.0, + 1422.0, + 1404.0, + 1455.0, + 297.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1450.0, + 1408.0, + 1450.0, + 1408.0, + 1488.0, + 294.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1480.0, + 1409.0, + 1480.0, + 1409.0, + 1520.0, + 293.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 1404.0, + 1511.0, + 1404.0, + 1551.0, + 293.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1543.0, + 972.0, + 1543.0, + 972.0, + 1580.0, + 295.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1587.0, + 786.0, + 1587.0, + 786.0, + 1626.0, + 293.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1587.0, + 1404.0, + 1587.0, + 1404.0, + 1626.0, + 809.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1620.0, + 338.0, + 1620.0, + 338.0, + 1654.0, + 295.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1620.0, + 1062.0, + 1620.0, + 1062.0, + 1654.0, + 360.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1620.0, + 1403.0, + 1620.0, + 1403.0, + 1654.0, + 1083.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1651.0, + 504.0, + 1651.0, + 504.0, + 1686.0, + 293.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 915.0, + 1406.0, + 915.0, + 1406.0, + 949.0, + 295.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 943.0, + 1169.0, + 943.0, + 1169.0, + 980.0, + 294.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 943.0, + 1404.0, + 943.0, + 1404.0, + 980.0, + 1244.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 974.0, + 894.0, + 974.0, + 894.0, + 1010.0, + 296.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 745.0, + 718.0, + 745.0, + 718.0, + 788.0, + 292.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 745.0, + 1404.0, + 745.0, + 1404.0, + 788.0, + 748.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 775.0, + 1394.0, + 775.0, + 1394.0, + 819.0, + 293.0, + 819.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 286, + 1405, + 286, + 1405, + 537, + 296, + 537 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1343, + 1405, + 1343, + 1405, + 1558, + 298, + 1558 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1820, + 827, + 1820, + 827, + 2034, + 299, + 2034 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1143, + 1404, + 1143, + 1404, + 1328, + 297, + 1328 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1650, + 1406, + 1650, + 1406, + 1805, + 298, + 1805 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 758, + 1404, + 758, + 1404, + 913, + 297, + 913 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 927, + 1404, + 927, + 1404, + 1052, + 298, + 1052 + ], + "score": 0.977 + }, + { + "category_id": 5, + "poly": [ + 906, + 1826, + 1334, + 1826, + 1334, + 1985, + 906, + 1985 + ], + "score": 0.974, + "html": "
SourceFlowsMOS
Real14.27 ± 0.13
Flowtron23.66 ± 0.16
Tacotron 213.52 ± 0.17
" + }, + { + "category_id": 8, + "poly": [ + 685, + 584, + 1012, + 584, + 1012, + 660, + 685, + 660 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 300, + 691, + 558, + 691, + 558, + 726, + 300, + 726 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 300, + 1087, + 570, + 1087, + 570, + 1118, + 300, + 1118 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 299, + 1593, + 900, + 1593, + 900, + 1626, + 299, + 1626 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1366, + 605, + 1400, + 605, + 1400, + 635, + 1366, + 635 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 300, + 229, + 647, + 229, + 647, + 261, + 300, + 261 + ], + "score": 0.881 + }, + { + "category_id": 6, + "poly": [ + 953, + 2008, + 1297, + 2008, + 1297, + 2039, + 953, + 2039 + ], + "score": 0.848 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.736 + }, + { + "category_id": 14, + "poly": [ + 685, + 582, + 1015, + 582, + 1015, + 659, + 685, + 659 + ], + "score": 0.94, + "latex": "\\pmb { \\mu } _ { p } = \\frac { \\frac { m } { \\lambda } \\bar { \\zeta } } { \\frac { m } { \\lambda } + 1 } \\quad \\pmb { \\Sigma } _ { p } = \\frac { 1 } { \\frac { m } { \\lambda } + 1 } \\pmb { I }" + }, + { + "category_id": 13, + "poly": [ + 1100, + 379, + 1262, + 379, + 1262, + 416, + 1100, + 416 + ], + "score": 0.93, + "latex": "\\zeta _ { i } = f ^ { - 1 } ( { \\pmb x } _ { i } )" + }, + { + "category_id": 13, + "poly": [ + 1223, + 317, + 1403, + 317, + 1403, + 352, + 1223, + 352 + ], + "score": 0.93, + "latex": "q ( z ) = \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 972, + 348, + 1087, + 348, + 1087, + 383, + 972, + 383 + ], + "score": 0.93, + "latex": "q ( z | \\zeta _ { 1 : m } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 443, + 577, + 443, + 577, + 478, + 298, + 478 + ], + "score": 0.92, + "latex": "q ( z | \\boldsymbol { \\zeta } _ { 1 : m } ) = \\mathcal { N } ( \\mu _ { p } , \\boldsymbol { \\Sigma } _ { p } )" + }, + { + "category_id": 13, + "poly": [ + 297, + 383, + 354, + 383, + 354, + 416, + 297, + 416 + ], + "score": 0.89, + "latex": "\\zeta _ { 1 : m }" + }, + { + "category_id": 13, + "poly": [ + 355, + 1850, + 410, + 1850, + 410, + 1881, + 355, + 1881 + ], + "score": 0.88, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 394, + 478, + 417, + 478, + 417, + 507, + 394, + 507 + ], + "score": 0.86, + "latex": "\\zeta _ { i }" + }, + { + "category_id": 13, + "poly": [ + 704, + 388, + 735, + 388, + 735, + 414, + 704, + 414 + ], + "score": 0.85, + "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }" + }, + { + "category_id": 13, + "poly": [ + 897, + 415, + 920, + 415, + 920, + 441, + 897, + 441 + ], + "score": 0.83, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 1309, + 442, + 1327, + 442, + 1327, + 476, + 1309, + 476 + ], + "score": 0.83, + "latex": "\\bar { \\zeta }" + }, + { + "category_id": 13, + "poly": [ + 457, + 323, + 477, + 323, + 477, + 346, + 457, + 346 + ], + "score": 0.76, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 535, + 478, + 555, + 478, + 555, + 503, + 535, + 503 + ], + "score": 0.75, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 478, + 388, + 506, + 388, + 506, + 412, + 478, + 412 + ], + "score": 0.74, + "latex": "m" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 688.0, + 562.0, + 688.0, + 562.0, + 732.0, + 292.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1084.0, + 574.0, + 1084.0, + 574.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1588.0, + 904.0, + 1588.0, + 904.0, + 1633.0, + 292.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 650.0, + 229.0, + 650.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 2002.0, + 1299.0, + 2002.0, + 1299.0, + 2044.0, + 949.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 286.0, + 1406.0, + 286.0, + 1406.0, + 324.0, + 294.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 316.0, + 456.0, + 316.0, + 456.0, + 354.0, + 294.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 316.0, + 1222.0, + 316.0, + 1222.0, + 354.0, + 478.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 348.0, + 971.0, + 348.0, + 971.0, + 386.0, + 294.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 348.0, + 1406.0, + 348.0, + 1406.0, + 386.0, + 1088.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 380.0, + 296.0, + 380.0, + 296.0, + 420.0, + 291.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 380.0, + 477.0, + 380.0, + 477.0, + 420.0, + 355.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 380.0, + 703.0, + 380.0, + 703.0, + 420.0, + 507.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 380.0, + 1099.0, + 380.0, + 1099.0, + 420.0, + 736.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 380.0, + 1407.0, + 380.0, + 1407.0, + 420.0, + 1263.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 413.0, + 896.0, + 413.0, + 896.0, + 449.0, + 293.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 449.0, + 921.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 441.0, + 297.0, + 441.0, + 297.0, + 481.0, + 293.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 441.0, + 1308.0, + 441.0, + 1308.0, + 481.0, + 578.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 441.0, + 1407.0, + 441.0, + 1407.0, + 481.0, + 1328.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 475.0, + 393.0, + 475.0, + 393.0, + 508.0, + 293.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 475.0, + 534.0, + 475.0, + 534.0, + 508.0, + 418.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 475.0, + 1406.0, + 475.0, + 1406.0, + 508.0, + 556.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 505.0, + 1408.0, + 505.0, + 1408.0, + 539.0, + 295.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1341.0, + 1403.0, + 1341.0, + 1403.0, + 1377.0, + 293.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1409.0, + 295.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1401.0, + 1405.0, + 1401.0, + 1405.0, + 1441.0, + 293.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1433.0, + 1407.0, + 1433.0, + 1407.0, + 1470.0, + 293.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1462.0, + 1408.0, + 1462.0, + 1408.0, + 1502.0, + 292.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1497.0, + 1407.0, + 1497.0, + 1407.0, + 1532.0, + 295.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1524.0, + 1342.0, + 1524.0, + 1342.0, + 1563.0, + 293.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1820.0, + 827.0, + 1820.0, + 827.0, + 1853.0, + 296.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1848.0, + 354.0, + 1848.0, + 354.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1848.0, + 828.0, + 1848.0, + 828.0, + 1885.0, + 411.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 831.0, + 1881.0, + 831.0, + 1914.0, + 293.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 826.0, + 1910.0, + 826.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1944.0, + 826.0, + 1944.0, + 826.0, + 1974.0, + 296.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 830.0, + 1972.0, + 830.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 815.0, + 2001.0, + 815.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1143.0, + 1404.0, + 1143.0, + 1404.0, + 1177.0, + 295.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1174.0, + 1406.0, + 1174.0, + 1406.0, + 1210.0, + 294.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1206.0, + 1402.0, + 1206.0, + 1402.0, + 1238.0, + 296.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1233.0, + 1404.0, + 1233.0, + 1404.0, + 1271.0, + 292.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1263.0, + 1408.0, + 1263.0, + 1408.0, + 1302.0, + 291.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1294.0, + 1166.0, + 1294.0, + 1166.0, + 1332.0, + 292.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1652.0, + 1404.0, + 1652.0, + 1404.0, + 1685.0, + 296.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1680.0, + 1407.0, + 1680.0, + 1407.0, + 1716.0, + 295.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1743.0, + 1404.0, + 1743.0, + 1404.0, + 1777.0, + 296.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1773.0, + 1036.0, + 1773.0, + 1036.0, + 1810.0, + 295.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 758.0, + 1406.0, + 758.0, + 1406.0, + 796.0, + 295.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 789.0, + 1406.0, + 789.0, + 1406.0, + 826.0, + 295.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 818.0, + 1406.0, + 818.0, + 1406.0, + 857.0, + 294.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 849.0, + 1405.0, + 849.0, + 1405.0, + 886.0, + 295.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 881.0, + 1166.0, + 881.0, + 1166.0, + 918.0, + 295.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 929.0, + 1404.0, + 929.0, + 1404.0, + 961.0, + 296.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 957.0, + 1406.0, + 957.0, + 1406.0, + 994.0, + 292.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 991.0, + 1405.0, + 991.0, + 1405.0, + 1023.0, + 294.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1020.0, + 1359.0, + 1020.0, + 1359.0, + 1055.0, + 293.0, + 1055.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1454, + 1404, + 1454, + 1404, + 1607, + 298, + 1607 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1692, + 1404, + 1692, + 1404, + 1907, + 298, + 1907 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1146, + 1404, + 1146, + 1404, + 1300, + 298, + 1300 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 495, + 1404, + 495, + 1404, + 619, + 298, + 619 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1314, + 1405, + 1314, + 1405, + 1438, + 297, + 1438 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 320, + 806, + 1427, + 806, + 1427, + 1051, + 320, + 1051 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 634, + 1405, + 634, + 1405, + 727, + 298, + 727 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 302, + 1922, + 1404, + 1922, + 1404, + 1985, + 302, + 1985 + ], + "score": 0.93 + }, + { + "category_id": 4, + "poly": [ + 300, + 1054, + 1404, + 1054, + 1404, + 1116, + 300, + 1116 + ], + "score": 0.928 + }, + { + "category_id": 2, + "poly": [ + 332, + 2004, + 927, + 2004, + 927, + 2033, + 332, + 2033 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 300, + 443, + 620, + 443, + 620, + 474, + 300, + 474 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 303, + 1641, + 820, + 1641, + 820, + 1671, + 303, + 1671 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.888 + }, + { + "category_id": 0, + "poly": [ + 300, + 230, + 626, + 230, + 626, + 261, + 300, + 261 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 299, + 289, + 1404, + 289, + 1404, + 410, + 299, + 410 + ], + "score": 0.795 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.789 + }, + { + "category_id": 13, + "poly": [ + 904, + 316, + 1072, + 316, + 1072, + 352, + 904, + 352 + ], + "score": 0.94, + "latex": "z \\sim \\mathcal { N } ( 0 , \\sigma ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1816, + 463, + 1816, + 463, + 1849, + 298, + 1849 + ], + "score": 0.93, + "latex": "z \\sim \\mathcal { N } ( 0 , 0 . 8 )" + }, + { + "category_id": 13, + "poly": [ + 458, + 525, + 690, + 525, + 690, + 560, + 458, + 560 + ], + "score": 0.92, + "latex": "\\sigma ^ { 2 } \\in \\{ 0 . 0 , 0 . 5 , 1 . 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 631, + 1238, + 868, + 1238, + 868, + 1271, + 631, + 1271 + ], + "score": 0.91, + "latex": "p \\in \\{ 0 . 4 5 , 0 . 5 , 0 . 5 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 635, + 347, + 719, + 347, + 719, + 376, + 635, + 376 + ], + "score": 0.91, + "latex": "\\sigma ^ { 2 } = { \\overset { \\cdot } { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 1318, + 1375, + 1402, + 1375, + 1402, + 1405, + 1318, + 1405 + ], + "score": 0.91, + "latex": "\\sigma ^ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 772, + 1923, + 944, + 1923, + 944, + 1957, + 772, + 1957 + ], + "score": 0.91, + "latex": "z \\sim \\mathcal { N } ( 0 , 0 . 5 )" + }, + { + "category_id": 13, + "poly": [ + 439, + 1269, + 587, + 1269, + 587, + 1302, + 439, + 1302 + ], + "score": 0.9, + "latex": "p \\in [ 0 . 3 , 0 . 8 ]" + }, + { + "category_id": 13, + "poly": [ + 584, + 1317, + 617, + 1317, + 617, + 1347, + 584, + 1347 + ], + "score": 0.89, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1082, + 1547, + 1115, + 1547, + 1115, + 1576, + 1082, + 1576 + ], + "score": 0.89, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1409, + 1253, + 1409, + 1253, + 1438, + 1219, + 1438 + ], + "score": 0.88, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 567, + 1408, + 600, + 1408, + 600, + 1438, + 567, + 1438 + ], + "score": 0.88, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 619, + 1146, + 652, + 1146, + 652, + 1176, + 619, + 1176 + ], + "score": 0.88, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 475, + 2002, + 552, + 2002, + 552, + 2031, + 475, + 2031 + ], + "score": 0.88, + "latex": "\\sigma ^ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 831, + 1516, + 864, + 1516, + 864, + 1546, + 831, + 1546 + ], + "score": 0.88, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 710, + 1348, + 743, + 1348, + 743, + 1377, + 710, + 1377 + ], + "score": 0.87, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1272, + 1052, + 1305, + 1052, + 1305, + 1081, + 1272, + 1081 + ], + "score": 0.87, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1205, + 316, + 1237, + 316, + 1237, + 347, + 1205, + 347 + ], + "score": 0.87, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 547, + 1348, + 579, + 1348, + 579, + 1377, + 547, + 1377 + ], + "score": 0.87, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 909, + 1082, + 942, + 1082, + 942, + 1111, + 909, + 1111 + ], + "score": 0.87, + "latex": "\\sigma ^ { \\tilde { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 593, + 377, + 626, + 377, + 626, + 406, + 593, + 406 + ], + "score": 0.87, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 415, + 1514, + 448, + 1514, + 448, + 1544, + 415, + 1544 + ], + "score": 0.87, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 902, + 1406, + 935, + 1406, + 935, + 1436, + 902, + 1436 + ], + "score": 0.86, + "latex": "\\sigma ^ { \\bar { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 742, + 495, + 775, + 495, + 775, + 524, + 742, + 524 + ], + "score": 0.86, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1309, + 1928, + 1341, + 1928, + 1341, + 1954, + 1309, + 1954 + ], + "score": 0.86, + "latex": "z _ { s }" + }, + { + "category_id": 13, + "poly": [ + 1324, + 556, + 1356, + 556, + 1356, + 586, + 1324, + 586 + ], + "score": 0.86, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1929, + 1255, + 1929, + 1255, + 1954, + 1219, + 1954 + ], + "score": 0.85, + "latex": "_ { z _ { h } }" + }, + { + "category_id": 13, + "poly": [ + 993, + 1520, + 1012, + 1520, + 1012, + 1548, + 993, + 1548 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1106, + 1928, + 1127, + 1928, + 1127, + 1951, + 1106, + 1951 + ], + "score": 0.75, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 825, + 1699, + 844, + 1699, + 844, + 1721, + 825, + 1721 + ], + "score": 0.74, + "latex": "_ { z }" + }, + { + "category_id": 13, + "poly": [ + 751, + 2011, + 769, + 2011, + 769, + 2029, + 751, + 2029 + ], + "score": 0.7, + "latex": "_ { z }" + }, + { + "category_id": 14, + "poly": [ + 826, + 1020, + 933, + 1020, + 933, + 1053, + 826, + 1053 + ], + "score": 0.69, + "latex": "\\ : \\sigma ^ { 2 } = 0 . 5 \\ :" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1021, + 1298, + 1021, + 1298, + 1051, + 1219, + 1051 + ], + "score": 0.68, + "latex": "\\sigma ^ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 1147, + 1347, + 1217, + 1347, + 1217, + 1376, + 1147, + 1376 + ], + "score": 0.52, + "latex": "8 0 \\mathrm { H z }" + }, + { + "category_id": 13, + "poly": [ + 1228, + 1347, + 1313, + 1347, + 1313, + 1376, + 1228, + 1376 + ], + "score": 0.49, + "latex": "4 0 0 \\mathrm { H z }" + }, + { + "category_id": 13, + "poly": [ + 465, + 1021, + 544, + 1021, + 544, + 1052, + 465, + 1052 + ], + "score": 0.4, + "latex": "\\sigma ^ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1755, + 1244, + 1755, + 1244, + 1783, + 1211, + 1783 + ], + "score": 0.26, + "latex": "^ { 6 6 } I t" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 851.0, + 339.0, + 851.0, + 339.0, + 859.0, + 325.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 851.0, + 1092.0, + 851.0, + 1092.0, + 859.0, + 1080.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 873.0, + 338.0, + 873.0, + 338.0, + 882.0, + 325.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 867.0, + 1011.0, + 867.0, + 1011.0, + 887.0, + 958.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 872.0, + 1092.0, + 872.0, + 1092.0, + 881.0, + 1080.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 895.0, + 338.0, + 895.0, + 338.0, + 903.0, + 327.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 917.0, + 338.0, + 917.0, + 338.0, + 926.0, + 325.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 917.0, + 1092.0, + 917.0, + 1092.0, + 926.0, + 1080.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 911.0, + 1383.0, + 911.0, + 1383.0, + 963.0, + 1340.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 940.0, + 338.0, + 940.0, + 338.0, + 949.0, + 325.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 940.0, + 1092.0, + 940.0, + 1092.0, + 949.0, + 1080.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 993.0, + 391.0, + 993.0, + 391.0, + 1003.0, + 378.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 992.0, + 432.0, + 992.0, + 432.0, + 1003.0, + 421.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 993.0, + 476.0, + 993.0, + 476.0, + 1003.0, + 462.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 993.0, + 516.0, + 993.0, + 516.0, + 1003.0, + 504.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 991.0, + 562.0, + 991.0, + 562.0, + 1004.0, + 543.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 991.0, + 604.0, + 991.0, + 604.0, + 1005.0, + 586.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 991.0, + 646.0, + 991.0, + 646.0, + 1004.0, + 626.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 991.0, + 767.0, + 991.0, + 767.0, + 1004.0, + 750.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 993.0, + 801.0, + 993.0, + 801.0, + 1003.0, + 791.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 993.0, + 840.0, + 993.0, + 840.0, + 1003.0, + 828.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 993.0, + 876.0, + 993.0, + 876.0, + 1003.0, + 865.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 993.0, + 918.0, + 993.0, + 918.0, + 1003.0, + 901.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 993.0, + 955.0, + 993.0, + 955.0, + 1003.0, + 939.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 991.0, + 993.0, + 991.0, + 993.0, + 1004.0, + 974.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 991.0, + 1031.0, + 991.0, + 1031.0, + 1004.0, + 1013.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 993.0, + 1149.0, + 993.0, + 1149.0, + 1003.0, + 1136.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 992.0, + 1193.0, + 992.0, + 1193.0, + 1003.0, + 1183.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 993.0, + 1241.0, + 993.0, + 1241.0, + 1003.0, + 1227.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 992.0, + 1285.0, + 992.0, + 1285.0, + 1003.0, + 1272.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 991.0, + 1335.0, + 991.0, + 1335.0, + 1005.0, + 1315.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 991.0, + 1380.0, + 991.0, + 1380.0, + 1004.0, + 1360.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1408.0, + 992.0, + 1420.0, + 992.0, + 1420.0, + 1003.0, + 1408.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1021.0, + 464.0, + 1021.0, + 464.0, + 1054.0, + 426.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1021.0, + 548.0, + 1021.0, + 548.0, + 1054.0, + 545.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1018.0, + 825.0, + 1018.0, + 825.0, + 1058.0, + 793.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1018.0, + 938.0, + 1018.0, + 938.0, + 1058.0, + 934.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1019.0, + 1218.0, + 1019.0, + 1218.0, + 1055.0, + 1179.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1019.0, + 1303.0, + 1019.0, + 1303.0, + 1055.0, + 1299.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.75, + 906.0, + 1008.75, + 906.0, + 1008.75, + 948.5, + 983.75, + 948.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 1271.0, + 1051.0, + 1271.0, + 1090.0, + 295.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1051.0, + 1406.0, + 1051.0, + 1406.0, + 1090.0, + 1306.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1082.0, + 908.0, + 1082.0, + 908.0, + 1118.0, + 295.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1082.0, + 1359.0, + 1082.0, + 1359.0, + 1118.0, + 943.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1998.0, + 474.0, + 1998.0, + 474.0, + 2038.0, + 331.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1998.0, + 750.0, + 1998.0, + 750.0, + 2038.0, + 553.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1998.0, + 930.0, + 1998.0, + 930.0, + 2038.0, + 770.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 443.0, + 625.0, + 443.0, + 625.0, + 476.0, + 295.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1640.0, + 824.0, + 1640.0, + 824.0, + 1673.0, + 296.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 230.0, + 631.0, + 230.0, + 631.0, + 263.0, + 295.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1451.0, + 1404.0, + 1451.0, + 1404.0, + 1490.0, + 295.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1482.0, + 1404.0, + 1482.0, + 1404.0, + 1519.0, + 296.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1515.0, + 414.0, + 1515.0, + 414.0, + 1552.0, + 294.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1515.0, + 830.0, + 1515.0, + 830.0, + 1552.0, + 449.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1515.0, + 992.0, + 1515.0, + 992.0, + 1552.0, + 865.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1515.0, + 1405.0, + 1515.0, + 1405.0, + 1552.0, + 1013.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1546.0, + 1081.0, + 1546.0, + 1081.0, + 1579.0, + 294.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 1546.0, + 1402.0, + 1546.0, + 1402.0, + 1579.0, + 1116.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1578.0, + 836.0, + 1578.0, + 836.0, + 1611.0, + 296.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1691.0, + 824.0, + 1691.0, + 824.0, + 1729.0, + 293.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1691.0, + 1404.0, + 1691.0, + 1404.0, + 1729.0, + 845.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1723.0, + 1405.0, + 1723.0, + 1405.0, + 1759.0, + 292.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1755.0, + 1210.0, + 1755.0, + 1210.0, + 1790.0, + 294.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1755.0, + 1404.0, + 1755.0, + 1404.0, + 1790.0, + 1245.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1780.0, + 1406.0, + 1780.0, + 1406.0, + 1825.0, + 291.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1815.0, + 297.0, + 1815.0, + 297.0, + 1850.0, + 294.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1815.0, + 1404.0, + 1815.0, + 1404.0, + 1850.0, + 464.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1845.0, + 1406.0, + 1845.0, + 1406.0, + 1881.0, + 292.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1875.0, + 397.0, + 1875.0, + 397.0, + 1909.0, + 290.0, + 1909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1146.0, + 618.0, + 1146.0, + 618.0, + 1179.0, + 296.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1146.0, + 1402.0, + 1146.0, + 1402.0, + 1179.0, + 653.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1178.0, + 1405.0, + 1178.0, + 1405.0, + 1211.0, + 296.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1208.0, + 1405.0, + 1208.0, + 1405.0, + 1242.0, + 296.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1238.0, + 630.0, + 1238.0, + 630.0, + 1275.0, + 294.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1238.0, + 1405.0, + 1238.0, + 1405.0, + 1275.0, + 869.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1269.0, + 438.0, + 1269.0, + 438.0, + 1302.0, + 296.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1269.0, + 1158.0, + 1269.0, + 1158.0, + 1302.0, + 588.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 492.0, + 741.0, + 492.0, + 741.0, + 531.0, + 292.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 492.0, + 1405.0, + 492.0, + 1405.0, + 531.0, + 776.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 524.0, + 457.0, + 524.0, + 457.0, + 562.0, + 293.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 524.0, + 1405.0, + 524.0, + 1405.0, + 562.0, + 691.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 556.0, + 1323.0, + 556.0, + 1323.0, + 592.0, + 292.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 556.0, + 1405.0, + 556.0, + 1405.0, + 592.0, + 1357.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 589.0, + 964.0, + 589.0, + 964.0, + 621.0, + 294.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1313.0, + 583.0, + 1313.0, + 583.0, + 1350.0, + 294.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1313.0, + 1408.0, + 1313.0, + 1408.0, + 1350.0, + 618.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1343.0, + 546.0, + 1343.0, + 546.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1343.0, + 709.0, + 1343.0, + 709.0, + 1381.0, + 580.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1343.0, + 1146.0, + 1343.0, + 1146.0, + 1381.0, + 744.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1343.0, + 1227.0, + 1343.0, + 1227.0, + 1381.0, + 1218.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1343.0, + 1406.0, + 1343.0, + 1406.0, + 1381.0, + 1314.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1376.0, + 1317.0, + 1376.0, + 1317.0, + 1410.0, + 292.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1376.0, + 1406.0, + 1376.0, + 1406.0, + 1410.0, + 1403.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1408.0, + 566.0, + 1408.0, + 566.0, + 1442.0, + 292.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1408.0, + 901.0, + 1408.0, + 901.0, + 1442.0, + 601.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1408.0, + 1218.0, + 1408.0, + 1218.0, + 1442.0, + 936.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1408.0, + 1364.0, + 1408.0, + 1364.0, + 1442.0, + 1254.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 631.0, + 1405.0, + 631.0, + 1405.0, + 671.0, + 293.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 665.0, + 1403.0, + 665.0, + 1403.0, + 699.0, + 295.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 696.0, + 1397.0, + 696.0, + 1397.0, + 730.0, + 296.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1920.0, + 771.0, + 1920.0, + 771.0, + 1960.0, + 296.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1920.0, + 1105.0, + 1920.0, + 1105.0, + 1960.0, + 945.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1920.0, + 1218.0, + 1920.0, + 1218.0, + 1960.0, + 1128.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 1920.0, + 1308.0, + 1920.0, + 1308.0, + 1960.0, + 1256.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1920.0, + 1406.0, + 1920.0, + 1406.0, + 1960.0, + 1342.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1951.0, + 1403.0, + 1951.0, + 1403.0, + 1990.0, + 296.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 288.0, + 826.0, + 288.0, + 826.0, + 321.0, + 295.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 317.0, + 903.0, + 317.0, + 903.0, + 351.0, + 293.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 317.0, + 1204.0, + 317.0, + 1204.0, + 351.0, + 1073.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 317.0, + 1404.0, + 317.0, + 1404.0, + 351.0, + 1238.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 346.0, + 634.0, + 346.0, + 634.0, + 384.0, + 293.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 346.0, + 1404.0, + 346.0, + 1404.0, + 384.0, + 720.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 376.0, + 592.0, + 376.0, + 592.0, + 414.0, + 293.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 376.0, + 1158.0, + 376.0, + 1158.0, + 414.0, + 627.0, + 414.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 910, + 1406, + 910, + 1406, + 1124, + 298, + 1124 + ], + "score": 0.982 + }, + { + "category_id": 5, + "poly": [ + 302, + 1617, + 1398, + 1617, + 1398, + 1838, + 302, + 1838 + ], + "score": 0.98, + "html": "
Pitch MeanPitch Standard Deviation
ModelStyle ExpressiveHigh PitchSurprisedExpressiveHigh PitchSurprised
Reference53.655.258.24.52.31.0
FTA Posterior53.453.355.52.52.33.0
FTA Baseline53.152.652.82.21.91.9
Tacotron 2 GST51.753.651.62.02.41.7
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1210, + 1404, + 1210, + 1404, + 1364, + 298, + 1364 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 693, + 1404, + 693, + 1404, + 816, + 299, + 816 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1378, + 1404, + 1378, + 1404, + 1594, + 297, + 1594 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 302, + 249, + 1399, + 249, + 1399, + 574, + 302, + 574 + ], + "score": 0.936 + }, + { + "category_id": 4, + "poly": [ + 298, + 579, + 1401, + 579, + 1401, + 640, + 298, + 640 + ], + "score": 0.935 + }, + { + "category_id": 0, + "poly": [ + 299, + 853, + 930, + 853, + 930, + 885, + 299, + 885 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 300, + 1157, + 570, + 1157, + 570, + 1188, + 300, + 1188 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 812, + 76, + 812, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.764 + }, + { + "category_id": 1, + "poly": [ + 296, + 1857, + 1405, + 1857, + 1405, + 2013, + 296, + 2013 + ], + "score": 0.567 + }, + { + "category_id": 6, + "poly": [ + 296, + 1857, + 1405, + 1857, + 1405, + 2013, + 296, + 2013 + ], + "score": 0.458 + }, + { + "category_id": 13, + "poly": [ + 602, + 547, + 680, + 547, + 680, + 576, + 602, + 576 + ], + "score": 0.89, + "latex": "\\sigma ^ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 931, + 699, + 966, + 699, + 966, + 725, + 931, + 725 + ], + "score": 0.87, + "latex": "z _ { h }" + }, + { + "category_id": 13, + "poly": [ + 1090, + 547, + 1312, + 547, + 1312, + 578, + 1090, + 578 + ], + "score": 0.87, + "latex": "p \\in \\{ 0 . 4 5 , 0 . 5 , 0 . 5 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 413, + 608, + 445, + 608, + 445, + 635, + 413, + 635 + ], + "score": 0.86, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1015, + 698, + 1046, + 698, + 1046, + 725, + 1015, + 725 + ], + "score": 0.86, + "latex": "z _ { s }" + }, + { + "category_id": 13, + "poly": [ + 407, + 579, + 442, + 579, + 442, + 607, + 407, + 607 + ], + "score": 0.84, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 495, + 611, + 513, + 611, + 513, + 640, + 495, + 640 + ], + "score": 0.78, + "latex": "p" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 253.0, + 333.0, + 253.0, + 333.0, + 268.0, + 316.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 253.0, + 884.0, + 253.0, + 884.0, + 268.0, + 869.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 269.0, + 333.0, + 269.0, + 333.0, + 284.0, + 314.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 267.0, + 888.0, + 267.0, + 888.0, + 286.0, + 867.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 282.0, + 334.0, + 282.0, + 334.0, + 300.0, + 313.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 285.0, + 530.0, + 285.0, + 530.0, + 294.0, + 519.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 281.0, + 888.0, + 281.0, + 888.0, + 300.0, + 867.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 297.0, + 334.0, + 297.0, + 334.0, + 319.0, + 303.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 294.0, + 644.0, + 294.0, + 644.0, + 304.0, + 634.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 314.0, + 332.0, + 314.0, + 332.0, + 328.0, + 314.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 314.0, + 887.0, + 314.0, + 887.0, + 328.0, + 868.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 331.0, + 331.0, + 331.0, + 331.0, + 341.0, + 319.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 331.0, + 885.0, + 331.0, + 885.0, + 341.0, + 873.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 352.0, + 367.0, + 352.0, + 367.0, + 367.0, + 348.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 354.0, + 409.0, + 354.0, + 409.0, + 367.0, + 389.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 354.0, + 450.0, + 354.0, + 450.0, + 367.0, + 430.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 354.0, + 493.0, + 354.0, + 493.0, + 367.0, + 473.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 354.0, + 533.0, + 354.0, + 533.0, + 367.0, + 513.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 354.0, + 574.0, + 354.0, + 574.0, + 367.0, + 554.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 352.0, + 615.0, + 352.0, + 615.0, + 367.0, + 596.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 354.0, + 656.0, + 354.0, + 656.0, + 368.0, + 637.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 351.0, + 700.0, + 351.0, + 700.0, + 369.0, + 676.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 354.0, + 738.0, + 354.0, + 738.0, + 367.0, + 718.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 352.0, + 781.0, + 352.0, + 781.0, + 367.0, + 761.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 352.0, + 822.0, + 352.0, + 822.0, + 367.0, + 802.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 352.0, + 920.0, + 352.0, + 920.0, + 367.0, + 902.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 354.0, + 963.0, + 354.0, + 963.0, + 367.0, + 943.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 352.0, + 1005.0, + 352.0, + 1005.0, + 367.0, + 984.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 354.0, + 1046.0, + 354.0, + 1046.0, + 367.0, + 1026.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 354.0, + 1090.0, + 354.0, + 1090.0, + 367.0, + 1069.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 354.0, + 1131.0, + 354.0, + 1131.0, + 367.0, + 1111.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 354.0, + 1172.0, + 354.0, + 1172.0, + 367.0, + 1152.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 354.0, + 1215.0, + 354.0, + 1215.0, + 368.0, + 1195.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 354.0, + 1257.0, + 354.0, + 1257.0, + 367.0, + 1236.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 354.0, + 1298.0, + 354.0, + 1298.0, + 367.0, + 1279.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 355.0, + 1339.0, + 355.0, + 1339.0, + 368.0, + 1322.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 361.0, + 604.0, + 361.0, + 604.0, + 378.0, + 574.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 361.0, + 1156.0, + 361.0, + 1156.0, + 378.0, + 1126.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 379.0, + 686.0, + 379.0, + 686.0, + 416.0, + 466.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 378.0, + 1250.0, + 378.0, + 1250.0, + 416.0, + 1006.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 418.0, + 332.0, + 418.0, + 332.0, + 432.0, + 316.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 418.0, + 885.0, + 418.0, + 885.0, + 432.0, + 869.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 434.0, + 329.0, + 434.0, + 329.0, + 445.0, + 317.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 432.0, + 885.0, + 432.0, + 885.0, + 448.0, + 868.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 448.0, + 332.0, + 448.0, + 332.0, + 463.0, + 314.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 445.0, + 888.0, + 445.0, + 888.0, + 463.0, + 867.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 478.0, + 332.0, + 478.0, + 332.0, + 492.0, + 314.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 475.0, + 888.0, + 475.0, + 888.0, + 494.0, + 867.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 495.0, + 332.0, + 495.0, + 332.0, + 506.0, + 319.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 495.0, + 885.0, + 495.0, + 885.0, + 506.0, + 873.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 517.0, + 368.0, + 517.0, + 368.0, + 531.0, + 348.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 517.0, + 409.0, + 517.0, + 409.0, + 531.0, + 388.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 517.0, + 450.0, + 517.0, + 450.0, + 531.0, + 430.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 517.0, + 491.0, + 517.0, + 491.0, + 531.0, + 471.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 517.0, + 533.0, + 517.0, + 533.0, + 531.0, + 511.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 517.0, + 572.0, + 517.0, + 572.0, + 531.0, + 552.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 517.0, + 612.0, + 517.0, + 612.0, + 531.0, + 595.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 517.0, + 655.0, + 517.0, + 655.0, + 531.0, + 636.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 517.0, + 696.0, + 517.0, + 696.0, + 531.0, + 677.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 517.0, + 737.0, + 517.0, + 737.0, + 531.0, + 717.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 518.0, + 778.0, + 518.0, + 778.0, + 531.0, + 757.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 518.0, + 821.0, + 518.0, + 821.0, + 531.0, + 798.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 511.0, + 887.0, + 511.0, + 887.0, + 519.0, + 877.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 517.0, + 922.0, + 517.0, + 922.0, + 531.0, + 902.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 518.0, + 968.0, + 518.0, + 968.0, + 531.0, + 948.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 517.0, + 1015.0, + 517.0, + 1015.0, + 531.0, + 995.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 515.0, + 1064.0, + 515.0, + 1064.0, + 533.0, + 1040.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 517.0, + 1110.0, + 517.0, + 1110.0, + 531.0, + 1089.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 512.0, + 1161.0, + 512.0, + 1161.0, + 544.0, + 1122.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 518.0, + 1203.0, + 518.0, + 1203.0, + 532.0, + 1183.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 515.0, + 1251.0, + 515.0, + 1251.0, + 533.0, + 1230.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 518.0, + 1297.0, + 518.0, + 1297.0, + 531.0, + 1279.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 518.0, + 1344.0, + 518.0, + 1344.0, + 531.0, + 1326.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 526.0, + 602.0, + 526.0, + 602.0, + 539.0, + 574.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 542.0, + 601.0, + 542.0, + 601.0, + 580.0, + 464.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 542.0, + 685.0, + 542.0, + 685.0, + 580.0, + 681.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 544.0, + 1089.0, + 544.0, + 1089.0, + 578.0, + 942.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.75, + 280.0, + 987.75, + 280.0, + 987.75, + 293.0, + 959.75, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 295.0, + 890.0, + 295.0, + 890.0, + 317.0, + 853.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.25, + 460.0, + 336.25, + 460.0, + 336.25, + 480.0, + 299.25, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 459.5, + 891.0, + 459.5, + 891.0, + 481.0, + 851.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 577.0, + 406.0, + 577.0, + 406.0, + 612.0, + 296.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 577.0, + 1406.0, + 577.0, + 1406.0, + 612.0, + 443.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 606.0, + 412.0, + 606.0, + 412.0, + 641.0, + 294.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 606.0, + 494.0, + 606.0, + 494.0, + 641.0, + 446.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 606.0, + 1292.0, + 606.0, + 1292.0, + 641.0, + 514.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 849.0, + 932.0, + 849.0, + 932.0, + 891.0, + 293.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1154.0, + 572.0, + 1154.0, + 572.0, + 1192.0, + 294.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1855.0, + 1406.0, + 1855.0, + 1406.0, + 1894.0, + 294.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1890.0, + 1403.0, + 1890.0, + 1403.0, + 1923.0, + 295.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1916.0, + 1407.0, + 1916.0, + 1407.0, + 1955.0, + 294.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1947.0, + 1406.0, + 1947.0, + 1406.0, + 1986.0, + 293.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1982.0, + 940.0, + 1982.0, + 940.0, + 2015.0, + 296.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 909.0, + 1404.0, + 909.0, + 1404.0, + 945.0, + 295.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 942.0, + 1406.0, + 942.0, + 1406.0, + 974.0, + 293.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 968.0, + 1409.0, + 968.0, + 1409.0, + 1010.0, + 292.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1002.0, + 1404.0, + 1002.0, + 1404.0, + 1037.0, + 296.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1033.0, + 1404.0, + 1033.0, + 1404.0, + 1067.0, + 295.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1064.0, + 1407.0, + 1064.0, + 1407.0, + 1098.0, + 293.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1094.0, + 1399.0, + 1094.0, + 1399.0, + 1129.0, + 295.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1208.0, + 1404.0, + 1208.0, + 1404.0, + 1246.0, + 292.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1241.0, + 1404.0, + 1241.0, + 1404.0, + 1278.0, + 294.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1271.0, + 1405.0, + 1271.0, + 1405.0, + 1308.0, + 294.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1296.0, + 1408.0, + 1296.0, + 1408.0, + 1340.0, + 291.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1331.0, + 1279.0, + 1331.0, + 1279.0, + 1368.0, + 293.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 693.0, + 930.0, + 693.0, + 930.0, + 729.0, + 294.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 693.0, + 1014.0, + 693.0, + 1014.0, + 729.0, + 967.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 693.0, + 1405.0, + 693.0, + 1405.0, + 729.0, + 1047.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 725.0, + 1406.0, + 725.0, + 1406.0, + 758.0, + 295.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 756.0, + 1403.0, + 756.0, + 1403.0, + 788.0, + 295.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 788.0, + 790.0, + 788.0, + 790.0, + 817.0, + 297.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1376.0, + 1406.0, + 1376.0, + 1406.0, + 1417.0, + 294.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1408.0, + 1409.0, + 1408.0, + 1409.0, + 1445.0, + 294.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1440.0, + 1406.0, + 1440.0, + 1406.0, + 1475.0, + 294.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1471.0, + 1405.0, + 1471.0, + 1405.0, + 1506.0, + 295.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1500.0, + 1404.0, + 1500.0, + 1404.0, + 1537.0, + 292.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1532.0, + 1404.0, + 1532.0, + 1404.0, + 1566.0, + 295.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1563.0, + 720.0, + 1563.0, + 720.0, + 1597.0, + 295.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1855.0, + 1406.0, + 1855.0, + 1406.0, + 1894.0, + 294.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1890.0, + 1403.0, + 1890.0, + 1403.0, + 1923.0, + 295.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1916.0, + 1407.0, + 1916.0, + 1407.0, + 1955.0, + 294.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1947.0, + 1406.0, + 1947.0, + 1406.0, + 1986.0, + 293.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1982.0, + 940.0, + 1982.0, + 940.0, + 2015.0, + 296.0, + 2015.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 883, + 1404, + 883, + 1404, + 1098, + 297, + 1098 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1833, + 1404, + 1833, + 1404, + 1956, + 299, + 1956 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 338, + 1154, + 1362, + 1154, + 1362, + 1569, + 338, + 1569 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 283, + 1403, + 283, + 1403, + 407, + 298, + 407 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 300, + 421, + 1402, + 421, + 1402, + 515, + 300, + 515 + ], + "score": 0.962 + }, + { + "category_id": 4, + "poly": [ + 295, + 1594, + 1406, + 1594, + 1406, + 1718, + 295, + 1718 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 297, + 1971, + 1400, + 1971, + 1400, + 2034, + 297, + 2034 + ], + "score": 0.944 + }, + { + "category_id": 0, + "poly": [ + 300, + 550, + 604, + 550, + 604, + 581, + 300, + 581 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 104, + 300, + 104 + ], + "score": 0.889 + }, + { + "category_id": 0, + "poly": [ + 301, + 230, + 827, + 230, + 827, + 261, + 301, + 261 + ], + "score": 0.869 + }, + { + "category_id": 0, + "poly": [ + 302, + 1775, + 803, + 1775, + 803, + 1807, + 302, + 1807 + ], + "score": 0.866 + }, + { + "category_id": 1, + "poly": [ + 297, + 602, + 1404, + 602, + 1404, + 788, + 297, + 788 + ], + "score": 0.86 + }, + { + "category_id": 0, + "poly": [ + 300, + 826, + 1084, + 826, + 1084, + 857, + 300, + 857 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.79 + }, + { + "category_id": 13, + "poly": [ + 728, + 1005, + 1016, + 1005, + 1016, + 1039, + 728, + 1039 + ], + "score": 0.91, + "latex": "\\lambda \\in \\{ 0 . 1 , \\bar { 0 } . 6 6 6 , \\bar { 1 . 0 } , 2 . 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 1149, + 455, + 1181, + 455, + 1181, + 484, + 1149, + 484 + ], + "score": 0.89, + "latex": "F _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 587, + 1539, + 653, + 1539, + 653, + 1565, + 587, + 1565 + ], + "score": 0.83, + "latex": "\\lambda = 2" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1318, + 1349, + 1318, + 1349, + 1346, + 1263, + 1346 + ], + "score": 0.83, + "latex": "\\lambda = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 907, + 1538, + 1018, + 1538, + 1018, + 1566, + 907, + 1566 + ], + "score": 0.79, + "latex": "\\lambda = 0 . 6 6 6" + }, + { + "category_id": 13, + "poly": [ + 702, + 1657, + 721, + 1657, + 721, + 1683, + 702, + 1683 + ], + "score": 0.78, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 930, + 1319, + 996, + 1319, + 996, + 1345, + 930, + 1345 + ], + "score": 0.76, + "latex": "\\lambda = 1" + }, + { + "category_id": 13, + "poly": [ + 464, + 916, + 485, + 916, + 485, + 943, + 464, + 943 + ], + "score": 0.74, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 1214, + 1626, + 1233, + 1626, + 1233, + 1652, + 1214, + 1652 + ], + "score": 0.73, + "latex": "\\lambda" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1175.0, + 1008.0, + 1175.0, + 1008.0, + 1189.0, + 994.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1175.0, + 1353.0, + 1175.0, + 1353.0, + 1190.0, + 1338.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1204.0, + 663.0, + 1204.0, + 663.0, + 1210.0, + 657.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1205.0, + 777.0, + 1205.0, + 777.0, + 1213.0, + 766.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1198.0, + 1008.0, + 1198.0, + 1008.0, + 1212.0, + 994.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1198.0, + 1353.0, + 1198.0, + 1353.0, + 1213.0, + 1337.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1220.0, + 1008.0, + 1220.0, + 1008.0, + 1234.0, + 993.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1220.0, + 1353.0, + 1220.0, + 1353.0, + 1235.0, + 1337.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1241.0, + 719.0, + 1241.0, + 719.0, + 1250.0, + 710.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1243.0, + 1008.0, + 1243.0, + 1008.0, + 1258.0, + 993.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1242.0, + 1355.0, + 1242.0, + 1355.0, + 1259.0, + 1334.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1267.0, + 1012.0, + 1267.0, + 1012.0, + 1276.0, + 996.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1266.0, + 1356.0, + 1266.0, + 1356.0, + 1279.0, + 1339.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1298.0, + 385.0, + 1298.0, + 385.0, + 1307.0, + 370.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1295.0, + 733.0, + 1295.0, + 733.0, + 1311.0, + 710.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1295.0, + 842.0, + 1295.0, + 842.0, + 1310.0, + 827.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1294.0, + 990.0, + 1294.0, + 990.0, + 1311.0, + 971.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1296.0, + 1076.0, + 1296.0, + 1076.0, + 1310.0, + 1056.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1299.0, + 1185.0, + 1299.0, + 1185.0, + 1307.0, + 1173.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1298.0, + 1331.0, + 1298.0, + 1331.0, + 1309.0, + 1320.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1318.0, + 620.0, + 1318.0, + 620.0, + 1348.0, + 392.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1317.0, + 929.0, + 1317.0, + 929.0, + 1348.0, + 700.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1317.0, + 1262.0, + 1317.0, + 1262.0, + 1348.0, + 997.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1382.0, + 1007.0, + 1382.0, + 1007.0, + 1390.0, + 998.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1397.0, + 1007.0, + 1397.0, + 1007.0, + 1409.0, + 998.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1396.0, + 1353.0, + 1396.0, + 1353.0, + 1410.0, + 1338.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1420.0, + 665.0, + 1420.0, + 665.0, + 1432.0, + 654.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1421.0, + 1007.0, + 1421.0, + 1007.0, + 1432.0, + 998.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1419.0, + 1353.0, + 1419.0, + 1353.0, + 1433.0, + 1338.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1430.0, + 491.0, + 1430.0, + 491.0, + 1452.0, + 464.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1443.0, + 665.0, + 1443.0, + 665.0, + 1452.0, + 656.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1436.0, + 886.0, + 1436.0, + 886.0, + 1455.0, + 831.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1443.0, + 1007.0, + 1443.0, + 1007.0, + 1452.0, + 998.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1439.0, + 1119.0, + 1439.0, + 1119.0, + 1447.0, + 1110.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1440.0, + 1355.0, + 1440.0, + 1355.0, + 1457.0, + 1335.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1462.0, + 376.0, + 1462.0, + 376.0, + 1471.0, + 368.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1464.0, + 666.0, + 1464.0, + 666.0, + 1479.0, + 650.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1464.0, + 1008.0, + 1464.0, + 1008.0, + 1478.0, + 994.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1455.0, + 1263.0, + 1455.0, + 1263.0, + 1474.0, + 1216.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1464.0, + 1352.0, + 1464.0, + 1352.0, + 1479.0, + 1336.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1483.0, + 672.0, + 1483.0, + 672.0, + 1501.0, + 649.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1488.0, + 1013.0, + 1488.0, + 1013.0, + 1497.0, + 998.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1486.0, + 1357.0, + 1486.0, + 1357.0, + 1499.0, + 1337.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1516.0, + 391.0, + 1516.0, + 391.0, + 1532.0, + 368.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1516.0, + 500.0, + 1516.0, + 500.0, + 1530.0, + 484.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1516.0, + 647.0, + 1516.0, + 647.0, + 1529.0, + 629.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1514.0, + 734.0, + 1514.0, + 734.0, + 1532.0, + 710.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1516.0, + 843.0, + 1516.0, + 843.0, + 1530.0, + 827.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1514.0, + 992.0, + 1514.0, + 992.0, + 1532.0, + 972.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1514.0, + 1077.0, + 1514.0, + 1077.0, + 1532.0, + 1053.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1519.0, + 1185.0, + 1519.0, + 1185.0, + 1528.0, + 1173.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1514.0, + 1337.0, + 1514.0, + 1337.0, + 1532.0, + 1315.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1536.0, + 586.0, + 1536.0, + 586.0, + 1568.0, + 356.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1536.0, + 906.0, + 1536.0, + 906.0, + 1568.0, + 654.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 1536.0, + 1022.0, + 1536.0, + 1022.0, + 1568.0, + 1019.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1534.0, + 1276.0, + 1534.0, + 1276.0, + 1574.0, + 1109.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1202.5, + 911.0, + 1202.5, + 911.0, + 1236.5, + 780.0, + 1236.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.75, + 1209.0, + 1155.75, + 1209.0, + 1155.75, + 1220.5, + 1122.75, + 1220.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1223.5, + 983.0, + 1223.5, + 983.0, + 1253.5, + 915.0, + 1253.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1226.0, + 1309.0, + 1226.0, + 1309.0, + 1258.0, + 1246.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.75, + 1245.0, + 638.75, + 1245.0, + 638.75, + 1274.0, + 593.75, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1424.5, + 832.0, + 1424.5, + 832.0, + 1447.5, + 783.0, + 1447.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.75, + 1442.0, + 985.75, + 1442.0, + 985.75, + 1477.5, + 906.75, + 1477.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1438.0, + 1227.0, + 1438.0, + 1227.0, + 1464.5, + 1119.0, + 1464.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1453.5, + 639.0, + 1453.5, + 639.0, + 1484.5, + 576.0, + 1484.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1592.0, + 1408.0, + 1592.0, + 1408.0, + 1630.0, + 293.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1626.0, + 1213.0, + 1626.0, + 1213.0, + 1658.0, + 294.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1626.0, + 1404.0, + 1626.0, + 1404.0, + 1658.0, + 1234.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1654.0, + 701.0, + 1654.0, + 701.0, + 1692.0, + 293.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1654.0, + 1406.0, + 1654.0, + 1406.0, + 1692.0, + 722.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1686.0, + 702.0, + 1686.0, + 702.0, + 1721.0, + 294.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 548.0, + 608.0, + 548.0, + 608.0, + 584.0, + 296.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 832.0, + 229.0, + 832.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1771.0, + 809.0, + 1771.0, + 809.0, + 1812.0, + 294.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 823.0, + 1087.0, + 823.0, + 1087.0, + 863.0, + 293.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 880.0, + 1405.0, + 880.0, + 1405.0, + 923.0, + 292.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 463.0, + 913.0, + 463.0, + 950.0, + 294.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 913.0, + 1404.0, + 913.0, + 1404.0, + 950.0, + 486.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 946.0, + 1404.0, + 946.0, + 1404.0, + 981.0, + 295.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 976.0, + 1405.0, + 976.0, + 1405.0, + 1013.0, + 292.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1005.0, + 727.0, + 1005.0, + 727.0, + 1040.0, + 295.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1005.0, + 1404.0, + 1005.0, + 1404.0, + 1040.0, + 1017.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1037.0, + 1405.0, + 1037.0, + 1405.0, + 1072.0, + 295.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1068.0, + 843.0, + 1068.0, + 843.0, + 1099.0, + 295.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1835.0, + 1408.0, + 1835.0, + 1408.0, + 1868.0, + 294.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1862.0, + 1405.0, + 1862.0, + 1405.0, + 1900.0, + 293.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1895.0, + 1405.0, + 1895.0, + 1405.0, + 1929.0, + 293.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1926.0, + 1118.0, + 1926.0, + 1118.0, + 1962.0, + 294.0, + 1962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 284.0, + 1403.0, + 284.0, + 1403.0, + 317.0, + 294.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 316.0, + 1404.0, + 316.0, + 1404.0, + 349.0, + 294.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 346.0, + 1403.0, + 346.0, + 1403.0, + 382.0, + 293.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 377.0, + 1016.0, + 377.0, + 1016.0, + 409.0, + 294.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 417.0, + 1408.0, + 417.0, + 1408.0, + 458.0, + 294.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 451.0, + 1148.0, + 451.0, + 1148.0, + 487.0, + 294.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 451.0, + 1406.0, + 451.0, + 1406.0, + 487.0, + 1182.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 484.0, + 683.0, + 484.0, + 683.0, + 518.0, + 295.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1967.0, + 1404.0, + 1967.0, + 1404.0, + 2010.0, + 292.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 604.0, + 1405.0, + 604.0, + 1405.0, + 636.0, + 295.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 635.0, + 1401.0, + 635.0, + 1401.0, + 667.0, + 296.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 662.0, + 1406.0, + 662.0, + 1406.0, + 699.0, + 294.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 692.0, + 1404.0, + 692.0, + 1404.0, + 729.0, + 292.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 723.0, + 1405.0, + 723.0, + 1405.0, + 761.0, + 292.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 756.0, + 492.0, + 756.0, + 492.0, + 793.0, + 290.0, + 793.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 379, + 1405, + 379, + 1405, + 570, + 297, + 570 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1283, + 1405, + 1283, + 1405, + 1498, + 297, + 1498 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1812, + 1404, + 1812, + 1404, + 1998, + 298, + 1998 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 583, + 1406, + 583, + 1406, + 783, + 296, + 783 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1611, + 1405, + 1611, + 1405, + 1797, + 298, + 1797 + ], + "score": 0.98 + }, + { + "category_id": 3, + "poly": [ + 579, + 817, + 1114, + 817, + 1114, + 1115, + 579, + 1115 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1401, + 229, + 1401, + 294, + 295, + 294 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 289, + 1145, + 1399, + 1145, + 1399, + 1179, + 289, + 1179 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1543, + 544, + 1543, + 544, + 1578, + 300, + 1578 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 104, + 300, + 104 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 301, + 1229, + 718, + 1229, + 718, + 1261, + 301, + 1261 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 301, + 326, + 727, + 326, + 727, + 357, + 301, + 357 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.794 + }, + { + "category_id": 13, + "poly": [ + 563, + 685, + 664, + 685, + 664, + 728, + 563, + 728 + ], + "score": 0.93, + "latex": "\\{ y _ { i j } \\} _ { j = 1 } ^ { N _ { i } }" + }, + { + "category_id": 13, + "poly": [ + 1141, + 614, + 1405, + 614, + 1405, + 658, + 1141, + 658 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { 1 } { M } \\sum _ { i = 1 } ^ { N } \\sum _ { j = 1 } ^ { N _ { i } } \\mathbb { 1 } y _ { i j } = } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1322, + 688, + 1397, + 688, + 1397, + 721, + 1322, + 721 + ], + "score": 0.9, + "latex": "8 2 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 367, + 691, + 394, + 691, + 394, + 723, + 367, + 723 + ], + "score": 0.88, + "latex": "\\hat { y } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 745, + 658, + 782, + 658, + 782, + 689, + 745, + 689 + ], + "score": 0.87, + "latex": "y _ { i j }" + }, + { + "category_id": 13, + "poly": [ + 957, + 441, + 1157, + 441, + 1157, + 479, + 957, + 479 + ], + "score": 0.87, + "latex": "\\cdot \\mathrm { g } \\operatorname* { m a x } _ { k } p ( \\hat { \\phi } _ { k } \\mid \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 406, + 1764, + 437, + 1764, + 437, + 1793, + 406, + 1793 + ], + "score": 0.86, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 655, + 322, + 655, + 322, + 687, + 297, + 687 + ], + "score": 0.85, + "latex": "\\hat { y } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 406, + 655, + 439, + 655, + 439, + 683, + 406, + 683 + ], + "score": 0.84, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1237, + 658, + 1253, + 658, + 1253, + 687, + 1237, + 687 + ], + "score": 0.82, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 297, + 693, + 310, + 693, + 310, + 719, + 297, + 719 + ], + "score": 0.75, + "latex": "i" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 815.0, + 1121.0, + 815.0, + 1121.0, + 847.0, + 606.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 881.0, + 641.0, + 881.0, + 641.0, + 912.0, + 601.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 906.0, + 610.0, + 906.0, + 610.0, + 1009.0, + 576.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 924.0, + 922.0, + 924.0, + 922.0, + 956.0, + 848.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 938.0, + 638.0, + 938.0, + 638.0, + 966.0, + 602.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 945.0, + 904.0, + 945.0, + 904.0, + 985.0, + 847.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 992.0, + 640.0, + 992.0, + 640.0, + 1024.0, + 601.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 1049.0, + 639.0, + 1049.0, + 639.0, + 1077.0, + 602.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1070.0, + 684.0, + 1070.0, + 684.0, + 1089.0, + 668.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1073.0, + 733.0, + 1073.0, + 733.0, + 1086.0, + 725.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1065.0, + 796.0, + 1065.0, + 796.0, + 1092.0, + 773.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1065.0, + 847.0, + 1065.0, + 847.0, + 1091.0, + 828.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1069.0, + 902.0, + 1069.0, + 902.0, + 1090.0, + 882.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1067.0, + 955.0, + 1067.0, + 955.0, + 1091.0, + 936.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1071.0, + 1008.0, + 1071.0, + 1008.0, + 1089.0, + 991.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1086.0, + 916.0, + 1086.0, + 916.0, + 1117.0, + 812.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1141.0, + 1404.0, + 1141.0, + 1404.0, + 1183.0, + 293.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1539.0, + 550.0, + 1539.0, + 550.0, + 1587.0, + 292.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1227.0, + 723.0, + 1227.0, + 723.0, + 1265.0, + 295.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 326.0, + 731.0, + 326.0, + 731.0, + 359.0, + 296.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 378.0, + 1408.0, + 378.0, + 1408.0, + 416.0, + 294.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 412.0, + 1403.0, + 412.0, + 1403.0, + 445.0, + 295.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 446.0, + 956.0, + 446.0, + 956.0, + 482.0, + 294.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 446.0, + 1406.0, + 446.0, + 1406.0, + 482.0, + 1158.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 477.0, + 1405.0, + 477.0, + 1405.0, + 511.0, + 294.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 507.0, + 1405.0, + 507.0, + 1405.0, + 543.0, + 294.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 538.0, + 965.0, + 538.0, + 965.0, + 573.0, + 294.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1285.0, + 1405.0, + 1285.0, + 1405.0, + 1319.0, + 295.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1315.0, + 1406.0, + 1315.0, + 1406.0, + 1350.0, + 294.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1346.0, + 1405.0, + 1346.0, + 1405.0, + 1381.0, + 295.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1375.0, + 1410.0, + 1375.0, + 1410.0, + 1413.0, + 292.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1406.0, + 1405.0, + 1406.0, + 1405.0, + 1441.0, + 295.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1435.0, + 1405.0, + 1435.0, + 1405.0, + 1472.0, + 292.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1468.0, + 515.0, + 1468.0, + 515.0, + 1500.0, + 295.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1811.0, + 1408.0, + 1811.0, + 1408.0, + 1848.0, + 295.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1841.0, + 1404.0, + 1841.0, + 1404.0, + 1878.0, + 293.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1872.0, + 1406.0, + 1872.0, + 1406.0, + 1909.0, + 293.0, + 1909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1905.0, + 1404.0, + 1905.0, + 1404.0, + 1937.0, + 296.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1934.0, + 1405.0, + 1934.0, + 1405.0, + 1968.0, + 292.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1964.0, + 1396.0, + 1964.0, + 1396.0, + 1999.0, + 296.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 580.0, + 296.0, + 580.0, + 296.0, + 682.0, + 278.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 580.0, + 405.0, + 580.0, + 405.0, + 682.0, + 323.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 580.0, + 1140.0, + 580.0, + 1140.0, + 682.0, + 440.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 654.0, + 405.0, + 654.0, + 405.0, + 691.0, + 323.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 654.0, + 744.0, + 654.0, + 744.0, + 691.0, + 440.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 654.0, + 1236.0, + 654.0, + 1236.0, + 691.0, + 783.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 654.0, + 1406.0, + 654.0, + 1406.0, + 691.0, + 1254.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 672.0, + 296.0, + 672.0, + 296.0, + 757.0, + 284.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 672.0, + 366.0, + 672.0, + 366.0, + 757.0, + 311.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 672.0, + 562.0, + 672.0, + 562.0, + 757.0, + 395.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 672.0, + 1321.0, + 672.0, + 1321.0, + 757.0, + 665.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 672.0, + 1416.0, + 672.0, + 1416.0, + 757.0, + 1398.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 751.0, + 683.0, + 751.0, + 683.0, + 786.0, + 293.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 695.5, + 654.0, + 695.5, + 654.0, + 718.5, + 608.0, + 718.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1613.0, + 1403.0, + 1613.0, + 1403.0, + 1645.0, + 296.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1643.0, + 1407.0, + 1643.0, + 1407.0, + 1678.0, + 295.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1672.0, + 1406.0, + 1672.0, + 1406.0, + 1707.0, + 292.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1701.0, + 1409.0, + 1701.0, + 1409.0, + 1740.0, + 293.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1733.0, + 1406.0, + 1733.0, + 1406.0, + 1773.0, + 293.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1763.0, + 405.0, + 1763.0, + 405.0, + 1801.0, + 292.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1763.0, + 450.0, + 1763.0, + 450.0, + 1801.0, + 438.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 1401.0, + 261.0, + 1401.0, + 298.0, + 295.0, + 298.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 301, + 1437, + 789, + 1437, + 789, + 1470, + 301, + 1470 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.816 + }, + { + "category_id": 0, + "poly": [ + 300, + 227, + 489, + 227, + 489, + 262, + 300, + 262 + ], + "score": 0.754 + }, + { + "category_id": 1, + "poly": [ + 291, + 590, + 1404, + 590, + 1404, + 655, + 291, + 655 + ], + "score": 0.717 + }, + { + "category_id": 1, + "poly": [ + 298, + 956, + 1400, + 956, + 1400, + 1021, + 298, + 1021 + ], + "score": 0.713 + }, + { + "category_id": 1, + "poly": [ + 294, + 1658, + 1401, + 1658, + 1401, + 1723, + 294, + 1723 + ], + "score": 0.7 + }, + { + "category_id": 1, + "poly": [ + 289, + 1741, + 1404, + 1741, + 1404, + 1807, + 289, + 1807 + ], + "score": 0.69 + }, + { + "category_id": 1, + "poly": [ + 299, + 674, + 1402, + 674, + 1402, + 769, + 299, + 769 + ], + "score": 0.685 + }, + { + "category_id": 1, + "poly": [ + 297, + 278, + 1401, + 278, + 1401, + 343, + 297, + 343 + ], + "score": 0.684 + }, + { + "category_id": 1, + "poly": [ + 299, + 1238, + 1402, + 1238, + 1402, + 1334, + 299, + 1334 + ], + "score": 0.68 + }, + { + "category_id": 1, + "poly": [ + 296, + 1352, + 1402, + 1352, + 1402, + 1418, + 296, + 1418 + ], + "score": 0.672 + }, + { + "category_id": 1, + "poly": [ + 299, + 476, + 1401, + 476, + 1401, + 571, + 299, + 571 + ], + "score": 0.668 + }, + { + "category_id": 1, + "poly": [ + 297, + 872, + 1399, + 872, + 1399, + 938, + 297, + 938 + ], + "score": 0.664 + }, + { + "category_id": 1, + "poly": [ + 292, + 1490, + 1403, + 1490, + 1403, + 1555, + 292, + 1555 + ], + "score": 0.66 + }, + { + "category_id": 1, + "poly": [ + 296, + 788, + 1400, + 788, + 1400, + 855, + 296, + 855 + ], + "score": 0.654 + }, + { + "category_id": 1, + "poly": [ + 300, + 362, + 1398, + 362, + 1398, + 457, + 300, + 457 + ], + "score": 0.644 + }, + { + "category_id": 1, + "poly": [ + 302, + 1124, + 1403, + 1124, + 1403, + 1220, + 302, + 1220 + ], + "score": 0.642 + }, + { + "category_id": 1, + "poly": [ + 291, + 1574, + 1401, + 1574, + 1401, + 1640, + 291, + 1640 + ], + "score": 0.626 + }, + { + "category_id": 1, + "poly": [ + 300, + 1825, + 1404, + 1825, + 1404, + 1921, + 300, + 1921 + ], + "score": 0.605 + }, + { + "category_id": 1, + "poly": [ + 294, + 1040, + 1404, + 1040, + 1404, + 1106, + 294, + 1106 + ], + "score": 0.565 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1404, + 1942, + 1404, + 2034, + 299, + 2034 + ], + "score": 0.497 + }, + { + "category_id": 1, + "poly": [ + 300, + 227, + 489, + 227, + 489, + 262, + 300, + 262 + ], + "score": 0.115 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1437.0, + 793.0, + 1437.0, + 793.0, + 1473.0, + 296.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 585.0, + 1405.0, + 585.0, + 1405.0, + 630.0, + 292.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 622.0, + 825.0, + 622.0, + 825.0, + 655.0, + 324.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 958.0, + 1404.0, + 958.0, + 1404.0, + 994.0, + 294.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 989.0, + 707.0, + 989.0, + 707.0, + 1022.0, + 319.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1653.0, + 1406.0, + 1653.0, + 1406.0, + 1697.0, + 293.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1691.0, + 597.0, + 1691.0, + 597.0, + 1721.0, + 322.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1742.0, + 1408.0, + 1742.0, + 1408.0, + 1778.0, + 296.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1774.0, + 774.0, + 1774.0, + 774.0, + 1806.0, + 322.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 673.0, + 1408.0, + 673.0, + 1408.0, + 712.0, + 293.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 707.0, + 1407.0, + 707.0, + 1407.0, + 741.0, + 322.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 739.0, + 775.0, + 739.0, + 775.0, + 770.0, + 323.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 273.0, + 1406.0, + 273.0, + 1406.0, + 318.0, + 294.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 311.0, + 1244.0, + 311.0, + 1244.0, + 344.0, + 322.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1236.0, + 1407.0, + 1236.0, + 1407.0, + 1276.0, + 294.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1270.0, + 1404.0, + 1270.0, + 1404.0, + 1304.0, + 323.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1301.0, + 895.0, + 1301.0, + 895.0, + 1336.0, + 322.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1350.0, + 1406.0, + 1350.0, + 1406.0, + 1392.0, + 294.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1385.0, + 850.0, + 1385.0, + 850.0, + 1417.0, + 323.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 476.0, + 1403.0, + 476.0, + 1403.0, + 513.0, + 295.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 508.0, + 1405.0, + 508.0, + 1405.0, + 544.0, + 322.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 541.0, + 623.0, + 541.0, + 623.0, + 570.0, + 323.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 870.0, + 1403.0, + 870.0, + 1403.0, + 910.0, + 293.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 905.0, + 895.0, + 905.0, + 895.0, + 939.0, + 322.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1492.0, + 1405.0, + 1492.0, + 1405.0, + 1528.0, + 295.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1521.0, + 1145.0, + 1521.0, + 1145.0, + 1556.0, + 321.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 788.0, + 1405.0, + 788.0, + 1405.0, + 825.0, + 295.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 822.0, + 1253.0, + 822.0, + 1253.0, + 855.0, + 321.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 358.0, + 1404.0, + 358.0, + 1404.0, + 401.0, + 294.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 395.0, + 1404.0, + 395.0, + 1404.0, + 429.0, + 324.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 425.0, + 960.0, + 425.0, + 960.0, + 459.0, + 324.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1407.0, + 1122.0, + 1407.0, + 1163.0, + 293.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1157.0, + 1404.0, + 1157.0, + 1404.0, + 1192.0, + 323.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1188.0, + 894.0, + 1188.0, + 894.0, + 1221.0, + 323.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1576.0, + 1406.0, + 1576.0, + 1406.0, + 1612.0, + 295.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1606.0, + 1309.0, + 1606.0, + 1309.0, + 1641.0, + 322.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1823.0, + 1408.0, + 1823.0, + 1408.0, + 1864.0, + 293.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1858.0, + 1404.0, + 1858.0, + 1404.0, + 1893.0, + 323.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1889.0, + 803.0, + 1889.0, + 803.0, + 1923.0, + 323.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1035.0, + 1405.0, + 1035.0, + 1405.0, + 1083.0, + 293.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1072.0, + 1135.0, + 1072.0, + 1135.0, + 1108.0, + 323.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1940.0, + 1404.0, + 1940.0, + 1404.0, + 1977.0, + 295.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2007.0, + 323.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 2001.0, + 681.0, + 2001.0, + 681.0, + 2038.0, + 324.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.827 + }, + { + "category_id": 1, + "poly": [ + 299, + 1343, + 1399, + 1343, + 1399, + 1408, + 299, + 1408 + ], + "score": 0.606 + }, + { + "category_id": 1, + "poly": [ + 297, + 1531, + 1403, + 1531, + 1403, + 1626, + 297, + 1626 + ], + "score": 0.601 + }, + { + "category_id": 1, + "poly": [ + 295, + 1423, + 1403, + 1423, + 1403, + 1517, + 295, + 1517 + ], + "score": 0.593 + }, + { + "category_id": 1, + "poly": [ + 298, + 1014, + 1401, + 1014, + 1401, + 1110, + 298, + 1110 + ], + "score": 0.582 + }, + { + "category_id": 1, + "poly": [ + 295, + 1263, + 1399, + 1263, + 1399, + 1329, + 295, + 1329 + ], + "score": 0.565 + }, + { + "category_id": 1, + "poly": [ + 298, + 1641, + 1403, + 1641, + 1403, + 1737, + 298, + 1737 + ], + "score": 0.538 + }, + { + "category_id": 1, + "poly": [ + 300, + 228, + 1405, + 228, + 1405, + 323, + 300, + 323 + ], + "score": 0.537 + }, + { + "category_id": 1, + "poly": [ + 297, + 935, + 1402, + 935, + 1402, + 1000, + 297, + 1000 + ], + "score": 0.537 + }, + { + "category_id": 1, + "poly": [ + 292, + 746, + 1400, + 746, + 1400, + 812, + 292, + 812 + ], + "score": 0.53 + }, + { + "category_id": 1, + "poly": [ + 296, + 825, + 1398, + 825, + 1398, + 921, + 296, + 921 + ], + "score": 0.524 + }, + { + "category_id": 1, + "poly": [ + 297, + 1752, + 1402, + 1752, + 1402, + 1847, + 297, + 1847 + ], + "score": 0.501 + }, + { + "category_id": 1, + "poly": [ + 295, + 636, + 1403, + 636, + 1403, + 733, + 295, + 733 + ], + "score": 0.486 + }, + { + "category_id": 1, + "poly": [ + 295, + 1861, + 1400, + 1861, + 1400, + 1926, + 295, + 1926 + ], + "score": 0.486 + }, + { + "category_id": 1, + "poly": [ + 300, + 528, + 1402, + 528, + 1402, + 621, + 300, + 621 + ], + "score": 0.481 + }, + { + "category_id": 1, + "poly": [ + 302, + 338, + 1399, + 338, + 1399, + 434, + 302, + 434 + ], + "score": 0.453 + }, + { + "category_id": 1, + "poly": [ + 296, + 1122, + 1405, + 1122, + 1405, + 1248, + 296, + 1248 + ], + "score": 0.413 + }, + { + "category_id": 1, + "poly": [ + 293, + 448, + 1402, + 448, + 1402, + 514, + 293, + 514 + ], + "score": 0.41 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1406, + 1942, + 1406, + 2033, + 299, + 2033 + ], + "score": 0.344 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1345.0, + 1403.0, + 1345.0, + 1403.0, + 1381.0, + 295.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1374.0, + 928.0, + 1374.0, + 928.0, + 1412.0, + 323.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1530.0, + 1407.0, + 1530.0, + 1407.0, + 1570.0, + 295.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1564.0, + 1409.0, + 1564.0, + 1409.0, + 1602.0, + 320.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1589.0, + 400.0, + 1589.0, + 400.0, + 1630.0, + 320.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1421.0, + 1408.0, + 1421.0, + 1408.0, + 1460.0, + 293.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1455.0, + 1405.0, + 1455.0, + 1405.0, + 1489.0, + 323.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1485.0, + 623.0, + 1485.0, + 623.0, + 1516.0, + 322.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1014.0, + 1407.0, + 1014.0, + 1407.0, + 1052.0, + 293.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1045.0, + 1405.0, + 1045.0, + 1405.0, + 1082.0, + 322.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1076.0, + 940.0, + 1076.0, + 940.0, + 1113.0, + 321.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1261.0, + 1404.0, + 1261.0, + 1404.0, + 1303.0, + 294.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1296.0, + 1357.0, + 1296.0, + 1357.0, + 1331.0, + 321.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1638.0, + 1405.0, + 1638.0, + 1405.0, + 1683.0, + 295.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1672.0, + 1405.0, + 1672.0, + 1405.0, + 1711.0, + 322.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1705.0, + 972.0, + 1705.0, + 972.0, + 1740.0, + 322.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1404.0, + 228.0, + 1404.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 262.0, + 1407.0, + 262.0, + 1407.0, + 296.0, + 323.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 288.0, + 398.0, + 288.0, + 398.0, + 325.0, + 320.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 933.0, + 1407.0, + 933.0, + 1407.0, + 974.0, + 293.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 966.0, + 506.0, + 966.0, + 506.0, + 1001.0, + 321.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 744.0, + 1405.0, + 744.0, + 1405.0, + 786.0, + 295.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 779.0, + 1300.0, + 779.0, + 1300.0, + 811.0, + 321.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 824.0, + 1403.0, + 824.0, + 1403.0, + 864.0, + 292.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 855.0, + 1404.0, + 855.0, + 1404.0, + 896.0, + 322.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 889.0, + 1241.0, + 889.0, + 1241.0, + 923.0, + 321.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1752.0, + 1406.0, + 1752.0, + 1406.0, + 1790.0, + 296.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1784.0, + 1406.0, + 1784.0, + 1406.0, + 1818.0, + 322.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1814.0, + 1102.0, + 1814.0, + 1102.0, + 1848.0, + 324.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 636.0, + 1406.0, + 636.0, + 1406.0, + 675.0, + 295.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 669.0, + 1406.0, + 669.0, + 1406.0, + 705.0, + 320.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 698.0, + 975.0, + 698.0, + 975.0, + 735.0, + 322.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1857.0, + 1405.0, + 1857.0, + 1405.0, + 1900.0, + 293.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1892.0, + 539.0, + 1892.0, + 539.0, + 1926.0, + 323.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 527.0, + 1404.0, + 527.0, + 1404.0, + 565.0, + 295.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 558.0, + 1404.0, + 558.0, + 1404.0, + 595.0, + 321.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 589.0, + 612.0, + 589.0, + 612.0, + 620.0, + 323.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 338.0, + 1405.0, + 338.0, + 1405.0, + 376.0, + 294.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 366.0, + 1405.0, + 366.0, + 1405.0, + 408.0, + 321.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 401.0, + 1035.0, + 401.0, + 1035.0, + 436.0, + 321.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1126.0, + 1405.0, + 1126.0, + 1405.0, + 1159.0, + 296.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1156.0, + 1407.0, + 1156.0, + 1407.0, + 1192.0, + 320.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1186.0, + 1408.0, + 1186.0, + 1408.0, + 1223.0, + 320.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1217.0, + 469.0, + 1217.0, + 469.0, + 1251.0, + 322.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 450.0, + 1403.0, + 450.0, + 1403.0, + 482.0, + 296.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 480.0, + 1145.0, + 480.0, + 1145.0, + 515.0, + 321.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1976.0, + 296.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1971.0, + 1408.0, + 1971.0, + 1408.0, + 2007.0, + 321.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1999.0, + 399.0, + 1999.0, + 399.0, + 2035.0, + 319.0, + 2035.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1291, + 1405, + 1291, + 1405, + 1538, + 298, + 1538 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1093, + 1406, + 1093, + 1406, + 1278, + 298, + 1278 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1787, + 1404, + 1787, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 295, + 387, + 1406, + 387, + 1406, + 511, + 295, + 511 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 301, + 1553, + 1402, + 1553, + 1402, + 1645, + 301, + 1645 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 632, + 575, + 1068, + 575, + 1068, + 807, + 632, + 807 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 300, + 1023, + 653, + 1023, + 653, + 1054, + 300, + 1054 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 301, + 1717, + 604, + 1717, + 604, + 1749, + 301, + 1749 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.848 + }, + { + "category_id": 0, + "poly": [ + 300, + 316, + 606, + 316, + 606, + 348, + 300, + 348 + ], + "score": 0.826 + }, + { + "category_id": 4, + "poly": [ + 293, + 836, + 1403, + 836, + 1403, + 899, + 293, + 899 + ], + "score": 0.778 + }, + { + "category_id": 0, + "poly": [ + 301, + 227, + 508, + 227, + 508, + 261, + 301, + 261 + ], + "score": 0.462 + }, + { + "category_id": 1, + "poly": [ + 301, + 227, + 508, + 227, + 508, + 261, + 301, + 261 + ], + "score": 0.454 + }, + { + "category_id": 1, + "poly": [ + 293, + 836, + 1403, + 836, + 1403, + 899, + 293, + 899 + ], + "score": 0.238 + }, + { + "category_id": 1, + "poly": [ + 300, + 316, + 606, + 316, + 606, + 348, + 300, + 348 + ], + "score": 0.096 + }, + { + "category_id": 13, + "poly": [ + 670, + 1294, + 716, + 1294, + 716, + 1329, + 670, + 1329 + ], + "score": 0.92, + "latex": "\\zeta _ { i , k }" + }, + { + "category_id": 13, + "poly": [ + 997, + 417, + 1081, + 417, + 1081, + 447, + 997, + 447 + ], + "score": 0.92, + "latex": "\\sigma ^ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1294, + 1187, + 1294, + 1187, + 1326, + 1127, + 1326 + ], + "score": 0.91, + "latex": "( i , k )" + }, + { + "category_id": 13, + "poly": [ + 368, + 1185, + 532, + 1185, + 532, + 1218, + 368, + 1218 + ], + "score": 0.91, + "latex": "( s \\sim \\mathcal { N } ( 0 , I ) )" + }, + { + "category_id": 13, + "poly": [ + 1008, + 1416, + 1098, + 1416, + 1098, + 1445, + 1008, + 1445 + ], + "score": 0.9, + "latex": "m * 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1149, + 1417, + 1218, + 1417, + 1218, + 1444, + 1149, + 1444 + ], + "score": 0.89, + "latex": "m * 4" + }, + { + "category_id": 13, + "poly": [ + 675, + 836, + 707, + 836, + 707, + 866, + 675, + 866 + ], + "score": 0.88, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 417, + 330, + 417, + 330, + 447, + 297, + 447 + ], + "score": 0.85, + "latex": "\\sigma ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 501, + 1290, + 521, + 1290, + 521, + 1326, + 501, + 1326 + ], + "score": 0.83, + "latex": "\\bar { \\zeta }" + }, + { + "category_id": 13, + "poly": [ + 766, + 1248, + 785, + 1248, + 785, + 1278, + 766, + 1278 + ], + "score": 0.81, + "latex": "\\zeta" + }, + { + "category_id": 13, + "poly": [ + 845, + 1447, + 865, + 1447, + 865, + 1473, + 845, + 1473 + ], + "score": 0.79, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 760, + 843, + 779, + 843, + 779, + 869, + 760, + 869 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 841, + 1294, + 873, + 1294, + 873, + 1326, + 841, + 1326 + ], + "score": 0.77, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 807, + 1477, + 827, + 1477, + 827, + 1503, + 807, + 1503 + ], + "score": 0.76, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 1305, + 1420, + 1333, + 1420, + 1333, + 1444, + 1305, + 1444 + ], + "score": 0.75, + "latex": "m" + }, + { + "category_id": 13, + "poly": [ + 804, + 1417, + 824, + 1417, + 824, + 1443, + 804, + 1443 + ], + "score": 0.75, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 1251, + 1330, + 1267, + 1330, + 1267, + 1351, + 1251, + 1351 + ], + "score": 0.32, + "latex": "\\mathbf { Z }" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 584.0, + 678.0, + 584.0, + 678.0, + 608.0, + 648.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 586.0, + 713.0, + 586.0, + 713.0, + 604.0, + 697.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 586.0, + 783.0, + 586.0, + 783.0, + 604.0, + 767.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 587.0, + 893.0, + 587.0, + 893.0, + 604.0, + 874.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 588.0, + 931.0, + 588.0, + 931.0, + 603.0, + 918.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 591.0, + 1010.0, + 591.0, + 1010.0, + 600.0, + 1002.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 587.0, + 1048.0, + 587.0, + 1048.0, + 604.0, + 1032.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 641.0, + 655.0, + 641.0, + 655.0, + 714.0, + 632.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 642.0, + 682.0, + 642.0, + 682.0, + 709.0, + 644.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 656.0, + 763.0, + 656.0, + 763.0, + 676.0, + 711.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 669.0, + 867.0, + 669.0, + 867.0, + 680.0, + 851.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 668.0, + 885.0, + 668.0, + 885.0, + 681.0, + 874.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 667.0, + 944.0, + 667.0, + 944.0, + 684.0, + 928.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 668.0, + 1019.0, + 668.0, + 1019.0, + 683.0, + 991.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 674.0, + 708.0, + 674.0, + 708.0, + 691.0, + 692.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 723.0, + 779.0, + 723.0, + 779.0, + 744.0, + 717.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 724.0, + 712.0, + 724.0, + 712.0, + 746.0, + 691.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 744.0, + 680.0, + 744.0, + 680.0, + 767.0, + 649.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 743.0, + 712.0, + 743.0, + 712.0, + 763.0, + 693.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 741.0, + 788.0, + 741.0, + 788.0, + 762.0, + 719.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 746.0, + 1042.0, + 746.0, + 1042.0, + 764.0, + 1025.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 768.0, + 707.0, + 768.0, + 707.0, + 793.0, + 678.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 762.0, + 769.0, + 762.0, + 769.0, + 795.0, + 736.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 768.0, + 828.0, + 768.0, + 828.0, + 791.0, + 796.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 768.0, + 886.0, + 768.0, + 886.0, + 793.0, + 858.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 769.0, + 948.0, + 769.0, + 948.0, + 793.0, + 919.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 768.0, + 1008.0, + 768.0, + 1008.0, + 793.0, + 978.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 768.0, + 1067.0, + 768.0, + 1067.0, + 792.0, + 1039.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 784.0, + 908.0, + 784.0, + 908.0, + 811.0, + 833.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 665.5, + 854.0, + 665.5, + 854.0, + 692.5, + 728.0, + 692.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1023.0, + 654.0, + 1023.0, + 654.0, + 1056.0, + 297.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1713.0, + 610.0, + 1713.0, + 610.0, + 1755.0, + 295.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 316.0, + 609.0, + 316.0, + 609.0, + 352.0, + 296.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 834.0, + 674.0, + 834.0, + 674.0, + 870.0, + 295.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 834.0, + 759.0, + 834.0, + 759.0, + 870.0, + 708.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 834.0, + 1401.0, + 834.0, + 1401.0, + 870.0, + 780.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 865.0, + 562.0, + 865.0, + 562.0, + 901.0, + 295.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 221.0, + 514.0, + 221.0, + 514.0, + 270.0, + 294.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1289.0, + 500.0, + 1289.0, + 500.0, + 1334.0, + 291.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1289.0, + 669.0, + 1289.0, + 669.0, + 1334.0, + 522.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1289.0, + 840.0, + 1289.0, + 840.0, + 1334.0, + 717.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1289.0, + 1126.0, + 1289.0, + 1126.0, + 1334.0, + 874.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1289.0, + 1408.0, + 1289.0, + 1408.0, + 1334.0, + 1188.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1324.0, + 1250.0, + 1324.0, + 1250.0, + 1358.0, + 293.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1324.0, + 1407.0, + 1324.0, + 1407.0, + 1358.0, + 1268.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1353.0, + 1406.0, + 1353.0, + 1406.0, + 1389.0, + 293.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1383.0, + 1407.0, + 1383.0, + 1407.0, + 1419.0, + 293.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 803.0, + 1416.0, + 803.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1416.0, + 1007.0, + 1416.0, + 1007.0, + 1449.0, + 825.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1416.0, + 1148.0, + 1416.0, + 1148.0, + 1449.0, + 1099.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1416.0, + 1304.0, + 1416.0, + 1304.0, + 1449.0, + 1219.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1449.0, + 1334.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1445.0, + 844.0, + 1445.0, + 844.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1445.0, + 1405.0, + 1445.0, + 1405.0, + 1480.0, + 866.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1476.0, + 806.0, + 1476.0, + 806.0, + 1510.0, + 295.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1476.0, + 1405.0, + 1476.0, + 1405.0, + 1510.0, + 828.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1507.0, + 649.0, + 1507.0, + 649.0, + 1540.0, + 294.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1094.0, + 1406.0, + 1094.0, + 1406.0, + 1130.0, + 295.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1407.0, + 1122.0, + 1407.0, + 1158.0, + 293.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1152.0, + 1408.0, + 1152.0, + 1408.0, + 1188.0, + 295.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1185.0, + 367.0, + 1185.0, + 367.0, + 1221.0, + 296.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1185.0, + 1406.0, + 1185.0, + 1406.0, + 1221.0, + 533.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1216.0, + 1407.0, + 1216.0, + 1407.0, + 1250.0, + 293.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1244.0, + 765.0, + 1244.0, + 765.0, + 1284.0, + 293.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1244.0, + 795.0, + 1244.0, + 795.0, + 1284.0, + 786.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1788.0, + 1404.0, + 1788.0, + 1404.0, + 1824.0, + 295.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1854.0, + 296.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1849.0, + 1408.0, + 1849.0, + 1408.0, + 1888.0, + 292.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1920.0, + 292.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1913.0, + 1406.0, + 1913.0, + 1406.0, + 1944.0, + 293.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1977.0, + 292.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1408.0, + 1972.0, + 1408.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 454.0, + 2001.0, + 454.0, + 2035.0, + 295.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 388.0, + 1406.0, + 388.0, + 1406.0, + 421.0, + 294.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 416.0, + 296.0, + 416.0, + 296.0, + 453.0, + 292.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 416.0, + 996.0, + 416.0, + 996.0, + 453.0, + 331.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 416.0, + 1406.0, + 416.0, + 1406.0, + 453.0, + 1082.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 450.0, + 1406.0, + 450.0, + 1406.0, + 483.0, + 293.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 482.0, + 742.0, + 482.0, + 742.0, + 512.0, + 294.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1552.0, + 1406.0, + 1552.0, + 1406.0, + 1589.0, + 296.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1585.0, + 1406.0, + 1585.0, + 1406.0, + 1618.0, + 296.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1615.0, + 684.0, + 1615.0, + 684.0, + 1648.0, + 295.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 221.0, + 514.0, + 221.0, + 514.0, + 270.0, + 294.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 834.0, + 674.0, + 834.0, + 674.0, + 870.0, + 295.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 834.0, + 759.0, + 834.0, + 759.0, + 870.0, + 708.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 834.0, + 1401.0, + 834.0, + 1401.0, + 870.0, + 780.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 865.0, + 562.0, + 865.0, + 562.0, + 901.0, + 295.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 316.0, + 609.0, + 316.0, + 609.0, + 352.0, + 296.0, + 352.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1334, + 1406, + 1334, + 1406, + 1456, + 298, + 1456 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1123, + 1404, + 1123, + 1404, + 1246, + 299, + 1246 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1544, + 1406, + 1544, + 1406, + 1668, + 299, + 1668 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 943, + 1405, + 943, + 1405, + 1037, + 299, + 1037 + ], + "score": 0.961 + }, + { + "category_id": 0, + "poly": [ + 299, + 1282, + 893, + 1282, + 893, + 1312, + 299, + 1312 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 301, + 890, + 632, + 890, + 632, + 922, + 301, + 922 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 300, + 1070, + 744, + 1070, + 744, + 1103, + 300, + 1103 + ], + "score": 0.885 + }, + { + "category_id": 0, + "poly": [ + 300, + 834, + 606, + 834, + 606, + 865, + 300, + 865 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.88 + }, + { + "category_id": 0, + "poly": [ + 302, + 1491, + 786, + 1491, + 786, + 1522, + 302, + 1522 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.84 + }, + { + "category_id": 1, + "poly": [ + 271, + 465, + 1092, + 465, + 1092, + 762, + 271, + 762 + ], + "score": 0.529 + }, + { + "category_id": 1, + "poly": [ + 295, + 278, + 1023, + 278, + 1023, + 337, + 295, + 337 + ], + "score": 0.478 + }, + { + "category_id": 1, + "poly": [ + 275, + 435, + 598, + 435, + 598, + 466, + 275, + 466 + ], + "score": 0.476 + }, + { + "category_id": 1, + "poly": [ + 301, + 238, + 778, + 238, + 778, + 270, + 301, + 270 + ], + "score": 0.365 + }, + { + "category_id": 1, + "poly": [ + 282, + 340, + 754, + 340, + 754, + 369, + 282, + 369 + ], + "score": 0.316 + }, + { + "category_id": 8, + "poly": [ + 310, + 373, + 736, + 373, + 736, + 407, + 310, + 407 + ], + "score": 0.276 + }, + { + "category_id": 0, + "poly": [ + 301, + 238, + 778, + 238, + 778, + 270, + 301, + 270 + ], + "score": 0.273 + }, + { + "category_id": 0, + "poly": [ + 301, + 1491, + 785, + 1491, + 785, + 1522, + 301, + 1522 + ], + "score": 0.108 + }, + { + "category_id": 13, + "poly": [ + 404, + 660, + 607, + 660, + 607, + 697, + 404, + 697 + ], + "score": 0.92, + "latex": "\\boldsymbol { Z } _ { p } \\sim \\mathcal { N } ( \\pmb { \\mu } _ { p } , \\pmb { \\Sigma } _ { p } )" + }, + { + "category_id": 13, + "poly": [ + 516, + 500, + 559, + 500, + 559, + 534, + 516, + 534 + ], + "score": 0.9, + "latex": "\\zeta _ { i , k }" + }, + { + "category_id": 13, + "poly": [ + 505, + 566, + 548, + 566, + 548, + 599, + 505, + 599 + ], + "score": 0.89, + "latex": "\\zeta _ { i , k }" + }, + { + "category_id": 13, + "poly": [ + 472, + 467, + 501, + 467, + 501, + 497, + 472, + 497 + ], + "score": 0.88, + "latex": "\\zeta _ { k }" + }, + { + "category_id": 13, + "poly": [ + 386, + 699, + 420, + 699, + 420, + 730, + 386, + 730 + ], + "score": 0.86, + "latex": "z _ { p }" + }, + { + "category_id": 13, + "poly": [ + 485, + 698, + 524, + 698, + 524, + 730, + 485, + 730 + ], + "score": 0.85, + "latex": "Z _ { p }" + }, + { + "category_id": 13, + "poly": [ + 735, + 730, + 767, + 730, + 767, + 761, + 735, + 761 + ], + "score": 0.85, + "latex": "z _ { p }" + }, + { + "category_id": 13, + "poly": [ + 663, + 278, + 684, + 278, + 684, + 309, + 663, + 309 + ], + "score": 0.8, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 777, + 501, + 797, + 501, + 797, + 528, + 777, + 528 + ], + "score": 0.77, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 964, + 282, + 1024, + 282, + 1024, + 309, + 964, + 309 + ], + "score": 0.76, + "latex": "{ \\boldsymbol { \\mathbf { \\mathit { x } } } } _ { 1 : m }" + }, + { + "category_id": 13, + "poly": [ + 400, + 338, + 546, + 338, + 546, + 370, + 400, + 370 + ], + "score": 0.73, + "latex": "m e l _ { i , k } , t e x t _ { i }" + }, + { + "category_id": 13, + "poly": [ + 868, + 567, + 888, + 567, + 888, + 593, + 868, + 593 + ], + "score": 0.71, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 311, + 371, + 606, + 371, + 606, + 408, + 311, + 408 + ], + "score": 0.68, + "latex": "\\zeta _ { i , k } \\gets f ^ { - 1 } ( m e l _ { i , k } , t e x t _ { i } ," + }, + { + "category_id": 13, + "poly": [ + 295, + 629, + 332, + 629, + 332, + 663, + 295, + 663 + ], + "score": 0.36, + "latex": "\\mu _ { p }" + }, + { + "category_id": 13, + "poly": [ + 339, + 497, + 408, + 497, + 408, + 533, + 339, + 533 + ], + "score": 0.35, + "latex": "\\bar { \\zeta _ { k } } \\gets" + }, + { + "category_id": 13, + "poly": [ + 343, + 627, + 419, + 627, + 419, + 662, + 343, + 662 + ], + "score": 0.34, + "latex": "\\Sigma _ { p } " + }, + { + "category_id": 13, + "poly": [ + 339, + 497, + 494, + 497, + 494, + 533, + 339, + 533 + ], + "score": 0.26, + "latex": "\\bar { \\zeta _ { k } } \\gets \\mathrm { C o m p u }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1281.0, + 894.0, + 1281.0, + 894.0, + 1315.0, + 296.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 890.0, + 633.0, + 890.0, + 633.0, + 926.0, + 297.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1071.0, + 747.0, + 1071.0, + 747.0, + 1106.0, + 296.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 833.0, + 609.0, + 833.0, + 609.0, + 868.0, + 296.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1490.0, + 789.0, + 1490.0, + 789.0, + 1525.0, + 298.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 235.0, + 782.0, + 235.0, + 782.0, + 275.0, + 297.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1490.0, + 790.0, + 1490.0, + 790.0, + 1525.0, + 297.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1331.0, + 1409.0, + 1331.0, + 1409.0, + 1371.0, + 292.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1364.0, + 1407.0, + 1364.0, + 1407.0, + 1401.0, + 292.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1397.0, + 1408.0, + 1397.0, + 1408.0, + 1430.0, + 295.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1426.0, + 901.0, + 1426.0, + 901.0, + 1459.0, + 295.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1125.0, + 1404.0, + 1125.0, + 1404.0, + 1158.0, + 295.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1152.0, + 1405.0, + 1152.0, + 1405.0, + 1191.0, + 292.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1186.0, + 1406.0, + 1186.0, + 1406.0, + 1222.0, + 294.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 498.0, + 1215.0, + 498.0, + 1247.0, + 294.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1542.0, + 1410.0, + 1542.0, + 1410.0, + 1580.0, + 293.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1577.0, + 1406.0, + 1577.0, + 1406.0, + 1610.0, + 296.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1607.0, + 1409.0, + 1607.0, + 1409.0, + 1640.0, + 296.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1638.0, + 902.0, + 1638.0, + 902.0, + 1670.0, + 296.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 944.0, + 1403.0, + 944.0, + 1403.0, + 978.0, + 295.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1006.0, + 1280.0, + 1006.0, + 1280.0, + 1040.0, + 295.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 472.0, + 292.0, + 472.0, + 292.0, + 495.0, + 274.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 462.0, + 471.0, + 462.0, + 471.0, + 502.0, + 335.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 462.0, + 1097.0, + 462.0, + 1097.0, + 502.0, + 502.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 509.0, + 291.0, + 509.0, + 291.0, + 529.0, + 275.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 499.0, + 515.0, + 499.0, + 515.0, + 535.0, + 495.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 499.0, + 776.0, + 499.0, + 776.0, + 535.0, + 560.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 499.0, + 802.0, + 499.0, + 802.0, + 535.0, + 798.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 271.0, + 532.0, + 349.0, + 532.0, + 349.0, + 566.0, + 271.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 272.0, + 564.0, + 504.0, + 564.0, + 504.0, + 600.0, + 272.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 564.0, + 867.0, + 564.0, + 867.0, + 600.0, + 549.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 564.0, + 892.0, + 564.0, + 892.0, + 600.0, + 889.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 598.0, + 347.0, + 598.0, + 347.0, + 626.0, + 275.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 270.0, + 625.0, + 294.0, + 625.0, + 294.0, + 661.0, + 270.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 625.0, + 342.0, + 625.0, + 342.0, + 661.0, + 333.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 625.0, + 956.0, + 625.0, + 956.0, + 661.0, + 420.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 271.0, + 659.0, + 403.0, + 659.0, + 403.0, + 698.0, + 271.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 659.0, + 611.0, + 659.0, + 611.0, + 698.0, + 608.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 264.0, + 690.0, + 385.0, + 690.0, + 385.0, + 734.0, + 264.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 690.0, + 484.0, + 690.0, + 484.0, + 734.0, + 421.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 690.0, + 530.0, + 690.0, + 530.0, + 734.0, + 525.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 724.0, + 734.0, + 724.0, + 734.0, + 760.0, + 265.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 724.0, + 982.0, + 724.0, + 982.0, + 760.0, + 768.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 269.0, + 662.0, + 269.0, + 662.0, + 317.0, + 294.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 269.0, + 963.0, + 269.0, + 963.0, + 317.0, + 685.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 307.0, + 599.0, + 307.0, + 599.0, + 341.0, + 297.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 434.0, + 598.0, + 434.0, + 598.0, + 467.0, + 274.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 235.0, + 782.0, + 235.0, + 782.0, + 275.0, + 297.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 330.0, + 399.0, + 330.0, + 399.0, + 378.0, + 276.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 330.0, + 759.0, + 330.0, + 759.0, + 378.0, + 547.0, + 378.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 295, + 219, + 1405, + 219, + 1405, + 667, + 295, + 667 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 306, + 1207, + 1407, + 1207, + 1407, + 1797, + 306, + 1797 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 296, + 1822, + 1405, + 1822, + 1405, + 1919, + 296, + 1919 + ], + "score": 0.952 + }, + { + "category_id": 4, + "poly": [ + 294, + 686, + 1400, + 686, + 1400, + 751, + 294, + 751 + ], + "score": 0.944 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 813, + 76, + 813, + 105, + 299, + 105 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 299, + 1005, + 705, + 1005, + 705, + 1035, + 299, + 1035 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.854 + }, + { + "category_id": 13, + "poly": [ + 458, + 1889, + 479, + 1889, + 479, + 1912, + 458, + 1912 + ], + "score": 0.74, + "latex": "_ z" + }, + { + "category_id": 13, + "poly": [ + 1042, + 1829, + 1063, + 1829, + 1063, + 1851, + 1042, + 1851 + ], + "score": 0.7, + "latex": "_ z" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 301.0, + 1285.0, + 301.0, + 1285.0, + 321.0, + 1261.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 328.0, + 521.0, + 328.0, + 521.0, + 359.0, + 488.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 326.0, + 847.0, + 326.0, + 847.0, + 354.0, + 777.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 326.0, + 1053.0, + 326.0, + 1053.0, + 354.0, + 982.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 324.0, + 1307.0, + 324.0, + 1307.0, + 357.0, + 1237.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 367.0, + 824.0, + 367.0, + 824.0, + 379.0, + 812.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 363.0, + 1030.0, + 363.0, + 1030.0, + 379.0, + 1013.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 366.0, + 1284.0, + 366.0, + 1284.0, + 383.0, + 1268.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 417.0, + 618.0, + 417.0, + 618.0, + 457.0, + 394.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 390.0, + 853.0, + 390.0, + 853.0, + 432.0, + 777.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 409.0, + 914.0, + 409.0, + 914.0, + 428.0, + 860.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 389.0, + 1055.0, + 389.0, + 1055.0, + 432.0, + 981.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 406.0, + 1120.0, + 406.0, + 1120.0, + 426.0, + 1066.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 414.0, + 1229.0, + 414.0, + 1229.0, + 433.0, + 1173.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 394.0, + 1309.0, + 394.0, + 1309.0, + 436.0, + 1234.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 466.0, + 850.0, + 466.0, + 850.0, + 494.0, + 772.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 463.0, + 1059.0, + 463.0, + 1059.0, + 496.0, + 980.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 467.0, + 1313.0, + 467.0, + 1313.0, + 496.0, + 1234.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 486.0, + 815.0, + 486.0, + 815.0, + 508.0, + 773.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 482.0, + 915.0, + 482.0, + 915.0, + 506.0, + 860.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 487.0, + 1024.0, + 487.0, + 1024.0, + 508.0, + 980.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 486.0, + 1120.0, + 486.0, + 1120.0, + 505.0, + 1066.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 492.0, + 1229.0, + 492.0, + 1229.0, + 511.0, + 1171.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 487.0, + 1277.0, + 487.0, + 1277.0, + 509.0, + 1236.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 503.0, + 387.0, + 503.0, + 387.0, + 535.0, + 357.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 549.0, + 677.0, + 549.0, + 677.0, + 584.0, + 351.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 540.0, + 842.0, + 540.0, + 842.0, + 577.0, + 777.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 552.0, + 903.0, + 552.0, + 903.0, + 569.0, + 887.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 545.0, + 1047.0, + 545.0, + 1047.0, + 574.0, + 982.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 552.0, + 1110.0, + 552.0, + 1110.0, + 568.0, + 1093.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 542.0, + 1302.0, + 542.0, + 1302.0, + 576.0, + 1237.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 552.0, + 1358.0, + 552.0, + 1358.0, + 567.0, + 1343.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 618.0, + 820.0, + 618.0, + 820.0, + 637.0, + 800.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 622.0, + 1020.0, + 622.0, + 1020.0, + 637.0, + 1005.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 625.0, + 1279.0, + 625.0, + 1279.0, + 639.0, + 1262.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1212.0, + 364.0, + 1212.0, + 364.0, + 1238.0, + 324.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1215.0, + 1397.0, + 1215.0, + 1397.0, + 1242.0, + 1349.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1239.0, + 362.0, + 1239.0, + 362.0, + 1264.0, + 326.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1231.0, + 1397.0, + 1231.0, + 1397.0, + 1262.0, + 1348.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1250.0, + 1396.0, + 1250.0, + 1396.0, + 1277.0, + 1351.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1266.0, + 362.0, + 1266.0, + 362.0, + 1289.0, + 326.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1288.0, + 333.0, + 1288.0, + 333.0, + 1331.0, + 310.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1291.0, + 364.0, + 1291.0, + 364.0, + 1318.0, + 322.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1316.0, + 364.0, + 1316.0, + 364.0, + 1343.0, + 325.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1345.0, + 362.0, + 1345.0, + 362.0, + 1369.0, + 326.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1370.0, + 362.0, + 1370.0, + 362.0, + 1394.0, + 326.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1404.0, + 431.0, + 1404.0, + 431.0, + 1427.0, + 394.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1402.0, + 501.0, + 1402.0, + 501.0, + 1428.0, + 460.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1402.0, + 570.0, + 1402.0, + 570.0, + 1428.0, + 530.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1401.0, + 639.0, + 1401.0, + 639.0, + 1428.0, + 599.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1402.0, + 709.0, + 1402.0, + 709.0, + 1428.0, + 668.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1404.0, + 775.0, + 1404.0, + 775.0, + 1427.0, + 738.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1404.0, + 843.0, + 1404.0, + 843.0, + 1428.0, + 806.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1402.0, + 914.0, + 1402.0, + 914.0, + 1430.0, + 874.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1402.0, + 984.0, + 1402.0, + 984.0, + 1428.0, + 943.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1402.0, + 1052.0, + 1402.0, + 1052.0, + 1430.0, + 1012.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1401.0, + 1122.0, + 1401.0, + 1122.0, + 1428.0, + 1080.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1402.0, + 1191.0, + 1402.0, + 1191.0, + 1428.0, + 1149.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1402.0, + 1259.0, + 1402.0, + 1259.0, + 1428.0, + 1219.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1404.0, + 1326.0, + 1404.0, + 1326.0, + 1427.0, + 1289.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1421.0, + 912.0, + 1421.0, + 912.0, + 1447.0, + 854.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1453.0, + 1175.0, + 1453.0, + 1175.0, + 1487.0, + 537.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1529.0, + 351.0, + 1529.0, + 351.0, + 1550.0, + 327.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1526.0, + 1397.0, + 1526.0, + 1397.0, + 1553.0, + 1351.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1543.0, + 1397.0, + 1543.0, + 1397.0, + 1570.0, + 1352.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1558.0, + 351.0, + 1558.0, + 351.0, + 1579.0, + 327.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1560.0, + 1397.0, + 1560.0, + 1397.0, + 1588.0, + 1351.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1587.0, + 351.0, + 1587.0, + 351.0, + 1609.0, + 327.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1602.0, + 331.0, + 1602.0, + 331.0, + 1643.0, + 309.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1616.0, + 351.0, + 1616.0, + 351.0, + 1638.0, + 327.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1647.0, + 351.0, + 1647.0, + 351.0, + 1668.0, + 327.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1676.0, + 351.0, + 1676.0, + 351.0, + 1697.0, + 327.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1705.0, + 350.0, + 1705.0, + 350.0, + 1725.0, + 327.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1715.0, + 422.0, + 1715.0, + 422.0, + 1742.0, + 381.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1715.0, + 490.0, + 1715.0, + 490.0, + 1742.0, + 449.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1714.0, + 559.0, + 1714.0, + 559.0, + 1742.0, + 519.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1714.0, + 628.0, + 1714.0, + 628.0, + 1742.0, + 586.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1715.0, + 695.0, + 1715.0, + 695.0, + 1742.0, + 654.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1715.0, + 764.0, + 1715.0, + 764.0, + 1742.0, + 723.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1715.0, + 830.0, + 1715.0, + 830.0, + 1742.0, + 790.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1715.0, + 900.0, + 1715.0, + 900.0, + 1743.0, + 859.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1715.0, + 968.0, + 1715.0, + 968.0, + 1742.0, + 927.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1714.0, + 1035.0, + 1714.0, + 1035.0, + 1742.0, + 994.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1714.0, + 1105.0, + 1714.0, + 1105.0, + 1742.0, + 1063.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1715.0, + 1174.0, + 1715.0, + 1174.0, + 1742.0, + 1133.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1714.0, + 1242.0, + 1714.0, + 1242.0, + 1742.0, + 1201.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1714.0, + 1309.0, + 1714.0, + 1309.0, + 1742.0, + 1268.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1731.0, + 907.0, + 1731.0, + 907.0, + 1761.0, + 850.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1766.0, + 1175.0, + 1766.0, + 1175.0, + 1798.0, + 537.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1041.0, + 1822.0, + 1041.0, + 1857.0, + 295.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1822.0, + 1405.0, + 1822.0, + 1405.0, + 1857.0, + 1064.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1853.0, + 1403.0, + 1853.0, + 1403.0, + 1888.0, + 295.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 457.0, + 1883.0, + 457.0, + 1921.0, + 295.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1883.0, + 1402.0, + 1883.0, + 1402.0, + 1921.0, + 480.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 686.0, + 1404.0, + 686.0, + 1404.0, + 722.0, + 296.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 717.0, + 1213.0, + 717.0, + 1213.0, + 752.0, + 294.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 71.0, + 815.0, + 71.0, + 815.0, + 108.0, + 294.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1003.0, + 709.0, + 1003.0, + 709.0, + 1038.0, + 296.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 307, + 1242, + 307, + 1242, + 1053, + 298, + 1053 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 317, + 1062, + 1279, + 1062, + 1279, + 2036, + 317, + 2036 + ], + "score": 0.959 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 813, + 76, + 813, + 104, + 300, + 104 + ], + "score": 0.868 + }, + { + "category_id": 0, + "poly": [ + 304, + 231, + 907, + 231, + 907, + 261, + 304, + 261 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.837 + }, + { + "category_id": 13, + "poly": [ + 1015, + 617, + 1067, + 617, + 1067, + 637, + 1015, + 637 + ], + "score": 0.7, + "latex": "_ { 3 } = 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 810, + 748, + 858, + 748, + 858, + 769, + 810, + 769 + ], + "score": 0.69, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 1050, + 1236, + 1102, + 1236, + 1102, + 1257, + 1050, + 1257 + ], + "score": 0.67, + "latex": "\\mathtt { s } = 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 810, + 617, + 858, + 617, + 858, + 637, + 810, + 637 + ], + "score": 0.66, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 1042, + 1303, + 1091, + 1303, + 1091, + 1323, + 1042, + 1323 + ], + "score": 0.63, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 832, + 1436, + 880, + 1436, + 880, + 1456, + 832, + 1456 + ], + "score": 0.62, + "latex": "_ { : = 6 4 0 }" + }, + { + "category_id": 13, + "poly": [ + 1204, + 1571, + 1219, + 1571, + 1219, + 1586, + 1204, + 1586 + ], + "score": 0.61, + "latex": "{ } , = { }" + }, + { + "category_id": 13, + "poly": [ + 1205, + 1637, + 1218, + 1637, + 1218, + 1652, + 1205, + 1652 + ], + "score": 0.6, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1018, + 749, + 1043, + 749, + 1043, + 768, + 1018, + 768 + ], + "score": 0.59, + "latex": ": = 1" + }, + { + "category_id": 13, + "poly": [ + 1030, + 1747, + 1054, + 1747, + 1054, + 1764, + 1030, + 1764 + ], + "score": 0.59, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 822, + 1106, + 837, + 1106, + 837, + 1121, + 822, + 1121 + ], + "score": 0.58, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 798, + 419, + 814, + 419, + 814, + 434, + 798, + 434 + ], + "score": 0.57, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 810, + 682, + 858, + 682, + 858, + 703, + 810, + 703 + ], + "score": 0.56, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 1170, + 1239, + 1184, + 1239, + 1184, + 1255, + 1170, + 1255 + ], + "score": 0.55, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 775, + 441, + 799, + 441, + 799, + 459, + 775, + 459 + ], + "score": 0.55, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1437, + 1067, + 1437, + 1067, + 1455, + 1043, + 1455 + ], + "score": 0.54, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 833, + 1947, + 848, + 1947, + 848, + 1963, + 833, + 1963 + ], + "score": 0.52, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 648, + 506, + 672, + 506, + 672, + 525, + 648, + 525 + ], + "score": 0.52, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 831, + 1303, + 880, + 1303, + 880, + 1323, + 831, + 1323 + ], + "score": 0.5, + "latex": "\\mathord { : } = 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 1017, + 682, + 1067, + 682, + 1067, + 703, + 1017, + 703 + ], + "score": 0.48, + "latex": "\\mathord { : } = 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 973, + 1105, + 988, + 1105, + 988, + 1122, + 973, + 1122 + ], + "score": 0.48, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1368, + 1090, + 1368, + 1090, + 1389, + 1043, + 1389 + ], + "score": 0.48, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 832, + 1369, + 879, + 1369, + 879, + 1389, + 832, + 1389 + ], + "score": 0.48, + "latex": "_ { : = 6 4 0 }" + }, + { + "category_id": 13, + "poly": [ + 1136, + 1438, + 1148, + 1438, + 1148, + 1453, + 1136, + 1453 + ], + "score": 0.44, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1158, + 1304, + 1179, + 1304, + 1179, + 1321, + 1158, + 1321 + ], + "score": 0.42, + "latex": "{ } = \\mathbb { E }" + }, + { + "category_id": 13, + "poly": [ + 984, + 1947, + 999, + 1947, + 999, + 1963, + 984, + 1963 + ], + "score": 0.41, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 811, + 1747, + 832, + 1747, + 832, + 1764, + 811, + 1764 + ], + "score": 0.4, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 1182, + 951, + 1197, + 951, + 1197, + 967, + 1182, + 967 + ], + "score": 0.38, + "latex": "= \"" + }, + { + "category_id": 13, + "poly": [ + 1052, + 948, + 1112, + 948, + 1112, + 969, + 1052, + 969 + ], + "score": 0.38, + "latex": "_ { \\mathrm { : = 1 0 2 4 } }" + }, + { + "category_id": 13, + "poly": [ + 1123, + 1747, + 1139, + 1747, + 1139, + 1764, + 1123, + 1764 + ], + "score": 0.35, + "latex": "= \"" + }, + { + "category_id": 13, + "poly": [ + 1030, + 549, + 1079, + 549, + 1079, + 570, + 1030, + 570 + ], + "score": 0.34, + "latex": "= 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 1136, + 619, + 1150, + 619, + 1150, + 635, + 1136, + 635 + ], + "score": 0.34, + "latex": "\\mathbf { \\tau } = \\dot { }" + }, + { + "category_id": 13, + "poly": [ + 828, + 949, + 892, + 949, + 892, + 969, + 828, + 969 + ], + "score": 0.33, + "latex": "\\mathord { \\mathrm { 3 } } = 1 0 2 4" + }, + { + "category_id": 13, + "poly": [ + 829, + 1236, + 892, + 1236, + 892, + 1257, + 829, + 1257 + ], + "score": 0.33, + "latex": "\\mathord { \\mathrm { 3 } } = 1 0 2 4" + }, + { + "category_id": 13, + "poly": [ + 857, + 1636, + 873, + 1636, + 873, + 1653, + 857, + 1653 + ], + "score": 0.31, + "latex": "= 1" + }, + { + "category_id": 13, + "poly": [ + 671, + 1192, + 695, + 1192, + 695, + 1211, + 671, + 1211 + ], + "score": 0.29, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 857, + 1570, + 872, + 1570, + 872, + 1587, + 857, + 1587 + ], + "score": 0.29, + "latex": "\\mathbf { \\Psi } = \\mathbf { \\dot { \\Psi } }" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1991, + 1056, + 1991, + 1056, + 2007, + 1043, + 2007 + ], + "score": 0.29, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1570, + 1093, + 1570, + 1093, + 1587, + 1077, + 1587 + ], + "score": 0.29, + "latex": "\\mathbf { \\Psi } = \\mathbf { \\dot { \\Psi } }" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1636, + 1092, + 1636, + 1092, + 1653, + 1077, + 1653 + ], + "score": 0.28, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 799, + 1127, + 823, + 1127, + 823, + 1144, + 799, + 1144 + ], + "score": 0.28, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 949, + 419, + 964, + 419, + 964, + 435, + 949, + 435 + ], + "score": 0.26, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 72.0, + 816.0, + 72.0, + 816.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 228.0, + 913.0, + 228.0, + 913.0, + 264.0, + 298.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 304.0, + 408.0, + 304.0, + 408.0, + 329.0, + 295.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 328.0, + 765.0, + 328.0, + 765.0, + 352.0, + 322.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 350.0, + 696.0, + 350.0, + 696.0, + 374.0, + 323.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 372.0, + 560.0, + 372.0, + 560.0, + 396.0, + 322.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 392.0, + 500.0, + 392.0, + 500.0, + 419.0, + 346.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 416.0, + 797.0, + 416.0, + 797.0, + 440.0, + 369.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 416.0, + 948.0, + 416.0, + 948.0, + 440.0, + 815.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 416.0, + 1020.0, + 416.0, + 1020.0, + 440.0, + 965.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 439.0, + 774.0, + 439.0, + 774.0, + 463.0, + 370.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 439.0, + 810.0, + 439.0, + 810.0, + 463.0, + 800.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 461.0, + 743.0, + 461.0, + 743.0, + 484.0, + 369.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 483.0, + 709.0, + 483.0, + 709.0, + 507.0, + 370.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 505.0, + 647.0, + 505.0, + 647.0, + 529.0, + 392.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 505.0, + 682.0, + 505.0, + 682.0, + 529.0, + 673.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 528.0, + 630.0, + 528.0, + 630.0, + 552.0, + 394.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 548.0, + 1029.0, + 548.0, + 1029.0, + 572.0, + 415.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 548.0, + 1228.0, + 548.0, + 1228.0, + 572.0, + 1080.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 575.0, + 404.0, + 575.0, + 404.0, + 591.0, + 390.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 593.0, + 606.0, + 593.0, + 606.0, + 617.0, + 392.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 616.0, + 809.0, + 616.0, + 809.0, + 640.0, + 417.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 616.0, + 1014.0, + 616.0, + 1014.0, + 640.0, + 859.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 616.0, + 1135.0, + 616.0, + 1135.0, + 640.0, + 1068.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 616.0, + 1217.0, + 616.0, + 1217.0, + 640.0, + 1151.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 640.0, + 405.0, + 640.0, + 405.0, + 660.0, + 388.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 660.0, + 630.0, + 660.0, + 630.0, + 684.0, + 392.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 681.0, + 809.0, + 681.0, + 809.0, + 705.0, + 415.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 681.0, + 1016.0, + 681.0, + 1016.0, + 705.0, + 859.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 681.0, + 1217.0, + 681.0, + 1217.0, + 705.0, + 1068.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 706.0, + 405.0, + 706.0, + 405.0, + 726.0, + 388.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 726.0, + 583.0, + 726.0, + 583.0, + 750.0, + 392.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 747.0, + 809.0, + 747.0, + 809.0, + 771.0, + 417.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 747.0, + 1017.0, + 747.0, + 1017.0, + 771.0, + 859.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 747.0, + 1195.0, + 747.0, + 1195.0, + 771.0, + 1044.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 774.0, + 405.0, + 774.0, + 405.0, + 795.0, + 387.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 796.0, + 383.0, + 796.0, + 383.0, + 815.0, + 367.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 817.0, + 673.0, + 817.0, + 673.0, + 838.0, + 371.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 835.0, + 640.0, + 835.0, + 640.0, + 860.0, + 393.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 857.0, + 606.0, + 857.0, + 606.0, + 882.0, + 415.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 880.0, + 1245.0, + 880.0, + 1245.0, + 905.0, + 439.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 906.0, + 430.0, + 906.0, + 430.0, + 925.0, + 412.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 926.0, + 606.0, + 926.0, + 606.0, + 949.0, + 414.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 948.0, + 827.0, + 948.0, + 827.0, + 972.0, + 439.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 948.0, + 1051.0, + 948.0, + 1051.0, + 972.0, + 893.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 948.0, + 1181.0, + 948.0, + 1181.0, + 972.0, + 1113.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 948.0, + 1244.0, + 948.0, + 1244.0, + 972.0, + 1198.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 974.0, + 427.0, + 974.0, + 427.0, + 990.0, + 414.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1058.0, + 557.0, + 1058.0, + 557.0, + 1082.0, + 347.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1080.0, + 596.0, + 1080.0, + 596.0, + 1106.0, + 369.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1103.0, + 821.0, + 1103.0, + 821.0, + 1127.0, + 392.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1103.0, + 972.0, + 1103.0, + 972.0, + 1127.0, + 838.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1103.0, + 1045.0, + 1103.0, + 1045.0, + 1127.0, + 989.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1125.0, + 798.0, + 1125.0, + 798.0, + 1150.0, + 393.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1125.0, + 835.0, + 1125.0, + 835.0, + 1150.0, + 824.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1146.0, + 765.0, + 1146.0, + 765.0, + 1171.0, + 392.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1169.0, + 734.0, + 1169.0, + 734.0, + 1193.0, + 393.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1192.0, + 670.0, + 1192.0, + 670.0, + 1213.0, + 417.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1192.0, + 705.0, + 1192.0, + 705.0, + 1213.0, + 696.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1213.0, + 653.0, + 1213.0, + 653.0, + 1238.0, + 416.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1236.0, + 828.0, + 1236.0, + 828.0, + 1260.0, + 440.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1236.0, + 1049.0, + 1236.0, + 1049.0, + 1260.0, + 893.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1236.0, + 1169.0, + 1236.0, + 1169.0, + 1260.0, + 1103.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 1236.0, + 1253.0, + 1236.0, + 1253.0, + 1260.0, + 1185.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1282.0, + 629.0, + 1282.0, + 629.0, + 1303.0, + 416.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1302.0, + 830.0, + 1302.0, + 830.0, + 1326.0, + 440.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1302.0, + 1041.0, + 1302.0, + 1041.0, + 1326.0, + 881.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1302.0, + 1157.0, + 1302.0, + 1157.0, + 1326.0, + 1092.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1302.0, + 1240.0, + 1302.0, + 1240.0, + 1326.0, + 1180.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1347.0, + 651.0, + 1347.0, + 651.0, + 1368.0, + 415.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1368.0, + 831.0, + 1368.0, + 831.0, + 1392.0, + 440.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 1368.0, + 1042.0, + 1368.0, + 1042.0, + 1392.0, + 880.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 1368.0, + 1240.0, + 1368.0, + 1240.0, + 1392.0, + 1091.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1395.0, + 429.0, + 1395.0, + 429.0, + 1410.0, + 414.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1415.0, + 604.0, + 1415.0, + 604.0, + 1436.0, + 416.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1436.0, + 831.0, + 1436.0, + 831.0, + 1457.0, + 441.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1436.0, + 1042.0, + 1436.0, + 1042.0, + 1457.0, + 881.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1436.0, + 1135.0, + 1436.0, + 1135.0, + 1457.0, + 1068.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1436.0, + 1216.0, + 1436.0, + 1216.0, + 1457.0, + 1149.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1461.0, + 427.0, + 1461.0, + 427.0, + 1476.0, + 414.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1486.0, + 402.0, + 1486.0, + 402.0, + 1498.0, + 393.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1501.0, + 698.0, + 1501.0, + 698.0, + 1526.0, + 392.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1524.0, + 663.0, + 1524.0, + 663.0, + 1546.0, + 417.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1545.0, + 630.0, + 1545.0, + 630.0, + 1568.0, + 438.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1567.0, + 856.0, + 1567.0, + 856.0, + 1592.0, + 462.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1567.0, + 1076.0, + 1567.0, + 1076.0, + 1592.0, + 873.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1567.0, + 1203.0, + 1567.0, + 1203.0, + 1592.0, + 1094.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1567.0, + 1276.0, + 1567.0, + 1276.0, + 1592.0, + 1220.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1593.0, + 453.0, + 1593.0, + 453.0, + 1612.0, + 436.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1612.0, + 629.0, + 1612.0, + 629.0, + 1635.0, + 438.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1632.0, + 856.0, + 1632.0, + 856.0, + 1659.0, + 461.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1632.0, + 1076.0, + 1632.0, + 1076.0, + 1659.0, + 874.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1632.0, + 1204.0, + 1632.0, + 1204.0, + 1659.0, + 1093.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1632.0, + 1277.0, + 1632.0, + 1277.0, + 1659.0, + 1219.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1706.0, + 403.0, + 1706.0, + 403.0, + 1720.0, + 391.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1724.0, + 687.0, + 1724.0, + 687.0, + 1747.0, + 392.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1745.0, + 810.0, + 1745.0, + 810.0, + 1769.0, + 416.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1745.0, + 1029.0, + 1745.0, + 1029.0, + 1769.0, + 833.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1745.0, + 1122.0, + 1745.0, + 1122.0, + 1769.0, + 1055.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1745.0, + 1195.0, + 1745.0, + 1195.0, + 1769.0, + 1140.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1772.0, + 403.0, + 1772.0, + 403.0, + 1787.0, + 390.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1838.0, + 333.0, + 1838.0, + 333.0, + 1853.0, + 321.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1853.0, + 548.0, + 1853.0, + 548.0, + 1878.0, + 322.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1878.0, + 664.0, + 1878.0, + 664.0, + 1901.0, + 346.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1901.0, + 558.0, + 1901.0, + 558.0, + 1923.0, + 370.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1923.0, + 558.0, + 1923.0, + 558.0, + 1944.0, + 393.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1941.0, + 832.0, + 1941.0, + 832.0, + 1970.0, + 414.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1941.0, + 983.0, + 1941.0, + 983.0, + 1970.0, + 849.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1941.0, + 1217.0, + 1941.0, + 1217.0, + 1970.0, + 1000.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1969.0, + 406.0, + 1969.0, + 406.0, + 1989.0, + 390.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1985.0, + 1042.0, + 1985.0, + 1042.0, + 2014.0, + 390.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1985.0, + 1284.0, + 1985.0, + 1284.0, + 2014.0, + 1057.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 2014.0, + 381.0, + 2014.0, + 381.0, + 2030.0, + 367.0, + 2030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.25, + 1263.0, + 422.25, + 1263.0, + 422.25, + 1276.0, + 417.25, + 1276.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 316, + 586, + 1380, + 586, + 1380, + 2040, + 316, + 2040 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 294, + 238, + 1429, + 238, + 1429, + 590, + 294, + 590 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 105, + 300, + 105 + ], + "score": 0.867 + }, + { + "category_id": 2, + "poly": [ + 835, + 2089, + 864, + 2089, + 864, + 2113, + 835, + 2113 + ], + "score": 0.837 + }, + { + "category_id": 13, + "poly": [ + 811, + 1790, + 846, + 1790, + 846, + 1810, + 811, + 1810 + ], + "score": 0.76, + "latex": "= 3 2" + }, + { + "category_id": 13, + "poly": [ + 784, + 1569, + 834, + 1569, + 834, + 1589, + 784, + 1589 + ], + "score": 0.75, + "latex": "\\mathord { 5 } = 6 4 0" + }, + { + "category_id": 13, + "poly": [ + 867, + 816, + 916, + 816, + 916, + 837, + 867, + 837 + ], + "score": 0.73, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 867, + 949, + 916, + 949, + 916, + 970, + 867, + 970 + ], + "score": 0.7, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 740, + 528, + 756, + 528, + 756, + 545, + 740, + 545 + ], + "score": 0.67, + "latex": "= \"" + }, + { + "category_id": 13, + "poly": [ + 810, + 1880, + 834, + 1880, + 834, + 1897, + 810, + 1897 + ], + "score": 0.67, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 995, + 1568, + 1044, + 1568, + 1044, + 1589, + 995, + 1589 + ], + "score": 0.66, + "latex": "^ { - 1 2 8 }" + }, + { + "category_id": 13, + "poly": [ + 997, + 1635, + 1020, + 1635, + 1020, + 1654, + 997, + 1654 + ], + "score": 0.65, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 811, + 1281, + 846, + 1281, + 846, + 1300, + 811, + 1300 + ], + "score": 0.64, + "latex": "= 8 0" + }, + { + "category_id": 13, + "poly": [ + 1008, + 819, + 1021, + 819, + 1021, + 834, + 1008, + 834 + ], + "score": 0.64, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 798, + 1726, + 814, + 1726, + 814, + 1742, + 798, + 1742 + ], + "score": 0.61, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 867, + 1082, + 916, + 1082, + 916, + 1103, + 867, + 1103 + ], + "score": 0.58, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1084, + 1021, + 1084, + 1021, + 1099, + 1007, + 1099 + ], + "score": 0.58, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 1124, + 1725, + 1139, + 1725, + 1139, + 1742, + 1124, + 1742 + ], + "score": 0.58, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1043, + 462, + 1056, + 462, + 1056, + 478, + 1043, + 478 + ], + "score": 0.57, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1007, + 952, + 1021, + 952, + 1021, + 966, + 1007, + 966 + ], + "score": 0.56, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 1043, + 330, + 1057, + 330, + 1057, + 346, + 1043, + 346 + ], + "score": 0.56, + "latex": "= ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 985, + 906, + 1000, + 906, + 1000, + 922, + 985, + 922 + ], + "score": 0.55, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 972, + 528, + 987, + 528, + 987, + 545, + 972, + 545 + ], + "score": 0.54, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 833, + 906, + 848, + 906, + 848, + 922, + 833, + 922 + ], + "score": 0.53, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1790, + 1055, + 1790, + 1055, + 1810, + 1007, + 1810 + ], + "score": 0.53, + "latex": "= 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 983, + 1923, + 1020, + 1923, + 1020, + 1943, + 983, + 1943 + ], + "score": 0.53, + "latex": "{ \\mathfrak { s } } = 8 0" + }, + { + "category_id": 13, + "poly": [ + 833, + 284, + 848, + 284, + 848, + 301, + 833, + 301 + ], + "score": 0.53, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 984, + 773, + 1000, + 773, + 1000, + 790, + 984, + 790 + ], + "score": 0.51, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 740, + 1150, + 754, + 1150, + 754, + 1165, + 740, + 1165 + ], + "score": 0.48, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 832, + 774, + 849, + 774, + 849, + 790, + 832, + 790 + ], + "score": 0.47, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 985, + 284, + 1000, + 284, + 1000, + 301, + 985, + 301 + ], + "score": 0.47, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 833, + 418, + 848, + 418, + 848, + 434, + 833, + 434 + ], + "score": 0.47, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 678, + 816, + 754, + 816, + 754, + 837, + 678, + 837 + ], + "score": 0.47, + "latex": "\\mathtt { s } = 1 \\mathtt { e } - 0 5" + }, + { + "category_id": 13, + "poly": [ + 985, + 418, + 999, + 418, + 999, + 434, + 985, + 434 + ], + "score": 0.46, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 784, + 1634, + 834, + 1634, + 834, + 1655, + 784, + 1655 + ], + "score": 0.45, + "latex": "\\mathord { \\left. \\mathrm { = } \\right.} 1 2 8 " + }, + { + "category_id": 13, + "poly": [ + 1090, + 1925, + 1104, + 1925, + 1104, + 1940, + 1090, + 1940 + ], + "score": 0.44, + "latex": "= \"" + }, + { + "category_id": 13, + "poly": [ + 1123, + 1792, + 1137, + 1792, + 1137, + 1808, + 1123, + 1808 + ], + "score": 0.43, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 682, + 1084, + 696, + 1084, + 696, + 1099, + 682, + 1099 + ], + "score": 0.42, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 671, + 949, + 754, + 949, + 754, + 970, + 671, + 970 + ], + "score": 0.42, + "latex": "\\mathsf { s } { = } 1 \\mathsf { e } { - } 0 5" + }, + { + "category_id": 13, + "poly": [ + 1123, + 1282, + 1138, + 1282, + 1138, + 1299, + 1123, + 1299 + ], + "score": 0.42, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1306, + 817, + 1324, + 817, + 1324, + 834, + 1306, + 834 + ], + "score": 0.41, + "latex": "s =" + }, + { + "category_id": 13, + "poly": [ + 764, + 1924, + 787, + 1924, + 787, + 1941, + 764, + 1941 + ], + "score": 0.4, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 984, + 1039, + 1000, + 1039, + 1000, + 1056, + 984, + 1056 + ], + "score": 0.39, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 960, + 1726, + 976, + 1726, + 976, + 1742, + 960, + 1742 + ], + "score": 0.38, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 901, + 459, + 951, + 459, + 951, + 480, + 901, + 480 + ], + "score": 0.38, + "latex": "\\ i { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 833, + 1039, + 848, + 1039, + 848, + 1056, + 833, + 1056 + ], + "score": 0.37, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1262, + 1726, + 1276, + 1726, + 1276, + 1741, + 1262, + 1741 + ], + "score": 0.37, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 1304, + 951, + 1324, + 951, + 1324, + 967, + 1304, + 967 + ], + "score": 0.37, + "latex": "{ \\mathfrak { s } } =" + }, + { + "category_id": 13, + "poly": [ + 901, + 327, + 951, + 327, + 951, + 349, + 901, + 349 + ], + "score": 0.36, + "latex": "\\ i { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 1112, + 1570, + 1127, + 1570, + 1127, + 1587, + 1112, + 1587 + ], + "score": 0.35, + "latex": "\\fallingdotseq" + }, + { + "category_id": 13, + "poly": [ + 1307, + 1084, + 1323, + 1084, + 1323, + 1099, + 1307, + 1099 + ], + "score": 0.34, + "latex": "; =" + }, + { + "category_id": 13, + "poly": [ + 960, + 1723, + 1017, + 1723, + 1017, + 1745, + 960, + 1745 + ], + "score": 0.32, + "latex": "\\mathbf { \\Omega } : = \\left( \\mathbb { 1 } , \\right)" + }, + { + "category_id": 13, + "poly": [ + 972, + 1150, + 986, + 1150, + 986, + 1165, + 972, + 1165 + ], + "score": 0.32, + "latex": "{ } = { }" + }, + { + "category_id": 13, + "poly": [ + 1147, + 906, + 1162, + 906, + 1162, + 923, + 1147, + 923 + ], + "score": 0.31, + "latex": "{ } = { }" + }, + { + "category_id": 13, + "poly": [ + 718, + 461, + 731, + 461, + 731, + 478, + 718, + 478 + ], + "score": 0.28, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 984, + 1990, + 1008, + 1990, + 1008, + 2007, + 984, + 2007 + ], + "score": 0.26, + "latex": "^ { = 1 }" + }, + { + "category_id": 13, + "poly": [ + 714, + 328, + 788, + 328, + 788, + 349, + 714, + 349 + ], + "score": 0.26, + "latex": "\\mathrm { 3 } { = } 1 \\mathrm { e } { - } 0 5" + }, + { + "category_id": 13, + "poly": [ + 1005, + 1280, + 1056, + 1280, + 1056, + 1301, + 1005, + 1301 + ], + "score": 0.26, + "latex": "_ { 5 } = 2 5 6" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 71.0, + 815.0, + 71.0, + 815.0, + 108.0, + 294.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 640.0, + 418.0, + 640.0, + 418.0, + 661.0, + 312.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 659.0, + 698.0, + 659.0, + 698.0, + 685.0, + 322.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 682.0, + 550.0, + 682.0, + 550.0, + 708.0, + 322.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 706.0, + 663.0, + 706.0, + 663.0, + 727.0, + 347.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 725.0, + 562.0, + 725.0, + 562.0, + 751.0, + 370.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 748.0, + 562.0, + 748.0, + 562.0, + 774.0, + 392.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 771.0, + 831.0, + 771.0, + 831.0, + 797.0, + 416.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 771.0, + 983.0, + 771.0, + 983.0, + 797.0, + 850.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 771.0, + 1220.0, + 771.0, + 1220.0, + 797.0, + 1001.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 798.0, + 407.0, + 798.0, + 407.0, + 813.0, + 392.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 811.0, + 677.0, + 811.0, + 677.0, + 844.0, + 388.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 811.0, + 866.0, + 811.0, + 866.0, + 844.0, + 755.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 811.0, + 1007.0, + 811.0, + 1007.0, + 844.0, + 917.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 811.0, + 1305.0, + 811.0, + 1305.0, + 844.0, + 1022.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 811.0, + 1382.0, + 811.0, + 1382.0, + 844.0, + 1325.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 842.0, + 382.0, + 842.0, + 382.0, + 858.0, + 370.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 858.0, + 562.0, + 858.0, + 562.0, + 884.0, + 370.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 881.0, + 564.0, + 881.0, + 564.0, + 907.0, + 393.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 903.0, + 832.0, + 903.0, + 832.0, + 929.0, + 416.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 903.0, + 984.0, + 903.0, + 984.0, + 929.0, + 849.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 903.0, + 1146.0, + 903.0, + 1146.0, + 929.0, + 1001.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 903.0, + 1220.0, + 903.0, + 1220.0, + 929.0, + 1163.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 929.0, + 410.0, + 929.0, + 410.0, + 949.0, + 390.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 949.0, + 670.0, + 949.0, + 670.0, + 975.0, + 392.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 949.0, + 866.0, + 949.0, + 866.0, + 975.0, + 755.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 949.0, + 1006.0, + 949.0, + 1006.0, + 975.0, + 917.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 949.0, + 1303.0, + 949.0, + 1303.0, + 975.0, + 1022.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 949.0, + 1380.0, + 949.0, + 1380.0, + 975.0, + 1325.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 975.0, + 382.0, + 975.0, + 382.0, + 991.0, + 369.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 992.0, + 562.0, + 992.0, + 562.0, + 1018.0, + 370.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1017.0, + 562.0, + 1017.0, + 562.0, + 1038.0, + 393.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1036.0, + 832.0, + 1036.0, + 832.0, + 1062.0, + 416.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1036.0, + 983.0, + 1036.0, + 983.0, + 1062.0, + 849.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1036.0, + 1220.0, + 1036.0, + 1220.0, + 1062.0, + 1001.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1060.0, + 408.0, + 1060.0, + 408.0, + 1082.0, + 390.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1082.0, + 681.0, + 1082.0, + 681.0, + 1107.0, + 392.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1082.0, + 866.0, + 1082.0, + 866.0, + 1107.0, + 697.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1082.0, + 1006.0, + 1082.0, + 1006.0, + 1107.0, + 917.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1082.0, + 1306.0, + 1082.0, + 1306.0, + 1107.0, + 1022.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1082.0, + 1380.0, + 1082.0, + 1380.0, + 1107.0, + 1324.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1107.0, + 382.0, + 1107.0, + 382.0, + 1124.0, + 369.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1130.0, + 360.0, + 1130.0, + 360.0, + 1146.0, + 345.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1146.0, + 739.0, + 1146.0, + 739.0, + 1172.0, + 345.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1146.0, + 971.0, + 1146.0, + 971.0, + 1172.0, + 755.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1146.0, + 1043.0, + 1146.0, + 1043.0, + 1172.0, + 987.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1174.0, + 335.0, + 1174.0, + 335.0, + 1188.0, + 322.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1190.0, + 550.0, + 1190.0, + 550.0, + 1216.0, + 321.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1213.0, + 549.0, + 1213.0, + 549.0, + 1239.0, + 345.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1235.0, + 620.0, + 1235.0, + 620.0, + 1261.0, + 370.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1258.0, + 585.0, + 1258.0, + 585.0, + 1284.0, + 392.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1279.0, + 810.0, + 1279.0, + 810.0, + 1305.0, + 416.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1279.0, + 1004.0, + 1279.0, + 1004.0, + 1305.0, + 847.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1279.0, + 1122.0, + 1279.0, + 1122.0, + 1305.0, + 1057.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1279.0, + 1207.0, + 1279.0, + 1207.0, + 1305.0, + 1139.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1303.0, + 408.0, + 1303.0, + 408.0, + 1324.0, + 390.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1324.0, + 585.0, + 1324.0, + 585.0, + 1350.0, + 390.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1345.0, + 1217.0, + 1345.0, + 1217.0, + 1371.0, + 416.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1373.0, + 407.0, + 1373.0, + 407.0, + 1387.0, + 390.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1396.0, + 382.0, + 1396.0, + 382.0, + 1410.0, + 369.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1434.0, + 765.0, + 1434.0, + 765.0, + 1460.0, + 345.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1457.0, + 689.0, + 1457.0, + 689.0, + 1483.0, + 347.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1477.0, + 679.0, + 1477.0, + 679.0, + 1506.0, + 367.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1501.0, + 1207.0, + 1501.0, + 1207.0, + 1527.0, + 393.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1528.0, + 382.0, + 1528.0, + 382.0, + 1543.0, + 369.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1544.0, + 689.0, + 1544.0, + 689.0, + 1570.0, + 370.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1566.0, + 783.0, + 1566.0, + 783.0, + 1596.0, + 392.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1566.0, + 994.0, + 1566.0, + 994.0, + 1596.0, + 835.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1566.0, + 1111.0, + 1566.0, + 1111.0, + 1596.0, + 1045.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1566.0, + 1197.0, + 1566.0, + 1197.0, + 1596.0, + 1128.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1595.0, + 382.0, + 1595.0, + 382.0, + 1611.0, + 367.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1612.0, + 564.0, + 1612.0, + 564.0, + 1638.0, + 369.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1632.0, + 783.0, + 1632.0, + 783.0, + 1658.0, + 393.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1632.0, + 996.0, + 1632.0, + 996.0, + 1658.0, + 835.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1632.0, + 1170.0, + 1632.0, + 1170.0, + 1658.0, + 1021.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1661.0, + 382.0, + 1661.0, + 382.0, + 1677.0, + 370.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1677.0, + 747.0, + 1677.0, + 747.0, + 1703.0, + 370.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1698.0, + 699.0, + 1698.0, + 699.0, + 1724.0, + 390.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1723.0, + 797.0, + 1723.0, + 797.0, + 1748.0, + 418.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1723.0, + 959.0, + 1723.0, + 959.0, + 1748.0, + 815.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1723.0, + 1123.0, + 1723.0, + 1123.0, + 1748.0, + 1018.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1723.0, + 1261.0, + 1723.0, + 1261.0, + 1748.0, + 1140.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1723.0, + 1347.0, + 1723.0, + 1347.0, + 1748.0, + 1277.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1750.0, + 407.0, + 1750.0, + 407.0, + 1766.0, + 392.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1769.0, + 734.0, + 1769.0, + 734.0, + 1791.0, + 393.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1789.0, + 810.0, + 1789.0, + 810.0, + 1815.0, + 416.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1789.0, + 1006.0, + 1789.0, + 1006.0, + 1815.0, + 847.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1789.0, + 1122.0, + 1789.0, + 1122.0, + 1815.0, + 1056.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1789.0, + 1207.0, + 1789.0, + 1207.0, + 1815.0, + 1138.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1816.0, + 407.0, + 1816.0, + 407.0, + 1831.0, + 390.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1837.0, + 382.0, + 1837.0, + 382.0, + 1854.0, + 367.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1878.0, + 809.0, + 1878.0, + 809.0, + 1904.0, + 345.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1878.0, + 846.0, + 1878.0, + 846.0, + 1904.0, + 835.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1896.0, + 726.0, + 1896.0, + 726.0, + 1928.0, + 342.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1922.0, + 763.0, + 1922.0, + 763.0, + 1948.0, + 370.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1922.0, + 982.0, + 1922.0, + 982.0, + 1948.0, + 788.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1922.0, + 1089.0, + 1922.0, + 1089.0, + 1948.0, + 1021.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1922.0, + 1160.0, + 1922.0, + 1160.0, + 1948.0, + 1105.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1949.0, + 357.0, + 1949.0, + 357.0, + 1964.0, + 345.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1964.0, + 645.0, + 1964.0, + 645.0, + 1993.0, + 342.0, + 1993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1988.0, + 983.0, + 1988.0, + 983.0, + 2014.0, + 370.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1988.0, + 1149.0, + 1988.0, + 1149.0, + 2014.0, + 1009.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 2016.0, + 359.0, + 2016.0, + 359.0, + 2032.0, + 344.0, + 2032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 238.0, + 560.0, + 238.0, + 560.0, + 262.0, + 370.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 260.0, + 560.0, + 260.0, + 560.0, + 284.0, + 392.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 279.0, + 832.0, + 279.0, + 832.0, + 309.0, + 414.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 279.0, + 984.0, + 279.0, + 984.0, + 309.0, + 849.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 279.0, + 1220.0, + 279.0, + 1220.0, + 309.0, + 1001.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 305.0, + 407.0, + 305.0, + 407.0, + 327.0, + 388.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 325.0, + 713.0, + 325.0, + 713.0, + 354.0, + 389.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 325.0, + 900.0, + 325.0, + 900.0, + 354.0, + 789.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 325.0, + 1042.0, + 325.0, + 1042.0, + 354.0, + 952.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 325.0, + 1429.0, + 325.0, + 1429.0, + 354.0, + 1058.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 354.0, + 382.0, + 354.0, + 382.0, + 370.0, + 368.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 370.0, + 560.0, + 370.0, + 560.0, + 395.0, + 367.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 393.0, + 560.0, + 393.0, + 560.0, + 418.0, + 392.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 415.0, + 832.0, + 415.0, + 832.0, + 440.0, + 415.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 415.0, + 984.0, + 415.0, + 984.0, + 440.0, + 849.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 415.0, + 1219.0, + 415.0, + 1219.0, + 440.0, + 1000.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 441.0, + 406.0, + 441.0, + 406.0, + 458.0, + 389.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 459.0, + 717.0, + 459.0, + 717.0, + 483.0, + 391.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 459.0, + 900.0, + 459.0, + 900.0, + 483.0, + 732.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 459.0, + 1042.0, + 459.0, + 1042.0, + 483.0, + 952.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 459.0, + 1428.0, + 459.0, + 1428.0, + 483.0, + 1057.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 485.0, + 382.0, + 485.0, + 382.0, + 505.0, + 366.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 510.0, + 357.0, + 510.0, + 357.0, + 524.0, + 347.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 526.0, + 739.0, + 526.0, + 739.0, + 550.0, + 346.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 526.0, + 971.0, + 526.0, + 971.0, + 550.0, + 757.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 526.0, + 1043.0, + 526.0, + 1043.0, + 550.0, + 988.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 550.0, + 334.0, + 550.0, + 334.0, + 569.0, + 319.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 576.0, + 308.0, + 576.0, + 308.0, + 589.0, + 299.0, + 589.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 306, + 228, + 1388, + 228, + 1388, + 1059, + 306, + 1059 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 105, + 300, + 105 + ], + "score": 0.854 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.833 + }, + { + "category_id": 13, + "poly": [ + 867, + 527, + 916, + 527, + 916, + 547, + 867, + 547 + ], + "score": 0.73, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 868, + 393, + 916, + 393, + 916, + 415, + 868, + 415 + ], + "score": 0.7, + "latex": "= 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 867, + 660, + 916, + 660, + 916, + 681, + 867, + 681 + ], + "score": 0.66, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 1008, + 530, + 1021, + 530, + 1021, + 545, + 1008, + 545 + ], + "score": 0.62, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1007, + 796, + 1021, + 796, + 1021, + 810, + 1007, + 810 + ], + "score": 0.59, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 1007, + 663, + 1021, + 663, + 1021, + 678, + 1007, + 678 + ], + "score": 0.56, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 985, + 617, + 1000, + 617, + 1000, + 633, + 985, + 633 + ], + "score": 0.55, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 973, + 351, + 988, + 351, + 988, + 368, + 973, + 368 + ], + "score": 0.55, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 833, + 617, + 848, + 617, + 848, + 633, + 833, + 633 + ], + "score": 0.54, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 984, + 485, + 1000, + 485, + 1000, + 501, + 984, + 501 + ], + "score": 0.54, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 832, + 485, + 849, + 485, + 849, + 501, + 832, + 501 + ], + "score": 0.53, + "latex": ": =" + }, + { + "category_id": 13, + "poly": [ + 681, + 527, + 754, + 527, + 754, + 547, + 681, + 547 + ], + "score": 0.51, + "latex": "= 1 \\mathrm { e } - 0 5" + }, + { + "category_id": 13, + "poly": [ + 822, + 352, + 837, + 352, + 837, + 368, + 822, + 368 + ], + "score": 0.5, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 997, + 927, + 1010, + 927, + 1010, + 943, + 997, + 943 + ], + "score": 0.48, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 677, + 660, + 754, + 660, + 754, + 680, + 677, + 680 + ], + "score": 0.47, + "latex": "\\mathtt { s } = 1 \\mathtt { e } - 0 5" + }, + { + "category_id": 13, + "poly": [ + 973, + 882, + 988, + 882, + 988, + 899, + 973, + 899 + ], + "score": 0.47, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 867, + 792, + 916, + 792, + 916, + 813, + 867, + 813 + ], + "score": 0.45, + "latex": "\\iota { = } 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 1136, + 352, + 1150, + 352, + 1150, + 368, + 1136, + 368 + ], + "score": 0.41, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1306, + 529, + 1325, + 529, + 1325, + 545, + 1306, + 545 + ], + "score": 0.4, + "latex": "\\bar { \\mathsf { z } } = \\mathsf { ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 822, + 883, + 836, + 883, + 836, + 899, + 822, + 899 + ], + "score": 0.39, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 833, + 750, + 848, + 750, + 848, + 766, + 833, + 766 + ], + "score": 0.39, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 985, + 750, + 999, + 750, + 999, + 766, + 985, + 766 + ], + "score": 0.37, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 682, + 795, + 696, + 795, + 696, + 810, + 682, + 810 + ], + "score": 0.37, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1008, + 397, + 1020, + 397, + 1020, + 411, + 1008, + 411 + ], + "score": 0.36, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1147, + 617, + 1163, + 617, + 1163, + 633, + 1147, + 633 + ], + "score": 0.36, + "latex": "{ } = { }" + }, + { + "category_id": 13, + "poly": [ + 1306, + 795, + 1323, + 795, + 1323, + 810, + 1306, + 810 + ], + "score": 0.34, + "latex": "s =" + }, + { + "category_id": 13, + "poly": [ + 1303, + 662, + 1324, + 662, + 1324, + 678, + 1303, + 678 + ], + "score": 0.33, + "latex": "{ \\tt S } =" + }, + { + "category_id": 13, + "poly": [ + 1298, + 927, + 1311, + 927, + 1311, + 943, + 1298, + 943 + ], + "score": 0.33, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1136, + 883, + 1150, + 883, + 1150, + 900, + 1136, + 900 + ], + "score": 0.32, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1322, + 396, + 1336, + 396, + 1336, + 412, + 1322, + 412 + ], + "score": 0.32, + "latex": "= ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 854, + 924, + 905, + 924, + 905, + 946, + 854, + 946 + ], + "score": 0.29, + "latex": "\\mathord { \\mathrm { 1 } } = 0 \\cdot 1" + }, + { + "category_id": 13, + "poly": [ + 1146, + 484, + 1163, + 484, + 1163, + 501, + 1146, + 501 + ], + "score": 0.29, + "latex": "\\bf \\tilde { \\tau } =" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 71.0, + 815.0, + 71.0, + 815.0, + 108.0, + 294.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 245.0, + 332.0, + 245.0, + 332.0, + 257.0, + 322.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 261.0, + 549.0, + 261.0, + 549.0, + 285.0, + 321.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 282.0, + 665.0, + 282.0, + 665.0, + 305.0, + 346.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 304.0, + 559.0, + 304.0, + 559.0, + 327.0, + 369.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 327.0, + 560.0, + 327.0, + 560.0, + 349.0, + 391.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 348.0, + 821.0, + 348.0, + 821.0, + 374.0, + 416.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 348.0, + 972.0, + 348.0, + 972.0, + 374.0, + 838.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 348.0, + 1135.0, + 348.0, + 1135.0, + 374.0, + 989.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 348.0, + 1207.0, + 348.0, + 1207.0, + 374.0, + 1151.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 376.0, + 406.0, + 376.0, + 406.0, + 393.0, + 390.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 394.0, + 867.0, + 394.0, + 867.0, + 417.0, + 391.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 394.0, + 1007.0, + 394.0, + 1007.0, + 417.0, + 917.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 394.0, + 1321.0, + 394.0, + 1321.0, + 417.0, + 1021.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 394.0, + 1391.0, + 394.0, + 1391.0, + 417.0, + 1337.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 422.0, + 380.0, + 422.0, + 380.0, + 434.0, + 369.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 437.0, + 559.0, + 437.0, + 559.0, + 460.0, + 368.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 458.0, + 559.0, + 458.0, + 559.0, + 482.0, + 391.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 483.0, + 831.0, + 483.0, + 831.0, + 507.0, + 416.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 483.0, + 983.0, + 483.0, + 983.0, + 507.0, + 850.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 483.0, + 1145.0, + 483.0, + 1145.0, + 507.0, + 1001.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 483.0, + 1218.0, + 483.0, + 1218.0, + 507.0, + 1164.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 511.0, + 403.0, + 511.0, + 403.0, + 523.0, + 391.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 523.0, + 680.0, + 523.0, + 680.0, + 550.0, + 390.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 523.0, + 866.0, + 523.0, + 866.0, + 550.0, + 755.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 523.0, + 1007.0, + 523.0, + 1007.0, + 550.0, + 917.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 523.0, + 1305.0, + 523.0, + 1305.0, + 550.0, + 1022.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 523.0, + 1380.0, + 523.0, + 1380.0, + 550.0, + 1326.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 555.0, + 380.0, + 555.0, + 380.0, + 567.0, + 369.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 571.0, + 559.0, + 571.0, + 559.0, + 594.0, + 369.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 591.0, + 560.0, + 591.0, + 560.0, + 614.0, + 394.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 616.0, + 832.0, + 616.0, + 832.0, + 639.0, + 417.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 616.0, + 984.0, + 616.0, + 984.0, + 639.0, + 849.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 616.0, + 1146.0, + 616.0, + 1146.0, + 639.0, + 1001.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 616.0, + 1218.0, + 616.0, + 1218.0, + 639.0, + 1164.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 641.0, + 406.0, + 641.0, + 406.0, + 658.0, + 390.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 658.0, + 676.0, + 658.0, + 676.0, + 685.0, + 390.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 658.0, + 866.0, + 658.0, + 866.0, + 685.0, + 755.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 658.0, + 1006.0, + 658.0, + 1006.0, + 685.0, + 917.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 658.0, + 1302.0, + 658.0, + 1302.0, + 685.0, + 1022.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 658.0, + 1381.0, + 658.0, + 1381.0, + 685.0, + 1325.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 688.0, + 380.0, + 688.0, + 380.0, + 699.0, + 369.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 702.0, + 561.0, + 702.0, + 561.0, + 728.0, + 368.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 726.0, + 560.0, + 726.0, + 560.0, + 749.0, + 393.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 748.0, + 832.0, + 748.0, + 832.0, + 771.0, + 417.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 748.0, + 984.0, + 748.0, + 984.0, + 771.0, + 849.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 748.0, + 1218.0, + 748.0, + 1218.0, + 771.0, + 1000.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 772.0, + 406.0, + 772.0, + 406.0, + 790.0, + 391.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 791.0, + 681.0, + 791.0, + 681.0, + 818.0, + 390.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 791.0, + 866.0, + 791.0, + 866.0, + 818.0, + 697.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 791.0, + 1006.0, + 791.0, + 1006.0, + 818.0, + 917.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 791.0, + 1305.0, + 791.0, + 1305.0, + 818.0, + 1022.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 791.0, + 1381.0, + 791.0, + 1381.0, + 818.0, + 1324.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 822.0, + 380.0, + 822.0, + 380.0, + 834.0, + 369.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 836.0, + 559.0, + 836.0, + 559.0, + 859.0, + 368.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 857.0, + 560.0, + 857.0, + 560.0, + 884.0, + 391.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 881.0, + 821.0, + 881.0, + 821.0, + 904.0, + 416.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 881.0, + 972.0, + 881.0, + 972.0, + 904.0, + 837.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 881.0, + 1135.0, + 881.0, + 1135.0, + 904.0, + 989.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 881.0, + 1205.0, + 881.0, + 1205.0, + 904.0, + 1151.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 925.0, + 853.0, + 925.0, + 853.0, + 948.0, + 391.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 925.0, + 996.0, + 925.0, + 996.0, + 948.0, + 906.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 925.0, + 1297.0, + 925.0, + 1297.0, + 948.0, + 1011.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 925.0, + 1369.0, + 925.0, + 1369.0, + 948.0, + 1312.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 956.0, + 377.0, + 956.0, + 377.0, + 962.0, + 371.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1012.0, + 766.0, + 1012.0, + 766.0, + 1037.0, + 321.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.75, + 905.5, + 407.75, + 905.5, + 407.75, + 923.5, + 387.75, + 923.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/O3bqkf_Puys/O3bqkf_Puys_layout.pdf b/parse/train/O3bqkf_Puys/O3bqkf_Puys_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a067efa9629c2527098ac8345f1b63d8a6b1d9ec --- /dev/null +++ b/parse/train/O3bqkf_Puys/O3bqkf_Puys_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2054c79e13a177d0dda7ae06e089018c0bc8ff4cd9ff8cd5e45877b0cbe637c +size 8493176 diff --git a/parse/train/O3bqkf_Puys/O3bqkf_Puys_origin.pdf b/parse/train/O3bqkf_Puys/O3bqkf_Puys_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3e82fd91d781b58cf9f4251c488324900e559a68 --- /dev/null +++ b/parse/train/O3bqkf_Puys/O3bqkf_Puys_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c71af5d42bee6aafd7b1458790023a8508a5cedd49552fb63e0d4ee562bb79a +size 8246930 diff --git a/parse/train/O3bqkf_Puys/O3bqkf_Puys_span.pdf b/parse/train/O3bqkf_Puys/O3bqkf_Puys_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6714ebf59e40bf99e531185db6609460c7c46659 --- /dev/null +++ b/parse/train/O3bqkf_Puys/O3bqkf_Puys_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b53a6b9953bbfafab8bd7347d69191ab33dde8f94e4d1efa716dd22c434c8f1c +size 8493733 diff --git a/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_layout.pdf b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d24b79f35e4e6b7a21a870da322304679dd5ec39 --- /dev/null +++ b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b305f569a3a2e21183df7e20b2bce33e3672f12f94db020a27adec6d9e0d593a +size 888378 diff --git a/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_origin.pdf b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7f7139e8c486f489b58e191ecd13e402e4524da2 --- /dev/null +++ b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f474c08daaa08622149876c883c6d7d38573bdde638c4040545eb5b3f72d14db +size 742702 diff --git a/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_span.pdf b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..93748bc5b6e4d882cb109f1e97f0decbfeed8ce3 --- /dev/null +++ b/parse/train/Q32U7dzWXpc/Q32U7dzWXpc_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1873760cbd5e2452c0039eb9ddad715a6d8f5de62f6caf0b073ca4da23fa1180 +size 890699 diff --git a/parse/train/R-616EWWKF5/R-616EWWKF5_layout.pdf b/parse/train/R-616EWWKF5/R-616EWWKF5_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..25f411718ad3ade555a4a95154a26b91b476364f --- /dev/null +++ b/parse/train/R-616EWWKF5/R-616EWWKF5_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:805f3de209aacca2e61b511a52ceb125da144703a5decd3f92addd92b2c2d88d +size 1637836 diff --git a/parse/train/R-616EWWKF5/R-616EWWKF5_origin.pdf b/parse/train/R-616EWWKF5/R-616EWWKF5_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6b61cc24d25373ed338a6ff44e2c08813ca22bdc --- /dev/null +++ b/parse/train/R-616EWWKF5/R-616EWWKF5_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74e24c523b5f089304ebe23d6cae95b47abbceb1c7b7df1dea436d6678cd0a4f +size 1533381 diff --git a/parse/train/R-616EWWKF5/R-616EWWKF5_span.pdf b/parse/train/R-616EWWKF5/R-616EWWKF5_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a8941837f9654c11f5c103b6f051f70080faf006 --- /dev/null +++ b/parse/train/R-616EWWKF5/R-616EWWKF5_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3445c17b80f8719564e96dfd4ca51b8339724dbf1a8dec50292973060dd0d1f2 +size 1642418 diff --git a/parse/train/S18Su--CW/S18Su--CW_layout.pdf b/parse/train/S18Su--CW/S18Su--CW_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..77012e9532319d3494af1107348ca6a5ae7fc459 --- /dev/null +++ b/parse/train/S18Su--CW/S18Su--CW_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2496c983f2efa517f4fd94cafb9080d6fb511bb1948bf72068237e043f599cab +size 2367806 diff --git a/parse/train/S18Su--CW/S18Su--CW_origin.pdf b/parse/train/S18Su--CW/S18Su--CW_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..32ec6bde2f8a561aa6566395a1f8a66dbb467101 --- /dev/null +++ b/parse/train/S18Su--CW/S18Su--CW_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b438dec170e0d5be569831680f9659e4740595c1bf6f16398c98e50f5472e8c8 +size 2136932 diff --git a/parse/train/S18Su--CW/S18Su--CW_span.pdf b/parse/train/S18Su--CW/S18Su--CW_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9de8eb1ae5e5c4444097abc5fc2b1f06202fd443 --- /dev/null +++ b/parse/train/S18Su--CW/S18Su--CW_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9a6f1dce3b1f826ed21fa1d32dd97a27d20ff141de41b39096a10ee604c5a63 +size 2370675 diff --git a/parse/train/S1AG8zYeg/S1AG8zYeg_layout.pdf b/parse/train/S1AG8zYeg/S1AG8zYeg_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..cd9dcc6e33104c9b5652609592d70cf194481124 --- /dev/null +++ b/parse/train/S1AG8zYeg/S1AG8zYeg_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78dc48afc133c2d74a212a41a3ac8fba8ad64d6e920f0ce44e0bca67c2ae6e5e +size 956175 diff --git a/parse/train/S1AG8zYeg/S1AG8zYeg_origin.pdf b/parse/train/S1AG8zYeg/S1AG8zYeg_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2ba306cc4d52328d65324f70ad9bbf017a850521 --- /dev/null +++ b/parse/train/S1AG8zYeg/S1AG8zYeg_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ca2da47aae70c93ec81ad194653569b03e03089524b19435fd9830cd349fe1c +size 769907 diff --git a/parse/train/S1AG8zYeg/S1AG8zYeg_span.pdf b/parse/train/S1AG8zYeg/S1AG8zYeg_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d20ec67f84c5fecd241c81ecf3a20b91bf361c43 --- /dev/null +++ b/parse/train/S1AG8zYeg/S1AG8zYeg_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5feeb950dc5a2468deaefa421222120bcf9ec52073dc30d345993bb32ce21a41 +size 953540 diff --git a/parse/train/S1erHoR5t7/S1erHoR5t7_layout.pdf b/parse/train/S1erHoR5t7/S1erHoR5t7_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..59c4fe89b4152df987995cbae032a8f8d589feb3 --- /dev/null +++ b/parse/train/S1erHoR5t7/S1erHoR5t7_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:744f3d4be1d4e77f2f84c9477738115f45d5d2030436c5fc2f30a70696c7666a +size 3123865 diff --git a/parse/train/S1erHoR5t7/S1erHoR5t7_origin.pdf b/parse/train/S1erHoR5t7/S1erHoR5t7_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..77142493f73a09fbfeb642755b0817b0af31cefb --- /dev/null +++ b/parse/train/S1erHoR5t7/S1erHoR5t7_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ee71fca0d7eaf341247c8aa3bc5e72251f371d2c70a4ad815c2a74b2d2a3fb8 +size 2823272 diff --git a/parse/train/S1erHoR5t7/S1erHoR5t7_span.pdf b/parse/train/S1erHoR5t7/S1erHoR5t7_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..28f3db6a98c9eaabfc8d7975452efc8abc6a545b --- /dev/null +++ b/parse/train/S1erHoR5t7/S1erHoR5t7_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35e221cd0cc4e995a2581a5103d9db2d5031cdd8eba26184bb5fe6b80d0fc77e +size 3123857 diff --git a/parse/train/S1gE6TEYDB/S1gE6TEYDB_layout.pdf b/parse/train/S1gE6TEYDB/S1gE6TEYDB_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4445ea962e17c7ef9094ed91514ddeaff75e25f8 --- /dev/null +++ b/parse/train/S1gE6TEYDB/S1gE6TEYDB_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2ff42135bfb0efb947f885c5e03562ebe582278f9ed7295ceddc25e065bd31a +size 4386237 diff --git a/parse/train/S1gE6TEYDB/S1gE6TEYDB_origin.pdf b/parse/train/S1gE6TEYDB/S1gE6TEYDB_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f82356dde4b21839ecb7f9efe0d132b40f122cb0 --- /dev/null +++ b/parse/train/S1gE6TEYDB/S1gE6TEYDB_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d5998934c592e7e907e8ee19caf8e0abdf5c040109301e40478641de42db608 +size 4267387 diff --git a/parse/train/S1gE6TEYDB/S1gE6TEYDB_span.pdf b/parse/train/S1gE6TEYDB/S1gE6TEYDB_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c3d882b4c0e83a01229268d822f5e2c41af3e2a4 --- /dev/null +++ b/parse/train/S1gE6TEYDB/S1gE6TEYDB_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5716d7b2c1eb8254f60b35879ee522a5c8283d0a0a2f94191f83a141fedf976 +size 4386700 diff --git a/parse/train/S1gmrxHFvB/S1gmrxHFvB_layout.pdf b/parse/train/S1gmrxHFvB/S1gmrxHFvB_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e5e2e5a05f0760a68896ac85331f1621bfa51452 --- /dev/null +++ b/parse/train/S1gmrxHFvB/S1gmrxHFvB_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ecfa50550aa5c3b462b2084b8eca4661b4e0b9d2b3b09ff07c48974e08f74fcc +size 6673031 diff --git a/parse/train/S1gmrxHFvB/S1gmrxHFvB_origin.pdf b/parse/train/S1gmrxHFvB/S1gmrxHFvB_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..147339305393f78e859dff2053cd00ae174b0c83 --- /dev/null +++ b/parse/train/S1gmrxHFvB/S1gmrxHFvB_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41f5810876809182972bcfe872cc42ac3718a94aa93ac8ab261c63e6975e8b80 +size 6527058 diff --git a/parse/train/S1gmrxHFvB/S1gmrxHFvB_span.pdf b/parse/train/S1gmrxHFvB/S1gmrxHFvB_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6f3f5630a8af75ef3cf15643cf91a59f4acf67fb --- /dev/null +++ b/parse/train/S1gmrxHFvB/S1gmrxHFvB_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa7796b42e715836e08d14a1a91e4302392616c88f1837e35a2c4a2479dce916 +size 6673936 diff --git a/parse/train/SJRpRfKxx/SJRpRfKxx_layout.pdf b/parse/train/SJRpRfKxx/SJRpRfKxx_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..836f0360c7d61f36d806759e496606645983eb69 --- /dev/null +++ b/parse/train/SJRpRfKxx/SJRpRfKxx_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c6d7c798cc005a5268262fd03a8f0df10697700c15e6136df99ef9b9812b7b6 +size 1032037 diff --git a/parse/train/SJRpRfKxx/SJRpRfKxx_origin.pdf b/parse/train/SJRpRfKxx/SJRpRfKxx_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6283de1cbb9507ee91ff1401dafdb44b56271733 --- /dev/null +++ b/parse/train/SJRpRfKxx/SJRpRfKxx_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4402adaa8a2a436a76ad0e94409ec03fe3cfc32120a3f59d99a28e210c311a02 +size 863596 diff --git a/parse/train/SJRpRfKxx/SJRpRfKxx_span.pdf b/parse/train/SJRpRfKxx/SJRpRfKxx_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8e3d9f347dc23abaf70de5375f9fcf2dc1e16f40 --- /dev/null +++ b/parse/train/SJRpRfKxx/SJRpRfKxx_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac4650a0d355aa7ec41f2b5e7cfe6fed5d16fc1ee6fd602ffbfbea50dc5a744a +size 1035791 diff --git a/parse/train/SJzMATlAZ/SJzMATlAZ_layout.pdf b/parse/train/SJzMATlAZ/SJzMATlAZ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..51f68027d844b18a99a10b131410642b1337e9c7 --- /dev/null +++ b/parse/train/SJzMATlAZ/SJzMATlAZ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed8076469fbb49dbcc54e27008503fce0efe4e8e4197cd049bfa74787132d347 +size 1172452 diff --git a/parse/train/SJzMATlAZ/SJzMATlAZ_origin.pdf b/parse/train/SJzMATlAZ/SJzMATlAZ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..038a1feaf958c0502b01272513355c8d1077805f --- /dev/null +++ b/parse/train/SJzMATlAZ/SJzMATlAZ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d8a352458041d5c252a1e6e2c3fc75fbab338407838e00abc7ddd386516f671 +size 1033744 diff --git a/parse/train/SJzMATlAZ/SJzMATlAZ_span.pdf b/parse/train/SJzMATlAZ/SJzMATlAZ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7fab2eb03bb5e61162a07c9c912e3bcd96c00b58 --- /dev/null +++ b/parse/train/SJzMATlAZ/SJzMATlAZ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfa1a6871b7c66f7652f9475a28f5e50fe7bf226235ec7426150e752fba5539e +size 1176989 diff --git a/parse/train/Skeke3C5Fm/Skeke3C5Fm_layout.pdf b/parse/train/Skeke3C5Fm/Skeke3C5Fm_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6bbfb9f127a7d7b7a3b6040d19bb68541fbeba88 --- /dev/null +++ b/parse/train/Skeke3C5Fm/Skeke3C5Fm_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24b68d3663a3d8a8ec41ea14fab54277483ab125aa0d427906b327439383f8cb +size 558717 diff --git a/parse/train/Skeke3C5Fm/Skeke3C5Fm_origin.pdf b/parse/train/Skeke3C5Fm/Skeke3C5Fm_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8e7fedf4073c8885a56912ec145b55f243432640 --- /dev/null +++ b/parse/train/Skeke3C5Fm/Skeke3C5Fm_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72c6afe8a1ef743f2d7d2ea006126ad4f41bbd2ddd475c6f67c795543fd6d6df +size 434918 diff --git a/parse/train/Skeke3C5Fm/Skeke3C5Fm_span.pdf b/parse/train/Skeke3C5Fm/Skeke3C5Fm_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4bd74fe9af4b6ac508d426c4cde02ccdfece7820 --- /dev/null +++ b/parse/train/Skeke3C5Fm/Skeke3C5Fm_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fae36b2e56e218cb508c8140557dfd4f70c01d973a4ce28add99b908347be5b +size 558166 diff --git a/parse/train/SkgEaj05t7/SkgEaj05t7_layout.pdf b/parse/train/SkgEaj05t7/SkgEaj05t7_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c78a2134ec0709c914cb1042028a22969c843c03 --- /dev/null +++ b/parse/train/SkgEaj05t7/SkgEaj05t7_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eea11e1d7833452368f57831e5b9149a2446d2fd61773d404bcd9932ac929322 +size 3212765 diff --git a/parse/train/SkgEaj05t7/SkgEaj05t7_origin.pdf b/parse/train/SkgEaj05t7/SkgEaj05t7_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d072ef646f869cf96a1a49ef8525110ef3286d6b --- /dev/null +++ b/parse/train/SkgEaj05t7/SkgEaj05t7_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c83261c524bf9d926c0619b677eb7cb5cc6dfda0d0a2f4c71b0eaa1468f36cdd +size 3015962 diff --git a/parse/train/SkgEaj05t7/SkgEaj05t7_span.pdf b/parse/train/SkgEaj05t7/SkgEaj05t7_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0be0bae341e4d359d915f96a37ad1539f0845e22 --- /dev/null +++ b/parse/train/SkgEaj05t7/SkgEaj05t7_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e0439b4f2ac817529ed2098bcec2e5cf29dc3d885b6bbfe277d9e92fe77554f +size 3220312 diff --git a/parse/train/SkhQHMW0W/SkhQHMW0W_layout.pdf b/parse/train/SkhQHMW0W/SkhQHMW0W_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2ef020d14a1bcf291526334ff110595e005538d5 --- /dev/null +++ b/parse/train/SkhQHMW0W/SkhQHMW0W_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6226ea10248b4cb0d3fab2cb8b9d9ba1064bc2bb75a83ff374dbefaf28d67ec3 +size 2285191 diff --git a/parse/train/SkhQHMW0W/SkhQHMW0W_origin.pdf b/parse/train/SkhQHMW0W/SkhQHMW0W_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5097bdde36dba52bfc8130607d4f0e809646439a --- /dev/null +++ b/parse/train/SkhQHMW0W/SkhQHMW0W_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a7e3fb458bd715f349f7232794a5969c1ca3a1b322f8105aa79f83f19cfaf93 +size 2125609 diff --git a/parse/train/SkhQHMW0W/SkhQHMW0W_span.pdf b/parse/train/SkhQHMW0W/SkhQHMW0W_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5f3b2c0d1214612deef30fc45d547ec02984b521 --- /dev/null +++ b/parse/train/SkhQHMW0W/SkhQHMW0W_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76faef1073b0c5f0125f29df2a9c1db8a8d6c6861eed67259a6552aaebd1d9d1 +size 2287305 diff --git a/parse/train/Skk3Jm96W/Skk3Jm96W_layout.pdf b/parse/train/Skk3Jm96W/Skk3Jm96W_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3c11bde515b2e0a3d1e214200f549dcf4b184f9e --- /dev/null +++ b/parse/train/Skk3Jm96W/Skk3Jm96W_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00d173f23aaa3ea373f4b5360601cbede94cdbba3ce6c3f1ec734aa67a640f8a +size 3930982 diff --git a/parse/train/Skk3Jm96W/Skk3Jm96W_origin.pdf b/parse/train/Skk3Jm96W/Skk3Jm96W_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e5023070060e904a044187cd3b24412588748363 --- /dev/null +++ b/parse/train/Skk3Jm96W/Skk3Jm96W_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df66eb0980e7564326ad2e301ad09f32fca09ab6b6068472fddc5272dd313edd +size 3817212 diff --git a/parse/train/Skk3Jm96W/Skk3Jm96W_span.pdf b/parse/train/Skk3Jm96W/Skk3Jm96W_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..11a84d69d1eb872d930671ec7e218ad8873aa6c8 --- /dev/null +++ b/parse/train/Skk3Jm96W/Skk3Jm96W_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0fa4494f48cbf20585d2375755ddb0aa9443613b2b22aadfb9522f9ba194439 +size 3935933 diff --git a/parse/train/SkpSlKIel/SkpSlKIel_layout.pdf b/parse/train/SkpSlKIel/SkpSlKIel_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b17c6c73d58fa39bc3b992c7ff3ef815b1d39a11 --- /dev/null +++ b/parse/train/SkpSlKIel/SkpSlKIel_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04bb72355d7d760efb5910fbfe0cf14d263bcedd5244009cb28ec5c2b981ff0a +size 1327641 diff --git a/parse/train/SkpSlKIel/SkpSlKIel_origin.pdf b/parse/train/SkpSlKIel/SkpSlKIel_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..609b57bff13dbe51c44f6a4e8917174593de90f5 --- /dev/null +++ b/parse/train/SkpSlKIel/SkpSlKIel_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4c253803b420e79135260e8ccaf83950283ac18e8ccdc7ab9eddd16b278dd12 +size 1002320 diff --git a/parse/train/SkpSlKIel/SkpSlKIel_span.pdf b/parse/train/SkpSlKIel/SkpSlKIel_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4127f094b72013fa79274bd76230771648b85e64 --- /dev/null +++ b/parse/train/SkpSlKIel/SkpSlKIel_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70b5d4604909d97059d4fb170dc7a570b42690721be69b0bd5589975e05eaa98 +size 1342565 diff --git a/parse/train/Sy1f0e-R-/Sy1f0e-R-.md b/parse/train/Sy1f0e-R-/Sy1f0e-R-.md new file mode 100644 index 0000000000000000000000000000000000000000..3f71175db0c94ca759397a5e6d72089cdd22793f --- /dev/null +++ b/parse/train/Sy1f0e-R-/Sy1f0e-R-.md @@ -0,0 +1,310 @@ +# AN EMPIRICAL STUDY ON EVALUATION METRICS OF GENERATIVE ADVERSARIAL NETWORKS + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +Despite the widespread interest in generative adversarial networks (GANs), few works have studied the metrics that quantitatively evaluate GANs’ performance. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the important problem of how to evaluate the evaluation metrics. We start with a few necessary conditions for metrics to produce meaningful scores, such as distinguishing real from generated samples, identifying mode dropping and mode collapsing, and detecting overfitting. Then with a series of carefully designed experiments, we are able to comprehensively investigate existing sample-based metrics and identify their strengths and limitations in practical settings. Based on these results, we observe that kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbour (1-NN) two-sample test seem to satisfy most of the desirable properties, provided that the distances between samples are computed in a suitable feature space. Our experiments also unveil interesting properties about the behavior of several popular GAN models, such as whether they are memorizing training samples, and how far these state-of-the-art GANs are from perfect. + +# 1 INTRODUCTION + +Generative adversarial networks (GANs) (Goodfellow et al., 2014) have been studied extensively in recent years. Besides producing surprisingly plausible images of faces (Radford et al., 2015; Larsen et al., 2015) and bedrooms (Radford et al., 2015; Arjovsky et al., 2017; Gulrajani et al., 2017), they have also been innovatively applied in, for example, semi-supervised learning (Odena, 2016; Makhzani et al., 2015), image-to-image translation (Isola et al., 2016; Zhu et al., 2017), and simulated image refinement (Shrivastava et al., 2016). However, despite the availability of a plethora of GAN models (Arjovsky et al., 2017; Qi, 2017; Radford et al., 2015; Zhao et al., 2016), their evaluation is still predominantly qualitative, very often resorting to manual inspection of the visual fidelity of generated images. Such evaluation is time-consuming, subjective and possibly misleading. Given the inherent limitations of qualitative evaluations, proper quantitative metrics are crucial for the development of GANs to avoid the human factors and guide the design of better models. + +Possibly the most popular metric is the Inception Score (Salimans et al., 2016), which measures the quality and diversity of the generated images using an external model, the Google Inception network (Szegedy et al., 2014), trained on the large scale ImageNet dataset (Deng et al., 2009). Some other metrics are less widely used but still very valuable. Wu et al. (2016) proposed a sampling method to estimate the log-likelihood of generative models, by assuming a Gaussian observation model with a fixed variance. Bounliphone et al. (2015) propose to use maximum mean discrepancies (MMDs) for model selection in generative models. Lopez-Paz & Oquab (2016) apply the classifier two-sample test, a well-studied tool in statistics, to assess the difference between the generated and target distribution. + +Although these evaluation metrics are shown to be effective on various tasks, it is unclear in which scenarios their scores are meaningful, and in which other scenarios, prone to misinterpretations. Given that evaluating GANs is already challenging, it can only be more difficult to evaluate the evaluation metrics themselves. Most existing works attempt to justify their proposed metrics by showing a strong correlation with human evaluation (Salimans et al., 2016; Lopez-Paz & Oquab, 2016). However, human evaluation tends to be biased towards the visual quality of generated samples and neglect the overall distributional characteristics, which are arguably just as important for unsupervised learning. + +![](images/49450f89442e39d6bd90bdd00539f5bc95c21f16ec6bfdf0c04ded989a0111e0.jpg) +Figure 1: A schematic layout of the typical approach for sample based GAN evaluation methods. + +In this paper we comprehensively examine the existing literature on sample-based quantitative evaluation of GANs. We address the challenge of evaluating the metrics themselves by carefully designing a series of experiments, through which we hope to answer the following important questions: 1.) What are reasonable characterizations of the behavior of existing sample-based metrics for GANs? 2.) What are the strengths and limitations of these metrics? 3.) Which metrics are preferred accordingly? 4.) How are the metrics helpful in understanding and improving GANs? + +Ultimately, we hope that this paper will establish good principles on choosing, applying, interpreting and designing evaluation metrics for GANs in practical settings. We will also release the source code for all experiments and metrics examined, providing the community with off-the-shelf tools to debug and improve their GAN algorithms. + +# 2 BACKGROUND + +We briefly review the original GAN framework proposed by Goodfellow et al. (2014). Description of the GAN variants used in our experiments is deferred to the Appendix A. + +# 2.1 GENERATIVE ADVERSARIAL NETWORKS + +Let $\mathcal { X } = \mathbb { R } ^ { d \times d }$ be the space of natural images. Given i.i.d. samples $S _ { r } = \{ \mathbf { x } _ { 1 } ^ { r } , \ldots , \mathbf { x } _ { n } ^ { r } \}$ drawn from a real distribution $\mathbb { P } _ { r }$ over $\mathcal { X }$ , we would like to learn a parameterized distribution $\mathbb { P } _ { g }$ that approximates the distribution $\mathbb { P } _ { r }$ . + +The setup of generative adversarial networks is as follows. We define two networks, the discriminator $D : \mathcal { X } [ 0 , 1 )$ and the generator $G : { \mathcal { Z } } \to { \mathcal { X } }$ , where $\mathcal { Z }$ is some latent space. Given a distribution $\mathbb { P } _ { z }$ over $\mathcal { Z }$ (usually an isotropic Gaussian), the distribution $\mathbb { P } _ { g }$ is defined as $G ( \mathbb { P } _ { z } )$ . Optimization is performed with respect to a joint loss for $D$ and $G$ + +$$ +\operatorname* { m i n } _ { G } \operatorname* { m a x } _ { D } L ( D , G ) = \mathbb { E } _ { \mathbf { x } \sim \mathbb { P } _ { r } } \log { [ D ( \mathbf { x } ) ] } + \mathbb { E } _ { \mathbf { z } \sim \mathbb { P } _ { z } } \left[ \log ( 1 - D ( G ( \mathbf { z } ) ) ) \right] . +$$ + +Intuitively, the discriminator $D$ outputs a probability for every $\mathbf { x } \in \mathcal { X }$ that corresponds to its likelihood of being drawn from $\mathbb { P } _ { r }$ , and the loss function encourages the generator $G$ to produce samples that maximize this probability. Practically, the loss is approximated with finite samples from $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ , and optimized with alternating steps for $D$ and $G$ using gradient descent. + +To evaluate the generator, we would like to design a metric $\rho$ that measures the “dissimilarity" between $\mathbb { P } _ { g }$ to $\mathbb { P } _ { r }$ .1 In theory, with both distributions known, common choices of $\rho$ include the Kullback-Leibler divergence (KLD), Jensen-Shannon divergence (JSD) and total variation. However, in practical scenarios, $\mathbb { P } _ { r }$ is unknown and only the finite samples in $S _ { r }$ are observed. Furthermore, it is almost always intractable to compute the exact density of $S _ { g } = \{ \mathbf { x } _ { 1 } ^ { g } , . . . , \mathbf { x } _ { m } ^ { \bar { g } } \} \sim \mathbb { P } _ { q } ^ { m }$ specially so for GANs). Given these limitations, we focus on empiricalof “dissimilarity" between samples from two distributions. $\mathbb { P } _ { g }$ , but much easier to sample $\hat { \rho } : \mathcal X ^ { n } \times \mathcal X ^ { m } \stackrel { \smile } { } \mathbb R$ + +# 2.2 SAMPLE BASED METRICS + +We mainly focus on sample based evaluation metrics that follow a common setup illustrated in Figure 1. The metric calculator is the key element, for which we briefly introduce five representative methods: Inception Score (Salimans et al., 2016), Mode Score (Che et al., 2016) , Kernel MMD (Gretton et al., 2007), Wasserstein distance, Fréchet Inception Distance (FID) (Heusel et al., 2017), and 1-nearest neighbor (1-NN)-based two sample test (Lopez-Paz & Oquab, 2016). All of them are model agnostic and require only finite samples from the generator. + +The Inception Score is arguably the most widely adopted metric in the literature. It uses a image classification model $\mathcal { M }$ , the Google Inception network (Szegedy et al., 2016), pre-trained on the ImageNet (Deng et al., 2009) dataset, to compute + +$$ +\mathrm { I S } ( \mathbb { P } _ { g } ) = e ^ { \mathbb { E } _ { \mathbf { x } \sim \mathbb { P } _ { g } } [ K L ( p _ { \mathcal { M } } ( y | \mathbf { x } ) | | p _ { \mathcal { M } } ( y ) ) ] } , +$$ + +where $p _ { \mathcal { M } } ( y | \mathbf { x } )$ denotes the label distribution of $\mathbf { x }$ as predicted by $\mathcal { M }$ , and $\begin{array} { r } { p _ { \mathcal M } ( y ) = \int _ { \mathbf x } p _ { \mathcal M } ( y | \mathbf x ) d \mathbb P _ { g } } \end{array}$ , i.e. the marginal of $p _ { \mathcal { M } } ( y | \mathbf { x } )$ over the probability measure $\mathbb { P } _ { g }$ . The expectation and the integral in $p _ { \mathcal { M } } ( y | \mathbf { x } )$ can be approximated with i.i.d. samples from $\mathbb { P } _ { g }$ . A higher IS has $p _ { \mathcal { M } } ( y | \mathbf { x } )$ close to a point mass, which happens when the Inception network is very confident that the image belongs to a particular ImageNet category, and has $p _ { \mathcal { M } } ( y )$ close to uniform, i.e. all categories are equally represented. This suggests that the generative model has both high quality and diversity. Salimans et al. (2016) show that the Inception Score has a reasonable correlation with human judgment of image quality. We would like to highlight two specific properties: 1) the distributions on both sides of the KL are dependent on $\mathcal { M }$ , and 2) the distribution of the real data $\mathbb { P } _ { r }$ , or even samples thereof, are not used anywhere. + +The Mode Score is an improved version of the Inception Score. Formally, it is given by + +$$ +\mathrm { M S } ( \mathbb { P } _ { g } ) = e ^ { \mathbb { E } _ { \mathbf { x } \sim \mathbb { P } _ { g } } [ K L ( p _ { \mathcal { M } } ( y | \mathbf { x } ) | | p _ { \mathcal { M } } ( y ) ) ] - K L ( p _ { \mathcal { M } } ( y ) | | p _ { \mathcal { M } } ( y ^ { \ast } ) ) } , +$$ + +where $\begin{array} { r } { p _ { \mathcal M } ( y ^ { \ast } ) = \int _ { \mathbf x } p _ { \mathcal M } ( y | \mathbf x ) d \mathbb P _ { r } } \end{array}$ is the marginal label distribution for the samples from the real data distribution. Unlike the Inception Score, it is able to measure the dissimilarity between the real distribution $\mathbb { P } _ { r }$ and generated distribution $\mathbb { P } _ { g }$ through the term $K L ( p _ { \mathcal { M } } ( y ) | | p _ { \mathcal { M } } ( \bar { y } ^ { * } ) )$ . + +The Kernel MMD (Maximum Mean Discrepancy), defined as + +$$ +\begin{array} { r } { \mathrm { M M D } ( \mathbb { P } _ { r } , \mathbb { P } _ { g } ) = \biggl ( \mathbb { E } _ { \mathbf { x } _ { r } , \mathbf { x } _ { r } ^ { \prime } \sim \mathbb { P } _ { r } , } \biggl [ k ( \mathbf { x } _ { r } , \mathbf { x } _ { r } ^ { \prime } ) - 2 k ( \mathbf { x } _ { r } , \mathbf { x } _ { g } ) + k ( \mathbf { x } _ { g } , \mathbf { x } _ { g } ^ { \prime } ) \biggr ] \biggr ) ^ { \frac { 1 } { 2 } } , } \end{array} +$$ + +measures the dissimilarity between $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ for some fixed kernel function $k$ . Given two sets of samples from $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ , the empirical MMD between the two distributions can be computed with finite sample approximation of the expectation. A lower MMD means that $\mathbb { P } _ { g }$ is closer to $\mathbb { P } _ { r }$ . The Parzen window estimate (Gretton et al., 2007) can be viewed as a specialization of Kernel MMD. + +The Wasserstein distance between $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ is defined as + +$$ +\operatorname { W D } ( \mathbb { P } _ { r } , \mathbb { P } _ { g } ) = \operatorname* { i n f } _ { \substack { \gamma \in \Gamma ( \mathbb { P } _ { r } , \mathbb { P } _ { g } ) } } \mathbb { E } _ { ( \mathbf { x } ^ { r } , \mathbf { x } ^ { g } ) \sim \gamma } \left[ d ( \mathbf { x } ^ { r } , \mathbf { x } ^ { g } ) \right] , +$$ + +where $\Gamma ( \mathbb { P } _ { r } , \mathbb { P } _ { g } )$ denotes the set of all joint distributions (i.e. probabilistic couplings) whose marginals are respectively $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ , and $d ( \mathbf { x } ^ { r } , \mathbf { x } ^ { g } )$ denotes the base distance between the two samples. For discrete distributions with densities $p _ { r }$ and $p _ { g }$ , the Wasserstein distance is often referred to as the Earth Mover’s Distance (EMD), and corresponds to the solution to the optimal transport problem + +$$ +N \mathrm { D } ( p _ { r } , p _ { g } ) = \operatorname* { m i n } _ { w \in \mathbb { R } ^ { n \times m } } \sum _ { i = 1 } ^ { n } \sum _ { j = 1 } ^ { m } w _ { i j } d ( \mathbf { x } _ { i } ^ { r } , \mathbf { x } _ { j } ^ { g } ) \quad \mathrm { s . t . } \quad \sum _ { j = 1 } ^ { m } w _ { i , j } = p _ { r } ( \mathbf { x } _ { i } ^ { r } ) \ \forall i , \sum _ { i = 1 } ^ { n } w _ { i , j } = p _ { g } ( \mathbf { x } _ { j } ^ { g } ) \ \forall j . +$$ + +This is the finite sample approximation of $\mathrm { W D } ( \mathbb { P } _ { r } , \mathbb { P } _ { g } )$ used in practice. Similar to MMD, the Wasserstein distance is lower when two distributions are more similar. + +The Fréchet Inception Distance (FID) was recently introduced by Heusel et al. (2017) to evaluate GANs. Formally, it is given by + +$$ +\mathrm { F I D } ( \mathbb { P } _ { r } , \mathbb { P } _ { g } ) = \| \mu _ { r } - \mu _ { g } \| + \operatorname { T r } ( \mathbf { C } _ { r } + \mathbf { C } _ { g } - 2 ( \mathbf { C } _ { r } \mathbf { C } _ { g } ) ^ { 1 / 2 } ) , +$$ + +where $\mu _ { r } \left( \mu _ { g } \right)$ and $\mathbf { C } _ { r }$ $( \mathbf { C } _ { g } )$ are the mean and covariance of the real (generated) distribution, respectively. Note that under the Gaussian assumption on both $\mathbb { P } _ { r }$ and $\mathbb { P } _ { g }$ , the Fréchet distance is equivalent to the Wasserstein-2 distance. + +The 1-Nearest Neighbor classifier is used in two-sample tests to assess whether two distributions are identical. Given two sets of samples $S _ { r } \sim \mathbb { P } _ { r } ^ { n }$ and $S _ { g } \sim \mathbb { P } _ { g } ^ { m }$ , with $| S _ { r } | = | S _ { g } |$ , one can compute the leave-one-out (LOO) accuracy of a 1-NN classifier trained on $S _ { r }$ and $S _ { g }$ with positive labels for $S _ { r }$ and negative labels for $S _ { g }$ . Different from the most common use of accuracy, here the 1-NN classifier should yield a $\sim 5 0 \%$ LOO accuracy when $| S _ { r } | = | S _ { g } |$ is large. This is achieved when the two distributions match. The LOO accuracy can be lower than $5 0 \%$ , which happens when the GAN overfits $\mathbb { P } _ { g }$ to $S _ { r }$ . In the (hypothetical) extreme case, if the GAN were to memorize every sample in $S _ { r }$ and re-generate it exactly, i.e. $S _ { g } = S _ { r }$ , the accuracy would be $0 \%$ , as every sample from $S _ { r }$ would have it nearest neighbour from $S _ { g }$ with zero distance. The 1-NN classifier belongs to the two-sample test family, for which any binary classifier can be adopted in principle. We will only consider the 1-NN classifier because it requires no special training and little hyperparameter tuning. + +Lopez-Paz & Oquab (2016) considered the 1-NN accuracy primarily as a statistic for two-sample testing. In fact, it is more informative to analyze it for the two classes separately. For example, a typical outcome of GANs is that for both real and generated images, the majority of their nearest neighbors are generated images due to mode collapse. In this case, the LOO 1-NN accuracy of the real images would be relatively low (desired): the mode(s) of the real distribution are usually well captured by the generative model, so a majority of real samples from $S _ { r }$ are surrounded by generated samples from $S _ { g }$ , leading to low LOO accuracy; whereas the LOO accuracy of the generated images is high (not desired): generative samples tend to collapse to a few mode centers, thus they are surrounded by samples from the same class, leading to high LOO accuracy. For the rest of the paper, we distinguish these two cases as 1-NN accuracy (real) and 1-NN accuracy (fake). + +# 2.3 OTHER METRICS + +All of the metrics above are, what we refer to as “model agnostic": they use the generator as a black box to sample the generated images $S _ { g }$ . Model agnostic metrics should not require a density estimation from the model. We choose to only experiment with model agnostic metrics, which allow us to support as many generative models as possible for evaluation without modification to their structure. We will briefly mention some other evaluation metrics not included in our experiments. + +Kernel density estimation (KDE, or Parzen window estimation) is a well-studied method for estimating the density function of a distribution from samples. For a probability kernel $K$ (most often an isotropic Gaussian) and i.i.d samples $\mathbf { x } _ { 1 } , \ldots , \mathbf { x } _ { n }$ , we can define the density function at $\mathbf { x }$ as $p ( \mathbf { x } ) \approx$ $\textstyle { \frac { 1 } { z } } \sum _ { i = 1 } ^ { n ^ { \cdot } } K ( \mathbf { x } - \mathbf { x } _ { i } )$ , where $z$ is a normalizing constant. This allows the use of classical metrics such as KLD and JSD. However, despite the widespread adoption of this technique to various applications, its suitability to estimating the density of $\mathbb { P } _ { r }$ or $\mathbb { P } _ { g }$ for GANs has been questioned by Theis et al. (2015) since the probability kernel depends on the Euclidean distance between images. + +More recently, Wu et al. (2016) applied annealed importance sampling (AIS) to estimate the marginal distribution $p ( \mathbf { x } )$ of a generative model. This method is most natural for models that define a conditional distribution $p ( \mathbf { x } | \mathbf { z } )$ where $\mathbf { z }$ is the latent code, which is not satisfied by most GAN models. Nevertheless, AIS has been applied to GAN evaluation by assuming a Gaussian observation model. We exclude this method from our experiments as it needs the access to the generative model to compute the likelihood, instead of only depending on a finite sample set $S _ { g }$ . + +# 3 EXPERIMENTS WITH GAN EVALUATION METRICS + +# 3.1 FEATURE SPACE + +All the metrics introduced in the previous section, except for the Inception Score and Mode Score, access the samples $\mathbf { x }$ only through pair-wise distances. The Kernel MMD requires a fixed kernel function $k$ , typically set to an isotopic Gaussian; the Wasserstein distance and 1-NN accuracy use the underlying distance metric $d$ directly; all of these methods are highly sensitive to the choice that distance. + +It is well-established that pixel representations of images do not induce meaningful Euclidean distances (Forsyth & Ponce, 2011). Small translations, rotations, or changes in illumination can increase distances dramatically with little effect on the image content. To quantity the similarity between distributions of images, it is therefore desirable to use distances invariant to such transformations. The choice of distance function can be re-interpreted as a choice of representation, by defining the distance in a more general form as $d ( \mathbf { x } , \mathbf { x } ^ { \prime } ) = \| \boldsymbol { \phi } ( \mathbf { x } ) - \boldsymbol { \phi } ( \mathbf { x } ^ { \prime } ) \| _ { 2 }$ , where $\phi ( \cdot )$ is some general mapping of the input into a semantically meaningful feature space. For Kernel MMD, this corresponds to computing the usual inner product in the feature space $\phi ( \mathcal { X } )$ . + +![](images/cab638a8370b4cd54a62381ec515f1f9af5798f9c29f188a4b0647fee588444c.jpg) +Figure 2: Distinguishing a set of real images (in the training set) from a mixed set of real images and GAN generated images. For the metric to be discriminative, its score should increase as the fraction of generated samples in the mix increases. RIS and RMS fail as they decrease with the fraction of generated samples in $S _ { g }$ on LSUN. Wasserstein and 1-NN accuracy (real) fail in pixel space as they do not increase. + +Inspired by recent works from Upchurch et al. (2017); Larsen et al. (2015) which show that convolutional neural networks may linearize the image manifold, we propose to operate in the feature space of an external model pre-trained on the ImageNet dataset. For efficiency, we use a 34-layer ResNet2 as the feature extractor. Our experiments show that other models such as VGG or Inception give very similar results. + +To illustrate our point, we show failure examples of the pixel space distance for evaluating GANs in this section, and highlight that using a proper feature space is key to obtaining meaningful results when applying the distance-based metrics. The usage of a well-suited feature space enables us to draw more optimistic conclusions on GAN evaluation metrics than in Theis et al. (2015). + +# 3.2 SETUP + +For the rest of this major section, we introduce what in our opinion are necessary conditions for good metrics for GANs. After the introduction of each condition, we use it as a criterion to judge the effectiveness of the metrics presented in Section 2, through carefully designed empirical experiments. + +The experiments are performed on two standard benchmark datasets for generative models, CelebA3and LSUN bedrooms4. To remove the degree of freedom induced by feature representation, we use the Inception Score (IS) and Mode Score (MS) computed from the softmax probabilities of the same ResNet-34 model as the other metrics, instead of the Inception model. We also compute the Inception Score over the real training data $S _ { r }$ as an upper bound, which we denote as $\mathrm { I S } _ { \mathrm { 0 } }$ . Moreover, to be consistent with other metrics where lower values correspond to better models, we report the relative inverse Inception Score $R I S = \left( 1 - { \mathrm { I S } } / { \mathrm { I S } } _ { 0 } \right)$ in tables and plots, after computing IS the Inception Score. We similarly report the relative inverse Mode Score (RMS). Although RIS and RMS operate in the softmax space, we always compare them together with other metrics in the convolutional space for simplicity. For all the plots in this paper, shaded areas denote the standard deviations, computed by running the same experiment 5 times with different random seeds. + +# 3.3 DISCRIMINABILITY + +Mixing of generated images. Arguably, the most important property of a metric $\hat { \rho }$ for measuring GANs is the ability to distinguish generated images from real images. To test this property, we sample a set $S _ { r }$ consisting of $n$ ( $n = 2 0 0 0$ if not otherwise specified) real images uniformly from the training set, and a set $S _ { g } ( t )$ of the same size $n$ consisting of a mix of real samples and generated images from a DCGAN (Radford et al., 2015) trained on the same training set, where $t$ denotes the ratio of generated images. + +The computed values of various metrics between $S _ { r }$ and $S _ { g } ( t )$ for $t \in [ 0 , 1 ]$ are shown in Figure 2. Since $\mathbb { P } _ { r }$ should serve as a lower bound for any metric, we expect that any reasonable $\hat { \rho }$ should increase as the ratio of generated images increases. This is indeed satisfied for all the metrics except: + +![](images/1c1ee6c3e031cabba0b7fe1ab9f3c6c30c09d9cf1e241dac2fc9e61e61743cb0.jpg) +Figure 3: Experiment on simulated mode collapsing. A metric score should increase to reflect the mismatch between true distribution and generated distribution as more modes are collapsed towards their cluster center. All metrics respond correctly in convolutional space. In pixel space, both Wasserstein distance and 1-NN accuracy (real) fail as they decrease in response to more collapsed clusters. + +1) RIS and RMS (red and green curves) on LSUN, which decrease as more fake samples are in the mix; 2) 1-NN accuracy of real samples (dotted magenta curve) computed in pixel space, which also appears to be a decreasing function; and 3) Wasserstein Distance (cyan curve), which almost remains unchanged when $t$ varies. + +The reason that RIS and RMS do not work well here is likely because they are not suitable for images beyond the ImageNet categories. Although other metrics operate in the convolutional feature space also depend on a network pretrained on ImageNet, the convolutional features are much more general than the specific softmax representation. The failure of Wasserstein Distance is possibly due to an insufficient number of samples, which we will discuss in more detail when we analyze the sample efficiency of various metrics in a latter subsection. The last paragraph of Section 2.2 explains why the 1-NN accuracy for real samples (dotted magenta curve) is always lower than that for generated samples (dashed magenta curve). In the pixel space, more than half of the samples from $S _ { r }$ have the nearest neighbor from $S _ { g } ( t )$ , indicating that the DCGAN is able to represent the modes in the pixel space quite well. + +We also conducted the same experiment using 1) random noise images and 2) images from an entirely different distribution (e.g. CIFAR-10), instead of DCGAN generated images to construct $S _ { g } ( t )$ . We call these injected samples as out-of-domain violations since they are not in $\mathcal { X }$ , the domain of the real images. These settings yield similar results as in Figure 2, thus we omit their plots. + +Mode collapsing and mode dropping. In realistic settings, $\mathbb { P } _ { r }$ is usually very diverse since natural images are inherently multimodal. Many have conjectured that $\mathbb { P } _ { g }$ differs from $\mathbb { P } _ { r }$ by reducing diversity, possibly due to the lack of model capacity or inadequate optimization (Arora et al., 2017). This is often manifested itself for generative models in a mix of two ways: mode dropping, where some hard-to-represent modes of $\mathbb { P } _ { r }$ are simply “ignored" by $\mathbb { P } _ { g }$ ; and mode collapsing, where several modes of $\mathbb { P } _ { r }$ are “averaged" by $\mathbb { P } _ { g }$ into a single mode, possibly located at a midpoint. An ideal metric should be sensitive to these two phenomena. + +To test for mode collapsing, we first randomly sample both $S _ { r }$ and $S _ { r } ^ { \prime }$ as two disjoint sets of 2000 real images. Next, we find 50 clusters in the whole training set with $k$ -means and progressively replace each cluster by its respective cluster center to simulate mode collapse. Figure 3 shows computed values of $\hat { \rho } ( S _ { r } , S _ { r } ^ { \prime } )$ as the number of replaced (collapsed) clusters, denoted as $C$ increases. Ideally, we expect the scores increase as $C$ grows. We first observe that all the metrics are able to respond correctly when distances are computed in the convolutional feature space. However, the Wasserstein metric (cyan curve) breaks down in pixel space, as it considers a collapsed sample set (with $C > 0$ ) being closer to the real sample set than another set of real images (with $C = 0$ ). Moreover, although the overall 1-NN accuracy (solid magenta curve) follows the desired trend, the real and fake parts follow opposite trends: 1-NN real accuracy (dotted magenta curve) decreases while 1-NN fake accuracy (dashed magenta curve) increases. Again, this is inline with our explanation given in the last paragraph of Section 2.2. + +To test for mode dropping, we take $S _ { r }$ as above and construct $S _ { r } ^ { \prime }$ by randomly removing clusters. To keep the size of $S _ { r } ^ { \prime }$ constant, we replace images from the removed cluster with images randomly selected from the remaining clusters. Figure 4 shows how different metrics react to the number of removed clusters, also denoted as $C$ . All scores effectively discriminate against mode dropping except the RIS and RMS - they remain almost indifferent when some modes are dropped. Again, this is perhaps caused by the fact that the Inception/Mode Score were originally designed for datasets with classes overlapping with the ImageNet dataset, and they do not generalize well to other datasets. + +![](images/27afa6c9ff1eb40623809e4d625a094bd3d23b7636c473490f6547ec4db34c1c.jpg) +Figure 4: Experiment on simulated mode dropping. A metric score should increase to reflect the mismatch between true distribution and generated distribution as more modes are dropped. All metrics except RIS and RMS respond correctly, as they only increase slightly in value even when almost all modes are dropped. + +![](images/a3f3cd6ed8e7a79b3ee018f9f4c752f97c498ad8d7aed192a7d48c76b1718aa3.jpg) +Figure 5: Experiment on robustness of each metric to small transformations (rotations and translations). All metrics should remain constant across all mixes of real and transformed real samples, since the transformations do not alter semantics of the image. All metrics respond correctly in convolutional space, but behave incorrectly in pixel space. This experiment illustrates the unsuitability of distances in pixel space. + +# 3.4 ROBUSTNESS TO TRANSFORMATIONS + +GANs are widely used for image datasets, which have the property that certain transformations to the input do not change its semantic meaning. Thus an ideal evaluation metric should be invariant to such transformations to some extent. For example, a generator trained on CelebA should not be penalized by a metric if its generated faces are shifted by a few pixels or rotated by a small angle. + +Figure 5 shows how the various metrics react to such small transformation to the images. In this experiment, $S _ { r }$ and $S _ { r } ^ { \prime }$ are two disjoint sets of 2000 real images sampled from the training data. However, a proportion of images from $S _ { r } ^ { \prime }$ are randomly shifted (up to 4 pixels) or rotated (up to 15 degrees). We can observe from the results that metrics operating in the convolutional space (or softmax space for RIS and RMS) are robust to these transformations, as all the curves are approximated horizontal as the ratio of transformed samples increases. This is not that surprising as convolutional networks are well know for being invariant to certain transformations (Mallat, 2016). In comparison, in the pixel space all the metrics consider the shifted/rotated images as drawn from a different distribution, highlighting the importance of computing distances in a proper feature space. + +![](images/3477b7f2e2783fbb5fa4506505a1f49b1b75ebf97b2bb2dd6d880f8cc60b969f.jpg) +Figure 7: Measurement of wall-clock time for computing various metrics as a function of the number of samples. All metrics are practical to compute for a sample of size 2000, but Wasserstein distance does not scale to large sample sizes. + +![](images/2fcde7509c9dec9a43fcff36da5c2ef1c139c751b252031a98486f573d1f95c4.jpg) +Figure 6: The score of various metrics as a function of the number of samples. An ideal metric should result in a large gap between the real-real (R-R; $\hat { \rho } ( S _ { r } , S _ { r } ^ { \prime } ) )$ ) and real-fake (R-G; $\hat { \rho } ( \bar { S } _ { r } , S _ { g } ) \rangle$ ) curves in order to distinguish between real and fake distributions using as few samples as possible. Compared with Wasserstein distance, MMD and 1-NN accuracy require much fewer samples to discriminate real and generated images, while RIS totally fails on LSUN as it scores generated images even better (lower) than real images. + +# 3.5 EFFICIENCY + +A practical GAN evaluation metric should be able to compute “accurate” scores from a reasonable number of samples and within an affordable computation cost, such that it can be computed, for example, after each training epoch to monitor the training process. + +Sample efficiency. Here we measure the sample efficiency of various metrics by investigating how many samples are needed for each of them in order to discriminate a set of generated samples $S _ { g }$ (from DCGAN) from a set of real samples $S _ { r } ^ { \prime }$ . To do this, we introduce a reference set $S _ { r }$ , which is also uniformly sampled from the real training data, but is disjoint with $S _ { r } ^ { \prime }$ . All three sample sets have the same size, i.e., $| S _ { r } | = | S _ { r } ^ { \prime } | = | S _ { g } | = n$ . We expect that an ideal metric $\rho$ should correctly score $\hat { \rho } ( S _ { r } , S _ { r } ^ { \prime } )$ lower than $\hat { \rho } ( S _ { r } , S _ { g } )$ with a relatively small $n$ . In other words, the number of samples $n$ needed for the metric to distinguish $S _ { r } ^ { \prime }$ and $S _ { g }$ can be viewed as its sample complexity. + +In Figure 6 we show the individual scores as a function of $n$ . We can observe that MMD, FID and 1-NN accuracy computed in convolution feature space are able to distinguish the two set of images $S _ { g }$ (solid blue and magenta curves) and $S _ { r } ^ { \prime }$ (dotted solid blue and magenta curves) with relatively few samples. The Wasserstein distance (cyan curves) is not discriminative with samples size less than 1000, while the RIS even considers the generated samples to be more “real” than the real samples on the LSUN dataset (the red curves in the third panel). The dotted lines in Figure 6 also quantify how fast the scores converge to their expectations as we increase the sample size. Note that MMD for $\hat { \rho } ( S _ { r } , S _ { r } ^ { \prime } )$ converges very quickly to zero and gives discriminative scores with few samples, making it a practical metric for comparing GAN models. + +Computational efficiency. Fast computation of the empirical metric is of practical concern as it helps researchers monitor the training process and diagnose problems early on, or perform early stopping. In Figure 7 we investigate the computational efficiency of the above metrics by showing the wall-clock time (in log scale) to compute them as a function of the number of samples. For a typical number of 2000 samples, it only takes about 8 seconds to compute each of these metrics on an NVIDIA TitanX GPU. In fact, the majority of time is spent on extracting features from the ResNet model. Only the Wasserstein distance becomes prohibitively slow for large sample sizes. + +# 3.6 DETECTING OVERFITTING + +Overfitting is an artifact of training with finite samples. If a GAN successfully memorizes the training images, i.e., $\mathbb { P } _ { g }$ is a uniform distribution over the training sample set $S _ { r } ^ { t r }$ , then the generated samples $S _ { g }$ becomes a uniformly drawn set of $n$ samples from $S _ { r } ^ { t r }$ , and any reasonable $\hat { \rho }$ should be close to 0. The Wasserstein distance, MMD and 1-NN two sample test are able to detect overfitting in the following sense: if we hold out a validation set $S _ { r } ^ { v a l }$ , then $\hat { \rho } ( S _ { g } , S _ { r } ^ { v a l } )$ should be significantly higher than $\hat { \rho } ( S _ { g } , S _ { r } ^ { t r } )$ when $\mathbb { P } _ { g }$ memorizes a part of $S _ { r } ^ { t r }$ . The difference between them can informally be viewed as a form of “generalization gap". + +We simulate the overfitting process by defining $S _ { r } ^ { \prime }$ as a mix of samples from the training set $S _ { r } ^ { t r }$ and a second holdout set, disjoint from both $S _ { r } ^ { t r }$ and $S _ { r } ^ { v a l }$ . Figure 8 shows the gap $\hat { \rho } ( S _ { g } , S _ { r } ^ { v a l } ) - \hat { \rho } ( \mathbf { \dot { } { } } S _ { g } , S _ { r } ^ { t r } )$ of the various metrics as a function of the overlapping ratio between $S _ { r } ^ { \prime }$ and $S _ { r } ^ { t r }$ . The left most point of each curve can be viewed as the score $\hat { \rho } ( S _ { r } ^ { \prime } , S _ { r } ^ { v a l } )$ computed on a validation set since the overlap ratio is 0. For better visualization, we normalize the Wasserstein distance and MMD by dividing their corresponding score when $S _ { r } ^ { \prime }$ and $S _ { r }$ have no overlap. As shown in Figure 8, all the metrics except RIS and RMS reflect that the “generalization gap" increases as $S _ { r } ^ { \prime }$ overfits more to $S _ { r }$ . The failure of RIS is not surprising: it totally ignores the real data distribution as we discussed in Section 2.2. While the reason that RMS also fails to detect overfitting may again be its lack of generalization to datasets with classes not contained in the ImageNet dataset. In addition, RMS operates in the softmax space, the features in which might be too specific compared to the features in the convolutional space. + +![](images/a7fbd0dd8d052615064f08210b23e53041fd3c73b5d0a1f9762bc4f7a90af568.jpg) +Figure 8: Experiment on detecting overfitting of generated samples. As more generated samples overlap with real samples from the training set, the gap between validation and training score should increase to signal overfitting. All metrics behave correctly except for RIS and RMS, as these two metrics do not increase when the fraction of overlapping samples increases. + +# 4 DISCUSSIONS AND CONCLUSION + +Based on the above analysis, we can summarize the advantages and inherent limitations of the five evaluation metrics, and conditions under which they produce meaningful results. With some of the metrics, we are able to study the problem of overfitting (see Appendix C), perform model selection on GAN models and compare GAN models without resorting to human evaluation based on cherry-picked samples (see Appendix D). + +The Inception Score does show a reasonable correlation with the quality and diversity of generated images, which explains the wide usage in practice. However, it is ill-posed mostly because it only evaluates $\mathbb { P } _ { g }$ as an image generation model rather than its similarity to $\mathbb { P } _ { r }$ . Blunt violations like mixing in natural images from an entirely different distribution completely deceives the Inception Score. As a result, it may encourage the models to simply learn sharp and diversified images (or even some adversarial noise), instead of $\mathbb { P } _ { r }$ . This also applies to the Mode Score. Moreover, the Inception Score is unable to detect overfitting since it cannot make use of a holdout validation set. + +Kernel MMD works surprising well when it operates in the feature space of a pre-trained ResNet. It is always able to identify generative/noise images from real images, and both its sample complexity and computational complexity are low. Given these advantages, even though MMD is biased, we recommend its use in practice. + +Wasserstein distance works well when the base distance is computed in a suitable feature space. However, it has a high sample complexity, a fact that has been independently observed by (Arora et al., 2017). Another key weakness is that computing the exact Wasserstein distance has a time complexity of $O ( n ^ { 3 } )$ , which is prohibitively expensive as sample size increases. Compared to other methods, Wasserstein distance is less appealing as a practical evaluation metric. + +Fréchet Inception Distance performs well in terms of discriminability, robustness and efficiency. It appears to be a good evaluation metric for GANs, even it only takes into consideration the first two order moments of the distributions. + +1-NN classifier seems to be an ideal metric for evaluating GANs. Not only does it enjoy all the advantages of the other metrics, it also outputs a score in the interval [0, 1], similar to the accuracy/error in classification problems. When the generative distribution perfectly match the true distribution, perfect score (i.e., $5 0 \%$ accuracy) is attainable with a reasonable (e.g., 1000) sample size. From Figure 2, we find that typical GAN models tend to achieve lower LOO accuracy for real samples (1-NN accuracy (real)), while higher LOO accuracy for generated samples (1-NN accuracy (fake)). This suggests that GANs are able to capture modes from the training distribution, such that the majority of training samples distributed around the mode centers have their nearest neighbor from the generated images, yet most of the generated images are still surrounded by generated images as they are collapsed. The observation indicates that the mode collapse problem is prevalent for typical GAN models. We also note that this problem, however, cannot be effectively detected by human evaluation or the widely used Inception Score. + +Overall, our empirical study suggests that the choice of feature space in which to compute various metrics is crucial. In the convolutional space of a ResNet pretrained on ImageNet, both MMD and 1-NN accuracy appear to be good metrics in terms of discriminability, robustness and efficiency. Wasserstein distance has very poor sample efficiency, while Inception Score and Mode Score appear to be unsuitable for datasets that are very different from ImageNet. We will release our source code for all these metrics, providing researchers with an off-the-shelf tool to compare and improve GAN algorithms. + +Based on the two most prominent metrics, MMD and 1-NN accuracy, we study the overfitting problem of DCGAN and WGAN (in Appendix C). Despite the widespread belief that GANs are overfitting to the training data, we find that this does not occur unless there are very few training samples. This raises an interesting question regarding the generalization of GANs in comparison to the supervised setting. We hope that future work can contribute to explaining this phenomenon. + +# REFERENCES + +Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein GAN. arXiv preprint arXiv:1701.07875, 2017. + +Sanjeev Arora, Rong Ge, Yingyu Liang, Tengyu Ma, and Yi Zhang. Generalization and equilibrium in generative adversarial nets (gans). arXiv preprint arXiv:1703.00573, 2017. + +Wacha Bounliphone, Eugene Belilovsky, Matthew B Blaschko, Ioannis Antonoglou, and Arthur Gretton. A test of relative similarity for model selection in generative models. arXiv preprint arXiv:1511.04581, 2015. + +Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative adversarial networks. arXiv preprint arXiv:1612.02136, 2016. + +Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255. IEEE, 2009. + +David Forsyth and Jean Ponce. Computer vision: a modern approach. Upper Saddle River, NJ; London: Prentice Hall, 2011. + +Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014. + +Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, Alexander J Smola, et al. A kernel method for the two-sample-problem. Advances in neural information processing systems, 19:513, 2007. + +Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. Improved training of wasserstein gans. arXiv preprint arXiv:1704.00028, 2017. + +Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Günter Klambauer, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a nash equilibrium. arXiv preprint arXiv:1706.08500, 2017. + +Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. arXiv preprint arXiv:1611.07004, 2016. + +Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. In Tech Report, 2009. + +Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, and Ole Winther. Autoencoding beyond pixels using a learned similarity metric. arXiv preprint arXiv:1512.09300, 2015. + +David Lopez-Paz and Maxime Oquab. Revisiting classifier two-sample tests. arXiv preprint arXiv:1610.06545, 2016. + +Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian Goodfellow, and Brendan Frey. Adversarial autoencoders. arXiv preprint arXiv:1511.05644, 2015. + +Stéphane Mallat. Understanding deep convolutional networks. Phil. Trans. R. Soc. A, 374(2065): 20150203, 2016. + +Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least squares generative adversarial networks. arXiv preprint ArXiv:1611.04076, 2016. + +Augustus Odena. Semi-supervised learning with generative adversarial networks. arXiv preprint arXiv:1606.01583, 2016. + +Guo-Jun Qi. Loss-sensitive generative adversarial networks on lipschitz densities. arXiv preprint arXiv:1701.06264, 2017. + +Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. + +Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training GANs. In Advances in Neural Information Processing Systems, pp. 2226–2234, 2016. + +Ashish Shrivastava, Tomas Pfister, Oncel Tuzel, Josh Susskind, Wenda Wang, and Russ Webb. Learning from simulated and unsupervised images through adversarial training. arXiv preprint arXiv:1612.07828, 2016. + +Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott E. Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. arXiv preprint arXiv:1409.4842, 2014. + +Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2818–2826, 2016. + +Lucas Theis, Aäron van den Oord, and Matthias Bethge. A note on the evaluation of generative models. arXiv preprint arXiv:1511.01844, 2015. + +Paul Upchurch, Jacob R. Gardner, Kavita Bala, Robert Pless, Noah Snavely, and Kilian Q. Weinberger. Deep feature interpolation for image content changes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. in press ..., 2017. + +Yuhuai Wu, Yuri Burda, Ruslan Salakhutdinov, and Roger Grosse. On the quantitative analysis of decoder-based generative models. arXiv preprint arXiv:1611.04273, 2016. + +Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016. + +Junbo Jake Zhao, Michaël Mathieu, and Yann LeCun. Energy-based generative adversarial network. arXiv preprint arXiv:1609.03126, 2016. + +Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. arXiv preprint arXiv:1703.10593, 2017. + +# A GAN VARIANTS USED IN OUR EXPERIMENTS + +Many GAN variants have been proposed recently. In this paper we consider several of them, which we briefly review in this section. + +DCGAN (Radford et al., 2015). The generator of a DCGAN takes a lower dimensional input from a uniform noise distribution, then projects and reshapes it to a small convolutional representation with many feature maps. After applying a series of four fractionally-strided convolutions, the generator converts this representation into a $6 4 \times 6 4$ pixel image. DCGAN is optimized by minimizing the Jensen-Shannon divergence between the real and generated images. + +WGAN (Arjovsky et al., 2017). A critic network that outputs unconstrained real values is used in place of the discriminator. When the critic is Lipschitz, this network approximates the Wasserstein distance between $S _ { r }$ and $S _ { g }$ . A Lipschitz condition is enforced by clipping the critic networks’ parameters to stay within a predefined bounding box. + +WGAN with gradient penalty (Gulrajani et al., 2017) improves upon WGAN by enforcing the Lipschitz condition with a gradient penalty term. This method significantly improves the convergence speed and the quality of the images generated by a WGAN. + +LSGAN (Mao et al., 2016). Least Squares GAN adopts the least squares loss function instead of the commonly used sigmoid cross entropy loss for the discriminator, essentially minimizing the Pearson $\chi ^ { 2 }$ divergence between the real distribution $\mathbb { P } _ { r }$ and generative distribution $\mathbb { P } _ { g }$ . + +# B THE CHOICE OF FEATURE SPACE + +The choice of features space is crucial for all these metrics. Here we consider several alternatives to the convolutional features from the 34-layer ResNet trained on ImageNet. In particular, we compute various metrics using the features extracted by (1) the VGG and Inception networks; (2) a 34-layer ResNet with random weights; (3) a ResNet classifier trained on the same dataset as the GAN models. We use the features extracted from these models to test all metrics in the discriminative experiments we performed in Section 3. All experimental settings are identical except for the third experiments, which is performed on the CIFAR-10 dataset (Krizhevsky & Hinton, 2009) instead as we need class labels to train the classifier. Note that we consider setting (3) only for analytical purposes. It is not a practical choice as GANs are mainly designed for unsupervised learning and we should not assume the existence of ground truth labels. + +![](images/eb751a4b07b47aa017a1aea8245697f431d986c743349d44ad91f1efadce7805.jpg) +Figure 9: Comparison of all metrics in different feature spaces. When using different trained networks, the trends of all metrics are very similar. Most metrics work well even in a random network, but Wasserstein distance has very high variance and the magnitude of increase for 1-NN accuracy is small. + +The results are shown in Figure 9 and Figure 10, from which several observations can be made: (1) switching from ResNet-34 to VGG or Inception has little effect to the metric scores; (2) the features from a random network still works for MMD, while it makes the Wasserstein distance unstable and 1-NN accuracy less discriminative. Not surprisingly, the Inception Score and Mode Score becomes meaningless if we use the softmax values from the random network; (3) features extracted from the classifier trained on the same dataset as the GAN model also offers high discriminability for these metrics, especially for the Wasserstein distance. However, this may be simply due to the fact that the feature dimensionality of the ResNet trained on CIFAR-10 is much smaller than that of the ResNet-34 trained on ImageNet (64 v.s. 512). + +![](images/2aec4fabff4f4efa259561ca43ac79e2d4419b14e514a7fa7e0dfdcdc789922b.jpg) +Figure 10: Using features extracted from a ResNet trained on CIFAR-10 (right plot) to evaluate a GAN model trained on the same dataset. Compared to using an extractor trained on ImageNet, the metrics appear to have lower variance. However, this may due to the feature dimensionality being smaller for CIFAR-10. + +![](images/2272bebca854ed5b1c95d0a22f8cf047e84ae9c0cb02aceed35de9a3d7a44221.jpg) +Figure 11: Training curves of DCGAN and WGAN on a large (left two panels), small (middle two panels) and tiny (right two panels) subsets of CelebA. Note that for the first four plots, blue (yellow) curves almost overlap with the red (green) curves, indicating no overfitting detected by the two metrics. Overfitting only observed on the tiny training set, with MMD score and 1-NN accuracy significantly worse (higher) on the validation set. + +# C ARE GANS OVERFITTING TO THE TRAINING DATA? + +We trained two representative GAN models, DCGAN (Radford et al., 2015) and WGAN (Arjovsky et al., 2017) on the CelebA dataset. Out of the ${ \sim } 2 0 0 { , } 0 0 0$ images in total, we holdout 20,000 images for validation, and the rest for training. As the training set is sufficiently large, which makes overfitting unlikely to occur, we also create a small training set and a tiny training set respectively with only 2000 and 10 images sampled from the full training set. + +The training setting for DCGAN and WGAN strictly follow their original implementation, except that we change the default number of training iterations such that both models are sufficiently updated. For each metric, we compute their score on 2000 real samples and 2000 generated samples, where the real samples are drawn from either the training set or the validation set, giving rise to training and validation scores. The results are shown in Figure 11, from which we can make several observations: + +• The training and validation scores almost overlap with each other with 2000 or $1 8 0 \mathrm { k }$ training samples, showing that both DCGAN and WGAN do not overfit to the training data under of these metrics. Even when using only 2000 training samples, there is still no significant difference between the training score and validation score. This shows that the training process of GANs behaves quite differently from those of supervised deep learning models, where a model can easily achieve 0 training error while behaving like random guess on the validation set (Zhang et al., 2016). 5 + +Table 1: Comparison of several GAN models on the LSUN dataset + +
RealDCGANWGANWGAN-GPLSGAN
Conv SpaceMMD0.0190.2050.2700.1940.232
1-NN Accuracy0.4990.8250.9200.8120.871
1-NN Accuracy (real)0.4950.7590.8800.7650.804
1-NN Accuracy (fake)0.5030.8920.9610.8600.938
+ +• DCGAN outperforms WGAN on the full training set under both metrics, and converges faster. However, WGAN is much more stable on the small training set, and converges to better positions. + +# D COMPARISON OF POPULAR GAN MODELS BASED ON QUANTITATIVE EVALUATION METRICS + +Based on our analysis, we chose MMD and 1-NN accuracy in the feature space of a 34-layer ResNet trained on ImageNet to compare several state-of-the-art GAN models. All scores are computed using 2000 samples from the holdout set and 2000 generated samples. The GAN models evaluated include DCGAN (Radford et al., 2015), WGAN (Arjovsky et al., 2017), WGAN with gradient penalty (WGAN-GP ) (Gulrajani et al., 2017), and LSGAN (Mao et al., 2016) , all trained on the CelebA dataset. The results are reported in Table 1, from which we highlight three observations: + +• WGAN-GP performs the best under most of the metrics. +DCGAN achieves 0.759 overall 1-NN accuracy on real samples, slightly better than 0.765 achieved by WGAN-GP; while the 1-NN accuracy on generated (fake) samples achieved by DCGAN is higher than that by WGAN-GP (0.892 v.s. 0.860). This seems to suggest that DCGAN is better at capturing modes in the training data distribution, while its generated samples are more collapsed compared to WGAN-GP. Such subtle difference is unlikely to be discovered by the Inception Score or human evaluation. +• The 1-NN accuracy for all evaluated GAN models are higher than 0.8 , far above the ground truth of 0.5. The MMD score of the four GAN models are also much larger than that of ground truth (0.019). This indicates that even state-of-the-art GAN models are far from learning the true distribution. \ No newline at end of file diff --git a/parse/train/Sy1f0e-R-/Sy1f0e-R-_content_list.json b/parse/train/Sy1f0e-R-/Sy1f0e-R-_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..1c652df7947486beba5e6cfacdbf6fe7cfa6b17b --- /dev/null +++ b/parse/train/Sy1f0e-R-/Sy1f0e-R-_content_list.json @@ -0,0 +1,1702 @@ +[ + { + "type": "text", + "text": "AN EMPIRICAL STUDY ON EVALUATION METRICS OF GENERATIVE ADVERSARIAL NETWORKS ", + "text_level": 1, + "bbox": [ + 176, + 101, + 823, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous authors Paper under double-blind review ", + "bbox": [ + 183, + 170, + 398, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 234, + 544, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Despite the widespread interest in generative adversarial networks (GANs), few works have studied the metrics that quantitatively evaluate GANs’ performance. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the important problem of how to evaluate the evaluation metrics. We start with a few necessary conditions for metrics to produce meaningful scores, such as distinguishing real from generated samples, identifying mode dropping and mode collapsing, and detecting overfitting. Then with a series of carefully designed experiments, we are able to comprehensively investigate existing sample-based metrics and identify their strengths and limitations in practical settings. Based on these results, we observe that kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbour (1-NN) two-sample test seem to satisfy most of the desirable properties, provided that the distances between samples are computed in a suitable feature space. Our experiments also unveil interesting properties about the behavior of several popular GAN models, such as whether they are memorizing training samples, and how far these state-of-the-art GANs are from perfect. ", + "bbox": [ + 233, + 268, + 766, + 476 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 507, + 336, + 522 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Generative adversarial networks (GANs) (Goodfellow et al., 2014) have been studied extensively in recent years. Besides producing surprisingly plausible images of faces (Radford et al., 2015; Larsen et al., 2015) and bedrooms (Radford et al., 2015; Arjovsky et al., 2017; Gulrajani et al., 2017), they have also been innovatively applied in, for example, semi-supervised learning (Odena, 2016; Makhzani et al., 2015), image-to-image translation (Isola et al., 2016; Zhu et al., 2017), and simulated image refinement (Shrivastava et al., 2016). However, despite the availability of a plethora of GAN models (Arjovsky et al., 2017; Qi, 2017; Radford et al., 2015; Zhao et al., 2016), their evaluation is still predominantly qualitative, very often resorting to manual inspection of the visual fidelity of generated images. Such evaluation is time-consuming, subjective and possibly misleading. Given the inherent limitations of qualitative evaluations, proper quantitative metrics are crucial for the development of GANs to avoid the human factors and guide the design of better models. ", + "bbox": [ + 174, + 534, + 825, + 686 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Possibly the most popular metric is the Inception Score (Salimans et al., 2016), which measures the quality and diversity of the generated images using an external model, the Google Inception network (Szegedy et al., 2014), trained on the large scale ImageNet dataset (Deng et al., 2009). Some other metrics are less widely used but still very valuable. Wu et al. (2016) proposed a sampling method to estimate the log-likelihood of generative models, by assuming a Gaussian observation model with a fixed variance. Bounliphone et al. (2015) propose to use maximum mean discrepancies (MMDs) for model selection in generative models. Lopez-Paz & Oquab (2016) apply the classifier two-sample test, a well-studied tool in statistics, to assess the difference between the generated and target distribution. ", + "bbox": [ + 174, + 694, + 825, + 819 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Although these evaluation metrics are shown to be effective on various tasks, it is unclear in which scenarios their scores are meaningful, and in which other scenarios, prone to misinterpretations. Given that evaluating GANs is already challenging, it can only be more difficult to evaluate the evaluation metrics themselves. Most existing works attempt to justify their proposed metrics by showing a strong correlation with human evaluation (Salimans et al., 2016; Lopez-Paz & Oquab, 2016). However, human evaluation tends to be biased towards the visual quality of generated samples and neglect the overall distributional characteristics, which are arguably just as important for unsupervised learning. ", + "bbox": [ + 174, + 827, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/49450f89442e39d6bd90bdd00539f5bc95c21f16ec6bfdf0c04ded989a0111e0.jpg", + "image_caption": [ + "Figure 1: A schematic layout of the typical approach for sample based GAN evaluation methods. " + ], + "image_footnote": [], + "bbox": [ + 230, + 102, + 772, + 191 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this paper we comprehensively examine the existing literature on sample-based quantitative evaluation of GANs. We address the challenge of evaluating the metrics themselves by carefully designing a series of experiments, through which we hope to answer the following important questions: 1.) What are reasonable characterizations of the behavior of existing sample-based metrics for GANs? 2.) What are the strengths and limitations of these metrics? 3.) Which metrics are preferred accordingly? 4.) How are the metrics helpful in understanding and improving GANs? ", + "bbox": [ + 173, + 229, + 825, + 314 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Ultimately, we hope that this paper will establish good principles on choosing, applying, interpreting and designing evaluation metrics for GANs in practical settings. We will also release the source code for all experiments and metrics examined, providing the community with off-the-shelf tools to debug and improve their GAN algorithms. ", + "bbox": [ + 174, + 320, + 825, + 377 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 BACKGROUND ", + "text_level": 1, + "bbox": [ + 174, + 391, + 326, + 407 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We briefly review the original GAN framework proposed by Goodfellow et al. (2014). Description of the GAN variants used in our experiments is deferred to the Appendix A. ", + "bbox": [ + 174, + 416, + 823, + 445 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1 GENERATIVE ADVERSARIAL NETWORKS ", + "text_level": 1, + "bbox": [ + 174, + 455, + 493, + 469 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Let $\\mathcal { X } = \\mathbb { R } ^ { d \\times d }$ be the space of natural images. Given i.i.d. samples $S _ { r } = \\{ \\mathbf { x } _ { 1 } ^ { r } , \\ldots , \\mathbf { x } _ { n } ^ { r } \\}$ drawn from a real distribution $\\mathbb { P } _ { r }$ over $\\mathcal { X }$ , we would like to learn a parameterized distribution $\\mathbb { P } _ { g }$ that approximates the distribution $\\mathbb { P } _ { r }$ . ", + "bbox": [ + 174, + 476, + 825, + 518 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The setup of generative adversarial networks is as follows. We define two networks, the discriminator $D : \\mathcal { X } [ 0 , 1 )$ and the generator $G : { \\mathcal { Z } } \\to { \\mathcal { X } }$ , where $\\mathcal { Z }$ is some latent space. Given a distribution $\\mathbb { P } _ { z }$ over $\\mathcal { Z }$ (usually an isotropic Gaussian), the distribution $\\mathbb { P } _ { g }$ is defined as $G ( \\mathbb { P } _ { z } )$ . Optimization is performed with respect to a joint loss for $D$ and $G$ ", + "bbox": [ + 173, + 525, + 825, + 582 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/b2535cc6daf8bf45cb707b25225a313cbe9c6bd2d1e1ec85d1ad919727e0664c.jpg", + "text": "$$\n\\operatorname* { m i n } _ { G } \\operatorname* { m a x } _ { D } L ( D , G ) = \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { r } } \\log { [ D ( \\mathbf { x } ) ] } + \\mathbb { E } _ { \\mathbf { z } \\sim \\mathbb { P } _ { z } } \\left[ \\log ( 1 - D ( G ( \\mathbf { z } ) ) ) \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 266, + 585, + 730, + 608 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Intuitively, the discriminator $D$ outputs a probability for every $\\mathbf { x } \\in \\mathcal { X }$ that corresponds to its likelihood of being drawn from $\\mathbb { P } _ { r }$ , and the loss function encourages the generator $G$ to produce samples that maximize this probability. Practically, the loss is approximated with finite samples from $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ , and optimized with alternating steps for $D$ and $G$ using gradient descent. ", + "bbox": [ + 174, + 613, + 826, + 669 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To evaluate the generator, we would like to design a metric $\\rho$ that measures the “dissimilarity\" between $\\mathbb { P } _ { g }$ to $\\mathbb { P } _ { r }$ .1 In theory, with both distributions known, common choices of $\\rho$ include the Kullback-Leibler divergence (KLD), Jensen-Shannon divergence (JSD) and total variation. However, in practical scenarios, $\\mathbb { P } _ { r }$ is unknown and only the finite samples in $S _ { r }$ are observed. Furthermore, it is almost always intractable to compute the exact density of $S _ { g } = \\{ \\mathbf { x } _ { 1 } ^ { g } , . . . , \\mathbf { x } _ { m } ^ { \\bar { g } } \\} \\sim \\mathbb { P } _ { q } ^ { m }$ specially so for GANs). Given these limitations, we focus on empiricalof “dissimilarity\" between samples from two distributions. $\\mathbb { P } _ { g }$ , but much easier to sample $\\hat { \\rho } : \\mathcal X ^ { n } \\times \\mathcal X ^ { m } \\stackrel { \\smile } { } \\mathbb R$ ", + "bbox": [ + 173, + 675, + 825, + 773 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.2 SAMPLE BASED METRICS ", + "text_level": 1, + "bbox": [ + 174, + 785, + 390, + 797 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We mainly focus on sample based evaluation metrics that follow a common setup illustrated in Figure 1. The metric calculator is the key element, for which we briefly introduce five representative methods: Inception Score (Salimans et al., 2016), Mode Score (Che et al., 2016) , Kernel MMD (Gretton et al., 2007), Wasserstein distance, Fréchet Inception Distance (FID) (Heusel et al., 2017), and 1-nearest neighbor (1-NN)-based two sample test (Lopez-Paz & Oquab, 2016). All of them are model agnostic and require only finite samples from the generator. ", + "bbox": [ + 174, + 804, + 825, + 888 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The Inception Score is arguably the most widely adopted metric in the literature. It uses a image classification model $\\mathcal { M }$ , the Google Inception network (Szegedy et al., 2016), pre-trained on the ImageNet (Deng et al., 2009) dataset, to compute ", + "bbox": [ + 173, + 103, + 823, + 146 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/4ec726016605aea0fafe96847d73ac263947992e772b07fd92d445d3821860a7.jpg", + "text": "$$\n\\mathrm { I S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] } ,\n$$", + "text_format": "latex", + "bbox": [ + 372, + 151, + 624, + 171 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $p _ { \\mathcal { M } } ( y | \\mathbf { x } )$ denotes the label distribution of $\\mathbf { x }$ as predicted by $\\mathcal { M }$ , and $\\begin{array} { r } { p _ { \\mathcal M } ( y ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { g } } \\end{array}$ , i.e. the marginal of $p _ { \\mathcal { M } } ( y | \\mathbf { x } )$ over the probability measure $\\mathbb { P } _ { g }$ . The expectation and the integral in $p _ { \\mathcal { M } } ( y | \\mathbf { x } )$ can be approximated with i.i.d. samples from $\\mathbb { P } _ { g }$ . A higher IS has $p _ { \\mathcal { M } } ( y | \\mathbf { x } )$ close to a point mass, which happens when the Inception network is very confident that the image belongs to a particular ImageNet category, and has $p _ { \\mathcal { M } } ( y )$ close to uniform, i.e. all categories are equally represented. This suggests that the generative model has both high quality and diversity. Salimans et al. (2016) show that the Inception Score has a reasonable correlation with human judgment of image quality. We would like to highlight two specific properties: 1) the distributions on both sides of the KL are dependent on $\\mathcal { M }$ , and 2) the distribution of the real data $\\mathbb { P } _ { r }$ , or even samples thereof, are not used anywhere. ", + "bbox": [ + 173, + 175, + 826, + 318 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The Mode Score is an improved version of the Inception Score. Formally, it is given by ", + "bbox": [ + 171, + 323, + 750, + 339 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/3a9fd389181ebcbf46f515563540c7cf132b0297f5785f5eac2de30fbe71678b.jpg", + "text": "$$\n\\mathrm { M S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] - K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( y ^ { \\ast } ) ) } ,\n$$", + "text_format": "latex", + "bbox": [ + 302, + 344, + 692, + 364 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { p _ { \\mathcal M } ( y ^ { \\ast } ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { r } } \\end{array}$ is the marginal label distribution for the samples from the real data distribution. Unlike the Inception Score, it is able to measure the dissimilarity between the real distribution $\\mathbb { P } _ { r }$ and generated distribution $\\mathbb { P } _ { g }$ through the term $K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( \\bar { y } ^ { * } ) )$ . ", + "bbox": [ + 174, + 368, + 825, + 412 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The Kernel MMD (Maximum Mean Discrepancy), defined as ", + "bbox": [ + 173, + 417, + 584, + 434 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/f3788bd082c1f55a957a0105ce3effdad449912b13c637778f4d639a70f7588e.jpg", + "text": "$$\n\\begin{array} { r } { \\mathrm { M M D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\biggl ( \\mathbb { E } _ { \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } \\sim \\mathbb { P } _ { r } , } \\biggl [ k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } ) - 2 k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { g } ) + k ( \\mathbf { x } _ { g } , \\mathbf { x } _ { g } ^ { \\prime } ) \\biggr ] \\biggr ) ^ { \\frac { 1 } { 2 } } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 251, + 439, + 745, + 488 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "measures the dissimilarity between $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ for some fixed kernel function $k$ . Given two sets of samples from $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ , the empirical MMD between the two distributions can be computed with finite sample approximation of the expectation. A lower MMD means that $\\mathbb { P } _ { g }$ is closer to $\\mathbb { P } _ { r }$ . The Parzen window estimate (Gretton et al., 2007) can be viewed as a specialization of Kernel MMD. ", + "bbox": [ + 173, + 492, + 825, + 549 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The Wasserstein distance between $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ is defined as ", + "bbox": [ + 174, + 555, + 565, + 570 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/9d816f4d0c7692fe62f46391ab21fad24983fbd57ad4d0b9d96457eff4b54a80.jpg", + "text": "$$\n\\operatorname { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\operatorname* { i n f } _ { \\substack { \\gamma \\in \\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) } } \\mathbb { E } _ { ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\sim \\gamma } \\left[ d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 333, + 577, + 665, + 602 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )$ denotes the set of all joint distributions (i.e. probabilistic couplings) whose marginals are respectively $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ , and $d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } )$ denotes the base distance between the two samples. For discrete distributions with densities $p _ { r }$ and $p _ { g }$ , the Wasserstein distance is often referred to as the Earth Mover’s Distance (EMD), and corresponds to the solution to the optimal transport problem ", + "bbox": [ + 174, + 608, + 825, + 665 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/fae6004402fd30eadb74ce6936dd7f6f8715a2b25aee2749b8d04e6103c21bdf.jpg", + "text": "$$\nN \\mathrm { D } ( p _ { r } , p _ { g } ) = \\operatorname* { m i n } _ { w \\in \\mathbb { R } ^ { n \\times m } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { m } w _ { i j } d ( \\mathbf { x } _ { i } ^ { r } , \\mathbf { x } _ { j } ^ { g } ) \\quad \\mathrm { s . t . } \\quad \\sum _ { j = 1 } ^ { m } w _ { i , j } = p _ { r } ( \\mathbf { x } _ { i } ^ { r } ) \\ \\forall i , \\sum _ { i = 1 } ^ { n } w _ { i , j } = p _ { g } ( \\mathbf { x } _ { j } ^ { g } ) \\ \\forall j .\n$$", + "text_format": "latex", + "bbox": [ + 181, + 670, + 825, + 713 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "This is the finite sample approximation of $\\mathrm { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )$ used in practice. Similar to MMD, the Wasserstein distance is lower when two distributions are more similar. ", + "bbox": [ + 174, + 723, + 825, + 753 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The Fréchet Inception Distance (FID) was recently introduced by Heusel et al. (2017) to evaluate GANs. Formally, it is given by ", + "bbox": [ + 173, + 758, + 821, + 787 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/e52bdd7bfba7d3d3be37ebb5b8a714ad2369afe19286234f8745756614f42ae5.jpg", + "text": "$$\n\\mathrm { F I D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\| \\mu _ { r } - \\mu _ { g } \\| + \\operatorname { T r } ( \\mathbf { C } _ { r } + \\mathbf { C } _ { g } - 2 ( \\mathbf { C } _ { r } \\mathbf { C } _ { g } ) ^ { 1 / 2 } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 300, + 794, + 696, + 814 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\mu _ { r } \\left( \\mu _ { g } \\right)$ and $\\mathbf { C } _ { r }$ $( \\mathbf { C } _ { g } )$ are the mean and covariance of the real (generated) distribution, respectively. Note that under the Gaussian assumption on both $\\mathbb { P } _ { r }$ and $\\mathbb { P } _ { g }$ , the Fréchet distance is equivalent to the Wasserstein-2 distance. ", + "bbox": [ + 174, + 818, + 826, + 861 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The 1-Nearest Neighbor classifier is used in two-sample tests to assess whether two distributions are identical. Given two sets of samples $S _ { r } \\sim \\mathbb { P } _ { r } ^ { n }$ and $S _ { g } \\sim \\mathbb { P } _ { g } ^ { m }$ , with $| S _ { r } | = | S _ { g } |$ , one can compute the leave-one-out (LOO) accuracy of a 1-NN classifier trained on $S _ { r }$ and $S _ { g }$ with positive labels for $S _ { r }$ and negative labels for $S _ { g }$ . Different from the most common use of accuracy, here the 1-NN classifier should yield a $\\sim 5 0 \\%$ LOO accuracy when $| S _ { r } | = | S _ { g } |$ is large. This is achieved when the two distributions match. The LOO accuracy can be lower than $5 0 \\%$ , which happens when the GAN overfits $\\mathbb { P } _ { g }$ to $S _ { r }$ . In the (hypothetical) extreme case, if the GAN were to memorize every sample in $S _ { r }$ and re-generate it exactly, i.e. $S _ { g } = S _ { r }$ , the accuracy would be $0 \\%$ , as every sample from $S _ { r }$ would have it nearest neighbour from $S _ { g }$ with zero distance. The 1-NN classifier belongs to the two-sample test family, for which any binary classifier can be adopted in principle. We will only consider the 1-NN classifier because it requires no special training and little hyperparameter tuning. ", + "bbox": [ + 173, + 867, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 202 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Lopez-Paz & Oquab (2016) considered the 1-NN accuracy primarily as a statistic for two-sample testing. In fact, it is more informative to analyze it for the two classes separately. For example, a typical outcome of GANs is that for both real and generated images, the majority of their nearest neighbors are generated images due to mode collapse. In this case, the LOO 1-NN accuracy of the real images would be relatively low (desired): the mode(s) of the real distribution are usually well captured by the generative model, so a majority of real samples from $S _ { r }$ are surrounded by generated samples from $S _ { g }$ , leading to low LOO accuracy; whereas the LOO accuracy of the generated images is high (not desired): generative samples tend to collapse to a few mode centers, thus they are surrounded by samples from the same class, leading to high LOO accuracy. For the rest of the paper, we distinguish these two cases as 1-NN accuracy (real) and 1-NN accuracy (fake). ", + "bbox": [ + 174, + 208, + 825, + 347 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "2.3 OTHER METRICS ", + "text_level": 1, + "bbox": [ + 176, + 367, + 330, + 382 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "All of the metrics above are, what we refer to as “model agnostic\": they use the generator as a black box to sample the generated images $S _ { g }$ . Model agnostic metrics should not require a density estimation from the model. We choose to only experiment with model agnostic metrics, which allow us to support as many generative models as possible for evaluation without modification to their structure. We will briefly mention some other evaluation metrics not included in our experiments. ", + "bbox": [ + 174, + 392, + 825, + 460 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Kernel density estimation (KDE, or Parzen window estimation) is a well-studied method for estimating the density function of a distribution from samples. For a probability kernel $K$ (most often an isotropic Gaussian) and i.i.d samples $\\mathbf { x } _ { 1 } , \\ldots , \\mathbf { x } _ { n }$ , we can define the density function at $\\mathbf { x }$ as $p ( \\mathbf { x } ) \\approx$ $\\textstyle { \\frac { 1 } { z } } \\sum _ { i = 1 } ^ { n ^ { \\cdot } } K ( \\mathbf { x } - \\mathbf { x } _ { i } )$ , where $z$ is a normalizing constant. This allows the use of classical metrics such as KLD and JSD. However, despite the widespread adoption of this technique to various applications, its suitability to estimating the density of $\\mathbb { P } _ { r }$ or $\\mathbb { P } _ { g }$ for GANs has been questioned by Theis et al. (2015) since the probability kernel depends on the Euclidean distance between images. ", + "bbox": [ + 174, + 468, + 825, + 565 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "More recently, Wu et al. (2016) applied annealed importance sampling (AIS) to estimate the marginal distribution $p ( \\mathbf { x } )$ of a generative model. This method is most natural for models that define a conditional distribution $p ( \\mathbf { x } | \\mathbf { z } )$ where $\\mathbf { z }$ is the latent code, which is not satisfied by most GAN models. Nevertheless, AIS has been applied to GAN evaluation by assuming a Gaussian observation model. We exclude this method from our experiments as it needs the access to the generative model to compute the likelihood, instead of only depending on a finite sample set $S _ { g }$ . ", + "bbox": [ + 174, + 571, + 825, + 657 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3 EXPERIMENTS WITH GAN EVALUATION METRICS ", + "text_level": 1, + "bbox": [ + 174, + 680, + 617, + 695 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1 FEATURE SPACE ", + "text_level": 1, + "bbox": [ + 176, + 712, + 326, + 726 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "All the metrics introduced in the previous section, except for the Inception Score and Mode Score, access the samples $\\mathbf { x }$ only through pair-wise distances. The Kernel MMD requires a fixed kernel function $k$ , typically set to an isotopic Gaussian; the Wasserstein distance and 1-NN accuracy use the underlying distance metric $d$ directly; all of these methods are highly sensitive to the choice that distance. ", + "bbox": [ + 174, + 734, + 825, + 805 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "It is well-established that pixel representations of images do not induce meaningful Euclidean distances (Forsyth & Ponce, 2011). Small translations, rotations, or changes in illumination can increase distances dramatically with little effect on the image content. To quantity the similarity between distributions of images, it is therefore desirable to use distances invariant to such transformations. The choice of distance function can be re-interpreted as a choice of representation, by defining the distance in a more general form as $d ( \\mathbf { x } , \\mathbf { x } ^ { \\prime } ) = \\| \\boldsymbol { \\phi } ( \\mathbf { x } ) - \\boldsymbol { \\phi } ( \\mathbf { x } ^ { \\prime } ) \\| _ { 2 }$ , where $\\phi ( \\cdot )$ is some general mapping of the input into a semantically meaningful feature space. For Kernel MMD, this corresponds to computing the usual inner product in the feature space $\\phi ( \\mathcal { X } )$ . ", + "bbox": [ + 174, + 811, + 825, + 924 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/cab638a8370b4cd54a62381ec515f1f9af5798f9c29f188a4b0647fee588444c.jpg", + "image_caption": [ + "Figure 2: Distinguishing a set of real images (in the training set) from a mixed set of real images and GAN generated images. For the metric to be discriminative, its score should increase as the fraction of generated samples in the mix increases. RIS and RMS fail as they decrease with the fraction of generated samples in $S _ { g }$ on LSUN. Wasserstein and 1-NN accuracy (real) fail in pixel space as they do not increase. " + ], + "image_footnote": [], + "bbox": [ + 176, + 98, + 823, + 246 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Inspired by recent works from Upchurch et al. (2017); Larsen et al. (2015) which show that convolutional neural networks may linearize the image manifold, we propose to operate in the feature space of an external model pre-trained on the ImageNet dataset. For efficiency, we use a 34-layer ResNet2 as the feature extractor. Our experiments show that other models such as VGG or Inception give very similar results. ", + "bbox": [ + 174, + 323, + 825, + 393 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To illustrate our point, we show failure examples of the pixel space distance for evaluating GANs in this section, and highlight that using a proper feature space is key to obtaining meaningful results when applying the distance-based metrics. The usage of a well-suited feature space enables us to draw more optimistic conclusions on GAN evaluation metrics than in Theis et al. (2015). ", + "bbox": [ + 174, + 400, + 825, + 455 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.2 SETUP ", + "text_level": 1, + "bbox": [ + 174, + 469, + 261, + 483 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "For the rest of this major section, we introduce what in our opinion are necessary conditions for good metrics for GANs. After the introduction of each condition, we use it as a criterion to judge the effectiveness of the metrics presented in Section 2, through carefully designed empirical experiments. ", + "bbox": [ + 178, + 489, + 823, + 531 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The experiments are performed on two standard benchmark datasets for generative models, CelebA3and LSUN bedrooms4. To remove the degree of freedom induced by feature representation, we use the Inception Score (IS) and Mode Score (MS) computed from the softmax probabilities of the same ResNet-34 model as the other metrics, instead of the Inception model. We also compute the Inception Score over the real training data $S _ { r }$ as an upper bound, which we denote as $\\mathrm { I S } _ { \\mathrm { 0 } }$ . Moreover, to be consistent with other metrics where lower values correspond to better models, we report the relative inverse Inception Score $R I S = \\left( 1 - { \\mathrm { I S } } / { \\mathrm { I S } } _ { 0 } \\right)$ in tables and plots, after computing IS the Inception Score. We similarly report the relative inverse Mode Score (RMS). Although RIS and RMS operate in the softmax space, we always compare them together with other metrics in the convolutional space for simplicity. For all the plots in this paper, shaded areas denote the standard deviations, computed by running the same experiment 5 times with different random seeds. ", + "bbox": [ + 174, + 537, + 825, + 691 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 DISCRIMINABILITY ", + "text_level": 1, + "bbox": [ + 176, + 704, + 349, + 718 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Mixing of generated images. Arguably, the most important property of a metric $\\hat { \\rho }$ for measuring GANs is the ability to distinguish generated images from real images. To test this property, we sample a set $S _ { r }$ consisting of $n$ ( $n = 2 0 0 0$ if not otherwise specified) real images uniformly from the training set, and a set $S _ { g } ( t )$ of the same size $n$ consisting of a mix of real samples and generated images from a DCGAN (Radford et al., 2015) trained on the same training set, where $t$ denotes the ratio of generated images. ", + "bbox": [ + 174, + 724, + 825, + 809 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The computed values of various metrics between $S _ { r }$ and $S _ { g } ( t )$ for $t \\in [ 0 , 1 ]$ are shown in Figure 2. Since $\\mathbb { P } _ { r }$ should serve as a lower bound for any metric, we expect that any reasonable $\\hat { \\rho }$ should increase as the ratio of generated images increases. This is indeed satisfied for all the metrics except: ", + "bbox": [ + 176, + 815, + 823, + 858 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/1c1ee6c3e031cabba0b7fe1ab9f3c6c30c09d9cf1e241dac2fc9e61e61743cb0.jpg", + "image_caption": [ + "Figure 3: Experiment on simulated mode collapsing. A metric score should increase to reflect the mismatch between true distribution and generated distribution as more modes are collapsed towards their cluster center. All metrics respond correctly in convolutional space. In pixel space, both Wasserstein distance and 1-NN accuracy (real) fail as they decrease in response to more collapsed clusters. " + ], + "image_footnote": [], + "bbox": [ + 176, + 99, + 823, + 238 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "1) RIS and RMS (red and green curves) on LSUN, which decrease as more fake samples are in the mix; 2) 1-NN accuracy of real samples (dotted magenta curve) computed in pixel space, which also appears to be a decreasing function; and 3) Wasserstein Distance (cyan curve), which almost remains unchanged when $t$ varies. ", + "bbox": [ + 174, + 318, + 825, + 373 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The reason that RIS and RMS do not work well here is likely because they are not suitable for images beyond the ImageNet categories. Although other metrics operate in the convolutional feature space also depend on a network pretrained on ImageNet, the convolutional features are much more general than the specific softmax representation. The failure of Wasserstein Distance is possibly due to an insufficient number of samples, which we will discuss in more detail when we analyze the sample efficiency of various metrics in a latter subsection. The last paragraph of Section 2.2 explains why the 1-NN accuracy for real samples (dotted magenta curve) is always lower than that for generated samples (dashed magenta curve). In the pixel space, more than half of the samples from $S _ { r }$ have the nearest neighbor from $S _ { g } ( t )$ , indicating that the DCGAN is able to represent the modes in the pixel space quite well. ", + "bbox": [ + 174, + 381, + 825, + 520 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We also conducted the same experiment using 1) random noise images and 2) images from an entirely different distribution (e.g. CIFAR-10), instead of DCGAN generated images to construct $S _ { g } ( t )$ . We call these injected samples as out-of-domain violations since they are not in $\\mathcal { X }$ , the domain of the real images. These settings yield similar results as in Figure 2, thus we omit their plots. ", + "bbox": [ + 174, + 526, + 823, + 583 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Mode collapsing and mode dropping. In realistic settings, $\\mathbb { P } _ { r }$ is usually very diverse since natural images are inherently multimodal. Many have conjectured that $\\mathbb { P } _ { g }$ differs from $\\mathbb { P } _ { r }$ by reducing diversity, possibly due to the lack of model capacity or inadequate optimization (Arora et al., 2017). This is often manifested itself for generative models in a mix of two ways: mode dropping, where some hard-to-represent modes of $\\mathbb { P } _ { r }$ are simply “ignored\" by $\\mathbb { P } _ { g }$ ; and mode collapsing, where several modes of $\\mathbb { P } _ { r }$ are “averaged\" by $\\mathbb { P } _ { g }$ into a single mode, possibly located at a midpoint. An ideal metric should be sensitive to these two phenomena. ", + "bbox": [ + 174, + 589, + 825, + 686 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To test for mode collapsing, we first randomly sample both $S _ { r }$ and $S _ { r } ^ { \\prime }$ as two disjoint sets of 2000 real images. Next, we find 50 clusters in the whole training set with $k$ -means and progressively replace each cluster by its respective cluster center to simulate mode collapse. Figure 3 shows computed values of $\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )$ as the number of replaced (collapsed) clusters, denoted as $C$ increases. Ideally, we expect the scores increase as $C$ grows. We first observe that all the metrics are able to respond correctly when distances are computed in the convolutional feature space. However, the Wasserstein metric (cyan curve) breaks down in pixel space, as it considers a collapsed sample set (with $C > 0$ ) being closer to the real sample set than another set of real images (with $C = 0$ ). Moreover, although the overall 1-NN accuracy (solid magenta curve) follows the desired trend, the real and fake parts follow opposite trends: 1-NN real accuracy (dotted magenta curve) decreases while 1-NN fake accuracy (dashed magenta curve) increases. Again, this is inline with our explanation given in the last paragraph of Section 2.2. ", + "bbox": [ + 174, + 694, + 825, + 861 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To test for mode dropping, we take $S _ { r }$ as above and construct $S _ { r } ^ { \\prime }$ by randomly removing clusters. To keep the size of $S _ { r } ^ { \\prime }$ constant, we replace images from the removed cluster with images randomly selected from the remaining clusters. Figure 4 shows how different metrics react to the number of removed clusters, also denoted as $C$ . All scores effectively discriminate against mode dropping except the RIS and RMS - they remain almost indifferent when some modes are dropped. Again, this is perhaps caused by the fact that the Inception/Mode Score were originally designed for datasets with classes overlapping with the ImageNet dataset, and they do not generalize well to other datasets. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/27afa6c9ff1eb40623809e4d625a094bd3d23b7636c473490f6547ec4db34c1c.jpg", + "image_caption": [ + "Figure 4: Experiment on simulated mode dropping. A metric score should increase to reflect the mismatch between true distribution and generated distribution as more modes are dropped. All metrics except RIS and RMS respond correctly, as they only increase slightly in value even when almost all modes are dropped. " + ], + "image_footnote": [], + "bbox": [ + 176, + 99, + 823, + 237 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/a3f3cd6ed8e7a79b3ee018f9f4c752f97c498ad8d7aed192a7d48c76b1718aa3.jpg", + "image_caption": [ + "Figure 5: Experiment on robustness of each metric to small transformations (rotations and translations). All metrics should remain constant across all mixes of real and transformed real samples, since the transformations do not alter semantics of the image. All metrics respond correctly in convolutional space, but behave incorrectly in pixel space. This experiment illustrates the unsuitability of distances in pixel space. " + ], + "image_footnote": [], + "bbox": [ + 176, + 291, + 825, + 482 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 558, + 825, + 601 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.4 ROBUSTNESS TO TRANSFORMATIONS ", + "text_level": 1, + "bbox": [ + 176, + 617, + 472, + 631 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "GANs are widely used for image datasets, which have the property that certain transformations to the input do not change its semantic meaning. Thus an ideal evaluation metric should be invariant to such transformations to some extent. For example, a generator trained on CelebA should not be penalized by a metric if its generated faces are shifted by a few pixels or rotated by a small angle. ", + "bbox": [ + 174, + 638, + 562, + 736 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Figure 5 shows how the various metrics react to such small transformation to the images. In this experiment, $S _ { r }$ and $S _ { r } ^ { \\prime }$ are two disjoint sets of 2000 real images sampled from the training data. However, a proportion of images from $S _ { r } ^ { \\prime }$ are randomly shifted (up to 4 pixels) or rotated (up to 15 degrees). We can observe from the results that metrics operating in the convolutional space (or softmax space for RIS and RMS) are robust to these transformations, as all the curves are approximated horizontal as the ratio of transformed samples increases. This is not that surprising as convolutional networks are well know for being invariant to certain transformations (Mallat, 2016). In comparison, in the pixel space all the metrics consider the shifted/rotated images as drawn from a different distribution, highlighting the importance of computing distances in a proper feature space. ", + "bbox": [ + 174, + 743, + 562, + 882 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/3477b7f2e2783fbb5fa4506505a1f49b1b75ebf97b2bb2dd6d880f8cc60b969f.jpg", + "image_caption": [ + "Figure 7: Measurement of wall-clock time for computing various metrics as a function of the number of samples. All metrics are practical to compute for a sample of size 2000, but Wasserstein distance does not scale to large sample sizes. " + ], + "image_footnote": [], + "bbox": [ + 580, + 643, + 816, + 791 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 882, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/2fcde7509c9dec9a43fcff36da5c2ef1c139c751b252031a98486f573d1f95c4.jpg", + "image_caption": [ + "Figure 6: The score of various metrics as a function of the number of samples. An ideal metric should result in a large gap between the real-real (R-R; $\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } ) )$ ) and real-fake (R-G; $\\hat { \\rho } ( \\bar { S } _ { r } , S _ { g } ) \\rangle$ ) curves in order to distinguish between real and fake distributions using as few samples as possible. Compared with Wasserstein distance, MMD and 1-NN accuracy require much fewer samples to discriminate real and generated images, while RIS totally fails on LSUN as it scores generated images even better (lower) than real images. " + ], + "image_footnote": [], + "bbox": [ + 174, + 99, + 823, + 244 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "3.5 EFFICIENCY ", + "text_level": 1, + "bbox": [ + 174, + 335, + 299, + 349 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "A practical GAN evaluation metric should be able to compute “accurate” scores from a reasonable number of samples and within an affordable computation cost, such that it can be computed, for example, after each training epoch to monitor the training process. ", + "bbox": [ + 174, + 358, + 825, + 400 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Sample efficiency. Here we measure the sample efficiency of various metrics by investigating how many samples are needed for each of them in order to discriminate a set of generated samples $S _ { g }$ (from DCGAN) from a set of real samples $S _ { r } ^ { \\prime }$ . To do this, we introduce a reference set $S _ { r }$ , which is also uniformly sampled from the real training data, but is disjoint with $S _ { r } ^ { \\prime }$ . All three sample sets have the same size, i.e., $| S _ { r } | = | S _ { r } ^ { \\prime } | = | S _ { g } | = n$ . We expect that an ideal metric $\\rho$ should correctly score $\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )$ lower than $\\hat { \\rho } ( S _ { r } , S _ { g } )$ with a relatively small $n$ . In other words, the number of samples $n$ needed for the metric to distinguish $S _ { r } ^ { \\prime }$ and $S _ { g }$ can be viewed as its sample complexity. ", + "bbox": [ + 173, + 406, + 825, + 505 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In Figure 6 we show the individual scores as a function of $n$ . We can observe that MMD, FID and 1-NN accuracy computed in convolution feature space are able to distinguish the two set of images $S _ { g }$ (solid blue and magenta curves) and $S _ { r } ^ { \\prime }$ (dotted solid blue and magenta curves) with relatively few samples. The Wasserstein distance (cyan curves) is not discriminative with samples size less than 1000, while the RIS even considers the generated samples to be more “real” than the real samples on the LSUN dataset (the red curves in the third panel). The dotted lines in Figure 6 also quantify how fast the scores converge to their expectations as we increase the sample size. Note that MMD for $\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )$ converges very quickly to zero and gives discriminative scores with few samples, making it a practical metric for comparing GAN models. ", + "bbox": [ + 174, + 511, + 825, + 637 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Computational efficiency. Fast computation of the empirical metric is of practical concern as it helps researchers monitor the training process and diagnose problems early on, or perform early stopping. In Figure 7 we investigate the computational efficiency of the above metrics by showing the wall-clock time (in log scale) to compute them as a function of the number of samples. For a typical number of 2000 samples, it only takes about 8 seconds to compute each of these metrics on an NVIDIA TitanX GPU. In fact, the majority of time is spent on extracting features from the ResNet model. Only the Wasserstein distance becomes prohibitively slow for large sample sizes. ", + "bbox": [ + 174, + 643, + 825, + 741 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "3.6 DETECTING OVERFITTING ", + "text_level": 1, + "bbox": [ + 176, + 755, + 395, + 770 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Overfitting is an artifact of training with finite samples. If a GAN successfully memorizes the training images, i.e., $\\mathbb { P } _ { g }$ is a uniform distribution over the training sample set $S _ { r } ^ { t r }$ , then the generated samples $S _ { g }$ becomes a uniformly drawn set of $n$ samples from $S _ { r } ^ { t r }$ , and any reasonable $\\hat { \\rho }$ should be close to 0. The Wasserstein distance, MMD and 1-NN two sample test are able to detect overfitting in the following sense: if we hold out a validation set $S _ { r } ^ { v a l }$ , then $\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } )$ should be significantly higher than $\\hat { \\rho } ( S _ { g } , S _ { r } ^ { t r } )$ when $\\mathbb { P } _ { g }$ memorizes a part of $S _ { r } ^ { t r }$ . The difference between them can informally be viewed as a form of “generalization gap\". ", + "bbox": [ + 173, + 776, + 825, + 876 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We simulate the overfitting process by defining $S _ { r } ^ { \\prime }$ as a mix of samples from the training set $S _ { r } ^ { t r }$ and a second holdout set, disjoint from both $S _ { r } ^ { t r }$ and $S _ { r } ^ { v a l }$ . Figure 8 shows the gap $\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } ) - \\hat { \\rho } ( \\mathbf { \\dot { } { } } S _ { g } , S _ { r } ^ { t r } )$ of the various metrics as a function of the overlapping ratio between $S _ { r } ^ { \\prime }$ and $S _ { r } ^ { t r }$ . The left most point of each curve can be viewed as the score $\\hat { \\rho } ( S _ { r } ^ { \\prime } , S _ { r } ^ { v a l } )$ computed on a validation set since the overlap ratio is 0. For better visualization, we normalize the Wasserstein distance and MMD by dividing their corresponding score when $S _ { r } ^ { \\prime }$ and $S _ { r }$ have no overlap. As shown in Figure 8, all the metrics except RIS and RMS reflect that the “generalization gap\" increases as $S _ { r } ^ { \\prime }$ overfits more to $S _ { r }$ . The failure of RIS is not surprising: it totally ignores the real data distribution as we discussed in Section 2.2. While the reason that RMS also fails to detect overfitting may again be its lack of generalization to datasets with classes not contained in the ImageNet dataset. In addition, RMS operates in the softmax space, the features in which might be too specific compared to the features in the convolutional space. ", + "bbox": [ + 176, + 882, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/a7fbd0dd8d052615064f08210b23e53041fd3c73b5d0a1f9762bc4f7a90af568.jpg", + "image_caption": [ + "Figure 8: Experiment on detecting overfitting of generated samples. As more generated samples overlap with real samples from the training set, the gap between validation and training score should increase to signal overfitting. All metrics behave correctly except for RIS and RMS, as these two metrics do not increase when the fraction of overlapping samples increases. " + ], + "image_footnote": [], + "bbox": [ + 176, + 99, + 823, + 231 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 296, + 825, + 410 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4 DISCUSSIONS AND CONCLUSION ", + "text_level": 1, + "bbox": [ + 176, + 428, + 478, + 444 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Based on the above analysis, we can summarize the advantages and inherent limitations of the five evaluation metrics, and conditions under which they produce meaningful results. With some of the metrics, we are able to study the problem of overfitting (see Appendix C), perform model selection on GAN models and compare GAN models without resorting to human evaluation based on cherry-picked samples (see Appendix D). ", + "bbox": [ + 174, + 455, + 825, + 526 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "The Inception Score does show a reasonable correlation with the quality and diversity of generated images, which explains the wide usage in practice. However, it is ill-posed mostly because it only evaluates $\\mathbb { P } _ { g }$ as an image generation model rather than its similarity to $\\mathbb { P } _ { r }$ . Blunt violations like mixing in natural images from an entirely different distribution completely deceives the Inception Score. As a result, it may encourage the models to simply learn sharp and diversified images (or even some adversarial noise), instead of $\\mathbb { P } _ { r }$ . This also applies to the Mode Score. Moreover, the Inception Score is unable to detect overfitting since it cannot make use of a holdout validation set. ", + "bbox": [ + 174, + 534, + 825, + 631 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Kernel MMD works surprising well when it operates in the feature space of a pre-trained ResNet. It is always able to identify generative/noise images from real images, and both its sample complexity and computational complexity are low. Given these advantages, even though MMD is biased, we recommend its use in practice. ", + "bbox": [ + 174, + 637, + 825, + 693 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Wasserstein distance works well when the base distance is computed in a suitable feature space. However, it has a high sample complexity, a fact that has been independently observed by (Arora et al., 2017). Another key weakness is that computing the exact Wasserstein distance has a time complexity of $O ( n ^ { 3 } )$ , which is prohibitively expensive as sample size increases. Compared to other methods, Wasserstein distance is less appealing as a practical evaluation metric. ", + "bbox": [ + 174, + 700, + 825, + 770 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Fréchet Inception Distance performs well in terms of discriminability, robustness and efficiency. It appears to be a good evaluation metric for GANs, even it only takes into consideration the first two order moments of the distributions. ", + "bbox": [ + 176, + 777, + 820, + 819 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "1-NN classifier seems to be an ideal metric for evaluating GANs. Not only does it enjoy all the advantages of the other metrics, it also outputs a score in the interval [0, 1], similar to the accuracy/error in classification problems. When the generative distribution perfectly match the true distribution, perfect score (i.e., $5 0 \\%$ accuracy) is attainable with a reasonable (e.g., 1000) sample size. From Figure 2, we find that typical GAN models tend to achieve lower LOO accuracy for real samples (1-NN accuracy (real)), while higher LOO accuracy for generated samples (1-NN accuracy (fake)). This suggests that GANs are able to capture modes from the training distribution, such that the majority of training samples distributed around the mode centers have their nearest neighbor from the generated images, yet most of the generated images are still surrounded by generated images as they are collapsed. The observation indicates that the mode collapse problem is prevalent for typical GAN models. We also note that this problem, however, cannot be effectively detected by human evaluation or the widely used Inception Score. ", + "bbox": [ + 173, + 825, + 825, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 172 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Overall, our empirical study suggests that the choice of feature space in which to compute various metrics is crucial. In the convolutional space of a ResNet pretrained on ImageNet, both MMD and 1-NN accuracy appear to be good metrics in terms of discriminability, robustness and efficiency. Wasserstein distance has very poor sample efficiency, while Inception Score and Mode Score appear to be unsuitable for datasets that are very different from ImageNet. We will release our source code for all these metrics, providing researchers with an off-the-shelf tool to compare and improve GAN algorithms. ", + "bbox": [ + 173, + 180, + 825, + 277 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Based on the two most prominent metrics, MMD and 1-NN accuracy, we study the overfitting problem of DCGAN and WGAN (in Appendix C). Despite the widespread belief that GANs are overfitting to the training data, we find that this does not occur unless there are very few training samples. This raises an interesting question regarding the generalization of GANs in comparison to the supervised setting. We hope that future work can contribute to explaining this phenomenon. ", + "bbox": [ + 174, + 285, + 825, + 354 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 377, + 285, + 391 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein GAN. arXiv preprint arXiv:1701.07875, 2017. ", + "bbox": [ + 174, + 400, + 823, + 428 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Sanjeev Arora, Rong Ge, Yingyu Liang, Tengyu Ma, and Yi Zhang. Generalization and equilibrium in generative adversarial nets (gans). arXiv preprint arXiv:1703.00573, 2017. ", + "bbox": [ + 173, + 438, + 823, + 467 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Wacha Bounliphone, Eugene Belilovsky, Matthew B Blaschko, Ioannis Antonoglou, and Arthur Gretton. A test of relative similarity for model selection in generative models. arXiv preprint arXiv:1511.04581, 2015. ", + "bbox": [ + 176, + 477, + 821, + 518 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative adversarial networks. arXiv preprint arXiv:1612.02136, 2016. ", + "bbox": [ + 171, + 529, + 823, + 559 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255. IEEE, 2009. ", + "bbox": [ + 174, + 568, + 823, + 611 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "David Forsyth and Jean Ponce. Computer vision: a modern approach. Upper Saddle River, NJ; London: Prentice Hall, 2011. ", + "bbox": [ + 174, + 621, + 825, + 650 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014. ", + "bbox": [ + 174, + 659, + 825, + 702 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, Alexander J Smola, et al. A kernel method for the two-sample-problem. Advances in neural information processing systems, 19:513, 2007. ", + "bbox": [ + 174, + 712, + 825, + 755 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. Improved training of wasserstein gans. arXiv preprint arXiv:1704.00028, 2017. ", + "bbox": [ + 169, + 765, + 825, + 794 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Günter Klambauer, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a nash equilibrium. arXiv preprint arXiv:1706.08500, 2017. ", + "bbox": [ + 176, + 804, + 825, + 847 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. arXiv preprint arXiv:1611.07004, 2016. ", + "bbox": [ + 169, + 856, + 823, + 885 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. In Tech Report, 2009. ", + "bbox": [ + 174, + 895, + 823, + 922 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, and Ole Winther. Autoencoding beyond pixels using a learned similarity metric. arXiv preprint arXiv:1512.09300, 2015. ", + "bbox": [ + 174, + 103, + 825, + 146 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "David Lopez-Paz and Maxime Oquab. Revisiting classifier two-sample tests. arXiv preprint arXiv:1610.06545, 2016. ", + "bbox": [ + 173, + 155, + 825, + 184 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian Goodfellow, and Brendan Frey. Adversarial autoencoders. arXiv preprint arXiv:1511.05644, 2015. ", + "bbox": [ + 173, + 193, + 825, + 222 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Stéphane Mallat. Understanding deep convolutional networks. Phil. Trans. R. Soc. A, 374(2065): 20150203, 2016. ", + "bbox": [ + 174, + 229, + 825, + 260 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least squares generative adversarial networks. arXiv preprint ArXiv:1611.04076, 2016. ", + "bbox": [ + 171, + 267, + 823, + 297 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Augustus Odena. Semi-supervised learning with generative adversarial networks. arXiv preprint arXiv:1606.01583, 2016. ", + "bbox": [ + 171, + 305, + 823, + 335 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Guo-Jun Qi. Loss-sensitive generative adversarial networks on lipschitz densities. arXiv preprint arXiv:1701.06264, 2017. ", + "bbox": [ + 171, + 343, + 823, + 372 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. ", + "bbox": [ + 174, + 381, + 823, + 410 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training GANs. In Advances in Neural Information Processing Systems, pp. 2226–2234, 2016. ", + "bbox": [ + 174, + 419, + 825, + 462 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ashish Shrivastava, Tomas Pfister, Oncel Tuzel, Josh Susskind, Wenda Wang, and Russ Webb. Learning from simulated and unsupervised images through adversarial training. arXiv preprint arXiv:1612.07828, 2016. ", + "bbox": [ + 173, + 469, + 825, + 513 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott E. Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. arXiv preprint arXiv:1409.4842, 2014. ", + "bbox": [ + 174, + 522, + 825, + 564 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2818–2826, 2016. ", + "bbox": [ + 174, + 573, + 823, + 617 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Lucas Theis, Aäron van den Oord, and Matthias Bethge. A note on the evaluation of generative models. arXiv preprint arXiv:1511.01844, 2015. ", + "bbox": [ + 173, + 625, + 823, + 655 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Paul Upchurch, Jacob R. Gardner, Kavita Bala, Robert Pless, Noah Snavely, and Kilian Q. Weinberger. Deep feature interpolation for image content changes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. in press ..., 2017. ", + "bbox": [ + 174, + 662, + 825, + 707 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yuhuai Wu, Yuri Burda, Ruslan Salakhutdinov, and Roger Grosse. On the quantitative analysis of decoder-based generative models. arXiv preprint arXiv:1611.04273, 2016. ", + "bbox": [ + 171, + 714, + 823, + 744 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016. ", + "bbox": [ + 171, + 752, + 823, + 781 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Junbo Jake Zhao, Michaël Mathieu, and Yann LeCun. Energy-based generative adversarial network. arXiv preprint arXiv:1609.03126, 2016. ", + "bbox": [ + 171, + 790, + 825, + 820 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. arXiv preprint arXiv:1703.10593, 2017. ", + "bbox": [ + 171, + 828, + 823, + 858 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A GAN VARIANTS USED IN OUR EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 174, + 102, + 584, + 117 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Many GAN variants have been proposed recently. In this paper we consider several of them, which we briefly review in this section. ", + "bbox": [ + 169, + 127, + 823, + 155 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "DCGAN (Radford et al., 2015). The generator of a DCGAN takes a lower dimensional input from a uniform noise distribution, then projects and reshapes it to a small convolutional representation with many feature maps. After applying a series of four fractionally-strided convolutions, the generator converts this representation into a $6 4 \\times 6 4$ pixel image. DCGAN is optimized by minimizing the Jensen-Shannon divergence between the real and generated images. ", + "bbox": [ + 174, + 161, + 825, + 232 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "WGAN (Arjovsky et al., 2017). A critic network that outputs unconstrained real values is used in place of the discriminator. When the critic is Lipschitz, this network approximates the Wasserstein distance between $S _ { r }$ and $S _ { g }$ . A Lipschitz condition is enforced by clipping the critic networks’ parameters to stay within a predefined bounding box. ", + "bbox": [ + 174, + 239, + 825, + 295 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "WGAN with gradient penalty (Gulrajani et al., 2017) improves upon WGAN by enforcing the Lipschitz condition with a gradient penalty term. This method significantly improves the convergence speed and the quality of the images generated by a WGAN. ", + "bbox": [ + 174, + 301, + 820, + 344 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "LSGAN (Mao et al., 2016). Least Squares GAN adopts the least squares loss function instead of the commonly used sigmoid cross entropy loss for the discriminator, essentially minimizing the Pearson $\\chi ^ { 2 }$ divergence between the real distribution $\\mathbb { P } _ { r }$ and generative distribution $\\mathbb { P } _ { g }$ . ", + "bbox": [ + 174, + 351, + 825, + 393 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "B THE CHOICE OF FEATURE SPACE ", + "text_level": 1, + "bbox": [ + 178, + 407, + 477, + 422 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "The choice of features space is crucial for all these metrics. Here we consider several alternatives to the convolutional features from the 34-layer ResNet trained on ImageNet. In particular, we compute various metrics using the features extracted by (1) the VGG and Inception networks; (2) a 34-layer ResNet with random weights; (3) a ResNet classifier trained on the same dataset as the GAN models. We use the features extracted from these models to test all metrics in the discriminative experiments we performed in Section 3. All experimental settings are identical except for the third experiments, which is performed on the CIFAR-10 dataset (Krizhevsky & Hinton, 2009) instead as we need class labels to train the classifier. Note that we consider setting (3) only for analytical purposes. It is not a practical choice as GANs are mainly designed for unsupervised learning and we should not assume the existence of ground truth labels. ", + "bbox": [ + 173, + 431, + 825, + 570 + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/eb751a4b07b47aa017a1aea8245697f431d986c743349d44ad91f1efadce7805.jpg", + "image_caption": [ + "Figure 9: Comparison of all metrics in different feature spaces. When using different trained networks, the trends of all metrics are very similar. Most metrics work well even in a random network, but Wasserstein distance has very high variance and the magnitude of increase for 1-NN accuracy is small. " + ], + "image_footnote": [], + "bbox": [ + 174, + 583, + 825, + 775 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "The results are shown in Figure 9 and Figure 10, from which several observations can be made: (1) switching from ResNet-34 to VGG or Inception has little effect to the metric scores; (2) the features from a random network still works for MMD, while it makes the Wasserstein distance unstable and 1-NN accuracy less discriminative. Not surprisingly, the Inception Score and Mode Score becomes meaningless if we use the softmax values from the random network; (3) features extracted from the classifier trained on the same dataset as the GAN model also offers high discriminability for these metrics, especially for the Wasserstein distance. However, this may be simply due to the fact that the feature dimensionality of the ResNet trained on CIFAR-10 is much smaller than that of the ResNet-34 trained on ImageNet (64 v.s. 512). ", + "bbox": [ + 173, + 839, + 825, + 924 + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/2aec4fabff4f4efa259561ca43ac79e2d4419b14e514a7fa7e0dfdcdc789922b.jpg", + "image_caption": [ + "Figure 10: Using features extracted from a ResNet trained on CIFAR-10 (right plot) to evaluate a GAN model trained on the same dataset. Compared to using an extractor trained on ImageNet, the metrics appear to have lower variance. However, this may due to the feature dimensionality being smaller for CIFAR-10. " + ], + "image_footnote": [], + "bbox": [ + 240, + 101, + 754, + 214 + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/2272bebca854ed5b1c95d0a22f8cf047e84ae9c0cb02aceed35de9a3d7a44221.jpg", + "image_caption": [ + "Figure 11: Training curves of DCGAN and WGAN on a large (left two panels), small (middle two panels) and tiny (right two panels) subsets of CelebA. Note that for the first four plots, blue (yellow) curves almost overlap with the red (green) curves, indicating no overfitting detected by the two metrics. Overfitting only observed on the tiny training set, with MMD score and 1-NN accuracy significantly worse (higher) on the validation set. " + ], + "image_footnote": [], + "bbox": [ + 183, + 284, + 813, + 460 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 555, + 825, + 598 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "C ARE GANS OVERFITTING TO THE TRAINING DATA? ", + "text_level": 1, + "bbox": [ + 174, + 616, + 635, + 632 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "We trained two representative GAN models, DCGAN (Radford et al., 2015) and WGAN (Arjovsky et al., 2017) on the CelebA dataset. Out of the ${ \\sim } 2 0 0 { , } 0 0 0$ images in total, we holdout 20,000 images for validation, and the rest for training. As the training set is sufficiently large, which makes overfitting unlikely to occur, we also create a small training set and a tiny training set respectively with only 2000 and 10 images sampled from the full training set. ", + "bbox": [ + 174, + 643, + 825, + 713 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "The training setting for DCGAN and WGAN strictly follow their original implementation, except that we change the default number of training iterations such that both models are sufficiently updated. For each metric, we compute their score on 2000 real samples and 2000 generated samples, where the real samples are drawn from either the training set or the validation set, giving rise to training and validation scores. The results are shown in Figure 11, from which we can make several observations: ", + "bbox": [ + 174, + 719, + 825, + 790 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "• The training and validation scores almost overlap with each other with 2000 or $1 8 0 \\mathrm { k }$ training samples, showing that both DCGAN and WGAN do not overfit to the training data under of these metrics. Even when using only 2000 training samples, there is still no significant difference between the training score and validation score. This shows that the training process of GANs behaves quite differently from those of supervised deep learning models, where a model can easily achieve 0 training error while behaving like random guess on the validation set (Zhang et al., 2016). 5 ", + "bbox": [ + 179, + 796, + 825, + 893 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/ab92d3d0286e87eafef72ed0918c3d6c9537d5ee1f44d2019ef0f4a7e4ed705e.jpg", + "table_caption": [ + "Table 1: Comparison of several GAN models on the LSUN dataset " + ], + "table_footnote": [], + "table_body": "
RealDCGANWGANWGAN-GPLSGAN
Conv SpaceMMD0.0190.2050.2700.1940.232
1-NN Accuracy0.4990.8250.9200.8120.871
1-NN Accuracy (real)0.4950.7590.8800.7650.804
1-NN Accuracy (fake)0.5030.8920.9610.8600.938
", + "bbox": [ + 202, + 125, + 792, + 193 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "• DCGAN outperforms WGAN on the full training set under both metrics, and converges faster. However, WGAN is much more stable on the small training set, and converges to better positions. ", + "bbox": [ + 179, + 208, + 826, + 236 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "D COMPARISON OF POPULAR GAN MODELS BASED ON QUANTITATIVE EVALUATION METRICS ", + "text_level": 1, + "bbox": [ + 178, + 256, + 776, + 290 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Based on our analysis, we chose MMD and 1-NN accuracy in the feature space of a 34-layer ResNet trained on ImageNet to compare several state-of-the-art GAN models. All scores are computed using 2000 samples from the holdout set and 2000 generated samples. The GAN models evaluated include DCGAN (Radford et al., 2015), WGAN (Arjovsky et al., 2017), WGAN with gradient penalty (WGAN-GP ) (Gulrajani et al., 2017), and LSGAN (Mao et al., 2016) , all trained on the CelebA dataset. The results are reported in Table 1, from which we highlight three observations: ", + "bbox": [ + 174, + 299, + 825, + 383 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "• WGAN-GP performs the best under most of the metrics. \nDCGAN achieves 0.759 overall 1-NN accuracy on real samples, slightly better than 0.765 achieved by WGAN-GP; while the 1-NN accuracy on generated (fake) samples achieved by DCGAN is higher than that by WGAN-GP (0.892 v.s. 0.860). This seems to suggest that DCGAN is better at capturing modes in the training data distribution, while its generated samples are more collapsed compared to WGAN-GP. Such subtle difference is unlikely to be discovered by the Inception Score or human evaluation. \n• The 1-NN accuracy for all evaluated GAN models are higher than 0.8 , far above the ground truth of 0.5. The MMD score of the four GAN models are also much larger than that of ground truth (0.019). This indicates that even state-of-the-art GAN models are far from learning the true distribution. ", + "bbox": [ + 215, + 395, + 825, + 559 + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/parse/train/Sy1f0e-R-/Sy1f0e-R-_middle.json b/parse/train/Sy1f0e-R-/Sy1f0e-R-_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..68fd8b78e6c2b0899ed59f3d49344bed7a7f21d4 --- /dev/null +++ b/parse/train/Sy1f0e-R-/Sy1f0e-R-_middle.json @@ -0,0 +1,41207 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 80, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 507, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 507, + 100 + ], + "score": 1.0, + "content": "AN EMPIRICAL STUDY ON EVALUATION METRICS OF", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 102, + 401, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 401, + 118 + ], + "score": 1.0, + "content": "GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 213, + 469, + 377 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 469, + 226 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 469, + 226 + ], + "score": 1.0, + "content": "Despite the widespread interest in generative adversarial networks (GANs), few", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 224, + 469, + 235 + ], + "spans": [ + { + "bbox": [ + 142, + 224, + 469, + 235 + ], + "score": 1.0, + "content": "works have studied the metrics that quantitatively evaluate GANs’ performance. In", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 235, + 470, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 235, + 470, + 247 + ], + "score": 1.0, + "content": "this paper, we revisit several representative sample-based evaluation metrics for", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 471, + 258 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 471, + 258 + ], + "score": 1.0, + "content": "GANs, and address the important problem of how to evaluate the evaluation metrics.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 257, + 470, + 269 + ], + "spans": [ + { + "bbox": [ + 141, + 257, + 470, + 269 + ], + "score": 1.0, + "content": "We start with a few necessary conditions for metrics to produce meaningful scores,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 470, + 280 + ], + "score": 1.0, + "content": "such as distinguishing real from generated samples, identifying mode dropping and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 279, + 470, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 279, + 470, + 291 + ], + "score": 1.0, + "content": "mode collapsing, and detecting overfitting. Then with a series of carefully designed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 290, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 290, + 470, + 302 + ], + "score": 1.0, + "content": "experiments, we are able to comprehensively investigate existing sample-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 300, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 470, + 312 + ], + "score": 1.0, + "content": "metrics and identify their strengths and limitations in practical settings. Based", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 312, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 141, + 312, + 470, + 324 + ], + "score": 1.0, + "content": "on these results, we observe that kernel Maximum Mean Discrepancy (MMD)", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "score": 1.0, + "content": "and the 1-Nearest-Neighbour (1-NN) two-sample test seem to satisfy most of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 334, + 469, + 345 + ], + "spans": [ + { + "bbox": [ + 142, + 334, + 469, + 345 + ], + "score": 1.0, + "content": "desirable properties, provided that the distances between samples are computed in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "score": 1.0, + "content": "a suitable feature space. Our experiments also unveil interesting properties about", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 354, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 470, + 369 + ], + "score": 1.0, + "content": "the behavior of several popular GAN models, such as whether they are memorizing", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 444, + 379 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 444, + 379 + ], + "score": 1.0, + "content": "training samples, and how far these state-of-the-art GANs are from perfect.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 402, + 206, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 208, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 208, + 417 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "Generative adversarial networks (GANs) (Goodfellow et al., 2014) have been studied extensively", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "in recent years. Besides producing surprisingly plausible images of faces (Radford et al., 2015;", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "Larsen et al., 2015) and bedrooms (Radford et al., 2015; Arjovsky et al., 2017; Gulrajani et al., 2017),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "they have also been innovatively applied in, for example, semi-supervised learning (Odena, 2016;", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "Makhzani et al., 2015), image-to-image translation (Isola et al., 2016; Zhu et al., 2017), and simulated", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 491 + ], + "score": 1.0, + "content": "image refinement (Shrivastava et al., 2016). However, despite the availability of a plethora of GAN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "models (Arjovsky et al., 2017; Qi, 2017; Radford et al., 2015; Zhao et al., 2016), their evaluation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "score": 1.0, + "content": "is still predominantly qualitative, very often resorting to manual inspection of the visual fidelity of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "score": 1.0, + "content": "generated images. Such evaluation is time-consuming, subjective and possibly misleading. Given", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "score": 1.0, + "content": "the inherent limitations of qualitative evaluations, proper quantitative metrics are crucial for the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 461, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 461, + 546 + ], + "score": 1.0, + "content": "development of GANs to avoid the human factors and guide the design of better models.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "Possibly the most popular metric is the Inception Score (Salimans et al., 2016), which measures", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "the quality and diversity of the generated images using an external model, the Google Inception", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "score": 1.0, + "content": "network (Szegedy et al., 2014), trained on the large scale ImageNet dataset (Deng et al., 2009). Some", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "other metrics are less widely used but still very valuable. Wu et al. (2016) proposed a sampling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "method to estimate the log-likelihood of generative models, by assuming a Gaussian observation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "model with a fixed variance. Bounliphone et al. (2015) propose to use maximum mean discrepancies", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "(MMDs) for model selection in generative models. Lopez-Paz & Oquab (2016) apply the classifier", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "two-sample test, a well-studied tool in statistics, to assess the difference between the generated and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 639, + 182, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 182, + 651 + ], + "score": 1.0, + "content": "target distribution.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Although these evaluation metrics are shown to be effective on various tasks, it is unclear in which", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "scenarios their scores are meaningful, and in which other scenarios, prone to misinterpretations. Given", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "that evaluating GANs is already challenging, it can only be more difficult to evaluate the evaluation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "score": 1.0, + "content": "metrics themselves. Most existing works attempt to justify their proposed metrics by showing a strong", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "score": 1.0, + "content": "correlation with human evaluation (Salimans et al., 2016; Lopez-Paz & Oquab, 2016). However,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "human evaluation tends to be biased towards the visual quality of generated samples and neglect the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 719, + 507, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 507, + 734 + ], + "score": 1.0, + "content": "overall distributional characteristics, which are arguably just as important for unsupervised learning.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 80, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 507, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 507, + 100 + ], + "score": 1.0, + "content": "AN EMPIRICAL STUDY ON EVALUATION METRICS OF", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 102, + 401, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 401, + 118 + ], + "score": 1.0, + "content": "GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 111, + 135, + 245, + 158 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 213, + 469, + 377 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 469, + 226 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 469, + 226 + ], + "score": 1.0, + "content": "Despite the widespread interest in generative adversarial networks (GANs), few", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 224, + 469, + 235 + ], + "spans": [ + { + "bbox": [ + 142, + 224, + 469, + 235 + ], + "score": 1.0, + "content": "works have studied the metrics that quantitatively evaluate GANs’ performance. In", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 235, + 470, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 235, + 470, + 247 + ], + "score": 1.0, + "content": "this paper, we revisit several representative sample-based evaluation metrics for", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 471, + 258 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 471, + 258 + ], + "score": 1.0, + "content": "GANs, and address the important problem of how to evaluate the evaluation metrics.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 257, + 470, + 269 + ], + "spans": [ + { + "bbox": [ + 141, + 257, + 470, + 269 + ], + "score": 1.0, + "content": "We start with a few necessary conditions for metrics to produce meaningful scores,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 470, + 280 + ], + "score": 1.0, + "content": "such as distinguishing real from generated samples, identifying mode dropping and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 279, + 470, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 279, + 470, + 291 + ], + "score": 1.0, + "content": "mode collapsing, and detecting overfitting. Then with a series of carefully designed", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 290, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 290, + 470, + 302 + ], + "score": 1.0, + "content": "experiments, we are able to comprehensively investigate existing sample-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 300, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 470, + 312 + ], + "score": 1.0, + "content": "metrics and identify their strengths and limitations in practical settings. Based", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 312, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 141, + 312, + 470, + 324 + ], + "score": 1.0, + "content": "on these results, we observe that kernel Maximum Mean Discrepancy (MMD)", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "score": 1.0, + "content": "and the 1-Nearest-Neighbour (1-NN) two-sample test seem to satisfy most of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 334, + 469, + 345 + ], + "spans": [ + { + "bbox": [ + 142, + 334, + 469, + 345 + ], + "score": 1.0, + "content": "desirable properties, provided that the distances between samples are computed in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "score": 1.0, + "content": "a suitable feature space. Our experiments also unveil interesting properties about", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 354, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 470, + 369 + ], + "score": 1.0, + "content": "the behavior of several popular GAN models, such as whether they are memorizing", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 444, + 379 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 444, + 379 + ], + "score": 1.0, + "content": "training samples, and how far these state-of-the-art GANs are from perfect.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 12, + "bbox_fs": [ + 141, + 212, + 471, + 379 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 402, + 206, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 208, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 208, + 417 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "Generative adversarial networks (GANs) (Goodfellow et al., 2014) have been studied extensively", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "in recent years. Besides producing surprisingly plausible images of faces (Radford et al., 2015;", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "Larsen et al., 2015) and bedrooms (Radford et al., 2015; Arjovsky et al., 2017; Gulrajani et al., 2017),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "they have also been innovatively applied in, for example, semi-supervised learning (Odena, 2016;", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "Makhzani et al., 2015), image-to-image translation (Isola et al., 2016; Zhu et al., 2017), and simulated", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 491 + ], + "score": 1.0, + "content": "image refinement (Shrivastava et al., 2016). However, despite the availability of a plethora of GAN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "models (Arjovsky et al., 2017; Qi, 2017; Radford et al., 2015; Zhao et al., 2016), their evaluation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "score": 1.0, + "content": "is still predominantly qualitative, very often resorting to manual inspection of the visual fidelity of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "score": 1.0, + "content": "generated images. Such evaluation is time-consuming, subjective and possibly misleading. Given", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "score": 1.0, + "content": "the inherent limitations of qualitative evaluations, proper quantitative metrics are crucial for the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 461, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 461, + 546 + ], + "score": 1.0, + "content": "development of GANs to avoid the human factors and guide the design of better models.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 424, + 506, + 546 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "Possibly the most popular metric is the Inception Score (Salimans et al., 2016), which measures", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "the quality and diversity of the generated images using an external model, the Google Inception", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "score": 1.0, + "content": "network (Szegedy et al., 2014), trained on the large scale ImageNet dataset (Deng et al., 2009). Some", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "other metrics are less widely used but still very valuable. Wu et al. (2016) proposed a sampling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "method to estimate the log-likelihood of generative models, by assuming a Gaussian observation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "model with a fixed variance. Bounliphone et al. (2015) propose to use maximum mean discrepancies", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "(MMDs) for model selection in generative models. Lopez-Paz & Oquab (2016) apply the classifier", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "two-sample test, a well-studied tool in statistics, to assess the difference between the generated and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 639, + 182, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 182, + 651 + ], + "score": 1.0, + "content": "target distribution.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 549, + 506, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Although these evaluation metrics are shown to be effective on various tasks, it is unclear in which", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "scenarios their scores are meaningful, and in which other scenarios, prone to misinterpretations. Given", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "that evaluating GANs is already challenging, it can only be more difficult to evaluate the evaluation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "score": 1.0, + "content": "metrics themselves. Most existing works attempt to justify their proposed metrics by showing a strong", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "score": 1.0, + "content": "correlation with human evaluation (Salimans et al., 2016; Lopez-Paz & Oquab, 2016). However,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "human evaluation tends to be biased towards the visual quality of generated samples and neglect the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 719, + 507, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 507, + 734 + ], + "score": 1.0, + "content": "overall distributional characteristics, which are arguably just as important for unsupervised learning.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 655, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 141, + 81, + 473, + 152 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 141, + 81, + 473, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 81, + 473, + 152 + ], + "spans": [ + { + "bbox": [ + 141, + 81, + 473, + 152 + ], + "score": 0.962, + "type": "image", + "image_path": "49450f89442e39d6bd90bdd00539f5bc95c21f16ec6bfdf0c04ded989a0111e0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 141, + 81, + 473, + 104.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 141, + 104.66666666666667, + 473, + 128.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 141, + 128.33333333333334, + 473, + 152.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 130, + 161, + 480, + 172 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 160, + 481, + 172 + ], + "spans": [ + { + "bbox": [ + 129, + 160, + 481, + 172 + ], + "score": 1.0, + "content": "Figure 1: A schematic layout of the typical approach for sample based GAN evaluation methods.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 182, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "In this paper we comprehensively examine the existing literature on sample-based quantitative", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "score": 1.0, + "content": "evaluation of GANs. We address the challenge of evaluating the metrics themselves by carefully", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "designing a series of experiments, through which we hope to answer the following important questions:", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "1.) What are reasonable characterizations of the behavior of existing sample-based metrics for GANs?", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "score": 1.0, + "content": "2.) What are the strengths and limitations of these metrics? 3.) Which metrics are preferred", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 451, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 451, + 250 + ], + "score": 1.0, + "content": "accordingly? 4.) How are the metrics helpful in understanding and improving GANs?", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 254, + 505, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 268 + ], + "score": 1.0, + "content": "Ultimately, we hope that this paper will establish good principles on choosing, applying, interpreting", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "and designing evaluation metrics for GANs in practical settings. We will also release the source code", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 276, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 290 + ], + "score": 1.0, + "content": "for all experiments and metrics examined, providing the community with off-the-shelf tools to debug", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 288, + 251, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 251, + 299 + ], + "score": 1.0, + "content": "and improve their GAN algorithms.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 107, + 310, + 200, + 323 + ], + "lines": [ + { + "bbox": [ + 104, + 308, + 202, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 202, + 326 + ], + "score": 1.0, + "content": "2 BACKGROUND", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 504, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "We briefly review the original GAN framework proposed by Goodfellow et al. (2014). Description of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 342, + 400, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 400, + 353 + ], + "score": 1.0, + "content": "the GAN variants used in our experiments is deferred to the Appendix A.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 302, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 303, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 303, + 374 + ], + "score": 1.0, + "content": "2.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 377, + 505, + 411 + ], + "lines": [ + { + "bbox": [ + 104, + 375, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 122, + 391 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 376, + 167, + 387 + ], + "score": 0.92, + "content": "\\mathcal { X } = \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 375, + 370, + 391 + ], + "score": 1.0, + "content": "be the space of natural images. Given i.i.d. samples", + "type": "text" + }, + { + "bbox": [ + 371, + 378, + 449, + 389 + ], + "score": 0.93, + "content": "S _ { r } = \\{ \\mathbf { x } _ { 1 } ^ { r } , \\ldots , \\mathbf { x } _ { n } ^ { r } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 375, + 506, + 391 + ], + "score": 1.0, + "content": "drawn from a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 387, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 171, + 402 + ], + "score": 1.0, + "content": "real distribution", + "type": "text" + }, + { + "bbox": [ + 171, + 389, + 183, + 399 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 387, + 203, + 402 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 204, + 389, + 213, + 398 + ], + "score": 0.82, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 387, + 419, + 402 + ], + "score": 1.0, + "content": ", we would like to learn a parameterized distribution", + "type": "text" + }, + { + "bbox": [ + 419, + 389, + 431, + 401 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 387, + 505, + 402 + ], + "score": 1.0, + "content": "that approximates", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 399, + 184, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 169, + 411 + ], + "score": 1.0, + "content": "the distribution", + "type": "text" + }, + { + "bbox": [ + 169, + 400, + 181, + 410 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 399, + 184, + 411 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "The setup of generative adversarial networks is as follows. We define two networks, the discriminator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 169, + 439 + ], + "score": 0.92, + "content": "D : \\mathcal { X } [ 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 426, + 244, + 440 + ], + "score": 1.0, + "content": "and the generator", + "type": "text" + }, + { + "bbox": [ + 244, + 428, + 294, + 438 + ], + "score": 0.91, + "content": "G : { \\mathcal { Z } } \\to { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 426, + 325, + 440 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 325, + 428, + 334, + 437 + ], + "score": 0.85, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "is some latent space. Given a distribution", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 118, + 449 + ], + "score": 0.86, + "content": "\\mathbb { P } _ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 437, + 140, + 451 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 140, + 439, + 149, + 448 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 437, + 343, + 451 + ], + "score": 1.0, + "content": "(usually an isotropic Gaussian), the distribution", + "type": "text" + }, + { + "bbox": [ + 343, + 438, + 355, + 451 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 437, + 408, + 451 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + }, + { + "bbox": [ + 409, + 438, + 435, + 450 + ], + "score": 0.94, + "content": "G ( \\mathbb { P } _ { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 437, + 506, + 451 + ], + "score": 1.0, + "content": ". Optimization is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 450, + 309, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 271, + 461 + ], + "score": 1.0, + "content": "performed with respect to a joint loss for", + "type": "text" + }, + { + "bbox": [ + 272, + 450, + 281, + 459 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 450, + 299, + 461 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 450, + 309, + 459 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 464, + 447, + 482 + ], + "lines": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "spans": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { G } \\operatorname* { m a x } _ { D } L ( D , G ) = \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { r } } \\log { [ D ( \\mathbf { x } ) ] } + \\mathbb { E } _ { \\mathbf { z } \\sim \\mathbb { P } _ { z } } \\left[ \\log ( 1 - D ( G ( \\mathbf { z } ) ) ) \\right] .", + "type": "interline_equation", + "image_path": "b2535cc6daf8bf45cb707b25225a313cbe9c6bd2d1e1ec85d1ad919727e0664c.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 506, + 530 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 218, + 498 + ], + "score": 1.0, + "content": "Intuitively, the discriminator", + "type": "text" + }, + { + "bbox": [ + 219, + 486, + 228, + 496 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 486, + 347, + 498 + ], + "score": 1.0, + "content": "outputs a probability for every", + "type": "text" + }, + { + "bbox": [ + 348, + 486, + 376, + 496 + ], + "score": 0.93, + "content": "\\mathbf { x } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "that corresponds to its likelihood", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 497, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 191, + 509 + ], + "score": 1.0, + "content": "of being drawn from", + "type": "text" + }, + { + "bbox": [ + 192, + 497, + 203, + 508 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 497, + 397, + 509 + ], + "score": 1.0, + "content": ", and the loss function encourages the generator", + "type": "text" + }, + { + "bbox": [ + 397, + 497, + 406, + 507 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 497, + 506, + 509 + ], + "score": 1.0, + "content": "to produce samples that", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 506, + 507, + 522 + ], + "spans": [ + { + "bbox": [ + 104, + 506, + 460, + 522 + ], + "score": 1.0, + "content": "maximize this probability. Practically, the loss is approximated with finite samples from", + "type": "text" + }, + { + "bbox": [ + 461, + 508, + 472, + 519 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 506, + 490, + 522 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 491, + 508, + 502, + 520 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 506, + 507, + 522 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 519, + 400, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 267, + 531 + ], + "score": 1.0, + "content": "and optimized with alternating steps for", + "type": "text" + }, + { + "bbox": [ + 268, + 519, + 277, + 528 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 519, + 295, + 531 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 296, + 519, + 304, + 528 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 519, + 400, + 531 + ], + "score": 1.0, + "content": "using gradient descent.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 504, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 358, + 549 + ], + "score": 1.0, + "content": "To evaluate the generator, we would like to design a metric", + "type": "text" + }, + { + "bbox": [ + 358, + 537, + 365, + 547 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 534, + 504, + 549 + ], + "score": 1.0, + "content": "that measures the “dissimilarity\"", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 545, + 143, + 559 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 144, + 547, + 155, + 559 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 545, + 168, + 559 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 168, + 547, + 180, + 558 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 545, + 448, + 559 + ], + "score": 1.0, + "content": ".1 In theory, with both distributions known, common choices of", + "type": "text" + }, + { + "bbox": [ + 448, + 549, + 455, + 558 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "include the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 557, + 507, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 507, + 570 + ], + "score": 1.0, + "content": "Kullback-Leibler divergence (KLD), Jensen-Shannon divergence (JSD) and total variation. However,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 568, + 507, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 197, + 581 + ], + "score": 1.0, + "content": "in practical scenarios,", + "type": "text" + }, + { + "bbox": [ + 197, + 569, + 208, + 579 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 568, + 380, + 581 + ], + "score": 1.0, + "content": "is unknown and only the finite samples in", + "type": "text" + }, + { + "bbox": [ + 381, + 569, + 392, + 579 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 568, + 507, + 581 + ], + "score": 1.0, + "content": "are observed. Furthermore,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 578, + 509, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 578, + 373, + 593 + ], + "score": 1.0, + "content": "it is almost always intractable to compute the exact density of", + "type": "text" + }, + { + "bbox": [ + 106, + 590, + 215, + 603 + ], + "score": 0.91, + "content": "S _ { g } = \\{ \\mathbf { x } _ { 1 } ^ { g } , . . . , \\mathbf { x } _ { m } ^ { \\bar { g } } \\} \\sim \\mathbb { P } _ { q } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 584, + 509, + 615 + ], + "score": 1.0, + "content": "specially so for GANs). Given these limitations, we focus on empiricalof “dissimilarity\" between samples from two distributions.", + "type": "text" + }, + { + "bbox": [ + 374, + 580, + 385, + 591 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 578, + 506, + 593 + ], + "score": 1.0, + "content": ", but much easier to sample", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 146, + 602, + 226, + 613 + ], + "spans": [ + { + "bbox": [ + 146, + 602, + 226, + 613 + ], + "score": 0.88, + "content": "\\hat { \\rho } : \\mathcal X ^ { n } \\times \\mathcal X ^ { m } \\stackrel { \\smile } { } \\mathbb R", + "type": "inline_equation" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 107, + 622, + 239, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 240, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 240, + 633 + ], + "score": 1.0, + "content": "2.2 SAMPLE BASED METRICS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "We mainly focus on sample based evaluation metrics that follow a common setup illustrated in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "Figure 1. The metric calculator is the key element, for which we briefly introduce five representative", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "methods: Inception Score (Salimans et al., 2016), Mode Score (Che et al., 2016) , Kernel MMD (Gret-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "ton et al., 2007), Wasserstein distance, Fréchet Inception Distance (FID) (Heusel et al., 2017), and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "score": 1.0, + "content": "1-nearest neighbor (1-NN)-based two sample test (Lopez-Paz & Oquab, 2016). All of them are model", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 692, + 347, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 347, + 705 + ], + "score": 1.0, + "content": "agnostic and require only finite samples from the generator.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 712, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 710, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 119, + 710, + 158, + 723 + ], + "score": 1.0, + "content": "1Note that", + "type": "text" + }, + { + "bbox": [ + 158, + 713, + 165, + 722 + ], + "score": 0.75, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 710, + 506, + 723 + ], + "score": 1.0, + "content": "does not need satisfy symmetry or triangle inequality, so it is not, mathematically speaking, a", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 720, + 452, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 195, + 733 + ], + "score": 1.0, + "content": "distance metric between", + "type": "text" + }, + { + "bbox": [ + 195, + 722, + 206, + 732 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 720, + 222, + 733 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 722, + 233, + 731 + ], + "score": 0.86, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 720, + 452, + 733 + ], + "score": 1.0, + "content": ". We still call it a metric throughout this paper for simplicity.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 141, + 81, + 473, + 152 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 141, + 81, + 473, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 81, + 473, + 152 + ], + "spans": [ + { + "bbox": [ + 141, + 81, + 473, + 152 + ], + "score": 0.962, + "type": "image", + "image_path": "49450f89442e39d6bd90bdd00539f5bc95c21f16ec6bfdf0c04ded989a0111e0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 141, + 81, + 473, + 104.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 141, + 104.66666666666667, + 473, + 128.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 141, + 128.33333333333334, + 473, + 152.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 130, + 161, + 480, + 172 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 160, + 481, + 172 + ], + "spans": [ + { + "bbox": [ + 129, + 160, + 481, + 172 + ], + "score": 1.0, + "content": "Figure 1: A schematic layout of the typical approach for sample based GAN evaluation methods.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 182, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "In this paper we comprehensively examine the existing literature on sample-based quantitative", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "score": 1.0, + "content": "evaluation of GANs. We address the challenge of evaluating the metrics themselves by carefully", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "designing a series of experiments, through which we hope to answer the following important questions:", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "1.) What are reasonable characterizations of the behavior of existing sample-based metrics for GANs?", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "score": 1.0, + "content": "2.) What are the strengths and limitations of these metrics? 3.) Which metrics are preferred", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 451, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 451, + 250 + ], + "score": 1.0, + "content": "accordingly? 4.) How are the metrics helpful in understanding and improving GANs?", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 183, + 506, + 250 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 254, + 505, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 268 + ], + "score": 1.0, + "content": "Ultimately, we hope that this paper will establish good principles on choosing, applying, interpreting", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "and designing evaluation metrics for GANs in practical settings. We will also release the source code", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 276, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 290 + ], + "score": 1.0, + "content": "for all experiments and metrics examined, providing the community with off-the-shelf tools to debug", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 288, + 251, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 251, + 299 + ], + "score": 1.0, + "content": "and improve their GAN algorithms.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 254, + 505, + 299 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 310, + 200, + 323 + ], + "lines": [ + { + "bbox": [ + 104, + 308, + 202, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 202, + 326 + ], + "score": 1.0, + "content": "2 BACKGROUND", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 504, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "We briefly review the original GAN framework proposed by Goodfellow et al. (2014). Description of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 342, + 400, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 400, + 353 + ], + "score": 1.0, + "content": "the GAN variants used in our experiments is deferred to the Appendix A.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 330, + 506, + 353 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 302, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 303, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 303, + 374 + ], + "score": 1.0, + "content": "2.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 377, + 505, + 411 + ], + "lines": [ + { + "bbox": [ + 104, + 375, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 122, + 391 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 376, + 167, + 387 + ], + "score": 0.92, + "content": "\\mathcal { X } = \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 375, + 370, + 391 + ], + "score": 1.0, + "content": "be the space of natural images. Given i.i.d. samples", + "type": "text" + }, + { + "bbox": [ + 371, + 378, + 449, + 389 + ], + "score": 0.93, + "content": "S _ { r } = \\{ \\mathbf { x } _ { 1 } ^ { r } , \\ldots , \\mathbf { x } _ { n } ^ { r } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 375, + 506, + 391 + ], + "score": 1.0, + "content": "drawn from a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 387, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 171, + 402 + ], + "score": 1.0, + "content": "real distribution", + "type": "text" + }, + { + "bbox": [ + 171, + 389, + 183, + 399 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 387, + 203, + 402 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 204, + 389, + 213, + 398 + ], + "score": 0.82, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 387, + 419, + 402 + ], + "score": 1.0, + "content": ", we would like to learn a parameterized distribution", + "type": "text" + }, + { + "bbox": [ + 419, + 389, + 431, + 401 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 387, + 505, + 402 + ], + "score": 1.0, + "content": "that approximates", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 399, + 184, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 169, + 411 + ], + "score": 1.0, + "content": "the distribution", + "type": "text" + }, + { + "bbox": [ + 169, + 400, + 181, + 410 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 399, + 184, + 411 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 104, + 375, + 506, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "The setup of generative adversarial networks is as follows. We define two networks, the discriminator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 169, + 439 + ], + "score": 0.92, + "content": "D : \\mathcal { X } [ 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 426, + 244, + 440 + ], + "score": 1.0, + "content": "and the generator", + "type": "text" + }, + { + "bbox": [ + 244, + 428, + 294, + 438 + ], + "score": 0.91, + "content": "G : { \\mathcal { Z } } \\to { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 426, + 325, + 440 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 325, + 428, + 334, + 437 + ], + "score": 0.85, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "is some latent space. Given a distribution", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 118, + 449 + ], + "score": 0.86, + "content": "\\mathbb { P } _ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 437, + 140, + 451 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 140, + 439, + 149, + 448 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 437, + 343, + 451 + ], + "score": 1.0, + "content": "(usually an isotropic Gaussian), the distribution", + "type": "text" + }, + { + "bbox": [ + 343, + 438, + 355, + 451 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 437, + 408, + 451 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + }, + { + "bbox": [ + 409, + 438, + 435, + 450 + ], + "score": 0.94, + "content": "G ( \\mathbb { P } _ { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 437, + 506, + 451 + ], + "score": 1.0, + "content": ". Optimization is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 450, + 309, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 271, + 461 + ], + "score": 1.0, + "content": "performed with respect to a joint loss for", + "type": "text" + }, + { + "bbox": [ + 272, + 450, + 281, + 459 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 450, + 299, + 461 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 450, + 309, + 459 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 415, + 506, + 461 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 464, + 447, + 482 + ], + "lines": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "spans": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { G } \\operatorname* { m a x } _ { D } L ( D , G ) = \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { r } } \\log { [ D ( \\mathbf { x } ) ] } + \\mathbb { E } _ { \\mathbf { z } \\sim \\mathbb { P } _ { z } } \\left[ \\log ( 1 - D ( G ( \\mathbf { z } ) ) ) \\right] .", + "type": "interline_equation", + "image_path": "b2535cc6daf8bf45cb707b25225a313cbe9c6bd2d1e1ec85d1ad919727e0664c.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 464, + 447, + 482 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 506, + 530 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 218, + 498 + ], + "score": 1.0, + "content": "Intuitively, the discriminator", + "type": "text" + }, + { + "bbox": [ + 219, + 486, + 228, + 496 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 486, + 347, + 498 + ], + "score": 1.0, + "content": "outputs a probability for every", + "type": "text" + }, + { + "bbox": [ + 348, + 486, + 376, + 496 + ], + "score": 0.93, + "content": "\\mathbf { x } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "that corresponds to its likelihood", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 497, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 191, + 509 + ], + "score": 1.0, + "content": "of being drawn from", + "type": "text" + }, + { + "bbox": [ + 192, + 497, + 203, + 508 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 497, + 397, + 509 + ], + "score": 1.0, + "content": ", and the loss function encourages the generator", + "type": "text" + }, + { + "bbox": [ + 397, + 497, + 406, + 507 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 497, + 506, + 509 + ], + "score": 1.0, + "content": "to produce samples that", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 506, + 507, + 522 + ], + "spans": [ + { + "bbox": [ + 104, + 506, + 460, + 522 + ], + "score": 1.0, + "content": "maximize this probability. Practically, the loss is approximated with finite samples from", + "type": "text" + }, + { + "bbox": [ + 461, + 508, + 472, + 519 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 506, + 490, + 522 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 491, + 508, + 502, + 520 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 506, + 507, + 522 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 519, + 400, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 267, + 531 + ], + "score": 1.0, + "content": "and optimized with alternating steps for", + "type": "text" + }, + { + "bbox": [ + 268, + 519, + 277, + 528 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 519, + 295, + 531 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 296, + 519, + 304, + 528 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 519, + 400, + 531 + ], + "score": 1.0, + "content": "using gradient descent.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 104, + 486, + 507, + 531 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 504, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 358, + 549 + ], + "score": 1.0, + "content": "To evaluate the generator, we would like to design a metric", + "type": "text" + }, + { + "bbox": [ + 358, + 537, + 365, + 547 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 534, + 504, + 549 + ], + "score": 1.0, + "content": "that measures the “dissimilarity\"", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 545, + 143, + 559 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 144, + 547, + 155, + 559 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 545, + 168, + 559 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 168, + 547, + 180, + 558 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 545, + 448, + 559 + ], + "score": 1.0, + "content": ".1 In theory, with both distributions known, common choices of", + "type": "text" + }, + { + "bbox": [ + 448, + 549, + 455, + 558 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "include the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 557, + 507, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 507, + 570 + ], + "score": 1.0, + "content": "Kullback-Leibler divergence (KLD), Jensen-Shannon divergence (JSD) and total variation. However,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 568, + 507, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 197, + 581 + ], + "score": 1.0, + "content": "in practical scenarios,", + "type": "text" + }, + { + "bbox": [ + 197, + 569, + 208, + 579 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 568, + 380, + 581 + ], + "score": 1.0, + "content": "is unknown and only the finite samples in", + "type": "text" + }, + { + "bbox": [ + 381, + 569, + 392, + 579 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 568, + 507, + 581 + ], + "score": 1.0, + "content": "are observed. Furthermore,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 578, + 509, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 578, + 373, + 593 + ], + "score": 1.0, + "content": "it is almost always intractable to compute the exact density of", + "type": "text" + }, + { + "bbox": [ + 106, + 590, + 215, + 603 + ], + "score": 0.91, + "content": "S _ { g } = \\{ \\mathbf { x } _ { 1 } ^ { g } , . . . , \\mathbf { x } _ { m } ^ { \\bar { g } } \\} \\sim \\mathbb { P } _ { q } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 584, + 509, + 615 + ], + "score": 1.0, + "content": "specially so for GANs). Given these limitations, we focus on empiricalof “dissimilarity\" between samples from two distributions.", + "type": "text" + }, + { + "bbox": [ + 374, + 580, + 385, + 591 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 578, + 506, + 593 + ], + "score": 1.0, + "content": ", but much easier to sample", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 146, + 602, + 226, + 613 + ], + "spans": [ + { + "bbox": [ + 146, + 602, + 226, + 613 + ], + "score": 0.88, + "content": "\\hat { \\rho } : \\mathcal X ^ { n } \\times \\mathcal X ^ { m } \\stackrel { \\smile } { } \\mathbb R", + "type": "inline_equation" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 534, + 509, + 615 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 622, + 239, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 240, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 240, + 633 + ], + "score": 1.0, + "content": "2.2 SAMPLE BASED METRICS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "We mainly focus on sample based evaluation metrics that follow a common setup illustrated in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "Figure 1. The metric calculator is the key element, for which we briefly introduce five representative", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "methods: Inception Score (Salimans et al., 2016), Mode Score (Che et al., 2016) , Kernel MMD (Gret-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "ton et al., 2007), Wasserstein distance, Fréchet Inception Distance (FID) (Heusel et al., 2017), and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "score": 1.0, + "content": "1-nearest neighbor (1-NN)-based two sample test (Lopez-Paz & Oquab, 2016). All of them are model", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 692, + 347, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 347, + 705 + ], + "score": 1.0, + "content": "agnostic and require only finite samples from the generator.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 637, + 506, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "The Inception Score is arguably the most widely adopted metric in the literature. It uses a image", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 191, + 106 + ], + "score": 1.0, + "content": "classification model", + "type": "text" + }, + { + "bbox": [ + 191, + 94, + 204, + 104 + ], + "score": 0.81, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 94, + 505, + 106 + ], + "score": 1.0, + "content": ", the Google Inception network (Szegedy et al., 2016), pre-trained on the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 306, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 306, + 117 + ], + "score": 1.0, + "content": "ImageNet (Deng et al., 2009) dataset, to compute", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 120, + 382, + 136 + ], + "lines": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "spans": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "score": 0.92, + "content": "\\mathrm { I S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] } ,", + "type": "interline_equation", + "image_path": "4ec726016605aea0fafe96847d73ac263947992e772b07fd92d445d3821860a7.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 139, + 506, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 507, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 132, + 156 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 140, + 169, + 153 + ], + "score": 0.93, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 137, + 294, + 156 + ], + "score": 1.0, + "content": "denotes the label distribution of", + "type": "text" + }, + { + "bbox": [ + 294, + 142, + 301, + 150 + ], + "score": 0.74, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 137, + 362, + 156 + ], + "score": 1.0, + "content": "as predicted by", + "type": "text" + }, + { + "bbox": [ + 362, + 141, + 375, + 150 + ], + "score": 0.82, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 137, + 394, + 156 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 394, + 140, + 502, + 154 + ], + "score": 0.92, + "content": "\\begin{array} { r } { p _ { \\mathcal M } ( y ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { g } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 137, + 507, + 156 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 186, + 166 + ], + "score": 1.0, + "content": "i.e. the marginal of", + "type": "text" + }, + { + "bbox": [ + 187, + 152, + 225, + 164 + ], + "score": 0.93, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 151, + 344, + 166 + ], + "score": 1.0, + "content": "over the probability measure", + "type": "text" + }, + { + "bbox": [ + 345, + 153, + 357, + 165 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 151, + 506, + 166 + ], + "score": 1.0, + "content": ". The expectation and the integral in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 144, + 176 + ], + "score": 0.92, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 163, + 337, + 177 + ], + "score": 1.0, + "content": "can be approximated with i.i.d. samples from", + "type": "text" + }, + { + "bbox": [ + 337, + 164, + 349, + 176 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 163, + 423, + 177 + ], + "score": 1.0, + "content": ". A higher IS has", + "type": "text" + }, + { + "bbox": [ + 424, + 164, + 461, + 175 + ], + "score": 0.95, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "close to a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 173, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 506, + 189 + ], + "score": 1.0, + "content": "point mass, which happens when the Inception network is very confident that the image belongs", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 281, + 198 + ], + "score": 1.0, + "content": "to a particular ImageNet category, and has", + "type": "text" + }, + { + "bbox": [ + 282, + 185, + 311, + 198 + ], + "score": 0.92, + "content": "p _ { \\mathcal { M } } ( y )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "close to uniform, i.e. all categories are equally", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "score": 1.0, + "content": "represented. This suggests that the generative model has both high quality and diversity. Salimans", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "et al. (2016) show that the Inception Score has a reasonable correlation with human judgment of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 218, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 505, + 230 + ], + "score": 1.0, + "content": "image quality. We would like to highlight two specific properties: 1) the distributions on both sides", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 229, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 219, + 242 + ], + "score": 1.0, + "content": "of the KL are dependent on", + "type": "text" + }, + { + "bbox": [ + 220, + 230, + 232, + 240 + ], + "score": 0.82, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 229, + 390, + 242 + ], + "score": 1.0, + "content": ", and 2) the distribution of the real data", + "type": "text" + }, + { + "bbox": [ + 390, + 230, + 402, + 240 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 229, + 506, + 242 + ], + "score": 1.0, + "content": ", or even samples thereof,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 241, + 201, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 201, + 253 + ], + "score": 1.0, + "content": "are not used anywhere.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 105, + 256, + 459, + 269 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 460, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 460, + 272 + ], + "score": 1.0, + "content": "The Mode Score is an improved version of the Inception Score. Formally, it is given by", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 273, + 424, + 289 + ], + "lines": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "spans": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "score": 0.91, + "content": "\\mathrm { M S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] - K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( y ^ { \\ast } ) ) } ,", + "type": "interline_equation", + "image_path": "3a9fd389181ebcbf46f515563540c7cf132b0297f5785f5eac2de30fbe71678b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 292, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 133, + 305 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 293, + 248, + 306 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\mathcal M } ( y ^ { \\ast } ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { r } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "is the marginal label distribution for the samples from the real", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "data distribution. Unlike the Inception Score, it is able to measure the dissimilarity between the real", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 314, + 451, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 154, + 328 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 316, + 166, + 326 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 314, + 274, + 328 + ], + "score": 1.0, + "content": "and generated distribution", + "type": "text" + }, + { + "bbox": [ + 274, + 316, + 286, + 327 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 314, + 356, + 328 + ], + "score": 1.0, + "content": "through the term", + "type": "text" + }, + { + "bbox": [ + 357, + 315, + 447, + 327 + ], + "score": 0.91, + "content": "K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( \\bar { y } ^ { * } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 314, + 451, + 328 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 358, + 344 + ], + "lines": [ + { + "bbox": [ + 104, + 329, + 360, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 360, + 347 + ], + "score": 1.0, + "content": "The Kernel MMD (Maximum Mean Discrepancy), defined as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 348, + 456, + 387 + ], + "lines": [ + { + "bbox": [ + 154, + 348, + 456, + 387 + ], + "spans": [ + { + "bbox": [ + 154, + 348, + 456, + 387 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathrm { M M D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\biggl ( \\mathbb { E } _ { \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } \\sim \\mathbb { P } _ { r } , } \\biggl [ k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } ) - 2 k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { g } ) + k ( \\mathbf { x } _ { g } , \\mathbf { x } _ { g } ^ { \\prime } ) \\biggr ] \\biggr ) ^ { \\frac { 1 } { 2 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "f3788bd082c1f55a957a0105ce3effdad449912b13c637778f4d639a70f7588e.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 154, + 348, + 456, + 361.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 154, + 361.0, + 456, + 374.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 154, + 374.0, + 456, + 387.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 251, + 403 + ], + "score": 1.0, + "content": "measures the dissimilarity between", + "type": "text" + }, + { + "bbox": [ + 252, + 391, + 263, + 402 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 390, + 281, + 403 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 282, + 391, + 293, + 403 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 390, + 420, + 403 + ], + "score": 1.0, + "content": "for some fixed kernel function", + "type": "text" + }, + { + "bbox": [ + 420, + 391, + 427, + 401 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 390, + 506, + 403 + ], + "score": 1.0, + "content": ". Given two sets of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 163, + 414 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 163, + 402, + 175, + 413 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 402, + 193, + 414 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 402, + 205, + 414 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 402, + 505, + 414 + ], + "score": 1.0, + "content": ", the empirical MMD between the two distributions can be computed with", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 412, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 410, + 426 + ], + "score": 1.0, + "content": "finite sample approximation of the expectation. A lower MMD means that", + "type": "text" + }, + { + "bbox": [ + 411, + 413, + 423, + 425 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 412, + 470, + 426 + ], + "score": 1.0, + "content": "is closer to", + "type": "text" + }, + { + "bbox": [ + 470, + 413, + 482, + 424 + ], + "score": 0.86, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 412, + 506, + 426 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 497, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 497, + 436 + ], + "score": 1.0, + "content": "Parzen window estimate (Gretton et al., 2007) can be viewed as a specialization of Kernel MMD.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 440, + 346, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 347, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 251, + 454 + ], + "score": 1.0, + "content": "The Wasserstein distance between", + "type": "text" + }, + { + "bbox": [ + 252, + 441, + 263, + 452 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 438, + 281, + 454 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 281, + 441, + 293, + 453 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 438, + 347, + 454 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 457, + 407, + 477 + ], + "lines": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "spans": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "score": 0.91, + "content": "\\operatorname { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\operatorname* { i n f } _ { \\substack { \\gamma \\in \\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) } } \\mathbb { E } _ { ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\sim \\gamma } \\left[ d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\right] ,", + "type": "interline_equation", + "image_path": "9d816f4d0c7692fe62f46391ab21fad24983fbd57ad4d0b9d96457eff4b54a80.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 132, + 494 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 482, + 172, + 495 + ], + "score": 0.93, + "content": "\\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "denotes the set of all joint distributions (i.e. probabilistic couplings) whose marginals", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 172, + 506 + ], + "score": 1.0, + "content": "are respectively", + "type": "text" + }, + { + "bbox": [ + 172, + 494, + 184, + 504 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 493, + 202, + 506 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 494, + 214, + 506 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 493, + 236, + 506 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 236, + 493, + 275, + 505 + ], + "score": 0.93, + "content": "d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "denotes the base distance between the two samples. For", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 253, + 518 + ], + "score": 1.0, + "content": "discrete distributions with densities", + "type": "text" + }, + { + "bbox": [ + 253, + 506, + 264, + 516 + ], + "score": 0.86, + "content": "p _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 504, + 282, + 518 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 506, + 293, + 517 + ], + "score": 0.87, + "content": "p _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 504, + 506, + 518 + ], + "score": 1.0, + "content": ", the Wasserstein distance is often referred to as the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 514, + 496, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 496, + 528 + ], + "score": 1.0, + "content": "Earth Mover’s Distance (EMD), and corresponds to the solution to the optimal transport problem", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 531, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 111, + 531, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 111, + 531, + 505, + 565 + ], + "score": 0.91, + "content": "N \\mathrm { D } ( p _ { r } , p _ { g } ) = \\operatorname* { m i n } _ { w \\in \\mathbb { R } ^ { n \\times m } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { m } w _ { i j } d ( \\mathbf { x } _ { i } ^ { r } , \\mathbf { x } _ { j } ^ { g } ) \\quad \\mathrm { s . t . } \\quad \\sum _ { j = 1 } ^ { m } w _ { i , j } = p _ { r } ( \\mathbf { x } _ { i } ^ { r } ) \\ \\forall i , \\sum _ { i = 1 } ^ { n } w _ { i , j } = p _ { g } ( \\mathbf { x } _ { j } ^ { g } ) \\ \\forall j .", + "type": "interline_equation", + "image_path": "fae6004402fd30eadb74ce6936dd7f6f8715a2b25aee2749b8d04e6103c21bdf.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 111, + 531, + 505, + 542.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 111, + 542.3333333333334, + 505, + 553.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 111, + 553.6666666666667, + 505, + 565.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 285, + 586 + ], + "score": 1.0, + "content": "This is the finite sample approximation of", + "type": "text" + }, + { + "bbox": [ + 286, + 574, + 337, + 586 + ], + "score": 0.92, + "content": "\\mathrm { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "used in practice. Similar to MMD, the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 585, + 388, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 388, + 597 + ], + "score": 1.0, + "content": "Wasserstein distance is lower when two distributions are more similar.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 503, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "The Fréchet Inception Distance (FID) was recently introduced by Heusel et al. (2017) to evaluate", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 232, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 232, + 626 + ], + "score": 1.0, + "content": "GANs. Formally, it is given by", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 629, + 426, + 645 + ], + "lines": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "spans": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "score": 0.88, + "content": "\\mathrm { F I D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\| \\mu _ { r } - \\mu _ { g } \\| + \\operatorname { T r } ( \\mathbf { C } _ { r } + \\mathbf { C } _ { g } - 2 ( \\mathbf { C } _ { r } \\mathbf { C } _ { g } ) ^ { 1 / 2 } ) ,", + "type": "interline_equation", + "image_path": "e52bdd7bfba7d3d3be37ebb5b8a714ad2369afe19286234f8745756614f42ae5.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 506, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 132, + 662 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 649, + 164, + 661 + ], + "score": 0.47, + "content": "\\mu _ { r } \\left( \\mu _ { g } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 648, + 182, + 662 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 182, + 649, + 196, + 660 + ], + "score": 0.7, + "content": "\\mathbf { C } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 649, + 217, + 661 + ], + "score": 0.62, + "content": "( \\mathbf { C } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "are the mean and covariance of the real (generated) distribution, respec-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 327, + 673 + ], + "score": 1.0, + "content": "tively. Note that under the Gaussian assumption on both", + "type": "text" + }, + { + "bbox": [ + 328, + 660, + 339, + 671 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 659, + 357, + 673 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 357, + 660, + 369, + 672 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 659, + 506, + 673 + ], + "score": 1.0, + "content": ", the Fréchet distance is equivalent", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 669, + 228, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 228, + 683 + ], + "score": 1.0, + "content": "to the Wasserstein-2 distance.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "The 1-Nearest Neighbor classifier is used in two-sample tests to assess whether two distributions", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 696, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 104, + 696, + 267, + 712 + ], + "score": 1.0, + "content": "are identical. Given two sets of samples", + "type": "text" + }, + { + "bbox": [ + 267, + 699, + 303, + 711 + ], + "score": 0.91, + "content": "S _ { r } \\sim \\mathbb { P } _ { r } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 696, + 322, + 712 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 322, + 699, + 361, + 712 + ], + "score": 0.9, + "content": "S _ { g } \\sim \\mathbb { P } _ { g } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 696, + 385, + 712 + ], + "score": 1.0, + "content": ", with", + "type": "text" + }, + { + "bbox": [ + 385, + 699, + 431, + 711 + ], + "score": 0.91, + "content": "| S _ { r } | = | S _ { g } |", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 696, + 506, + 712 + ], + "score": 1.0, + "content": ", one can compute", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 378, + 722 + ], + "score": 1.0, + "content": "the leave-one-out (LOO) accuracy of a 1-NN classifier trained on", + "type": "text" + }, + { + "bbox": [ + 379, + 711, + 390, + 721 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 709, + 409, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 410, + 711, + 421, + 722 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "with positive labels", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 120, + 733 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 721, + 132, + 732 + ], + "score": 0.89, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 720, + 227, + 733 + ], + "score": 1.0, + "content": "and negative labels for", + "type": "text" + }, + { + "bbox": [ + 227, + 721, + 239, + 733 + ], + "score": 0.89, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 720, + 505, + 733 + ], + "score": 1.0, + "content": ". Different from the most common use of accuracy, here the 1-NN", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "The Inception Score is arguably the most widely adopted metric in the literature. It uses a image", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 191, + 106 + ], + "score": 1.0, + "content": "classification model", + "type": "text" + }, + { + "bbox": [ + 191, + 94, + 204, + 104 + ], + "score": 0.81, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 94, + 505, + 106 + ], + "score": 1.0, + "content": ", the Google Inception network (Szegedy et al., 2016), pre-trained on the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 306, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 306, + 117 + ], + "score": 1.0, + "content": "ImageNet (Deng et al., 2009) dataset, to compute", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 506, + 117 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 120, + 382, + 136 + ], + "lines": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "spans": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "score": 0.92, + "content": "\\mathrm { I S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] } ,", + "type": "interline_equation", + "image_path": "4ec726016605aea0fafe96847d73ac263947992e772b07fd92d445d3821860a7.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 228, + 120, + 382, + 136 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 139, + 506, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 507, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 132, + 156 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 140, + 169, + 153 + ], + "score": 0.93, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 137, + 294, + 156 + ], + "score": 1.0, + "content": "denotes the label distribution of", + "type": "text" + }, + { + "bbox": [ + 294, + 142, + 301, + 150 + ], + "score": 0.74, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 137, + 362, + 156 + ], + "score": 1.0, + "content": "as predicted by", + "type": "text" + }, + { + "bbox": [ + 362, + 141, + 375, + 150 + ], + "score": 0.82, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 137, + 394, + 156 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 394, + 140, + 502, + 154 + ], + "score": 0.92, + "content": "\\begin{array} { r } { p _ { \\mathcal M } ( y ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { g } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 137, + 507, + 156 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 186, + 166 + ], + "score": 1.0, + "content": "i.e. the marginal of", + "type": "text" + }, + { + "bbox": [ + 187, + 152, + 225, + 164 + ], + "score": 0.93, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 151, + 344, + 166 + ], + "score": 1.0, + "content": "over the probability measure", + "type": "text" + }, + { + "bbox": [ + 345, + 153, + 357, + 165 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 151, + 506, + 166 + ], + "score": 1.0, + "content": ". The expectation and the integral in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 144, + 176 + ], + "score": 0.92, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 163, + 337, + 177 + ], + "score": 1.0, + "content": "can be approximated with i.i.d. samples from", + "type": "text" + }, + { + "bbox": [ + 337, + 164, + 349, + 176 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 163, + 423, + 177 + ], + "score": 1.0, + "content": ". A higher IS has", + "type": "text" + }, + { + "bbox": [ + 424, + 164, + 461, + 175 + ], + "score": 0.95, + "content": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "close to a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 173, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 506, + 189 + ], + "score": 1.0, + "content": "point mass, which happens when the Inception network is very confident that the image belongs", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 281, + 198 + ], + "score": 1.0, + "content": "to a particular ImageNet category, and has", + "type": "text" + }, + { + "bbox": [ + 282, + 185, + 311, + 198 + ], + "score": 0.92, + "content": "p _ { \\mathcal { M } } ( y )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "close to uniform, i.e. all categories are equally", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "score": 1.0, + "content": "represented. This suggests that the generative model has both high quality and diversity. Salimans", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "et al. (2016) show that the Inception Score has a reasonable correlation with human judgment of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 218, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 505, + 230 + ], + "score": 1.0, + "content": "image quality. We would like to highlight two specific properties: 1) the distributions on both sides", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 229, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 219, + 242 + ], + "score": 1.0, + "content": "of the KL are dependent on", + "type": "text" + }, + { + "bbox": [ + 220, + 230, + 232, + 240 + ], + "score": 0.82, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 229, + 390, + 242 + ], + "score": 1.0, + "content": ", and 2) the distribution of the real data", + "type": "text" + }, + { + "bbox": [ + 390, + 230, + 402, + 240 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 229, + 506, + 242 + ], + "score": 1.0, + "content": ", or even samples thereof,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 241, + 201, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 201, + 253 + ], + "score": 1.0, + "content": "are not used anywhere.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 137, + 507, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 256, + 459, + 269 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 460, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 460, + 272 + ], + "score": 1.0, + "content": "The Mode Score is an improved version of the Inception Score. Formally, it is given by", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 254, + 460, + 272 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 273, + 424, + 289 + ], + "lines": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "spans": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "score": 0.91, + "content": "\\mathrm { M S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] - K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( y ^ { \\ast } ) ) } ,", + "type": "interline_equation", + "image_path": "3a9fd389181ebcbf46f515563540c7cf132b0297f5785f5eac2de30fbe71678b.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 185, + 273, + 424, + 289 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 292, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 133, + 305 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 293, + 248, + 306 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\mathcal M } ( y ^ { \\ast } ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { r } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "is the marginal label distribution for the samples from the real", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "data distribution. Unlike the Inception Score, it is able to measure the dissimilarity between the real", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 314, + 451, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 154, + 328 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 316, + 166, + 326 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 314, + 274, + 328 + ], + "score": 1.0, + "content": "and generated distribution", + "type": "text" + }, + { + "bbox": [ + 274, + 316, + 286, + 327 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 314, + 356, + 328 + ], + "score": 1.0, + "content": "through the term", + "type": "text" + }, + { + "bbox": [ + 357, + 315, + 447, + 327 + ], + "score": 0.91, + "content": "K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( \\bar { y } ^ { * } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 314, + 451, + 328 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 292, + 506, + 328 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 358, + 344 + ], + "lines": [ + { + "bbox": [ + 104, + 329, + 360, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 360, + 347 + ], + "score": 1.0, + "content": "The Kernel MMD (Maximum Mean Discrepancy), defined as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 104, + 329, + 360, + 347 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 348, + 456, + 387 + ], + "lines": [ + { + "bbox": [ + 154, + 348, + 456, + 387 + ], + "spans": [ + { + "bbox": [ + 154, + 348, + 456, + 387 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathrm { M M D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\biggl ( \\mathbb { E } _ { \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } \\sim \\mathbb { P } _ { r } , } \\biggl [ k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } ) - 2 k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { g } ) + k ( \\mathbf { x } _ { g } , \\mathbf { x } _ { g } ^ { \\prime } ) \\biggr ] \\biggr ) ^ { \\frac { 1 } { 2 } } , } \\end{array}", + "type": "interline_equation", + "image_path": "f3788bd082c1f55a957a0105ce3effdad449912b13c637778f4d639a70f7588e.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 154, + 348, + 456, + 361.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 154, + 361.0, + 456, + 374.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 154, + 374.0, + 456, + 387.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 251, + 403 + ], + "score": 1.0, + "content": "measures the dissimilarity between", + "type": "text" + }, + { + "bbox": [ + 252, + 391, + 263, + 402 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 390, + 281, + 403 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 282, + 391, + 293, + 403 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 390, + 420, + 403 + ], + "score": 1.0, + "content": "for some fixed kernel function", + "type": "text" + }, + { + "bbox": [ + 420, + 391, + 427, + 401 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 390, + 506, + 403 + ], + "score": 1.0, + "content": ". Given two sets of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 163, + 414 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 163, + 402, + 175, + 413 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 402, + 193, + 414 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 402, + 205, + 414 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 402, + 505, + 414 + ], + "score": 1.0, + "content": ", the empirical MMD between the two distributions can be computed with", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 412, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 410, + 426 + ], + "score": 1.0, + "content": "finite sample approximation of the expectation. A lower MMD means that", + "type": "text" + }, + { + "bbox": [ + 411, + 413, + 423, + 425 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 412, + 470, + 426 + ], + "score": 1.0, + "content": "is closer to", + "type": "text" + }, + { + "bbox": [ + 470, + 413, + 482, + 424 + ], + "score": 0.86, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 412, + 506, + 426 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 497, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 497, + 436 + ], + "score": 1.0, + "content": "Parzen window estimate (Gretton et al., 2007) can be viewed as a specialization of Kernel MMD.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 390, + 506, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 440, + 346, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 347, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 251, + 454 + ], + "score": 1.0, + "content": "The Wasserstein distance between", + "type": "text" + }, + { + "bbox": [ + 252, + 441, + 263, + 452 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 438, + 281, + 454 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 281, + 441, + 293, + 453 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 438, + 347, + 454 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 438, + 347, + 454 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 457, + 407, + 477 + ], + "lines": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "spans": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "score": 0.91, + "content": "\\operatorname { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\operatorname* { i n f } _ { \\substack { \\gamma \\in \\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) } } \\mathbb { E } _ { ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\sim \\gamma } \\left[ d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\right] ,", + "type": "interline_equation", + "image_path": "9d816f4d0c7692fe62f46391ab21fad24983fbd57ad4d0b9d96457eff4b54a80.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 204, + 457, + 407, + 477 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 132, + 494 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 482, + 172, + 495 + ], + "score": 0.93, + "content": "\\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "denotes the set of all joint distributions (i.e. probabilistic couplings) whose marginals", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 172, + 506 + ], + "score": 1.0, + "content": "are respectively", + "type": "text" + }, + { + "bbox": [ + 172, + 494, + 184, + 504 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 493, + 202, + 506 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 494, + 214, + 506 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 493, + 236, + 506 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 236, + 493, + 275, + 505 + ], + "score": 0.93, + "content": "d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "denotes the base distance between the two samples. For", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 253, + 518 + ], + "score": 1.0, + "content": "discrete distributions with densities", + "type": "text" + }, + { + "bbox": [ + 253, + 506, + 264, + 516 + ], + "score": 0.86, + "content": "p _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 504, + 282, + 518 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 506, + 293, + 517 + ], + "score": 0.87, + "content": "p _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 504, + 506, + 518 + ], + "score": 1.0, + "content": ", the Wasserstein distance is often referred to as the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 514, + 496, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 496, + 528 + ], + "score": 1.0, + "content": "Earth Mover’s Distance (EMD), and corresponds to the solution to the optimal transport problem", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 482, + 506, + 528 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 531, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 111, + 531, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 111, + 531, + 505, + 565 + ], + "score": 0.91, + "content": "N \\mathrm { D } ( p _ { r } , p _ { g } ) = \\operatorname* { m i n } _ { w \\in \\mathbb { R } ^ { n \\times m } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { m } w _ { i j } d ( \\mathbf { x } _ { i } ^ { r } , \\mathbf { x } _ { j } ^ { g } ) \\quad \\mathrm { s . t . } \\quad \\sum _ { j = 1 } ^ { m } w _ { i , j } = p _ { r } ( \\mathbf { x } _ { i } ^ { r } ) \\ \\forall i , \\sum _ { i = 1 } ^ { n } w _ { i , j } = p _ { g } ( \\mathbf { x } _ { j } ^ { g } ) \\ \\forall j .", + "type": "interline_equation", + "image_path": "fae6004402fd30eadb74ce6936dd7f6f8715a2b25aee2749b8d04e6103c21bdf.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 111, + 531, + 505, + 542.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 111, + 542.3333333333334, + 505, + 553.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 111, + 553.6666666666667, + 505, + 565.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 285, + 586 + ], + "score": 1.0, + "content": "This is the finite sample approximation of", + "type": "text" + }, + { + "bbox": [ + 286, + 574, + 337, + 586 + ], + "score": 0.92, + "content": "\\mathrm { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "used in practice. Similar to MMD, the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 585, + 388, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 388, + 597 + ], + "score": 1.0, + "content": "Wasserstein distance is lower when two distributions are more similar.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 573, + 505, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 503, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "The Fréchet Inception Distance (FID) was recently introduced by Heusel et al. (2017) to evaluate", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 232, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 232, + 626 + ], + "score": 1.0, + "content": "GANs. Formally, it is given by", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 602, + 505, + 626 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 629, + 426, + 645 + ], + "lines": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "spans": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "score": 0.88, + "content": "\\mathrm { F I D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\| \\mu _ { r } - \\mu _ { g } \\| + \\operatorname { T r } ( \\mathbf { C } _ { r } + \\mathbf { C } _ { g } - 2 ( \\mathbf { C } _ { r } \\mathbf { C } _ { g } ) ^ { 1 / 2 } ) ,", + "type": "interline_equation", + "image_path": "e52bdd7bfba7d3d3be37ebb5b8a714ad2369afe19286234f8745756614f42ae5.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 184, + 629, + 426, + 645 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 506, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 132, + 662 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 649, + 164, + 661 + ], + "score": 0.47, + "content": "\\mu _ { r } \\left( \\mu _ { g } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 648, + 182, + 662 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 182, + 649, + 196, + 660 + ], + "score": 0.7, + "content": "\\mathbf { C } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 649, + 217, + 661 + ], + "score": 0.62, + "content": "( \\mathbf { C } _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "are the mean and covariance of the real (generated) distribution, respec-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 327, + 673 + ], + "score": 1.0, + "content": "tively. Note that under the Gaussian assumption on both", + "type": "text" + }, + { + "bbox": [ + 328, + 660, + 339, + 671 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 659, + 357, + 673 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 357, + 660, + 369, + 672 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 659, + 506, + 673 + ], + "score": 1.0, + "content": ", the Fréchet distance is equivalent", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 669, + 228, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 228, + 683 + ], + "score": 1.0, + "content": "to the Wasserstein-2 distance.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 648, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "The 1-Nearest Neighbor classifier is used in two-sample tests to assess whether two distributions", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 696, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 104, + 696, + 267, + 712 + ], + "score": 1.0, + "content": "are identical. Given two sets of samples", + "type": "text" + }, + { + "bbox": [ + 267, + 699, + 303, + 711 + ], + "score": 0.91, + "content": "S _ { r } \\sim \\mathbb { P } _ { r } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 696, + 322, + 712 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 322, + 699, + 361, + 712 + ], + "score": 0.9, + "content": "S _ { g } \\sim \\mathbb { P } _ { g } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 696, + 385, + 712 + ], + "score": 1.0, + "content": ", with", + "type": "text" + }, + { + "bbox": [ + 385, + 699, + 431, + 711 + ], + "score": 0.91, + "content": "| S _ { r } | = | S _ { g } |", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 696, + 506, + 712 + ], + "score": 1.0, + "content": ", one can compute", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 378, + 722 + ], + "score": 1.0, + "content": "the leave-one-out (LOO) accuracy of a 1-NN classifier trained on", + "type": "text" + }, + { + "bbox": [ + 379, + 711, + 390, + 721 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 709, + 409, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 410, + 711, + 421, + 722 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "with positive labels", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 120, + 733 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 721, + 132, + 732 + ], + "score": 0.89, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 720, + 227, + 733 + ], + "score": 1.0, + "content": "and negative labels for", + "type": "text" + }, + { + "bbox": [ + 227, + 721, + 239, + 733 + ], + "score": 0.89, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 720, + 505, + 733 + ], + "score": 1.0, + "content": ". Different from the most common use of accuracy, here the 1-NN", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 202, + 97 + ], + "score": 1.0, + "content": "classifier should yield a", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 203, + 82, + 231, + 93 + ], + "score": 0.89, + "content": "\\sim 5 0 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 231, + 82, + 317, + 97 + ], + "score": 1.0, + "content": "LOO accuracy when", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 317, + 82, + 364, + 95 + ], + "score": 0.92, + "content": "| S _ { r } | = | S _ { g } |", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 364, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "is large. This is achieved when the", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 357, + 106 + ], + "score": 1.0, + "content": "two distributions match. The LOO accuracy can be lower than", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 357, + 94, + 376, + 104 + ], + "score": 0.88, + "content": "5 0 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 377, + 93, + 505, + 106 + ], + "score": 1.0, + "content": ", which happens when the GAN", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 139, + 117 + ], + "score": 1.0, + "content": "overfits", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 139, + 105, + 151, + 117 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 152, + 105, + 163, + 117 + ], + "score": 1.0, + "content": "to", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 163, + 105, + 175, + 116 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 175, + 105, + 505, + 117 + ], + "score": 1.0, + "content": ". In the (hypothetical) extreme case, if the GAN were to memorize every sample", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 117, + 128 + ], + "score": 1.0, + "content": "in", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 118, + 116, + 129, + 127 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 129, + 115, + 257, + 128 + ], + "score": 1.0, + "content": "and re-generate it exactly, i.e.", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 258, + 115, + 295, + 127 + ], + "score": 0.91, + "content": "S _ { g } = S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 295, + 115, + 395, + 128 + ], + "score": 1.0, + "content": ", the accuracy would be", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 395, + 115, + 410, + 126 + ], + "score": 0.88, + "content": "0 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 410, + 115, + 505, + 128 + ], + "score": 1.0, + "content": ", as every sample from", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 107, + 127, + 118, + 137 + ], + "score": 0.86, + "content": "S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 118, + 126, + 272, + 140 + ], + "score": 1.0, + "content": "would have it nearest neighbour from", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 272, + 127, + 285, + 138 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 285, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "with zero distance. The 1-NN classifier belongs to the", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "two-sample test family, for which any binary classifier can be adopted in principle. We will only", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 162 + ], + "score": 1.0, + "content": "consider the 1-NN classifier because it requires no special training and little hyperparameter tuning.", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 45.5, + "bbox_fs": [ + 104, + 686, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 202, + 97 + ], + "score": 1.0, + "content": "classifier should yield a", + "type": "text" + }, + { + "bbox": [ + 203, + 82, + 231, + 93 + ], + "score": 0.89, + "content": "\\sim 5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 82, + 317, + 97 + ], + "score": 1.0, + "content": "LOO accuracy when", + "type": "text" + }, + { + "bbox": [ + 317, + 82, + 364, + 95 + ], + "score": 0.92, + "content": "| S _ { r } | = | S _ { g } |", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "is large. This is achieved when the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 357, + 106 + ], + "score": 1.0, + "content": "two distributions match. The LOO accuracy can be lower than", + "type": "text" + }, + { + "bbox": [ + 357, + 94, + 376, + 104 + ], + "score": 0.88, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 93, + 505, + 106 + ], + "score": 1.0, + "content": ", which happens when the GAN", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 139, + 117 + ], + "score": 1.0, + "content": "overfits", + "type": "text" + }, + { + "bbox": [ + 139, + 105, + 151, + 117 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 105, + 163, + 117 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 163, + 105, + 175, + 116 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 105, + 505, + 117 + ], + "score": 1.0, + "content": ". In the (hypothetical) extreme case, if the GAN were to memorize every sample", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 117, + 128 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 118, + 116, + 129, + 127 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 115, + 257, + 128 + ], + "score": 1.0, + "content": "and re-generate it exactly, i.e.", + "type": "text" + }, + { + "bbox": [ + 258, + 115, + 295, + 127 + ], + "score": 0.91, + "content": "S _ { g } = S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 115, + 395, + 128 + ], + "score": 1.0, + "content": ", the accuracy would be", + "type": "text" + }, + { + "bbox": [ + 395, + 115, + 410, + 126 + ], + "score": 0.88, + "content": "0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 115, + 505, + 128 + ], + "score": 1.0, + "content": ", as every sample from", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 107, + 127, + 118, + 137 + ], + "score": 0.86, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 126, + 272, + 140 + ], + "score": 1.0, + "content": "would have it nearest neighbour from", + "type": "text" + }, + { + "bbox": [ + 272, + 127, + 285, + 138 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "with zero distance. The 1-NN classifier belongs to the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "two-sample test family, for which any binary classifier can be adopted in principle. We will only", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 162 + ], + "score": 1.0, + "content": "consider the 1-NN classifier because it requires no special training and little hyperparameter tuning.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 165, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "score": 1.0, + "content": "Lopez-Paz & Oquab (2016) considered the 1-NN accuracy primarily as a statistic for two-sample", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 506, + 190 + ], + "score": 1.0, + "content": "testing. In fact, it is more informative to analyze it for the two classes separately. For example, a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "typical outcome of GANs is that for both real and generated images, the majority of their nearest", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "neighbors are generated images due to mode collapse. In this case, the LOO 1-NN accuracy of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "real images would be relatively low (desired): the mode(s) of the real distribution are usually well", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 378, + 233 + ], + "score": 1.0, + "content": "captured by the generative model, so a majority of real samples from", + "type": "text" + }, + { + "bbox": [ + 379, + 221, + 390, + 231 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "are surrounded by generated", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 162, + 244 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 162, + 231, + 174, + 243 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 231, + 506, + 244 + ], + "score": 1.0, + "content": ", leading to low LOO accuracy; whereas the LOO accuracy of the generated images", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "is high (not desired): generative samples tend to collapse to a few mode centers, thus they are", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 507, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 507, + 267 + ], + "score": 1.0, + "content": "surrounded by samples from the same class, leading to high LOO accuracy. For the rest of the paper,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 436, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 436, + 276 + ], + "score": 1.0, + "content": "we distinguish these two cases as 1-NN accuracy (real) and 1-NN accuracy (fake).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 108, + 291, + 202, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 204, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 204, + 304 + ], + "score": 1.0, + "content": "2.3 OTHER METRICS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 310, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 506, + 323 + ], + "score": 1.0, + "content": "All of the metrics above are, what we refer to as “model agnostic\": they use the generator as a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 277, + 335 + ], + "score": 1.0, + "content": "black box to sample the generated images", + "type": "text" + }, + { + "bbox": [ + 277, + 322, + 289, + 334 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 320, + 505, + 335 + ], + "score": 1.0, + "content": ". Model agnostic metrics should not require a density", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "estimation from the model. We choose to only experiment with model agnostic metrics, which allow", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "score": 1.0, + "content": "us to support as many generative models as possible for evaluation without modification to their", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 355, + 497, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 497, + 367 + ], + "score": 1.0, + "content": "structure. We will briefly mention some other evaluation metrics not included in our experiments.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "Kernel density estimation (KDE, or Parzen window estimation) is a well-studied method for estimating", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 382, + 506, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 429, + 395 + ], + "score": 1.0, + "content": "the density function of a distribution from samples. For a probability kernel", + "type": "text" + }, + { + "bbox": [ + 429, + 383, + 440, + 393 + ], + "score": 0.75, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 382, + 506, + 395 + ], + "score": 1.0, + "content": "(most often an", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 255, + 406 + ], + "score": 1.0, + "content": "isotropic Gaussian) and i.i.d samples", + "type": "text" + }, + { + "bbox": [ + 255, + 395, + 301, + 405 + ], + "score": 0.9, + "content": "\\mathbf { x } _ { 1 } , \\ldots , \\mathbf { x } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 393, + 453, + 406 + ], + "score": 1.0, + "content": ", we can define the density function at", + "type": "text" + }, + { + "bbox": [ + 454, + 395, + 462, + 403 + ], + "score": 0.48, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 393, + 473, + 406 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 473, + 393, + 505, + 405 + ], + "score": 0.91, + "content": "p ( \\mathbf { x } ) \\approx", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 107, + 403, + 186, + 417 + ], + "score": 0.89, + "content": "\\textstyle { \\frac { 1 } { z } } \\sum _ { i = 1 } ^ { n ^ { \\cdot } } K ( \\mathbf { x } - \\mathbf { x } _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 404, + 216, + 416 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 216, + 406, + 223, + 414 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "is a normalizing constant. This allows the use of classical metrics such", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 415, + 507, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 507, + 428 + ], + "score": 1.0, + "content": "as KLD and JSD. However, despite the widespread adoption of this technique to various applications,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 426, + 507, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 277, + 440 + ], + "score": 1.0, + "content": "its suitability to estimating the density of", + "type": "text" + }, + { + "bbox": [ + 278, + 426, + 289, + 437 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 426, + 302, + 440 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 302, + 426, + 315, + 438 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 426, + 507, + 440 + ], + "score": 1.0, + "content": "for GANs has been questioned by Theis et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 438, + 454, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 454, + 450 + ], + "score": 1.0, + "content": "(2015) since the probability kernel depends on the Euclidean distance between images.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "More recently, Wu et al. (2016) applied annealed importance sampling (AIS) to estimate the marginal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 464, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 156, + 478 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 157, + 465, + 177, + 477 + ], + "score": 0.91, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 464, + 506, + 478 + ], + "score": 1.0, + "content": "of a generative model. This method is most natural for models that define a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 200, + 488 + ], + "score": 1.0, + "content": "conditional distribution", + "type": "text" + }, + { + "bbox": [ + 200, + 476, + 227, + 488 + ], + "score": 0.93, + "content": "p ( \\mathbf { x } | \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 475, + 254, + 488 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 255, + 478, + 262, + 486 + ], + "score": 0.66, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "is the latent code, which is not satisfied by most GAN models.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Nevertheless, AIS has been applied to GAN evaluation by assuming a Gaussian observation model.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "We exclude this method from our experiments as it needs the access to the generative model to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 507, + 411, + 524 + ], + "spans": [ + { + "bbox": [ + 104, + 507, + 395, + 524 + ], + "score": 1.0, + "content": "compute the likelihood, instead of only depending on a finite sample set", + "type": "text" + }, + { + "bbox": [ + 396, + 509, + 407, + 522 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 507, + 411, + 524 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 107, + 539, + 378, + 551 + ], + "lines": [ + { + "bbox": [ + 104, + 538, + 380, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 380, + 554 + ], + "score": 1.0, + "content": "3 EXPERIMENTS WITH GAN EVALUATION METRICS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 108, + 564, + 200, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 201, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 201, + 576 + ], + "score": 1.0, + "content": "3.1 FEATURE SPACE", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 582, + 507, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 507, + 596 + ], + "score": 1.0, + "content": "All the metrics introduced in the previous section, except for the Inception Score and Mode Score,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 185, + 606 + ], + "score": 1.0, + "content": "access the samples", + "type": "text" + }, + { + "bbox": [ + 186, + 596, + 194, + 604 + ], + "score": 0.39, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "only through pair-wise distances. The Kernel MMD requires a fixed kernel", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 142, + 618 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 143, + 605, + 150, + 615 + ], + "score": 0.74, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 605, + 506, + 618 + ], + "score": 1.0, + "content": ", typically set to an isotopic Gaussian; the Wasserstein distance and 1-NN accuracy use", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 229, + 629 + ], + "score": 1.0, + "content": "the underlying distance metric", + "type": "text" + }, + { + "bbox": [ + 230, + 617, + 236, + 626 + ], + "score": 0.77, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "directly; all of these methods are highly sensitive to the choice that", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 626, + 145, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 145, + 640 + ], + "score": 1.0, + "content": "distance.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "It is well-established that pixel representations of images do not induce meaningful Euclidean dis-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "tances (Forsyth & Ponce, 2011). Small translations, rotations, or changes in illumination can increase", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "distances dramatically with little effect on the image content. To quantity the similarity between", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "distributions of images, it is therefore desirable to use distances invariant to such transformations.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The choice of distance function can be re-interpreted as a choice of representation, by defining the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 697, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 104, + 697, + 243, + 713 + ], + "score": 1.0, + "content": "distance in a more general form as", + "type": "text" + }, + { + "bbox": [ + 243, + 699, + 357, + 711 + ], + "score": 0.91, + "content": "d ( \\mathbf { x } , \\mathbf { x } ^ { \\prime } ) = \\| \\boldsymbol { \\phi } ( \\mathbf { x } ) - \\boldsymbol { \\phi } ( \\mathbf { x } ^ { \\prime } ) \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 697, + 387, + 713 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 387, + 699, + 404, + 711 + ], + "score": 0.9, + "content": "\\phi ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 697, + 506, + 713 + ], + "score": 1.0, + "content": "is some general mapping", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "of the input into a semantically meaningful feature space. For Kernel MMD, this corresponds to", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 353, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 325, + 734 + ], + "score": 1.0, + "content": "computing the usual inner product in the feature space", + "type": "text" + }, + { + "bbox": [ + 326, + 721, + 348, + 732 + ], + "score": 0.92, + "content": "\\phi ( \\mathcal { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 720, + 353, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 46.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 160 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 105, + 82, + 506, + 162 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 165, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "score": 1.0, + "content": "Lopez-Paz & Oquab (2016) considered the 1-NN accuracy primarily as a statistic for two-sample", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 506, + 190 + ], + "score": 1.0, + "content": "testing. In fact, it is more informative to analyze it for the two classes separately. For example, a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "typical outcome of GANs is that for both real and generated images, the majority of their nearest", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "neighbors are generated images due to mode collapse. In this case, the LOO 1-NN accuracy of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "real images would be relatively low (desired): the mode(s) of the real distribution are usually well", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 378, + 233 + ], + "score": 1.0, + "content": "captured by the generative model, so a majority of real samples from", + "type": "text" + }, + { + "bbox": [ + 379, + 221, + 390, + 231 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "are surrounded by generated", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 162, + 244 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 162, + 231, + 174, + 243 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 231, + 506, + 244 + ], + "score": 1.0, + "content": ", leading to low LOO accuracy; whereas the LOO accuracy of the generated images", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "is high (not desired): generative samples tend to collapse to a few mode centers, thus they are", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 507, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 507, + 267 + ], + "score": 1.0, + "content": "surrounded by samples from the same class, leading to high LOO accuracy. For the rest of the paper,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 436, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 436, + 276 + ], + "score": 1.0, + "content": "we distinguish these two cases as 1-NN accuracy (real) and 1-NN accuracy (fake).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 165, + 507, + 276 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 291, + 202, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 204, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 204, + 304 + ], + "score": 1.0, + "content": "2.3 OTHER METRICS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 310, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 506, + 323 + ], + "score": 1.0, + "content": "All of the metrics above are, what we refer to as “model agnostic\": they use the generator as a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 277, + 335 + ], + "score": 1.0, + "content": "black box to sample the generated images", + "type": "text" + }, + { + "bbox": [ + 277, + 322, + 289, + 334 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 320, + 505, + 335 + ], + "score": 1.0, + "content": ". Model agnostic metrics should not require a density", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "estimation from the model. We choose to only experiment with model agnostic metrics, which allow", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "score": 1.0, + "content": "us to support as many generative models as possible for evaluation without modification to their", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 355, + 497, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 497, + 367 + ], + "score": 1.0, + "content": "structure. We will briefly mention some other evaluation metrics not included in our experiments.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 310, + 506, + 367 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "Kernel density estimation (KDE, or Parzen window estimation) is a well-studied method for estimating", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 382, + 506, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 429, + 395 + ], + "score": 1.0, + "content": "the density function of a distribution from samples. For a probability kernel", + "type": "text" + }, + { + "bbox": [ + 429, + 383, + 440, + 393 + ], + "score": 0.75, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 382, + 506, + 395 + ], + "score": 1.0, + "content": "(most often an", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 255, + 406 + ], + "score": 1.0, + "content": "isotropic Gaussian) and i.i.d samples", + "type": "text" + }, + { + "bbox": [ + 255, + 395, + 301, + 405 + ], + "score": 0.9, + "content": "\\mathbf { x } _ { 1 } , \\ldots , \\mathbf { x } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 393, + 453, + 406 + ], + "score": 1.0, + "content": ", we can define the density function at", + "type": "text" + }, + { + "bbox": [ + 454, + 395, + 462, + 403 + ], + "score": 0.48, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 393, + 473, + 406 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 473, + 393, + 505, + 405 + ], + "score": 0.91, + "content": "p ( \\mathbf { x } ) \\approx", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 107, + 403, + 186, + 417 + ], + "score": 0.89, + "content": "\\textstyle { \\frac { 1 } { z } } \\sum _ { i = 1 } ^ { n ^ { \\cdot } } K ( \\mathbf { x } - \\mathbf { x } _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 404, + 216, + 416 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 216, + 406, + 223, + 414 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "is a normalizing constant. This allows the use of classical metrics such", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 415, + 507, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 507, + 428 + ], + "score": 1.0, + "content": "as KLD and JSD. However, despite the widespread adoption of this technique to various applications,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 426, + 507, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 277, + 440 + ], + "score": 1.0, + "content": "its suitability to estimating the density of", + "type": "text" + }, + { + "bbox": [ + 278, + 426, + 289, + 437 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 426, + 302, + 440 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 302, + 426, + 315, + 438 + ], + "score": 0.9, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 426, + 507, + 440 + ], + "score": 1.0, + "content": "for GANs has been questioned by Theis et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 438, + 454, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 454, + 450 + ], + "score": 1.0, + "content": "(2015) since the probability kernel depends on the Euclidean distance between images.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 370, + 507, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "More recently, Wu et al. (2016) applied annealed importance sampling (AIS) to estimate the marginal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 464, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 156, + 478 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 157, + 465, + 177, + 477 + ], + "score": 0.91, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 464, + 506, + 478 + ], + "score": 1.0, + "content": "of a generative model. This method is most natural for models that define a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 200, + 488 + ], + "score": 1.0, + "content": "conditional distribution", + "type": "text" + }, + { + "bbox": [ + 200, + 476, + 227, + 488 + ], + "score": 0.93, + "content": "p ( \\mathbf { x } | \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 475, + 254, + 488 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 255, + 478, + 262, + 486 + ], + "score": 0.66, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "is the latent code, which is not satisfied by most GAN models.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Nevertheless, AIS has been applied to GAN evaluation by assuming a Gaussian observation model.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "We exclude this method from our experiments as it needs the access to the generative model to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 507, + 411, + 524 + ], + "spans": [ + { + "bbox": [ + 104, + 507, + 395, + 524 + ], + "score": 1.0, + "content": "compute the likelihood, instead of only depending on a finite sample set", + "type": "text" + }, + { + "bbox": [ + 396, + 509, + 407, + 522 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 507, + 411, + 524 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 453, + 506, + 524 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 539, + 378, + 551 + ], + "lines": [ + { + "bbox": [ + 104, + 538, + 380, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 380, + 554 + ], + "score": 1.0, + "content": "3 EXPERIMENTS WITH GAN EVALUATION METRICS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 108, + 564, + 200, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 201, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 201, + 576 + ], + "score": 1.0, + "content": "3.1 FEATURE SPACE", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 582, + 507, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 507, + 596 + ], + "score": 1.0, + "content": "All the metrics introduced in the previous section, except for the Inception Score and Mode Score,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 185, + 606 + ], + "score": 1.0, + "content": "access the samples", + "type": "text" + }, + { + "bbox": [ + 186, + 596, + 194, + 604 + ], + "score": 0.39, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "only through pair-wise distances. The Kernel MMD requires a fixed kernel", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 142, + 618 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 143, + 605, + 150, + 615 + ], + "score": 0.74, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 605, + 506, + 618 + ], + "score": 1.0, + "content": ", typically set to an isotopic Gaussian; the Wasserstein distance and 1-NN accuracy use", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 229, + 629 + ], + "score": 1.0, + "content": "the underlying distance metric", + "type": "text" + }, + { + "bbox": [ + 230, + 617, + 236, + 626 + ], + "score": 0.77, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "directly; all of these methods are highly sensitive to the choice that", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 626, + 145, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 145, + 640 + ], + "score": 1.0, + "content": "distance.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 582, + 507, + 640 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "It is well-established that pixel representations of images do not induce meaningful Euclidean dis-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "tances (Forsyth & Ponce, 2011). Small translations, rotations, or changes in illumination can increase", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "distances dramatically with little effect on the image content. To quantity the similarity between", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "distributions of images, it is therefore desirable to use distances invariant to such transformations.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The choice of distance function can be re-interpreted as a choice of representation, by defining the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 697, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 104, + 697, + 243, + 713 + ], + "score": 1.0, + "content": "distance in a more general form as", + "type": "text" + }, + { + "bbox": [ + 243, + 699, + 357, + 711 + ], + "score": 0.91, + "content": "d ( \\mathbf { x } , \\mathbf { x } ^ { \\prime } ) = \\| \\boldsymbol { \\phi } ( \\mathbf { x } ) - \\boldsymbol { \\phi } ( \\mathbf { x } ^ { \\prime } ) \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 697, + 387, + 713 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 387, + 699, + 404, + 711 + ], + "score": 0.9, + "content": "\\phi ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 697, + 506, + 713 + ], + "score": 1.0, + "content": "is some general mapping", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "of the input into a semantically meaningful feature space. For Kernel MMD, this corresponds to", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 353, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 325, + 734 + ], + "score": 1.0, + "content": "computing the usual inner product in the feature space", + "type": "text" + }, + { + "bbox": [ + 326, + 721, + 348, + 732 + ], + "score": 0.92, + "content": "\\phi ( \\mathcal { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 720, + 353, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 46.5, + "bbox_fs": [ + 104, + 644, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 78, + 504, + 195 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 78, + 504, + 195 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 78, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 108, + 78, + 504, + 195 + ], + "score": 0.961, + "type": "image", + "image_path": "cab638a8370b4cd54a62381ec515f1f9af5798f9c29f188a4b0647fee588444c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 78, + 504, + 117.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 117.0, + 504, + 156.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 156.0, + 504, + 195.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 203, + 505, + 243 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 201, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 216 + ], + "score": 1.0, + "content": "Figure 2: Distinguishing a set of real images (in the training set) from a mixed set of real images and GAN", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 212, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 225 + ], + "score": 1.0, + "content": "generated images. For the metric to be discriminative, its score should increase as the fraction of generated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 221, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 481, + 235 + ], + "score": 1.0, + "content": "samples in the mix increases. RIS and RMS fail as they decrease with the fraction of generated samples in", + "type": "text" + }, + { + "bbox": [ + 482, + 223, + 493, + 234 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 221, + 506, + 235 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 231, + 424, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 424, + 244 + ], + "score": 1.0, + "content": "LSUN. Wasserstein and 1-NN accuracy (real) fail in pixel space as they do not increase.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 256, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "Inspired by recent works from Upchurch et al. (2017); Larsen et al. (2015) which show that convolu-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 266, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 281 + ], + "score": 1.0, + "content": "tional neural networks may linearize the image manifold, we propose to operate in the feature space", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "of an external model pre-trained on the ImageNet dataset. For efficiency, we use a 34-layer ResNet2", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 287, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 104, + 287, + 505, + 304 + ], + "score": 1.0, + "content": "as the feature extractor. Our experiments show that other models such as VGG or Inception give very", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 300, + 168, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 168, + 312 + ], + "score": 1.0, + "content": "similar results.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "To illustrate our point, we show failure examples of the pixel space distance for evaluating GANs in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "score": 1.0, + "content": "this section, and highlight that using a proper feature space is key to obtaining meaningful results", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "score": 1.0, + "content": "when applying the distance-based metrics. The usage of a well-suited feature space enables us to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 351, + 463, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 463, + 362 + ], + "score": 1.0, + "content": "draw more optimistic conclusions on GAN evaluation metrics than in Theis et al. (2015).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 107, + 372, + 160, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 162, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 162, + 384 + ], + "score": 1.0, + "content": "3.2 SETUP", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 109, + 388, + 504, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "For the rest of this major section, we introduce what in our opinion are necessary conditions for good", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "metrics for GANs. After the introduction of each condition, we use it as a criterion to judge the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 107, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "effectiveness of the metrics presented in Section 2, through carefully designed empirical experiments.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "The experiments are performed on two standard benchmark datasets for generative models, Cele-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "bA3and LSUN bedrooms4. To remove the degree of freedom induced by feature representation, we", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "use the Inception Score (IS) and Mode Score (MS) computed from the softmax probabilities of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "same ResNet-34 model as the other metrics, instead of the Inception model. We also compute the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 469, + 507, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 274, + 484 + ], + "score": 1.0, + "content": "Inception Score over the real training data", + "type": "text" + }, + { + "bbox": [ + 275, + 471, + 286, + 482 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 469, + 444, + 484 + ], + "score": 1.0, + "content": "as an upper bound, which we denote as", + "type": "text" + }, + { + "bbox": [ + 444, + 471, + 459, + 482 + ], + "score": 0.87, + "content": "\\mathrm { I S } _ { \\mathrm { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 469, + 507, + 484 + ], + "score": 1.0, + "content": ". Moreover,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 482, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 506, + 495 + ], + "score": 1.0, + "content": "to be consistent with other metrics where lower values correspond to better models, we report the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 240, + 505 + ], + "score": 1.0, + "content": "relative inverse Inception Score", + "type": "text" + }, + { + "bbox": [ + 240, + 493, + 326, + 505 + ], + "score": 0.93, + "content": "R I S = \\left( 1 - { \\mathrm { I S } } / { \\mathrm { I S } } _ { 0 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "in tables and plots, after computing IS the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "Inception Score. We similarly report the relative inverse Mode Score (RMS). Although RIS and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "RMS operate in the softmax space, we always compare them together with other metrics in the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "convolutional space for simplicity. For all the plots in this paper, shaded areas denote the standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 537, + 472, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 472, + 549 + ], + "score": 1.0, + "content": "deviations, computed by running the same experiment 5 times with different random seeds.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 214, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 557, + 216, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 216, + 570 + ], + "score": 1.0, + "content": "3.3 DISCRIMINABILITY", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 437, + 588 + ], + "score": 1.0, + "content": "Mixing of generated images. Arguably, the most important property of a metric", + "type": "text" + }, + { + "bbox": [ + 438, + 576, + 444, + 586 + ], + "score": 0.84, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 573, + 505, + 588 + ], + "score": 1.0, + "content": "for measuring", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "GANs is the ability to distinguish generated images from real images. To test this property, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 597, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 159, + 609 + ], + "score": 1.0, + "content": "sample a set", + "type": "text" + }, + { + "bbox": [ + 159, + 597, + 171, + 608 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 597, + 227, + 609 + ], + "score": 1.0, + "content": "consisting of", + "type": "text" + }, + { + "bbox": [ + 227, + 598, + 234, + 606 + ], + "score": 0.61, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 597, + 239, + 609 + ], + "score": 1.0, + "content": "(", + "type": "text" + }, + { + "bbox": [ + 239, + 597, + 280, + 607 + ], + "score": 0.87, + "content": "n = 2 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 597, + 505, + 609 + ], + "score": 1.0, + "content": "if not otherwise specified) real images uniformly from", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 208, + 620 + ], + "score": 1.0, + "content": "the training set, and a set", + "type": "text" + }, + { + "bbox": [ + 208, + 608, + 231, + 619 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 607, + 298, + 620 + ], + "score": 1.0, + "content": "of the same size", + "type": "text" + }, + { + "bbox": [ + 299, + 609, + 306, + 617 + ], + "score": 0.72, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "consisting of a mix of real samples and generated", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 451, + 631 + ], + "score": 1.0, + "content": "images from a DCGAN (Radford et al., 2015) trained on the same training set, where", + "type": "text" + }, + { + "bbox": [ + 451, + 619, + 456, + 628 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 628, + 213, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 213, + 642 + ], + "score": 1.0, + "content": "ratio of generated images.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 108, + 646, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 306, + 659 + ], + "score": 1.0, + "content": "The computed values of various metrics between", + "type": "text" + }, + { + "bbox": [ + 307, + 646, + 318, + 657 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 646, + 336, + 659 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 337, + 646, + 360, + 659 + ], + "score": 0.93, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 646, + 375, + 659 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 376, + 646, + 412, + 658 + ], + "score": 0.92, + "content": "t \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 646, + 506, + 659 + ], + "score": 1.0, + "content": "are shown in Figure 2.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 131, + 671 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 658, + 144, + 668 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 657, + 466, + 671 + ], + "score": 1.0, + "content": "should serve as a lower bound for any metric, we expect that any reasonable", + "type": "text" + }, + { + "bbox": [ + 467, + 658, + 474, + 669 + ], + "score": 0.84, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "should", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "increase as the ratio of generated images increases. This is indeed satisfied for all the metrics except:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 689, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 118, + 687, + 495, + 703 + ], + "spans": [ + { + "bbox": [ + 118, + 687, + 495, + 703 + ], + "score": 1.0, + "content": "2We use the official pre-trained ResNet-34 model from PyTorch http://tinyurl.com/pytorch.", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 698, + 493, + 714 + ], + "spans": [ + { + "bbox": [ + 118, + 698, + 493, + 714 + ], + "score": 1.0, + "content": "3CelebA is a large-scale high resolution face dataset with more than 200,000 centered celebrity images.", + "type": "text" + } + ] + }, + { + "bbox": [ + 117, + 708, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 117, + 708, + 505, + 725 + ], + "score": 1.0, + "content": "4LSUN consists of around one million images for each of 10 scene classes. Following standard practice, we", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 720, + 213, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 213, + 734 + ], + "score": 1.0, + "content": "only take the bedroom scene.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 78, + 504, + 195 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 78, + 504, + 195 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 78, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 108, + 78, + 504, + 195 + ], + "score": 0.961, + "type": "image", + "image_path": "cab638a8370b4cd54a62381ec515f1f9af5798f9c29f188a4b0647fee588444c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 78, + 504, + 117.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 117.0, + 504, + 156.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 156.0, + 504, + 195.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 203, + 505, + 243 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 201, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 216 + ], + "score": 1.0, + "content": "Figure 2: Distinguishing a set of real images (in the training set) from a mixed set of real images and GAN", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 212, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 225 + ], + "score": 1.0, + "content": "generated images. For the metric to be discriminative, its score should increase as the fraction of generated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 221, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 481, + 235 + ], + "score": 1.0, + "content": "samples in the mix increases. RIS and RMS fail as they decrease with the fraction of generated samples in", + "type": "text" + }, + { + "bbox": [ + 482, + 223, + 493, + 234 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 221, + 506, + 235 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 231, + 424, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 424, + 244 + ], + "score": 1.0, + "content": "LSUN. Wasserstein and 1-NN accuracy (real) fail in pixel space as they do not increase.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 256, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "Inspired by recent works from Upchurch et al. (2017); Larsen et al. (2015) which show that convolu-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 266, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 281 + ], + "score": 1.0, + "content": "tional neural networks may linearize the image manifold, we propose to operate in the feature space", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "of an external model pre-trained on the ImageNet dataset. For efficiency, we use a 34-layer ResNet2", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 287, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 104, + 287, + 505, + 304 + ], + "score": 1.0, + "content": "as the feature extractor. Our experiments show that other models such as VGG or Inception give very", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 300, + 168, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 168, + 312 + ], + "score": 1.0, + "content": "similar results.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9, + "bbox_fs": [ + 104, + 256, + 506, + 312 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "To illustrate our point, we show failure examples of the pixel space distance for evaluating GANs in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "score": 1.0, + "content": "this section, and highlight that using a proper feature space is key to obtaining meaningful results", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "score": 1.0, + "content": "when applying the distance-based metrics. The usage of a well-suited feature space enables us to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 351, + 463, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 463, + 362 + ], + "score": 1.0, + "content": "draw more optimistic conclusions on GAN evaluation metrics than in Theis et al. (2015).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 317, + 506, + 362 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 372, + 160, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 162, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 162, + 384 + ], + "score": 1.0, + "content": "3.2 SETUP", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 109, + 388, + 504, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "For the rest of this major section, we introduce what in our opinion are necessary conditions for good", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "metrics for GANs. After the introduction of each condition, we use it as a criterion to judge the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 107, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "effectiveness of the metrics presented in Section 2, through carefully designed empirical experiments.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 388, + 506, + 423 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "The experiments are performed on two standard benchmark datasets for generative models, Cele-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "bA3and LSUN bedrooms4. To remove the degree of freedom induced by feature representation, we", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "use the Inception Score (IS) and Mode Score (MS) computed from the softmax probabilities of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "same ResNet-34 model as the other metrics, instead of the Inception model. We also compute the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 469, + 507, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 274, + 484 + ], + "score": 1.0, + "content": "Inception Score over the real training data", + "type": "text" + }, + { + "bbox": [ + 275, + 471, + 286, + 482 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 469, + 444, + 484 + ], + "score": 1.0, + "content": "as an upper bound, which we denote as", + "type": "text" + }, + { + "bbox": [ + 444, + 471, + 459, + 482 + ], + "score": 0.87, + "content": "\\mathrm { I S } _ { \\mathrm { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 469, + 507, + 484 + ], + "score": 1.0, + "content": ". Moreover,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 482, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 506, + 495 + ], + "score": 1.0, + "content": "to be consistent with other metrics where lower values correspond to better models, we report the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 240, + 505 + ], + "score": 1.0, + "content": "relative inverse Inception Score", + "type": "text" + }, + { + "bbox": [ + 240, + 493, + 326, + 505 + ], + "score": 0.93, + "content": "R I S = \\left( 1 - { \\mathrm { I S } } / { \\mathrm { I S } } _ { 0 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "in tables and plots, after computing IS the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "Inception Score. We similarly report the relative inverse Mode Score (RMS). Although RIS and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "RMS operate in the softmax space, we always compare them together with other metrics in the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "convolutional space for simplicity. For all the plots in this paper, shaded areas denote the standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 537, + 472, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 472, + 549 + ], + "score": 1.0, + "content": "deviations, computed by running the same experiment 5 times with different random seeds.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 427, + 507, + 549 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 214, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 557, + 216, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 216, + 570 + ], + "score": 1.0, + "content": "3.3 DISCRIMINABILITY", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 437, + 588 + ], + "score": 1.0, + "content": "Mixing of generated images. Arguably, the most important property of a metric", + "type": "text" + }, + { + "bbox": [ + 438, + 576, + 444, + 586 + ], + "score": 0.84, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 573, + 505, + 588 + ], + "score": 1.0, + "content": "for measuring", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "GANs is the ability to distinguish generated images from real images. To test this property, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 597, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 159, + 609 + ], + "score": 1.0, + "content": "sample a set", + "type": "text" + }, + { + "bbox": [ + 159, + 597, + 171, + 608 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 597, + 227, + 609 + ], + "score": 1.0, + "content": "consisting of", + "type": "text" + }, + { + "bbox": [ + 227, + 598, + 234, + 606 + ], + "score": 0.61, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 597, + 239, + 609 + ], + "score": 1.0, + "content": "(", + "type": "text" + }, + { + "bbox": [ + 239, + 597, + 280, + 607 + ], + "score": 0.87, + "content": "n = 2 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 597, + 505, + 609 + ], + "score": 1.0, + "content": "if not otherwise specified) real images uniformly from", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 208, + 620 + ], + "score": 1.0, + "content": "the training set, and a set", + "type": "text" + }, + { + "bbox": [ + 208, + 608, + 231, + 619 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 607, + 298, + 620 + ], + "score": 1.0, + "content": "of the same size", + "type": "text" + }, + { + "bbox": [ + 299, + 609, + 306, + 617 + ], + "score": 0.72, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "consisting of a mix of real samples and generated", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 451, + 631 + ], + "score": 1.0, + "content": "images from a DCGAN (Radford et al., 2015) trained on the same training set, where", + "type": "text" + }, + { + "bbox": [ + 451, + 619, + 456, + 628 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 628, + 213, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 213, + 642 + ], + "score": 1.0, + "content": "ratio of generated images.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 573, + 505, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 646, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 306, + 659 + ], + "score": 1.0, + "content": "The computed values of various metrics between", + "type": "text" + }, + { + "bbox": [ + 307, + 646, + 318, + 657 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 646, + 336, + 659 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 337, + 646, + 360, + 659 + ], + "score": 0.93, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 646, + 375, + 659 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 376, + 646, + 412, + 658 + ], + "score": 0.92, + "content": "t \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 646, + 506, + 659 + ], + "score": 1.0, + "content": "are shown in Figure 2.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 131, + 671 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 658, + 144, + 668 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 657, + 466, + 671 + ], + "score": 1.0, + "content": "should serve as a lower bound for any metric, we expect that any reasonable", + "type": "text" + }, + { + "bbox": [ + 467, + 658, + 474, + 669 + ], + "score": 0.84, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "should", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "increase as the ratio of generated images increases. This is indeed satisfied for all the metrics except:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 646, + 506, + 681 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 189 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 189 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 189 + ], + "score": 0.963, + "type": "image", + "image_path": "1c1ee6c3e031cabba0b7fe1ab9f3c6c30c09d9cf1e241dac2fc9e61e61743cb0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 115.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 115.66666666666666, + 504, + 152.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 152.33333333333331, + 504, + 188.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 196, + 505, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "Figure 3: Experiment on simulated mode collapsing. A metric score should increase to reflect the mismatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "between true distribution and generated distribution as more modes are collapsed towards their cluster center. All", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "metrics respond correctly in convolutional space. In pixel space, both Wasserstein distance and 1-NN accuracy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 226, + 343, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 343, + 237 + ], + "score": 1.0, + "content": "(real) fail as they decrease in response to more collapsed clusters.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 505, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 253, + 504, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 504, + 264 + ], + "score": 1.0, + "content": "1) RIS and RMS (red and green curves) on LSUN, which decrease as more fake samples are in the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 275 + ], + "score": 1.0, + "content": "mix; 2) 1-NN accuracy of real samples (dotted magenta curve) computed in pixel space, which also", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 275, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 287 + ], + "score": 1.0, + "content": "appears to be a decreasing function; and 3) Wasserstein Distance (cyan curve), which almost remains", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 285, + 211, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 176, + 298 + ], + "score": 1.0, + "content": "unchanged when", + "type": "text" + }, + { + "bbox": [ + 176, + 286, + 181, + 295 + ], + "score": 0.55, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 285, + 211, + 298 + ], + "score": 1.0, + "content": "varies.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 505, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "score": 1.0, + "content": "The reason that RIS and RMS do not work well here is likely because they are not suitable for images", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "beyond the ImageNet categories. Although other metrics operate in the convolutional feature space", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 324, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 337 + ], + "score": 1.0, + "content": "also depend on a network pretrained on ImageNet, the convolutional features are much more general", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "than the specific softmax representation. The failure of Wasserstein Distance is possibly due to an", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "insufficient number of samples, which we will discuss in more detail when we analyze the sample", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "efficiency of various metrics in a latter subsection. The last paragraph of Section 2.2 explains why", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "score": 1.0, + "content": "the 1-NN accuracy for real samples (dotted magenta curve) is always lower than that for generated", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 379, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 456, + 392 + ], + "score": 1.0, + "content": "samples (dashed magenta curve). In the pixel space, more than half of the samples from", + "type": "text" + }, + { + "bbox": [ + 457, + 379, + 468, + 390 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 379, + 505, + 392 + ], + "score": 1.0, + "content": "have the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 197, + 403 + ], + "score": 1.0, + "content": "nearest neighbor from", + "type": "text" + }, + { + "bbox": [ + 198, + 390, + 220, + 402 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 390, + 505, + 403 + ], + "score": 1.0, + "content": ", indicating that the DCGAN is able to represent the modes in the pixel", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 402, + 176, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 176, + 413 + ], + "score": 1.0, + "content": "space quite well.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 504, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "We also conducted the same experiment using 1) random noise images and 2) images from an entirely", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 428, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 462, + 441 + ], + "score": 1.0, + "content": "different distribution (e.g. CIFAR-10), instead of DCGAN generated images to construct", + "type": "text" + }, + { + "bbox": [ + 462, + 429, + 485, + 441 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 428, + 506, + 441 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 403, + 452 + ], + "score": 1.0, + "content": "call these injected samples as out-of-domain violations since they are not in", + "type": "text" + }, + { + "bbox": [ + 403, + 440, + 413, + 450 + ], + "score": 0.8, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ", the domain of the real", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 451, + 439, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 439, + 462 + ], + "score": 1.0, + "content": "images. These settings yield similar results as in Figure 2, thus we omit their plots.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 347, + 480 + ], + "score": 1.0, + "content": "Mode collapsing and mode dropping. In realistic settings,", + "type": "text" + }, + { + "bbox": [ + 347, + 468, + 359, + 479 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "is usually very diverse since natural", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 372, + 492 + ], + "score": 1.0, + "content": "images are inherently multimodal. Many have conjectured that", + "type": "text" + }, + { + "bbox": [ + 372, + 479, + 384, + 491 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 477, + 439, + 492 + ], + "score": 1.0, + "content": "differs from", + "type": "text" + }, + { + "bbox": [ + 439, + 479, + 451, + 489 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "by reducing", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "diversity, possibly due to the lack of model capacity or inadequate optimization (Arora et al., 2017).", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "This is often manifested itself for generative models in a mix of two ways: mode dropping, where", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 237, + 525 + ], + "score": 1.0, + "content": "some hard-to-represent modes of", + "type": "text" + }, + { + "bbox": [ + 238, + 512, + 249, + 522 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 511, + 347, + 525 + ], + "score": 1.0, + "content": "are simply “ignored\" by", + "type": "text" + }, + { + "bbox": [ + 347, + 512, + 359, + 523 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "; and mode collapsing, where several", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 144, + 536 + ], + "score": 1.0, + "content": "modes of", + "type": "text" + }, + { + "bbox": [ + 145, + 523, + 156, + 533 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 522, + 229, + 536 + ], + "score": 1.0, + "content": "are “averaged\" by", + "type": "text" + }, + { + "bbox": [ + 230, + 523, + 242, + 534 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 522, + 506, + 536 + ], + "score": 1.0, + "content": "into a single mode, possibly located at a midpoint. An ideal metric", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 286, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 286, + 546 + ], + "score": 1.0, + "content": "should be sensitive to these two phenomena.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 337, + 563 + ], + "score": 1.0, + "content": "To test for mode collapsing, we first randomly sample both", + "type": "text" + }, + { + "bbox": [ + 337, + 550, + 349, + 561 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 549, + 366, + 563 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 367, + 550, + 378, + 562 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 549, + 506, + 563 + ], + "score": 1.0, + "content": "as two disjoint sets of 2000 real", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 363, + 574 + ], + "score": 1.0, + "content": "images. Next, we find 50 clusters in the whole training set with", + "type": "text" + }, + { + "bbox": [ + 363, + 562, + 370, + 571 + ], + "score": 0.82, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "-means and progressively replace", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "each cluster by its respective cluster center to simulate mode collapse. Figure 3 shows computed", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 145, + 595 + ], + "score": 1.0, + "content": "values of", + "type": "text" + }, + { + "bbox": [ + 145, + 583, + 185, + 595 + ], + "score": 0.94, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 583, + 421, + 595 + ], + "score": 1.0, + "content": "as the number of replaced (collapsed) clusters, denoted as", + "type": "text" + }, + { + "bbox": [ + 421, + 583, + 430, + 593 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "increases. Ideally,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 239, + 607 + ], + "score": 1.0, + "content": "we expect the scores increase as", + "type": "text" + }, + { + "bbox": [ + 239, + 594, + 248, + 604 + ], + "score": 0.77, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "grows. We first observe that all the metrics are able to respond", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "correctly when distances are computed in the convolutional feature space. However, the Wasserstein", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 477, + 628 + ], + "score": 1.0, + "content": "metric (cyan curve) breaks down in pixel space, as it considers a collapsed sample set (with", + "type": "text" + }, + { + "bbox": [ + 477, + 616, + 502, + 627 + ], + "score": 0.88, + "content": "C > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 616, + 505, + 628 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 392, + 640 + ], + "score": 1.0, + "content": "being closer to the real sample set than another set of real images (with", + "type": "text" + }, + { + "bbox": [ + 392, + 627, + 417, + 637 + ], + "score": 0.89, + "content": "C = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "). Moreover, although", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "the overall 1-NN accuracy (solid magenta curve) follows the desired trend, the real and fake parts", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "follow opposite trends: 1-NN real accuracy (dotted magenta curve) decreases while 1-NN fake", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 674 + ], + "score": 1.0, + "content": "accuracy (dashed magenta curve) increases. Again, this is inline with our explanation given in the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 671, + 226, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 226, + 683 + ], + "score": 1.0, + "content": "last paragraph of Section 2.2.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 252, + 700 + ], + "score": 1.0, + "content": "To test for mode dropping, we take", + "type": "text" + }, + { + "bbox": [ + 253, + 688, + 264, + 699 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 687, + 361, + 700 + ], + "score": 1.0, + "content": "as above and construct", + "type": "text" + }, + { + "bbox": [ + 361, + 688, + 372, + 700 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 687, + 507, + 700 + ], + "score": 1.0, + "content": "by randomly removing clusters.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 183, + 711 + ], + "score": 1.0, + "content": "To keep the size of", + "type": "text" + }, + { + "bbox": [ + 184, + 699, + 195, + 711 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "constant, we replace images from the removed cluster with images randomly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "selected from the remaining clusters. Figure 4 shows how different metrics react to the number of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 249, + 734 + ], + "score": 1.0, + "content": "removed clusters, also denoted as", + "type": "text" + }, + { + "bbox": [ + 249, + 722, + 257, + 730 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 720, + 505, + 734 + ], + "score": 1.0, + "content": ". All scores effectively discriminate against mode dropping", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 189 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 189 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 189 + ], + "score": 0.963, + "type": "image", + "image_path": "1c1ee6c3e031cabba0b7fe1ab9f3c6c30c09d9cf1e241dac2fc9e61e61743cb0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 115.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 115.66666666666666, + 504, + 152.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 152.33333333333331, + 504, + 188.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 196, + 505, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "Figure 3: Experiment on simulated mode collapsing. A metric score should increase to reflect the mismatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "between true distribution and generated distribution as more modes are collapsed towards their cluster center. All", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "metrics respond correctly in convolutional space. In pixel space, both Wasserstein distance and 1-NN accuracy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 226, + 343, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 343, + 237 + ], + "score": 1.0, + "content": "(real) fail as they decrease in response to more collapsed clusters.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 505, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 253, + 504, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 504, + 264 + ], + "score": 1.0, + "content": "1) RIS and RMS (red and green curves) on LSUN, which decrease as more fake samples are in the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 275 + ], + "score": 1.0, + "content": "mix; 2) 1-NN accuracy of real samples (dotted magenta curve) computed in pixel space, which also", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 275, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 287 + ], + "score": 1.0, + "content": "appears to be a decreasing function; and 3) Wasserstein Distance (cyan curve), which almost remains", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 285, + 211, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 176, + 298 + ], + "score": 1.0, + "content": "unchanged when", + "type": "text" + }, + { + "bbox": [ + 176, + 286, + 181, + 295 + ], + "score": 0.55, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 285, + 211, + 298 + ], + "score": 1.0, + "content": "varies.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 253, + 506, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 505, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "score": 1.0, + "content": "The reason that RIS and RMS do not work well here is likely because they are not suitable for images", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "beyond the ImageNet categories. Although other metrics operate in the convolutional feature space", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 324, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 337 + ], + "score": 1.0, + "content": "also depend on a network pretrained on ImageNet, the convolutional features are much more general", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "than the specific softmax representation. The failure of Wasserstein Distance is possibly due to an", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "insufficient number of samples, which we will discuss in more detail when we analyze the sample", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "efficiency of various metrics in a latter subsection. The last paragraph of Section 2.2 explains why", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "score": 1.0, + "content": "the 1-NN accuracy for real samples (dotted magenta curve) is always lower than that for generated", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 379, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 456, + 392 + ], + "score": 1.0, + "content": "samples (dashed magenta curve). In the pixel space, more than half of the samples from", + "type": "text" + }, + { + "bbox": [ + 457, + 379, + 468, + 390 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 379, + 505, + 392 + ], + "score": 1.0, + "content": "have the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 197, + 403 + ], + "score": 1.0, + "content": "nearest neighbor from", + "type": "text" + }, + { + "bbox": [ + 198, + 390, + 220, + 402 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 390, + 505, + 403 + ], + "score": 1.0, + "content": ", indicating that the DCGAN is able to represent the modes in the pixel", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 402, + 176, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 176, + 413 + ], + "score": 1.0, + "content": "space quite well.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 301, + 506, + 413 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 504, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "We also conducted the same experiment using 1) random noise images and 2) images from an entirely", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 428, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 462, + 441 + ], + "score": 1.0, + "content": "different distribution (e.g. CIFAR-10), instead of DCGAN generated images to construct", + "type": "text" + }, + { + "bbox": [ + 462, + 429, + 485, + 441 + ], + "score": 0.92, + "content": "S _ { g } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 428, + 506, + 441 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 403, + 452 + ], + "score": 1.0, + "content": "call these injected samples as out-of-domain violations since they are not in", + "type": "text" + }, + { + "bbox": [ + 403, + 440, + 413, + 450 + ], + "score": 0.8, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ", the domain of the real", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 451, + 439, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 439, + 462 + ], + "score": 1.0, + "content": "images. These settings yield similar results as in Figure 2, thus we omit their plots.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 417, + 506, + 462 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 347, + 480 + ], + "score": 1.0, + "content": "Mode collapsing and mode dropping. In realistic settings,", + "type": "text" + }, + { + "bbox": [ + 347, + 468, + 359, + 479 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "is usually very diverse since natural", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 372, + 492 + ], + "score": 1.0, + "content": "images are inherently multimodal. Many have conjectured that", + "type": "text" + }, + { + "bbox": [ + 372, + 479, + 384, + 491 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 477, + 439, + 492 + ], + "score": 1.0, + "content": "differs from", + "type": "text" + }, + { + "bbox": [ + 439, + 479, + 451, + 489 + ], + "score": 0.87, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "by reducing", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "diversity, possibly due to the lack of model capacity or inadequate optimization (Arora et al., 2017).", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "This is often manifested itself for generative models in a mix of two ways: mode dropping, where", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 237, + 525 + ], + "score": 1.0, + "content": "some hard-to-represent modes of", + "type": "text" + }, + { + "bbox": [ + 238, + 512, + 249, + 522 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 511, + 347, + 525 + ], + "score": 1.0, + "content": "are simply “ignored\" by", + "type": "text" + }, + { + "bbox": [ + 347, + 512, + 359, + 523 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "; and mode collapsing, where several", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 144, + 536 + ], + "score": 1.0, + "content": "modes of", + "type": "text" + }, + { + "bbox": [ + 145, + 523, + 156, + 533 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 522, + 229, + 536 + ], + "score": 1.0, + "content": "are “averaged\" by", + "type": "text" + }, + { + "bbox": [ + 230, + 523, + 242, + 534 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 522, + 506, + 536 + ], + "score": 1.0, + "content": "into a single mode, possibly located at a midpoint. An ideal metric", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 286, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 286, + 546 + ], + "score": 1.0, + "content": "should be sensitive to these two phenomena.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 467, + 506, + 546 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 337, + 563 + ], + "score": 1.0, + "content": "To test for mode collapsing, we first randomly sample both", + "type": "text" + }, + { + "bbox": [ + 337, + 550, + 349, + 561 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 549, + 366, + 563 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 367, + 550, + 378, + 562 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 549, + 506, + 563 + ], + "score": 1.0, + "content": "as two disjoint sets of 2000 real", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 363, + 574 + ], + "score": 1.0, + "content": "images. Next, we find 50 clusters in the whole training set with", + "type": "text" + }, + { + "bbox": [ + 363, + 562, + 370, + 571 + ], + "score": 0.82, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "-means and progressively replace", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "each cluster by its respective cluster center to simulate mode collapse. Figure 3 shows computed", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 145, + 595 + ], + "score": 1.0, + "content": "values of", + "type": "text" + }, + { + "bbox": [ + 145, + 583, + 185, + 595 + ], + "score": 0.94, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 583, + 421, + 595 + ], + "score": 1.0, + "content": "as the number of replaced (collapsed) clusters, denoted as", + "type": "text" + }, + { + "bbox": [ + 421, + 583, + 430, + 593 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "increases. Ideally,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 239, + 607 + ], + "score": 1.0, + "content": "we expect the scores increase as", + "type": "text" + }, + { + "bbox": [ + 239, + 594, + 248, + 604 + ], + "score": 0.77, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "grows. We first observe that all the metrics are able to respond", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "correctly when distances are computed in the convolutional feature space. However, the Wasserstein", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 477, + 628 + ], + "score": 1.0, + "content": "metric (cyan curve) breaks down in pixel space, as it considers a collapsed sample set (with", + "type": "text" + }, + { + "bbox": [ + 477, + 616, + 502, + 627 + ], + "score": 0.88, + "content": "C > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 616, + 505, + 628 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 392, + 640 + ], + "score": 1.0, + "content": "being closer to the real sample set than another set of real images (with", + "type": "text" + }, + { + "bbox": [ + 392, + 627, + 417, + 637 + ], + "score": 0.89, + "content": "C = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "). Moreover, although", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "the overall 1-NN accuracy (solid magenta curve) follows the desired trend, the real and fake parts", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "follow opposite trends: 1-NN real accuracy (dotted magenta curve) decreases while 1-NN fake", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 659, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 674 + ], + "score": 1.0, + "content": "accuracy (dashed magenta curve) increases. Again, this is inline with our explanation given in the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 671, + 226, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 226, + 683 + ], + "score": 1.0, + "content": "last paragraph of Section 2.2.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 549, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 252, + 700 + ], + "score": 1.0, + "content": "To test for mode dropping, we take", + "type": "text" + }, + { + "bbox": [ + 253, + 688, + 264, + 699 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 687, + 361, + 700 + ], + "score": 1.0, + "content": "as above and construct", + "type": "text" + }, + { + "bbox": [ + 361, + 688, + 372, + 700 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 687, + 507, + 700 + ], + "score": 1.0, + "content": "by randomly removing clusters.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 183, + 711 + ], + "score": 1.0, + "content": "To keep the size of", + "type": "text" + }, + { + "bbox": [ + 184, + 699, + 195, + 711 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "constant, we replace images from the removed cluster with images randomly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "selected from the remaining clusters. Figure 4 shows how different metrics react to the number of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 249, + 734 + ], + "score": 1.0, + "content": "removed clusters, also denoted as", + "type": "text" + }, + { + "bbox": [ + 249, + 722, + 257, + 730 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 720, + 505, + 734 + ], + "score": 1.0, + "content": ". All scores effectively discriminate against mode dropping", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "except the RIS and RMS - they remain almost indifferent when some modes are dropped. Again, this", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "is perhaps caused by the fact that the Inception/Mode Score were originally designed for datasets", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 464, + 507, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 507, + 477 + ], + "score": 1.0, + "content": "with classes overlapping with the ImageNet dataset, and they do not generalize well to other datasets.", + "type": "text", + "cross_page": true + } + ], + "index": 15 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 687, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 188 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 188 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 188 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 188 + ], + "score": 0.966, + "type": "image", + "image_path": "27afa6c9ff1eb40623809e4d625a094bd3d23b7636c473490f6547ec4db34c1c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 115.33333333333334, + 504, + 151.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 151.66666666666669, + 504, + 188.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 191, + 504, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "Figure 4: Experiment on simulated mode dropping. A metric score should increase to reflect the mismatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "between true distribution and generated distribution as more modes are dropped. All metrics except RIS and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 211, + 482, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 482, + 223 + ], + "score": 1.0, + "content": "RMS respond correctly, as they only increase slightly in value even when almost all modes are dropped.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 108, + 231, + 505, + 382 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 231, + 505, + 382 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 231, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 108, + 231, + 505, + 382 + ], + "score": 0.971, + "type": "image", + "image_path": "a3f3cd6ed8e7a79b3ee018f9f4c752f97c498ad8d7aed192a7d48c76b1718aa3.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 108, + 231, + 505, + 281.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 281.3333333333333, + 505, + 331.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 331.66666666666663, + 505, + 381.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 386, + 506, + 427 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "score": 1.0, + "content": "Figure 5: Experiment on robustness of each metric to small transformations (rotations and translations). All", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "metrics should remain constant across all mixes of real and transformed real samples, since the transformations", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "do not alter semantics of the image. All metrics respond correctly in convolutional space, but behave incorrectly", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 416, + 417, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 417, + 428 + ], + "score": 1.0, + "content": "in pixel space. This experiment illustrates the unsuitability of distances in pixel space.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + } + ], + "index": 8.75 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 505, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "except the RIS and RMS - they remain almost indifferent when some modes are dropped. Again, this", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 454, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 466 + ], + "score": 1.0, + "content": "is perhaps caused by the fact that the Inception/Mode Score were originally designed for datasets", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 464, + 507, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 507, + 477 + ], + "score": 1.0, + "content": "with classes overlapping with the ImageNet dataset, and they do not generalize well to other datasets.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 489, + 289, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 291, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 291, + 501 + ], + "score": 1.0, + "content": "3.4 ROBUSTNESS TO TRANSFORMATIONS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 344, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 344, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 344, + 518 + ], + "score": 1.0, + "content": "GANs are widely used for image datasets, which have the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 517, + 345, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 345, + 529 + ], + "score": 1.0, + "content": "property that certain transformations to the input do not", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 528, + 345, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 345, + 540 + ], + "score": 1.0, + "content": "change its semantic meaning. Thus an ideal evaluation met-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 539, + 346, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 346, + 551 + ], + "score": 1.0, + "content": "ric should be invariant to such transformations to some ex-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 550, + 345, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 345, + 561 + ], + "score": 1.0, + "content": "tent. For example, a generator trained on CelebA should not", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 561, + 345, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 345, + 573 + ], + "score": 1.0, + "content": "be penalized by a metric if its generated faces are shifted by", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 572, + 267, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 267, + 585 + ], + "score": 1.0, + "content": "a few pixels or rotated by a small angle.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 344, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 345, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 345, + 600 + ], + "score": 1.0, + "content": "Figure 5 shows how the various metrics react to such small", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 600, + 345, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 313, + 612 + ], + "score": 1.0, + "content": "transformation to the images. In this experiment,", + "type": "text" + }, + { + "bbox": [ + 313, + 600, + 325, + 611 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 600, + 345, + 612 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 610, + 346, + 624 + ], + "spans": [ + { + "bbox": [ + 107, + 611, + 118, + 623 + ], + "score": 0.87, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 610, + 346, + 624 + ], + "score": 1.0, + "content": "are two disjoint sets of 2000 real images sampled from", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 622, + 345, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 345, + 634 + ], + "score": 1.0, + "content": "the training data. However, a proportion of images from", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 107, + 632, + 345, + 645 + ], + "spans": [ + { + "bbox": [ + 107, + 633, + 118, + 645 + ], + "score": 0.87, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 632, + 345, + 645 + ], + "score": 1.0, + "content": "are randomly shifted (up to 4 pixels) or rotated (up to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 644, + 346, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 346, + 656 + ], + "score": 1.0, + "content": "15 degrees). We can observe from the results that metric-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 655, + 345, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 345, + 667 + ], + "score": 1.0, + "content": "s operating in the convolutional space (or softmax space", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 665, + 345, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 345, + 678 + ], + "score": 1.0, + "content": "for RIS and RMS) are robust to these transformations, as", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 677, + 345, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 345, + 689 + ], + "score": 1.0, + "content": "all the curves are approximated horizontal as the ratio of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 686, + 345, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 345, + 701 + ], + "score": 1.0, + "content": "transformed samples increases. This is not that surprising as", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 30.5 + }, + { + "type": "image", + "bbox": [ + 355, + 510, + 500, + 627 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 355, + 510, + 500, + 627 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 355, + 510, + 500, + 627 + ], + "spans": [ + { + "bbox": [ + 355, + 510, + 500, + 627 + ], + "score": 0.964, + "type": "image", + "image_path": "3477b7f2e2783fbb5fa4506505a1f49b1b75ebf97b2bb2dd6d880f8cc60b969f.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 355, + 510, + 500, + 568.5 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 355, + 568.5, + 500, + 627.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 352, + 635, + 504, + 695 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 351, + 635, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 351, + 635, + 505, + 646 + ], + "score": 1.0, + "content": "Figure 7: Measurement of wall-clock time", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 351, + 645, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 351, + 645, + 505, + 655 + ], + "score": 1.0, + "content": "for computing various metrics as a function", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 351, + 655, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 351, + 655, + 505, + 666 + ], + "score": 1.0, + "content": "of the number of samples. All metrics are", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 351, + 665, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 351, + 665, + 505, + 676 + ], + "score": 1.0, + "content": "practical to compute for a sample of size", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 351, + 675, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 351, + 675, + 505, + 686 + ], + "score": 1.0, + "content": "2000, but Wasserstein distance does not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 352, + 685, + 451, + 695 + ], + "spans": [ + { + "bbox": [ + 352, + 685, + 451, + 695 + ], + "score": 1.0, + "content": "scale to large sample sizes.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.0 + } + ], + "index": 31.25 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "convolutional networks are well know for being invariant to certain transformations (Mallat, 2016).", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "In comparison, in the pixel space all the metrics consider the shifted/rotated images as drawn from a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "different distribution, highlighting the importance of computing distances in a proper feature space.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 188 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 188 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 188 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 188 + ], + "score": 0.966, + "type": "image", + "image_path": "27afa6c9ff1eb40623809e4d625a094bd3d23b7636c473490f6547ec4db34c1c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 115.33333333333334, + 504, + 151.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 151.66666666666669, + 504, + 188.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 191, + 504, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "Figure 4: Experiment on simulated mode dropping. A metric score should increase to reflect the mismatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "between true distribution and generated distribution as more modes are dropped. All metrics except RIS and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 211, + 482, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 482, + 223 + ], + "score": 1.0, + "content": "RMS respond correctly, as they only increase slightly in value even when almost all modes are dropped.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 108, + 231, + 505, + 382 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 231, + 505, + 382 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 231, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 108, + 231, + 505, + 382 + ], + "score": 0.971, + "type": "image", + "image_path": "a3f3cd6ed8e7a79b3ee018f9f4c752f97c498ad8d7aed192a7d48c76b1718aa3.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 108, + 231, + 505, + 281.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 281.3333333333333, + 505, + 331.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 331.66666666666663, + 505, + 381.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 386, + 506, + 427 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "score": 1.0, + "content": "Figure 5: Experiment on robustness of each metric to small transformations (rotations and translations). All", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "metrics should remain constant across all mixes of real and transformed real samples, since the transformations", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "do not alter semantics of the image. All metrics respond correctly in convolutional space, but behave incorrectly", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 416, + 417, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 417, + 428 + ], + "score": 1.0, + "content": "in pixel space. This experiment illustrates the unsuitability of distances in pixel space.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + } + ], + "index": 8.75 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 505, + 476 + ], + "lines": [], + "index": 14, + "bbox_fs": [ + 105, + 442, + 507, + 477 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 489, + 289, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 291, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 291, + 501 + ], + "score": 1.0, + "content": "3.4 ROBUSTNESS TO TRANSFORMATIONS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 344, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 344, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 344, + 518 + ], + "score": 1.0, + "content": "GANs are widely used for image datasets, which have the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 517, + 345, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 345, + 529 + ], + "score": 1.0, + "content": "property that certain transformations to the input do not", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 528, + 345, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 345, + 540 + ], + "score": 1.0, + "content": "change its semantic meaning. Thus an ideal evaluation met-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 539, + 346, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 346, + 551 + ], + "score": 1.0, + "content": "ric should be invariant to such transformations to some ex-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 550, + 345, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 345, + 561 + ], + "score": 1.0, + "content": "tent. For example, a generator trained on CelebA should not", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 561, + 345, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 345, + 573 + ], + "score": 1.0, + "content": "be penalized by a metric if its generated faces are shifted by", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 572, + 267, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 267, + 585 + ], + "score": 1.0, + "content": "a few pixels or rotated by a small angle.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 506, + 346, + 585 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 344, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 345, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 345, + 600 + ], + "score": 1.0, + "content": "Figure 5 shows how the various metrics react to such small", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 600, + 345, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 313, + 612 + ], + "score": 1.0, + "content": "transformation to the images. In this experiment,", + "type": "text" + }, + { + "bbox": [ + 313, + 600, + 325, + 611 + ], + "score": 0.87, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 600, + 345, + 612 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 610, + 346, + 624 + ], + "spans": [ + { + "bbox": [ + 107, + 611, + 118, + 623 + ], + "score": 0.87, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 610, + 346, + 624 + ], + "score": 1.0, + "content": "are two disjoint sets of 2000 real images sampled from", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 622, + 345, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 345, + 634 + ], + "score": 1.0, + "content": "the training data. However, a proportion of images from", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 107, + 632, + 345, + 645 + ], + "spans": [ + { + "bbox": [ + 107, + 633, + 118, + 645 + ], + "score": 0.87, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 632, + 345, + 645 + ], + "score": 1.0, + "content": "are randomly shifted (up to 4 pixels) or rotated (up to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 644, + 346, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 346, + 656 + ], + "score": 1.0, + "content": "15 degrees). We can observe from the results that metric-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 655, + 345, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 345, + 667 + ], + "score": 1.0, + "content": "s operating in the convolutional space (or softmax space", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 665, + 345, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 345, + 678 + ], + "score": 1.0, + "content": "for RIS and RMS) are robust to these transformations, as", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 677, + 345, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 345, + 689 + ], + "score": 1.0, + "content": "all the curves are approximated horizontal as the ratio of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 686, + 345, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 345, + 701 + ], + "score": 1.0, + "content": "transformed samples increases. This is not that surprising as", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "convolutional networks are well know for being invariant to certain transformations (Mallat, 2016).", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "In comparison, in the pixel space all the metrics consider the shifted/rotated images as drawn from a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "different distribution, highlighting the importance of computing distances in a proper feature space.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 589, + 346, + 701 + ] + }, + { + "type": "image", + "bbox": [ + 355, + 510, + 500, + 627 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 355, + 510, + 500, + 627 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 355, + 510, + 500, + 627 + ], + "spans": [ + { + "bbox": [ + 355, + 510, + 500, + 627 + ], + "score": 0.964, + "type": "image", + "image_path": "3477b7f2e2783fbb5fa4506505a1f49b1b75ebf97b2bb2dd6d880f8cc60b969f.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 355, + 510, + 500, + 568.5 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 355, + 568.5, + 500, + 627.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 352, + 635, + 504, + 695 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 351, + 635, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 351, + 635, + 505, + 646 + ], + "score": 1.0, + "content": "Figure 7: Measurement of wall-clock time", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 351, + 645, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 351, + 645, + 505, + 655 + ], + "score": 1.0, + "content": "for computing various metrics as a function", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 351, + 655, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 351, + 655, + 505, + 666 + ], + "score": 1.0, + "content": "of the number of samples. All metrics are", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 351, + 665, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 351, + 665, + 505, + 676 + ], + "score": 1.0, + "content": "practical to compute for a sample of size", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 351, + 675, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 351, + 675, + 505, + 686 + ], + "score": 1.0, + "content": "2000, but Wasserstein distance does not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 352, + 685, + 451, + 695 + ], + "spans": [ + { + "bbox": [ + 352, + 685, + 451, + 695 + ], + "score": 1.0, + "content": "scale to large sample sizes.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.0 + } + ], + "index": 31.25 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [], + "index": 43, + "bbox_fs": [ + 105, + 699, + 506, + 733 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 504, + 194 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 504, + 194 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 504, + 194 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 504, + 194 + ], + "score": 0.969, + "type": "image", + "image_path": "2fcde7509c9dec9a43fcff36da5c2ef1c139c751b252031a98486f573d1f95c4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 504, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 117.33333333333334, + 504, + 155.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 155.66666666666669, + 504, + 194.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 197, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 197, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 506, + 208 + ], + "score": 1.0, + "content": "Figure 6: The score of various metrics as a function of the number of samples. An ideal metric should result in a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 242, + 219 + ], + "score": 1.0, + "content": "large gap between the real-real (R-R;", + "type": "text" + }, + { + "bbox": [ + 243, + 207, + 281, + 218 + ], + "score": 0.91, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 207, + 356, + 219 + ], + "score": 1.0, + "content": ") and real-fake (R-G;", + "type": "text" + }, + { + "bbox": [ + 356, + 207, + 394, + 218 + ], + "score": 0.91, + "content": "\\hat { \\rho } ( \\bar { S } _ { r } , S _ { g } ) \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 207, + 505, + 219 + ], + "score": 1.0, + "content": ") curves in order to distinguish", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 216, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 506, + 229 + ], + "score": 1.0, + "content": "between real and fake distributions using as few samples as possible. Compared with Wasserstein distance,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "MMD and 1-NN accuracy require much fewer samples to discriminate real and generated images, while RIS", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 236, + 426, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 426, + 249 + ], + "score": 1.0, + "content": "totally fails on LSUN as it scores generated images even better (lower) than real images.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 183, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 184, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 184, + 280 + ], + "score": 1.0, + "content": "3.5 EFFICIENCY", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "A practical GAN evaluation metric should be able to compute “accurate” scores from a reasonable", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "number of samples and within an affordable computation cost, such that it can be computed, for", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 306, + 374, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 374, + 318 + ], + "score": 1.0, + "content": "example, after each training epoch to monitor the training process.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "Sample efficiency. Here we measure the sample efficiency of various metrics by investigating how", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 331, + 504, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 491, + 349 + ], + "score": 1.0, + "content": "many samples are needed for each of them in order to discriminate a set of generated samples", + "type": "text" + }, + { + "bbox": [ + 492, + 334, + 504, + 346 + ], + "score": 0.86, + "content": "S _ { g }", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 278, + 357 + ], + "score": 1.0, + "content": "(from DCGAN) from a set of real samples", + "type": "text" + }, + { + "bbox": [ + 279, + 345, + 290, + 357 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 344, + 453, + 357 + ], + "score": 1.0, + "content": ". To do this, we introduce a reference set", + "type": "text" + }, + { + "bbox": [ + 453, + 345, + 465, + 356 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 344, + 505, + 357 + ], + "score": 1.0, + "content": ", which is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 384, + 369 + ], + "score": 1.0, + "content": "also uniformly sampled from the real training data, but is disjoint with", + "type": "text" + }, + { + "bbox": [ + 384, + 356, + 396, + 367 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 355, + 505, + 369 + ], + "score": 1.0, + "content": ". All three sample sets have", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 183, + 380 + ], + "score": 1.0, + "content": "the same size, i.e.,", + "type": "text" + }, + { + "bbox": [ + 183, + 367, + 278, + 379 + ], + "score": 0.9, + "content": "| S _ { r } | = | S _ { r } ^ { \\prime } | = | S _ { g } | = n", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 366, + 406, + 380 + ], + "score": 1.0, + "content": ". We expect that an ideal metric", + "type": "text" + }, + { + "bbox": [ + 406, + 368, + 413, + 378 + ], + "score": 0.83, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "should correctly score", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 377, + 504, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 378, + 146, + 389 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 377, + 193, + 391 + ], + "score": 1.0, + "content": "lower than", + "type": "text" + }, + { + "bbox": [ + 193, + 379, + 232, + 390 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 377, + 326, + 391 + ], + "score": 1.0, + "content": "with a relatively small", + "type": "text" + }, + { + "bbox": [ + 327, + 381, + 333, + 387 + ], + "score": 0.73, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 377, + 496, + 391 + ], + "score": 1.0, + "content": ". In other words, the number of samples", + "type": "text" + }, + { + "bbox": [ + 497, + 380, + 504, + 387 + ], + "score": 0.72, + "content": "n", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 388, + 456, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 250, + 402 + ], + "score": 1.0, + "content": "needed for the metric to distinguish", + "type": "text" + }, + { + "bbox": [ + 251, + 389, + 262, + 401 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 388, + 280, + 402 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 281, + 389, + 292, + 401 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 388, + 456, + 402 + ], + "score": 1.0, + "content": "can be viewed as its sample complexity.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 343, + 417 + ], + "score": 1.0, + "content": "In Figure 6 we show the individual scores as a function of", + "type": "text" + }, + { + "bbox": [ + 343, + 408, + 350, + 415 + ], + "score": 0.71, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 406, + 505, + 417 + ], + "score": 1.0, + "content": ". We can observe that MMD, FID and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 415, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 429 + ], + "score": 1.0, + "content": "1-NN accuracy computed in convolution feature space are able to distinguish the two set of images", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 427, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 118, + 439 + ], + "score": 0.87, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 427, + 263, + 440 + ], + "score": 1.0, + "content": "(solid blue and magenta curves) and", + "type": "text" + }, + { + "bbox": [ + 263, + 428, + 275, + 439 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 427, + 506, + 440 + ], + "score": 1.0, + "content": "(dotted solid blue and magenta curves) with relatively few", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "samples. The Wasserstein distance (cyan curves) is not discriminative with samples size less than", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 449, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 506, + 462 + ], + "score": 1.0, + "content": "1000, while the RIS even considers the generated samples to be more “real” than the real samples on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "the LSUN dataset (the red curves in the third panel). The dotted lines in Figure 6 also quantify how", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "fast the scores converge to their expectations as we increase the sample size. Note that MMD for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 481, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 107, + 482, + 146, + 494 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 481, + 505, + 496 + ], + "score": 1.0, + "content": "converges very quickly to zero and gives discriminative scores with few samples, making", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 493, + 303, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 303, + 506 + ], + "score": 1.0, + "content": "it a practical metric for comparing GAN models.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "Computational efficiency. Fast computation of the empirical metric is of practical concern as it", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "helps researchers monitor the training process and diagnose problems early on, or perform early", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 546 + ], + "score": 1.0, + "content": "stopping. In Figure 7 we investigate the computational efficiency of the above metrics by showing", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 543, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "the wall-clock time (in log scale) to compute them as a function of the number of samples. For a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "typical number of 2000 samples, it only takes about 8 seconds to compute each of these metrics on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 565, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 577 + ], + "score": 1.0, + "content": "an NVIDIA TitanX GPU. In fact, the majority of time is spent on extracting features from the ResNet", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 462, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 462, + 588 + ], + "score": 1.0, + "content": "model. Only the Wasserstein distance becomes prohibitively slow for large sample sizes.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31 + }, + { + "type": "title", + "bbox": [ + 108, + 598, + 242, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 598, + 244, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 244, + 611 + ], + "score": 1.0, + "content": "3.6 DETECTING OVERFITTING", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "Overfitting is an artifact of training with finite samples. If a GAN successfully memorizes the training", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 625, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 155, + 640 + ], + "score": 1.0, + "content": "images, i.e.,", + "type": "text" + }, + { + "bbox": [ + 156, + 627, + 168, + 639 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 625, + 377, + 640 + ], + "score": 1.0, + "content": "is a uniform distribution over the training sample set", + "type": "text" + }, + { + "bbox": [ + 378, + 627, + 392, + 639 + ], + "score": 0.9, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 625, + 506, + 640 + ], + "score": 1.0, + "content": ", then the generated samples", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 635, + 507, + 652 + ], + "spans": [ + { + "bbox": [ + 107, + 639, + 118, + 650 + ], + "score": 0.87, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 635, + 259, + 652 + ], + "score": 1.0, + "content": "becomes a uniformly drawn set of", + "type": "text" + }, + { + "bbox": [ + 259, + 640, + 266, + 648 + ], + "score": 0.77, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 635, + 324, + 652 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 325, + 638, + 339, + 650 + ], + "score": 0.9, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 635, + 422, + 652 + ], + "score": 1.0, + "content": ", and any reasonable", + "type": "text" + }, + { + "bbox": [ + 423, + 639, + 429, + 650 + ], + "score": 0.85, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 635, + 507, + 652 + ], + "score": 1.0, + "content": "should be close to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "0. The Wasserstein distance, MMD and 1-NN two sample test are able to detect overfitting in the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 103, + 654, + 508, + 676 + ], + "spans": [ + { + "bbox": [ + 103, + 654, + 294, + 676 + ], + "score": 1.0, + "content": "following sense: if we hold out a validation set", + "type": "text" + }, + { + "bbox": [ + 294, + 660, + 313, + 672 + ], + "score": 0.89, + "content": "S _ { r } ^ { v a l }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 654, + 337, + 676 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 337, + 659, + 384, + 672 + ], + "score": 0.93, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 654, + 508, + 676 + ], + "score": 1.0, + "content": "should be significantly higher", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 127, + 684 + ], + "score": 1.0, + "content": "than", + "type": "text" + }, + { + "bbox": [ + 127, + 671, + 169, + 683 + ], + "score": 0.95, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { t r } )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 669, + 195, + 684 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 195, + 671, + 207, + 683 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 669, + 292, + 684 + ], + "score": 1.0, + "content": "memorizes a part of", + "type": "text" + }, + { + "bbox": [ + 292, + 672, + 307, + 683 + ], + "score": 0.88, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 669, + 506, + 684 + ], + "score": 1.0, + "content": ". The difference between them can informally be", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 680, + 275, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 275, + 695 + ], + "score": 1.0, + "content": "viewed as a form of “generalization gap\".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 292, + 712 + ], + "score": 1.0, + "content": "We simulate the overfitting process by defining", + "type": "text" + }, + { + "bbox": [ + 292, + 700, + 303, + 711 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 697, + 465, + 712 + ], + "score": 1.0, + "content": "as a mix of samples from the training set", + "type": "text" + }, + { + "bbox": [ + 466, + 698, + 480, + 710 + ], + "score": 0.87, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 697, + 507, + 712 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 707, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 707, + 256, + 723 + ], + "score": 1.0, + "content": "second holdout set, disjoint from both", + "type": "text" + }, + { + "bbox": [ + 257, + 711, + 271, + 722 + ], + "score": 0.91, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 707, + 289, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 290, + 711, + 308, + 722 + ], + "score": 0.9, + "content": "S _ { r } ^ { v a l }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 707, + 405, + 723 + ], + "score": 1.0, + "content": ". Figure 8 shows the gap", + "type": "text" + }, + { + "bbox": [ + 405, + 709, + 505, + 722 + ], + "score": 0.9, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } ) - \\hat { \\rho } ( \\mathbf { \\dot { } { } } S _ { g } , S _ { r } ^ { t r } )", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 718, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 378, + 734 + ], + "score": 1.0, + "content": "of the various metrics as a function of the overlapping ratio between", + "type": "text" + }, + { + "bbox": [ + 378, + 721, + 389, + 732 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 718, + 407, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 721, + 423, + 732 + ], + "score": 0.87, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 718, + 506, + 734 + ], + "score": 1.0, + "content": ". The left most point", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 504, + 194 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 504, + 194 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 504, + 194 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 504, + 194 + ], + "score": 0.969, + "type": "image", + "image_path": "2fcde7509c9dec9a43fcff36da5c2ef1c139c751b252031a98486f573d1f95c4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 504, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 117.33333333333334, + 504, + 155.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 155.66666666666669, + 504, + 194.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 197, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 197, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 506, + 208 + ], + "score": 1.0, + "content": "Figure 6: The score of various metrics as a function of the number of samples. An ideal metric should result in a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 242, + 219 + ], + "score": 1.0, + "content": "large gap between the real-real (R-R;", + "type": "text" + }, + { + "bbox": [ + 243, + 207, + 281, + 218 + ], + "score": 0.91, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 207, + 356, + 219 + ], + "score": 1.0, + "content": ") and real-fake (R-G;", + "type": "text" + }, + { + "bbox": [ + 356, + 207, + 394, + 218 + ], + "score": 0.91, + "content": "\\hat { \\rho } ( \\bar { S } _ { r } , S _ { g } ) \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 207, + 505, + 219 + ], + "score": 1.0, + "content": ") curves in order to distinguish", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 216, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 506, + 229 + ], + "score": 1.0, + "content": "between real and fake distributions using as few samples as possible. Compared with Wasserstein distance,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "MMD and 1-NN accuracy require much fewer samples to discriminate real and generated images, while RIS", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 236, + 426, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 426, + 249 + ], + "score": 1.0, + "content": "totally fails on LSUN as it scores generated images even better (lower) than real images.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 183, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 184, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 184, + 280 + ], + "score": 1.0, + "content": "3.5 EFFICIENCY", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "A practical GAN evaluation metric should be able to compute “accurate” scores from a reasonable", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "number of samples and within an affordable computation cost, such that it can be computed, for", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 306, + 374, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 374, + 318 + ], + "score": 1.0, + "content": "example, after each training epoch to monitor the training process.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 284, + 505, + 318 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "Sample efficiency. Here we measure the sample efficiency of various metrics by investigating how", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 331, + 504, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 491, + 349 + ], + "score": 1.0, + "content": "many samples are needed for each of them in order to discriminate a set of generated samples", + "type": "text" + }, + { + "bbox": [ + 492, + 334, + 504, + 346 + ], + "score": 0.86, + "content": "S _ { g }", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 278, + 357 + ], + "score": 1.0, + "content": "(from DCGAN) from a set of real samples", + "type": "text" + }, + { + "bbox": [ + 279, + 345, + 290, + 357 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 344, + 453, + 357 + ], + "score": 1.0, + "content": ". To do this, we introduce a reference set", + "type": "text" + }, + { + "bbox": [ + 453, + 345, + 465, + 356 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 344, + 505, + 357 + ], + "score": 1.0, + "content": ", which is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 384, + 369 + ], + "score": 1.0, + "content": "also uniformly sampled from the real training data, but is disjoint with", + "type": "text" + }, + { + "bbox": [ + 384, + 356, + 396, + 367 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 355, + 505, + 369 + ], + "score": 1.0, + "content": ". All three sample sets have", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 183, + 380 + ], + "score": 1.0, + "content": "the same size, i.e.,", + "type": "text" + }, + { + "bbox": [ + 183, + 367, + 278, + 379 + ], + "score": 0.9, + "content": "| S _ { r } | = | S _ { r } ^ { \\prime } | = | S _ { g } | = n", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 366, + 406, + 380 + ], + "score": 1.0, + "content": ". We expect that an ideal metric", + "type": "text" + }, + { + "bbox": [ + 406, + 368, + 413, + 378 + ], + "score": 0.83, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "should correctly score", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 377, + 504, + 391 + ], + "spans": [ + { + "bbox": [ + 107, + 378, + 146, + 389 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 377, + 193, + 391 + ], + "score": 1.0, + "content": "lower than", + "type": "text" + }, + { + "bbox": [ + 193, + 379, + 232, + 390 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { g } )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 377, + 326, + 391 + ], + "score": 1.0, + "content": "with a relatively small", + "type": "text" + }, + { + "bbox": [ + 327, + 381, + 333, + 387 + ], + "score": 0.73, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 377, + 496, + 391 + ], + "score": 1.0, + "content": ". In other words, the number of samples", + "type": "text" + }, + { + "bbox": [ + 497, + 380, + 504, + 387 + ], + "score": 0.72, + "content": "n", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 388, + 456, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 250, + 402 + ], + "score": 1.0, + "content": "needed for the metric to distinguish", + "type": "text" + }, + { + "bbox": [ + 251, + 389, + 262, + 401 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 388, + 280, + 402 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 281, + 389, + 292, + 401 + ], + "score": 0.88, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 388, + 456, + 402 + ], + "score": 1.0, + "content": "can be viewed as its sample complexity.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15, + "bbox_fs": [ + 104, + 323, + 505, + 402 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 343, + 417 + ], + "score": 1.0, + "content": "In Figure 6 we show the individual scores as a function of", + "type": "text" + }, + { + "bbox": [ + 343, + 408, + 350, + 415 + ], + "score": 0.71, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 406, + 505, + 417 + ], + "score": 1.0, + "content": ". We can observe that MMD, FID and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 415, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 429 + ], + "score": 1.0, + "content": "1-NN accuracy computed in convolution feature space are able to distinguish the two set of images", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 427, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 118, + 439 + ], + "score": 0.87, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 427, + 263, + 440 + ], + "score": 1.0, + "content": "(solid blue and magenta curves) and", + "type": "text" + }, + { + "bbox": [ + 263, + 428, + 275, + 439 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 427, + 506, + 440 + ], + "score": 1.0, + "content": "(dotted solid blue and magenta curves) with relatively few", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "samples. The Wasserstein distance (cyan curves) is not discriminative with samples size less than", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 449, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 506, + 462 + ], + "score": 1.0, + "content": "1000, while the RIS even considers the generated samples to be more “real” than the real samples on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "the LSUN dataset (the red curves in the third panel). The dotted lines in Figure 6 also quantify how", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "fast the scores converge to their expectations as we increase the sample size. Note that MMD for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 481, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 107, + 482, + 146, + 494 + ], + "score": 0.92, + "content": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 481, + 505, + 496 + ], + "score": 1.0, + "content": "converges very quickly to zero and gives discriminative scores with few samples, making", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 493, + 303, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 303, + 506 + ], + "score": 1.0, + "content": "it a practical metric for comparing GAN models.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 406, + 506, + 506 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "Computational efficiency. Fast computation of the empirical metric is of practical concern as it", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "helps researchers monitor the training process and diagnose problems early on, or perform early", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 546 + ], + "score": 1.0, + "content": "stopping. In Figure 7 we investigate the computational efficiency of the above metrics by showing", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 543, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "the wall-clock time (in log scale) to compute them as a function of the number of samples. For a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "typical number of 2000 samples, it only takes about 8 seconds to compute each of these metrics on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 565, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 577 + ], + "score": 1.0, + "content": "an NVIDIA TitanX GPU. In fact, the majority of time is spent on extracting features from the ResNet", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 462, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 462, + 588 + ], + "score": 1.0, + "content": "model. Only the Wasserstein distance becomes prohibitively slow for large sample sizes.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 510, + 506, + 588 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 598, + 242, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 598, + 244, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 244, + 611 + ], + "score": 1.0, + "content": "3.6 DETECTING OVERFITTING", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "Overfitting is an artifact of training with finite samples. If a GAN successfully memorizes the training", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 625, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 155, + 640 + ], + "score": 1.0, + "content": "images, i.e.,", + "type": "text" + }, + { + "bbox": [ + 156, + 627, + 168, + 639 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 625, + 377, + 640 + ], + "score": 1.0, + "content": "is a uniform distribution over the training sample set", + "type": "text" + }, + { + "bbox": [ + 378, + 627, + 392, + 639 + ], + "score": 0.9, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 625, + 506, + 640 + ], + "score": 1.0, + "content": ", then the generated samples", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 635, + 507, + 652 + ], + "spans": [ + { + "bbox": [ + 107, + 639, + 118, + 650 + ], + "score": 0.87, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 635, + 259, + 652 + ], + "score": 1.0, + "content": "becomes a uniformly drawn set of", + "type": "text" + }, + { + "bbox": [ + 259, + 640, + 266, + 648 + ], + "score": 0.77, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 635, + 324, + 652 + ], + "score": 1.0, + "content": "samples from", + "type": "text" + }, + { + "bbox": [ + 325, + 638, + 339, + 650 + ], + "score": 0.9, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 635, + 422, + 652 + ], + "score": 1.0, + "content": ", and any reasonable", + "type": "text" + }, + { + "bbox": [ + 423, + 639, + 429, + 650 + ], + "score": 0.85, + "content": "\\hat { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 635, + 507, + 652 + ], + "score": 1.0, + "content": "should be close to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "0. The Wasserstein distance, MMD and 1-NN two sample test are able to detect overfitting in the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 103, + 654, + 508, + 676 + ], + "spans": [ + { + "bbox": [ + 103, + 654, + 294, + 676 + ], + "score": 1.0, + "content": "following sense: if we hold out a validation set", + "type": "text" + }, + { + "bbox": [ + 294, + 660, + 313, + 672 + ], + "score": 0.89, + "content": "S _ { r } ^ { v a l }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 654, + 337, + 676 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 337, + 659, + 384, + 672 + ], + "score": 0.93, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 654, + 508, + 676 + ], + "score": 1.0, + "content": "should be significantly higher", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 127, + 684 + ], + "score": 1.0, + "content": "than", + "type": "text" + }, + { + "bbox": [ + 127, + 671, + 169, + 683 + ], + "score": 0.95, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { t r } )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 669, + 195, + 684 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 195, + 671, + 207, + 683 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 669, + 292, + 684 + ], + "score": 1.0, + "content": "memorizes a part of", + "type": "text" + }, + { + "bbox": [ + 292, + 672, + 307, + 683 + ], + "score": 0.88, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 669, + 506, + 684 + ], + "score": 1.0, + "content": ". The difference between them can informally be", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 680, + 275, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 275, + 695 + ], + "score": 1.0, + "content": "viewed as a form of “generalization gap\".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39, + "bbox_fs": [ + 103, + 614, + 508, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 292, + 712 + ], + "score": 1.0, + "content": "We simulate the overfitting process by defining", + "type": "text" + }, + { + "bbox": [ + 292, + 700, + 303, + 711 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 697, + 465, + 712 + ], + "score": 1.0, + "content": "as a mix of samples from the training set", + "type": "text" + }, + { + "bbox": [ + 466, + 698, + 480, + 710 + ], + "score": 0.87, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 697, + 507, + 712 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 707, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 707, + 256, + 723 + ], + "score": 1.0, + "content": "second holdout set, disjoint from both", + "type": "text" + }, + { + "bbox": [ + 257, + 711, + 271, + 722 + ], + "score": 0.91, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 707, + 289, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 290, + 711, + 308, + 722 + ], + "score": 0.9, + "content": "S _ { r } ^ { v a l }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 707, + 405, + 723 + ], + "score": 1.0, + "content": ". Figure 8 shows the gap", + "type": "text" + }, + { + "bbox": [ + 405, + 709, + 505, + 722 + ], + "score": 0.9, + "content": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } ) - \\hat { \\rho } ( \\mathbf { \\dot { } { } } S _ { g } , S _ { r } ^ { t r } )", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 718, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 378, + 734 + ], + "score": 1.0, + "content": "of the various metrics as a function of the overlapping ratio between", + "type": "text" + }, + { + "bbox": [ + 378, + 721, + 389, + 732 + ], + "score": 0.89, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 718, + 407, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 721, + 423, + 732 + ], + "score": 0.87, + "content": "S _ { r } ^ { t r }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 718, + 506, + 734 + ], + "score": 1.0, + "content": ". The left most point", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 234, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 104, + 234, + 271, + 250 + ], + "score": 1.0, + "content": "of each curve can be viewed as the score", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 271, + 236, + 317, + 249 + ], + "score": 0.93, + "content": "\\hat { \\rho } ( S _ { r } ^ { \\prime } , S _ { r } ^ { v a l } )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 318, + 234, + 506, + 250 + ], + "score": 1.0, + "content": "computed on a validation set since the overlap", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 260 + ], + "score": 1.0, + "content": "ratio is 0. For better visualization, we normalize the Wasserstein distance and MMD by dividing their", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 214, + 272 + ], + "score": 1.0, + "content": "corresponding score when", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 214, + 258, + 226, + 270 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 226, + 258, + 244, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 244, + 259, + 256, + 270 + ], + "score": 0.89, + "content": "S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 256, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "have no overlap. As shown in Figure 8, all the metrics except", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 362, + 282 + ], + "score": 1.0, + "content": "RIS and RMS reflect that the “generalization gap\" increases as", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 363, + 270, + 374, + 281 + ], + "score": 0.9, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 374, + 269, + 442, + 282 + ], + "score": 1.0, + "content": "overfits more to", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 442, + 270, + 453, + 280 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 453, + 269, + 506, + 282 + ], + "score": 1.0, + "content": ". The failure", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "of RIS is not surprising: it totally ignores the real data distribution as we discussed in Section 2.2.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "score": 1.0, + "content": "While the reason that RMS also fails to detect overfitting may again be its lack of generalization to", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "datasets with classes not contained in the ImageNet dataset. In addition, RMS operates in the softmax", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 314, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 506, + 326 + ], + "score": 1.0, + "content": "space, the features in which might be too specific compared to the features in the convolutional space.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + } + ], + "index": 44, + "bbox_fs": [ + 104, + 697, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 183 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 183 + ], + "score": 0.964, + "type": "image", + "image_path": "a7fbd0dd8d052615064f08210b23e53041fd3c73b5d0a1f9762bc4f7a90af568.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 113.66666666666666, + 504, + 148.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 148.33333333333331, + 504, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 506, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 185, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 505, + 197 + ], + "score": 1.0, + "content": "Figure 8: Experiment on detecting overfitting of generated samples. As more generated samples overlap with real", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "samples from the training set, the gap between validation and training score should increase to signal overfitting.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "All metrics behave correctly except for RIS and RMS, as these two metrics do not increase when the fraction of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 215, + 220, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 220, + 227 + ], + "score": 1.0, + "content": "overlapping samples increases.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 505, + 325 + ], + "lines": [ + { + "bbox": [ + 104, + 234, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 104, + 234, + 271, + 250 + ], + "score": 1.0, + "content": "of each curve can be viewed as the score", + "type": "text" + }, + { + "bbox": [ + 271, + 236, + 317, + 249 + ], + "score": 0.93, + "content": "\\hat { \\rho } ( S _ { r } ^ { \\prime } , S _ { r } ^ { v a l } )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 234, + 506, + 250 + ], + "score": 1.0, + "content": "computed on a validation set since the overlap", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 260 + ], + "score": 1.0, + "content": "ratio is 0. For better visualization, we normalize the Wasserstein distance and MMD by dividing their", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 214, + 272 + ], + "score": 1.0, + "content": "corresponding score when", + "type": "text" + }, + { + "bbox": [ + 214, + 258, + 226, + 270 + ], + "score": 0.88, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 258, + 244, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 244, + 259, + 256, + 270 + ], + "score": 0.89, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "have no overlap. As shown in Figure 8, all the metrics except", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 362, + 282 + ], + "score": 1.0, + "content": "RIS and RMS reflect that the “generalization gap\" increases as", + "type": "text" + }, + { + "bbox": [ + 363, + 270, + 374, + 281 + ], + "score": 0.9, + "content": "S _ { r } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 269, + 442, + 282 + ], + "score": 1.0, + "content": "overfits more to", + "type": "text" + }, + { + "bbox": [ + 442, + 270, + 453, + 280 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 269, + 506, + 282 + ], + "score": 1.0, + "content": ". The failure", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "of RIS is not surprising: it totally ignores the real data distribution as we discussed in Section 2.2.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "score": 1.0, + "content": "While the reason that RMS also fails to detect overfitting may again be its lack of generalization to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "datasets with classes not contained in the ImageNet dataset. In addition, RMS operates in the softmax", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 314, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 506, + 326 + ], + "score": 1.0, + "content": "space, the features in which might be too specific compared to the features in the convolutional space.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 108, + 339, + 293, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 295, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 295, + 355 + ], + "score": 1.0, + "content": "4 DISCUSSIONS AND CONCLUSION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 361, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "Based on the above analysis, we can summarize the advantages and inherent limitations of the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "five evaluation metrics, and conditions under which they produce meaningful results. With some", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "of the metrics, we are able to study the problem of overfitting (see Appendix C), perform model", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "selection on GAN models and compare GAN models without resorting to human evaluation based on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 406, + 274, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 274, + 419 + ], + "score": 1.0, + "content": "cherry-picked samples (see Appendix D).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "score": 1.0, + "content": "The Inception Score does show a reasonable correlation with the quality and diversity of generated", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "images, which explains the wide usage in practice. However, it is ill-posed mostly because it only", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 146, + 457 + ], + "score": 1.0, + "content": "evaluates", + "type": "text" + }, + { + "bbox": [ + 146, + 445, + 159, + 457 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 444, + 400, + 457 + ], + "score": 1.0, + "content": "as an image generation model rather than its similarity to", + "type": "text" + }, + { + "bbox": [ + 400, + 445, + 412, + 456 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 444, + 505, + 457 + ], + "score": 1.0, + "content": ". Blunt violations like", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "mixing in natural images from an entirely different distribution completely deceives the Inception", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "Score. As a result, it may encourage the models to simply learn sharp and diversified images (or even", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 244, + 490 + ], + "score": 1.0, + "content": "some adversarial noise), instead of", + "type": "text" + }, + { + "bbox": [ + 244, + 478, + 255, + 488 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 478, + 505, + 490 + ], + "score": 1.0, + "content": ". This also applies to the Mode Score. Moreover, the Inception", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 488, + 459, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 459, + 501 + ], + "score": 1.0, + "content": "Score is unable to detect overfitting since it cannot make use of a holdout validation set.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "Kernel MMD works surprising well when it operates in the feature space of a pre-trained ResNet. It", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "is always able to identify generative/noise images from real images, and both its sample complexity", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "and computational complexity are low. Given these advantages, even though MMD is biased, we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 231, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 231, + 551 + ], + "score": 1.0, + "content": "recommend its use in practice.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 556, + 507, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 507, + 568 + ], + "score": 1.0, + "content": "Wasserstein distance works well when the base distance is computed in a suitable feature space.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "However, it has a high sample complexity, a fact that has been independently observed by (Arora", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "et al., 2017). Another key weakness is that computing the exact Wasserstein distance has a time", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 587, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 164, + 601 + ], + "score": 1.0, + "content": "complexity of", + "type": "text" + }, + { + "bbox": [ + 165, + 587, + 191, + 600 + ], + "score": 0.92, + "content": "O ( n ^ { 3 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 587, + 506, + 601 + ], + "score": 1.0, + "content": ", which is prohibitively expensive as sample size increases. Compared to other", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 599, + 429, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 429, + 612 + ], + "score": 1.0, + "content": "methods, Wasserstein distance is less appealing as a practical evaluation metric.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 108, + 616, + 502, + 649 + ], + "lines": [ + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "Fréchet Inception Distance performs well in terms of discriminability, robustness and efficiency. It", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "appears to be a good evaluation metric for GANs, even it only takes into consideration the first two", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 639, + 249, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 249, + 649 + ], + "score": 1.0, + "content": "order moments of the distributions.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "1-NN classifier seems to be an ideal metric for evaluating GANs. Not only does it enjoy all the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "advantages of the other metrics, it also outputs a score in the interval [0, 1], similar to the accuracy/error", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "in classification problems. When the generative distribution perfectly match the true distribution,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 183, + 700 + ], + "score": 1.0, + "content": "perfect score (i.e.,", + "type": "text" + }, + { + "bbox": [ + 183, + 687, + 204, + 699 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "accuracy) is attainable with a reasonable (e.g., 1000) sample size. From", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "Figure 2, we find that typical GAN models tend to achieve lower LOO accuracy for real samples", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "(1-NN accuracy (real)), while higher LOO accuracy for generated samples (1-NN accuracy (fake)).", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "This suggests that GANs are able to capture modes from the training distribution, such that the", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 183 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 183 + ], + "score": 0.964, + "type": "image", + "image_path": "a7fbd0dd8d052615064f08210b23e53041fd3c73b5d0a1f9762bc4f7a90af568.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 113.66666666666666, + 504, + 148.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 148.33333333333331, + 504, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 506, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 185, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 505, + 197 + ], + "score": 1.0, + "content": "Figure 8: Experiment on detecting overfitting of generated samples. As more generated samples overlap with real", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "samples from the training set, the gap between validation and training score should increase to signal overfitting.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "All metrics behave correctly except for RIS and RMS, as these two metrics do not increase when the fraction of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 215, + 220, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 220, + 227 + ], + "score": 1.0, + "content": "overlapping samples increases.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 505, + 325 + ], + "lines": [], + "index": 10.5, + "bbox_fs": [ + 104, + 234, + 506, + 326 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 339, + 293, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 295, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 295, + 355 + ], + "score": 1.0, + "content": "4 DISCUSSIONS AND CONCLUSION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 361, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "Based on the above analysis, we can summarize the advantages and inherent limitations of the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "five evaluation metrics, and conditions under which they produce meaningful results. With some", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "of the metrics, we are able to study the problem of overfitting (see Appendix C), perform model", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "selection on GAN models and compare GAN models without resorting to human evaluation based on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 406, + 274, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 274, + 419 + ], + "score": 1.0, + "content": "cherry-picked samples (see Appendix D).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 362, + 505, + 419 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "score": 1.0, + "content": "The Inception Score does show a reasonable correlation with the quality and diversity of generated", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "images, which explains the wide usage in practice. However, it is ill-posed mostly because it only", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 146, + 457 + ], + "score": 1.0, + "content": "evaluates", + "type": "text" + }, + { + "bbox": [ + 146, + 445, + 159, + 457 + ], + "score": 0.89, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 444, + 400, + 457 + ], + "score": 1.0, + "content": "as an image generation model rather than its similarity to", + "type": "text" + }, + { + "bbox": [ + 400, + 445, + 412, + 456 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 444, + 505, + 457 + ], + "score": 1.0, + "content": ". Blunt violations like", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "mixing in natural images from an entirely different distribution completely deceives the Inception", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "Score. As a result, it may encourage the models to simply learn sharp and diversified images (or even", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 244, + 490 + ], + "score": 1.0, + "content": "some adversarial noise), instead of", + "type": "text" + }, + { + "bbox": [ + 244, + 478, + 255, + 488 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 478, + 505, + 490 + ], + "score": 1.0, + "content": ". This also applies to the Mode Score. Moreover, the Inception", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 488, + 459, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 459, + 501 + ], + "score": 1.0, + "content": "Score is unable to detect overfitting since it cannot make use of a holdout validation set.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 422, + 505, + 501 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "Kernel MMD works surprising well when it operates in the feature space of a pre-trained ResNet. It", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "is always able to identify generative/noise images from real images, and both its sample complexity", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "and computational complexity are low. Given these advantages, even though MMD is biased, we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 231, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 231, + 551 + ], + "score": 1.0, + "content": "recommend its use in practice.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 505, + 506, + 551 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 556, + 507, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 507, + 568 + ], + "score": 1.0, + "content": "Wasserstein distance works well when the base distance is computed in a suitable feature space.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "However, it has a high sample complexity, a fact that has been independently observed by (Arora", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "et al., 2017). Another key weakness is that computing the exact Wasserstein distance has a time", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 587, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 164, + 601 + ], + "score": 1.0, + "content": "complexity of", + "type": "text" + }, + { + "bbox": [ + 165, + 587, + 191, + 600 + ], + "score": 0.92, + "content": "O ( n ^ { 3 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 587, + 506, + 601 + ], + "score": 1.0, + "content": ", which is prohibitively expensive as sample size increases. Compared to other", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 599, + 429, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 429, + 612 + ], + "score": 1.0, + "content": "methods, Wasserstein distance is less appealing as a practical evaluation metric.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 556, + 507, + 612 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 616, + 502, + 649 + ], + "lines": [ + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "Fréchet Inception Distance performs well in terms of discriminability, robustness and efficiency. It", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "appears to be a good evaluation metric for GANs, even it only takes into consideration the first two", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 639, + 249, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 249, + 649 + ], + "score": 1.0, + "content": "order moments of the distributions.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 106, + 616, + 505, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "1-NN classifier seems to be an ideal metric for evaluating GANs. Not only does it enjoy all the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "advantages of the other metrics, it also outputs a score in the interval [0, 1], similar to the accuracy/error", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "in classification problems. When the generative distribution perfectly match the true distribution,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 183, + 700 + ], + "score": 1.0, + "content": "perfect score (i.e.,", + "type": "text" + }, + { + "bbox": [ + 183, + 687, + 204, + 699 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "accuracy) is attainable with a reasonable (e.g., 1000) sample size. From", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "Figure 2, we find that typical GAN models tend to achieve lower LOO accuracy for real samples", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "(1-NN accuracy (real)), while higher LOO accuracy for generated samples (1-NN accuracy (fake)).", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "This suggests that GANs are able to capture modes from the training distribution, such that the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "majority of training samples distributed around the mode centers have their nearest neighbor from the", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 505, + 107 + ], + "score": 1.0, + "content": "generated images, yet most of the generated images are still surrounded by generated images as they", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "are collapsed. The observation indicates that the mode collapse problem is prevalent for typical GAN", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "models. We also note that this problem, however, cannot be effectively detected by human evaluation", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 250, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 250, + 139 + ], + "score": 1.0, + "content": "or the widely used Inception Score.", + "type": "text", + "cross_page": true + } + ], + "index": 4 + } + ], + "index": 43, + "bbox_fs": [ + 104, + 655, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "majority of training samples distributed around the mode centers have their nearest neighbor from the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 505, + 107 + ], + "score": 1.0, + "content": "generated images, yet most of the generated images are still surrounded by generated images as they", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "are collapsed. The observation indicates that the mode collapse problem is prevalent for typical GAN", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "models. We also note that this problem, however, cannot be effectively detected by human evaluation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 250, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 250, + 139 + ], + "score": 1.0, + "content": "or the widely used Inception Score.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Overall, our empirical study suggests that the choice of feature space in which to compute various", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "metrics is crucial. In the convolutional space of a ResNet pretrained on ImageNet, both MMD and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 164, + 507, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 164, + 507, + 179 + ], + "score": 1.0, + "content": "1-NN accuracy appear to be good metrics in terms of discriminability, robustness and efficiency.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "Wasserstein distance has very poor sample efficiency, while Inception Score and Mode Score appear", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "score": 1.0, + "content": "to be unsuitable for datasets that are very different from ImageNet. We will release our source code", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "for all these metrics, providing researchers with an off-the-shelf tool to compare and improve GAN", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 155, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 155, + 223 + ], + "score": 1.0, + "content": "algorithms.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "score": 1.0, + "content": "Based on the two most prominent metrics, MMD and 1-NN accuracy, we study the overfitting", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 236, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 505, + 250 + ], + "score": 1.0, + "content": "problem of DCGAN and WGAN (in Appendix C). Despite the widespread belief that GANs are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 246, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 262 + ], + "score": 1.0, + "content": "overfitting to the training data, we find that this does not occur unless there are very few training", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "samples. This raises an interesting question regarding the generalization of GANs in comparison to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 269, + 488, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 488, + 282 + ], + "score": 1.0, + "content": "the supervised setting. We hope that future work can contribute to explaining this phenomenon.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 107, + 299, + 175, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 176, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 176, + 312 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein GAN. arXiv preprint arX-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 328, + 205, + 339 + ], + "spans": [ + { + "bbox": [ + 116, + 328, + 205, + 339 + ], + "score": 1.0, + "content": "iv:1701.07875, 2017.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 347, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 504, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 504, + 361 + ], + "score": 1.0, + "content": "Sanjeev Arora, Rong Ge, Yingyu Liang, Tengyu Ma, and Yi Zhang. Generalization and equilibrium", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 359, + 429, + 371 + ], + "spans": [ + { + "bbox": [ + 115, + 359, + 429, + 371 + ], + "score": 1.0, + "content": "in generative adversarial nets (gans). arXiv preprint arXiv:1703.00573, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 108, + 378, + 503, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Wacha Bounliphone, Eugene Belilovsky, Matthew B Blaschko, Ioannis Antonoglou, and Arthur", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 116, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "Gretton. A test of relative similarity for model selection in generative models. arXiv preprint", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 401, + 219, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 401, + 219, + 411 + ], + "score": 1.0, + "content": "arXiv:1511.04581, 2015.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 105, + 419, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 430, + 367, + 443 + ], + "spans": [ + { + "bbox": [ + 115, + 430, + 367, + 443 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv:1612.02136, 2016.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 504, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 473, + 309, + 485 + ], + "spans": [ + { + "bbox": [ + 116, + 473, + 309, + 485 + ], + "score": 1.0, + "content": "IEEE Conference on, pp. 248–255. IEEE, 2009.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "score": 1.0, + "content": "David Forsyth and Jean Ponce. Computer vision: a modern approach. Upper Saddle River, NJ;", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 504, + 236, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 236, + 515 + ], + "score": 1.0, + "content": "London: Prentice Hall, 2011.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 507, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 507, + 536 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 534, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 534, + 506, + 546 + ], + "score": 1.0, + "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural informa-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 545, + 305, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 305, + 557 + ], + "score": 1.0, + "content": "tion processing systems, pp. 2672–2680, 2014.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 507, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 507, + 577 + ], + "score": 1.0, + "content": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, Alexander J Smola, et al.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 574, + 507, + 590 + ], + "spans": [ + { + "bbox": [ + 114, + 574, + 507, + 590 + ], + "score": 1.0, + "content": "A kernel method for the two-sample-problem. Advances in neural information processing systems,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 587, + 173, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 587, + 173, + 598 + ], + "score": 1.0, + "content": "19:513, 2007.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 104, + 606, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. Improved", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 617, + 396, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 396, + 630 + ], + "score": 1.0, + "content": "training of wasserstein gans. arXiv preprint arXiv:1704.00028, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 637, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Günter Klambauer, and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a nash equilibrium.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 659, + 279, + 671 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 279, + 671 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1706.08500, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 689, + 414, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 414, + 702 + ], + "score": 1.0, + "content": "conditional adversarial networks. arXiv preprint arXiv:1611.07004, 2016.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. In", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 194, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 194, + 732 + ], + "score": 1.0, + "content": "Tech Report, 2009.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [], + "index": 2, + "bbox_fs": [ + 105, + 83, + 505, + 139 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Overall, our empirical study suggests that the choice of feature space in which to compute various", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "metrics is crucial. In the convolutional space of a ResNet pretrained on ImageNet, both MMD and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 164, + 507, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 164, + 507, + 179 + ], + "score": 1.0, + "content": "1-NN accuracy appear to be good metrics in terms of discriminability, robustness and efficiency.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "Wasserstein distance has very poor sample efficiency, while Inception Score and Mode Score appear", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "score": 1.0, + "content": "to be unsuitable for datasets that are very different from ImageNet. We will release our source code", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "for all these metrics, providing researchers with an off-the-shelf tool to compare and improve GAN", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 155, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 155, + 223 + ], + "score": 1.0, + "content": "algorithms.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8, + "bbox_fs": [ + 104, + 143, + 507, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "score": 1.0, + "content": "Based on the two most prominent metrics, MMD and 1-NN accuracy, we study the overfitting", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 236, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 505, + 250 + ], + "score": 1.0, + "content": "problem of DCGAN and WGAN (in Appendix C). Despite the widespread belief that GANs are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 246, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 262 + ], + "score": 1.0, + "content": "overfitting to the training data, we find that this does not occur unless there are very few training", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "samples. This raises an interesting question regarding the generalization of GANs in comparison to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 269, + 488, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 488, + 282 + ], + "score": 1.0, + "content": "the supervised setting. We hope that future work can contribute to explaining this phenomenon.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 225, + 505, + 282 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 299, + 175, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 176, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 176, + 312 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein GAN. arXiv preprint arX-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 328, + 205, + 339 + ], + "spans": [ + { + "bbox": [ + 116, + 328, + 205, + 339 + ], + "score": 1.0, + "content": "iv:1701.07875, 2017.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 316, + 506, + 339 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 347, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 504, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 504, + 361 + ], + "score": 1.0, + "content": "Sanjeev Arora, Rong Ge, Yingyu Liang, Tengyu Ma, and Yi Zhang. Generalization and equilibrium", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 359, + 429, + 371 + ], + "spans": [ + { + "bbox": [ + 115, + 359, + 429, + 371 + ], + "score": 1.0, + "content": "in generative adversarial nets (gans). arXiv preprint arXiv:1703.00573, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 348, + 504, + 371 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 378, + 503, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Wacha Bounliphone, Eugene Belilovsky, Matthew B Blaschko, Ioannis Antonoglou, and Arthur", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 116, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "Gretton. A test of relative similarity for model selection in generative models. arXiv preprint", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 401, + 219, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 401, + 219, + 411 + ], + "score": 1.0, + "content": "arXiv:1511.04581, 2015.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 378, + 505, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 419, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 430, + 367, + 443 + ], + "spans": [ + { + "bbox": [ + 115, + 430, + 367, + 443 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv:1612.02136, 2016.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 420, + 505, + 443 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 504, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 473, + 309, + 485 + ], + "spans": [ + { + "bbox": [ + 116, + 473, + 309, + 485 + ], + "score": 1.0, + "content": "IEEE Conference on, pp. 248–255. IEEE, 2009.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 451, + 505, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "score": 1.0, + "content": "David Forsyth and Jean Ponce. Computer vision: a modern approach. Upper Saddle River, NJ;", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 504, + 236, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 236, + 515 + ], + "score": 1.0, + "content": "London: Prentice Hall, 2011.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 492, + 506, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 507, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 507, + 536 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 534, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 534, + 506, + 546 + ], + "score": 1.0, + "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural informa-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 545, + 305, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 305, + 557 + ], + "score": 1.0, + "content": "tion processing systems, pp. 2672–2680, 2014.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 522, + 507, + 557 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 507, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 507, + 577 + ], + "score": 1.0, + "content": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, Alexander J Smola, et al.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 574, + 507, + 590 + ], + "spans": [ + { + "bbox": [ + 114, + 574, + 507, + 590 + ], + "score": 1.0, + "content": "A kernel method for the two-sample-problem. Advances in neural information processing systems,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 587, + 173, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 587, + 173, + 598 + ], + "score": 1.0, + "content": "19:513, 2007.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 564, + 507, + 598 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 606, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. Improved", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 617, + 396, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 396, + 630 + ], + "score": 1.0, + "content": "training of wasserstein gans. arXiv preprint arXiv:1704.00028, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 606, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 637, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Günter Klambauer, and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a nash equilibrium.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 659, + 279, + 671 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 279, + 671 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1706.08500, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 106, + 637, + 505, + 671 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 689, + 414, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 414, + 702 + ], + "score": 1.0, + "content": "conditional adversarial networks. arXiv preprint arXiv:1611.07004, 2016.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 678, + 505, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. In", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 194, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 194, + 732 + ], + "score": 1.0, + "content": "Tech Report, 2009.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 708, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, and Ole Winther. Au-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "toencoding beyond pixels using a learned similarity metric. arXiv preprint arXiv:1512.09300,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 143, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 143, + 117 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 505, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "David Lopez-Paz and Maxime Oquab. Revisiting classifier two-sample tests. arXiv preprint arX-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 204, + 145 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 204, + 145 + ], + "score": 1.0, + "content": "iv:1610.06545, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian Goodfellow, and Brendan Frey. Adversarial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 338, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 338, + 176 + ], + "score": 1.0, + "content": "autoencoders. arXiv preprint arXiv:1511.05644, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 206 + ], + "lines": [ + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "score": 1.0, + "content": "Stéphane Mallat. Understanding deep convolutional networks. Phil. Trans. R. Soc. A, 374(2065):", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 186, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 186, + 206 + ], + "score": 1.0, + "content": "20150203, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 105, + 212, + 504, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 444, + 236 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 444, + 236 + ], + "score": 1.0, + "content": "squares generative adversarial networks. arXiv preprint ArXiv:1611.04076, 2016.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 105, + 242, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "Augustus Odena. Semi-supervised learning with generative adversarial networks. arXiv preprint", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 254, + 219, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 254, + 219, + 265 + ], + "score": 1.0, + "content": "arXiv:1606.01583, 2016.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 286 + ], + "score": 1.0, + "content": "Guo-Jun Qi. Loss-sensitive generative adversarial networks on lipschitz densities. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 283, + 220, + 294 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 220, + 294 + ], + "score": 1.0, + "content": "arXiv:1701.06264, 2017.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 314, + 468, + 326 + ], + "spans": [ + { + "bbox": [ + 116, + 314, + 468, + 326 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "Improved techniques for training GANs. In Advances in Neural Information Processing Systems,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 354, + 207, + 366 + ], + "spans": [ + { + "bbox": [ + 114, + 354, + 207, + 366 + ], + "score": 1.0, + "content": "pp. 2226–2234, 2016.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 385 + ], + "score": 1.0, + "content": "Ashish Shrivastava, Tomas Pfister, Oncel Tuzel, Josh Susskind, Wenda Wang, and Russ Webb.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 115, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Learning from simulated and unsupervised images through adversarial training. arXiv preprint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 396, + 219, + 406 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 219, + 406 + ], + "score": 1.0, + "content": "arXiv:1612.07828, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 414, + 505, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 428 + ], + "score": 1.0, + "content": "Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott E. Reed, Dragomir Anguelov,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 115, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 117, + 436, + 273, + 448 + ], + "spans": [ + { + "bbox": [ + 117, + 436, + 273, + 448 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1409.4842, 2014.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 504, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 468 + ], + "score": 1.0, + "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 476, + 335, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 335, + 490 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition, pp. 2818–2826, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 504, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Lucas Theis, Aäron van den Oord, and Matthias Bethge. A note on the evaluation of generative", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 507, + 314, + 519 + ], + "spans": [ + { + "bbox": [ + 115, + 507, + 314, + 519 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:1511.01844, 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 507, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 507, + 540 + ], + "score": 1.0, + "content": "Paul Upchurch, Jacob R. Gardner, Kavita Bala, Robert Pless, Noah Snavely, and Kilian Q. Weinberger.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Deep feature interpolation for image content changes. In Proceedings of the IEEE Conference on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 546, + 374, + 561 + ], + "spans": [ + { + "bbox": [ + 116, + 546, + 374, + 561 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. in press ..., 2017.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 105, + 566, + 504, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Yuhuai Wu, Yuri Burda, Ruslan Salakhutdinov, and Roger Grosse. On the quantitative analysis of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 578, + 417, + 589 + ], + "spans": [ + { + "bbox": [ + 116, + 578, + 417, + 589 + ], + "score": 1.0, + "content": "decoder-based generative models. arXiv preprint arXiv:1611.04273, 2016.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 506, + 610 + ], + "score": 1.0, + "content": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 606, + 475, + 620 + ], + "spans": [ + { + "bbox": [ + 116, + 606, + 475, + 620 + ], + "score": 1.0, + "content": "deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 105, + 626, + 505, + 650 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "Junbo Jake Zhao, Michaël Mathieu, and Yann LeCun. Energy-based generative adversarial network.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 637, + 279, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 637, + 279, + 650 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1609.03126, 2016.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 105, + 656, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 667, + 458, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 667, + 458, + 680 + ], + "score": 1.0, + "content": "using cycle-consistent adversarial networks. arXiv preprint arXiv:1703.10593, 2017.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, and Ole Winther. Au-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "toencoding beyond pixels using a learned similarity metric. arXiv preprint arXiv:1512.09300,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 143, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 143, + 117 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 506, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 505, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "David Lopez-Paz and Maxime Oquab. Revisiting classifier two-sample tests. arXiv preprint arX-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 204, + 145 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 204, + 145 + ], + "score": 1.0, + "content": "iv:1610.06545, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 123, + 506, + 145 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian Goodfellow, and Brendan Frey. Adversarial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 338, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 338, + 176 + ], + "score": 1.0, + "content": "autoencoders. arXiv preprint arXiv:1511.05644, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 153, + 505, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 206 + ], + "lines": [ + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "score": 1.0, + "content": "Stéphane Mallat. Understanding deep convolutional networks. Phil. Trans. R. Soc. A, 374(2065):", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 186, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 186, + 206 + ], + "score": 1.0, + "content": "20150203, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 182, + 506, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 212, + 504, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 444, + 236 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 444, + 236 + ], + "score": 1.0, + "content": "squares generative adversarial networks. arXiv preprint ArXiv:1611.04076, 2016.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 213, + 505, + 236 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 242, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "Augustus Odena. Semi-supervised learning with generative adversarial networks. arXiv preprint", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 254, + 219, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 254, + 219, + 265 + ], + "score": 1.0, + "content": "arXiv:1606.01583, 2016.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 242, + 505, + 265 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 286 + ], + "score": 1.0, + "content": "Guo-Jun Qi. Loss-sensitive generative adversarial networks on lipschitz densities. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 283, + 220, + 294 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 220, + 294 + ], + "score": 1.0, + "content": "arXiv:1701.06264, 2017.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 271, + 505, + 294 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 314, + 468, + 326 + ], + "spans": [ + { + "bbox": [ + 116, + 314, + 468, + 326 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 302, + 505, + 326 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "Improved techniques for training GANs. In Advances in Neural Information Processing Systems,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 354, + 207, + 366 + ], + "spans": [ + { + "bbox": [ + 114, + 354, + 207, + 366 + ], + "score": 1.0, + "content": "pp. 2226–2234, 2016.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 332, + 506, + 366 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 385 + ], + "score": 1.0, + "content": "Ashish Shrivastava, Tomas Pfister, Oncel Tuzel, Josh Susskind, Wenda Wang, and Russ Webb.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 115, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Learning from simulated and unsupervised images through adversarial training. arXiv preprint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 396, + 219, + 406 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 219, + 406 + ], + "score": 1.0, + "content": "arXiv:1612.07828, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 373, + 506, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 414, + 505, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 428 + ], + "score": 1.0, + "content": "Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott E. Reed, Dragomir Anguelov,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 115, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 117, + 436, + 273, + 448 + ], + "spans": [ + { + "bbox": [ + 117, + 436, + 273, + 448 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1409.4842, 2014.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 413, + 506, + 448 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 504, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 468 + ], + "score": 1.0, + "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 476, + 335, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 335, + 490 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition, pp. 2818–2826, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 454, + 505, + 490 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 504, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Lucas Theis, Aäron van den Oord, and Matthias Bethge. A note on the evaluation of generative", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 507, + 314, + 519 + ], + "spans": [ + { + "bbox": [ + 115, + 507, + 314, + 519 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:1511.01844, 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 495, + 505, + 519 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 507, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 507, + 540 + ], + "score": 1.0, + "content": "Paul Upchurch, Jacob R. Gardner, Kavita Bala, Robert Pless, Noah Snavely, and Kilian Q. Weinberger.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Deep feature interpolation for image content changes. In Proceedings of the IEEE Conference on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 546, + 374, + 561 + ], + "spans": [ + { + "bbox": [ + 116, + 546, + 374, + 561 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. in press ..., 2017.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 524, + 507, + 561 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 566, + 504, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Yuhuai Wu, Yuri Burda, Ruslan Salakhutdinov, and Roger Grosse. On the quantitative analysis of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 578, + 417, + 589 + ], + "spans": [ + { + "bbox": [ + 116, + 578, + 417, + 589 + ], + "score": 1.0, + "content": "decoder-based generative models. arXiv preprint arXiv:1611.04273, 2016.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 565, + 506, + 589 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 506, + 610 + ], + "score": 1.0, + "content": "Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 606, + 475, + 620 + ], + "spans": [ + { + "bbox": [ + 116, + 606, + 475, + 620 + ], + "score": 1.0, + "content": "deep learning requires rethinking generalization. arXiv preprint arXiv:1611.03530, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 595, + 506, + 620 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 626, + 505, + 650 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "Junbo Jake Zhao, Michaël Mathieu, and Yann LeCun. Energy-based generative adversarial network.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 637, + 279, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 637, + 279, + 650 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1609.03126, 2016.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 626, + 506, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 656, + 504, + 680 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 667, + 458, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 667, + 458, + 680 + ], + "score": 1.0, + "content": "using cycle-consistent adversarial networks. arXiv preprint arXiv:1703.10593, 2017.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 655, + 505, + 680 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 358, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 359, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 359, + 95 + ], + "score": 1.0, + "content": "A GAN VARIANTS USED IN OUR EXPERIMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 101, + 504, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 114 + ], + "score": 1.0, + "content": "Many GAN variants have been proposed recently. In this paper we consider several of them, which", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 112, + 238, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 238, + 124 + ], + "score": 1.0, + "content": "we briefly review in this section.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 128, + 505, + 184 + ], + "lines": [ + { + "bbox": [ + 106, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "DCGAN (Radford et al., 2015). The generator of a DCGAN takes a lower dimensional input from a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 505, + 153 + ], + "score": 1.0, + "content": "uniform noise distribution, then projects and reshapes it to a small convolutional representation with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 164 + ], + "score": 1.0, + "content": "many feature maps. After applying a series of four fractionally-strided convolutions, the generator", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 161, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 246, + 176 + ], + "score": 1.0, + "content": "converts this representation into a", + "type": "text" + }, + { + "bbox": [ + 246, + 162, + 281, + 173 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 161, + 505, + 176 + ], + "score": 1.0, + "content": "pixel image. DCGAN is optimized by minimizing the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 173, + 378, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 378, + 186 + ], + "score": 1.0, + "content": "Jensen-Shannon divergence between the real and generated images.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 190, + 505, + 234 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "WGAN (Arjovsky et al., 2017). A critic network that outputs unconstrained real values is used in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "place of the discriminator. When the critic is Lipschitz, this network approximates the Wasserstein", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 211, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 180, + 225 + ], + "score": 1.0, + "content": "distance between", + "type": "text" + }, + { + "bbox": [ + 180, + 212, + 192, + 223 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 211, + 212, + 225 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 212, + 212, + 224, + 224 + ], + "score": 0.89, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 211, + 505, + 225 + ], + "score": 1.0, + "content": ". A Lipschitz condition is enforced by clipping the critic networks’", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 223, + 321, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 321, + 235 + ], + "score": 1.0, + "content": "parameters to stay within a predefined bounding box.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 502, + 273 + ], + "lines": [ + { + "bbox": [ + 104, + 238, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 104, + 238, + 505, + 253 + ], + "score": 1.0, + "content": "WGAN with gradient penalty (Gulrajani et al., 2017) improves upon WGAN by enforcing the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 250, + 504, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 504, + 263 + ], + "score": 1.0, + "content": "Lipschitz condition with a gradient penalty term. This method significantly improves the convergence", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 262, + 347, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 347, + 274 + ], + "score": 1.0, + "content": "speed and the quality of the images generated by a WGAN.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "LSGAN (Mao et al., 2016). Least Squares GAN adopts the least squares loss function instead of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "commonly used sigmoid cross entropy loss for the discriminator, essentially minimizing the Pearson", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 299, + 419, + 315 + ], + "spans": [ + { + "bbox": [ + 107, + 300, + 118, + 312 + ], + "score": 0.87, + "content": "\\chi ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 299, + 281, + 315 + ], + "score": 1.0, + "content": "divergence between the real distribution", + "type": "text" + }, + { + "bbox": [ + 281, + 301, + 293, + 312 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 299, + 402, + 315 + ], + "score": 1.0, + "content": "and generative distribution", + "type": "text" + }, + { + "bbox": [ + 403, + 301, + 415, + 313 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 299, + 419, + 315 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 109, + 323, + 292, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 294, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 294, + 337 + ], + "score": 1.0, + "content": "B THE CHOICE OF FEATURE SPACE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 505, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "score": 1.0, + "content": "The choice of features space is crucial for all these metrics. Here we consider several alternatives to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "the convolutional features from the 34-layer ResNet trained on ImageNet. In particular, we compute", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "various metrics using the features extracted by (1) the VGG and Inception networks; (2) a 34-layer", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 376, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 506, + 388 + ], + "score": 1.0, + "content": "ResNet with random weights; (3) a ResNet classifier trained on the same dataset as the GAN models.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "We use the features extracted from these models to test all metrics in the discriminative experiments", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "we performed in Section 3. All experimental settings are identical except for the third experiments,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "which is performed on the CIFAR-10 dataset (Krizhevsky & Hinton, 2009) instead as we need class", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "labels to train the classifier. Note that we consider setting (3) only for analytical purposes. It is not a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "practical choice as GANs are mainly designed for unsupervised learning and we should not assume", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 442, + 251, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 251, + 453 + ], + "score": 1.0, + "content": "the existence of ground truth labels.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5 + }, + { + "type": "image", + "bbox": [ + 107, + 462, + 505, + 614 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 462, + 505, + 614 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 462, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 107, + 462, + 505, + 614 + ], + "score": 0.97, + "type": "image", + "image_path": "eb751a4b07b47aa017a1aea8245697f431d986c743349d44ad91f1efadce7805.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 107, + 462, + 505, + 512.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 107, + 512.6666666666666, + 505, + 563.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 107, + 563.3333333333333, + 505, + 613.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 624, + 505, + 654 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "Figure 9: Comparison of all metrics in different feature spaces. When using different trained networks, the trends", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 635, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 505, + 645 + ], + "score": 1.0, + "content": "of all metrics are very similar. Most metrics work well even in a random network, but Wasserstein distance has", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 644, + 387, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 387, + 655 + ], + "score": 1.0, + "content": "very high variance and the magnitude of increase for 1-NN accuracy is small.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "The results are shown in Figure 9 and Figure 10, from which several observations can be made: (1)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "switching from ResNet-34 to VGG or Inception has little effect to the metric scores; (2) the features", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "from a random network still works for MMD, while it makes the Wasserstein distance unstable and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "1-NN accuracy less discriminative. Not surprisingly, the Inception Score and Mode Score becomes", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "meaningless if we use the softmax values from the random network; (3) features extracted from the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "classifier trained on the same dataset as the GAN model also offers high discriminability for these", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 358, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 359, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 359, + 95 + ], + "score": 1.0, + "content": "A GAN VARIANTS USED IN OUR EXPERIMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 101, + 504, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 114 + ], + "score": 1.0, + "content": "Many GAN variants have been proposed recently. In this paper we consider several of them, which", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 112, + 238, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 238, + 124 + ], + "score": 1.0, + "content": "we briefly review in this section.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 100, + 505, + 124 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 128, + 505, + 184 + ], + "lines": [ + { + "bbox": [ + 106, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "DCGAN (Radford et al., 2015). The generator of a DCGAN takes a lower dimensional input from a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 505, + 153 + ], + "score": 1.0, + "content": "uniform noise distribution, then projects and reshapes it to a small convolutional representation with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 164 + ], + "score": 1.0, + "content": "many feature maps. After applying a series of four fractionally-strided convolutions, the generator", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 161, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 246, + 176 + ], + "score": 1.0, + "content": "converts this representation into a", + "type": "text" + }, + { + "bbox": [ + 246, + 162, + 281, + 173 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 161, + 505, + 176 + ], + "score": 1.0, + "content": "pixel image. DCGAN is optimized by minimizing the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 173, + 378, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 378, + 186 + ], + "score": 1.0, + "content": "Jensen-Shannon divergence between the real and generated images.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 128, + 506, + 186 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 190, + 505, + 234 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "WGAN (Arjovsky et al., 2017). A critic network that outputs unconstrained real values is used in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "place of the discriminator. When the critic is Lipschitz, this network approximates the Wasserstein", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 211, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 180, + 225 + ], + "score": 1.0, + "content": "distance between", + "type": "text" + }, + { + "bbox": [ + 180, + 212, + 192, + 223 + ], + "score": 0.88, + "content": "S _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 211, + 212, + 225 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 212, + 212, + 224, + 224 + ], + "score": 0.89, + "content": "S _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 211, + 505, + 225 + ], + "score": 1.0, + "content": ". A Lipschitz condition is enforced by clipping the critic networks’", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 223, + 321, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 321, + 235 + ], + "score": 1.0, + "content": "parameters to stay within a predefined bounding box.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 190, + 505, + 235 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 502, + 273 + ], + "lines": [ + { + "bbox": [ + 104, + 238, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 104, + 238, + 505, + 253 + ], + "score": 1.0, + "content": "WGAN with gradient penalty (Gulrajani et al., 2017) improves upon WGAN by enforcing the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 250, + 504, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 504, + 263 + ], + "score": 1.0, + "content": "Lipschitz condition with a gradient penalty term. This method significantly improves the convergence", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 262, + 347, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 347, + 274 + ], + "score": 1.0, + "content": "speed and the quality of the images generated by a WGAN.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 104, + 238, + 505, + 274 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "LSGAN (Mao et al., 2016). Least Squares GAN adopts the least squares loss function instead of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "commonly used sigmoid cross entropy loss for the discriminator, essentially minimizing the Pearson", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 299, + 419, + 315 + ], + "spans": [ + { + "bbox": [ + 107, + 300, + 118, + 312 + ], + "score": 0.87, + "content": "\\chi ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 299, + 281, + 315 + ], + "score": 1.0, + "content": "divergence between the real distribution", + "type": "text" + }, + { + "bbox": [ + 281, + 301, + 293, + 312 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 299, + 402, + 315 + ], + "score": 1.0, + "content": "and generative distribution", + "type": "text" + }, + { + "bbox": [ + 403, + 301, + 415, + 313 + ], + "score": 0.88, + "content": "\\mathbb { P } _ { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 299, + 419, + 315 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 278, + 505, + 315 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 323, + 292, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 294, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 294, + 337 + ], + "score": 1.0, + "content": "B THE CHOICE OF FEATURE SPACE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 505, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "score": 1.0, + "content": "The choice of features space is crucial for all these metrics. Here we consider several alternatives to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "the convolutional features from the 34-layer ResNet trained on ImageNet. In particular, we compute", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "various metrics using the features extracted by (1) the VGG and Inception networks; (2) a 34-layer", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 376, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 506, + 388 + ], + "score": 1.0, + "content": "ResNet with random weights; (3) a ResNet classifier trained on the same dataset as the GAN models.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "We use the features extracted from these models to test all metrics in the discriminative experiments", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "we performed in Section 3. All experimental settings are identical except for the third experiments,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "which is performed on the CIFAR-10 dataset (Krizhevsky & Hinton, 2009) instead as we need class", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "labels to train the classifier. Note that we consider setting (3) only for analytical purposes. It is not a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "practical choice as GANs are mainly designed for unsupervised learning and we should not assume", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 442, + 251, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 251, + 453 + ], + "score": 1.0, + "content": "the existence of ground truth labels.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 343, + 506, + 453 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 462, + 505, + 614 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 462, + 505, + 614 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 462, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 107, + 462, + 505, + 614 + ], + "score": 0.97, + "type": "image", + "image_path": "eb751a4b07b47aa017a1aea8245697f431d986c743349d44ad91f1efadce7805.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 107, + 462, + 505, + 512.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 107, + 512.6666666666666, + 505, + 563.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 107, + 563.3333333333333, + 505, + 613.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 624, + 505, + 654 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "Figure 9: Comparison of all metrics in different feature spaces. When using different trained networks, the trends", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 635, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 505, + 645 + ], + "score": 1.0, + "content": "of all metrics are very similar. Most metrics work well even in a random network, but Wasserstein distance has", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 644, + 387, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 387, + 655 + ], + "score": 1.0, + "content": "very high variance and the magnitude of increase for 1-NN accuracy is small.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "The results are shown in Figure 9 and Figure 10, from which several observations can be made: (1)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "switching from ResNet-34 to VGG or Inception has little effect to the metric scores; (2) the features", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "from a random network still works for MMD, while it makes the Wasserstein distance unstable and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "1-NN accuracy less discriminative. Not surprisingly, the Inception Score and Mode Score becomes", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "meaningless if we use the softmax values from the random network; (3) features extracted from the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "classifier trained on the same dataset as the GAN model also offers high discriminability for these", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "score": 1.0, + "content": "metrics, especially for the Wasserstein distance. However, this may be simply due to the fact that the", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 452, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 505, + 462 + ], + "score": 1.0, + "content": "feature dimensionality of the ResNet trained on CIFAR-10 is much smaller than that of the ResNet-34", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 462, + 246, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 246, + 475 + ], + "score": 1.0, + "content": "trained on ImageNet (64 v.s. 512).", + "type": "text", + "cross_page": true + } + ], + "index": 15 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 665, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 147, + 80, + 462, + 170 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 147, + 80, + 462, + 170 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 80, + 462, + 170 + ], + "spans": [ + { + "bbox": [ + 147, + 80, + 462, + 170 + ], + "score": 0.961, + "type": "image", + "image_path": "2aec4fabff4f4efa259561ca43ac79e2d4419b14e514a7fa7e0dfdcdc789922b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 80, + 462, + 110.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 110.0, + 462, + 140.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 140.0, + 462, + 170.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 180, + 506, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 191 + ], + "score": 1.0, + "content": "Figure 10: Using features extracted from a ResNet trained on CIFAR-10 (right plot) to evaluate a GAN model", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 191, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 505, + 201 + ], + "score": 1.0, + "content": "trained on the same dataset. Compared to using an extractor trained on ImageNet, the metrics appear to have", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 199, + 460, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 460, + 211 + ], + "score": 1.0, + "content": "lower variance. However, this may due to the feature dimensionality being smaller for CIFAR-10.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 112, + 225, + 498, + 365 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 225, + 498, + 365 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 225, + 498, + 365 + ], + "spans": [ + { + "bbox": [ + 112, + 225, + 498, + 365 + ], + "score": 0.973, + "type": "image", + "image_path": "2272bebca854ed5b1c95d0a22f8cf047e84ae9c0cb02aceed35de9a3d7a44221.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 112, + 225, + 498, + 271.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 271.6666666666667, + 498, + 318.33333333333337 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 318.33333333333337, + 498, + 365.00000000000006 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 376, + 505, + 417 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "Figure 11: Training curves of DCGAN and WGAN on a large (left two panels), small (middle two panels) and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "score": 1.0, + "content": "tiny (right two panels) subsets of CelebA. Note that for the first four plots, blue (yellow) curves almost overlap", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "with the red (green) curves, indicating no overfitting detected by the two metrics. Overfitting only observed on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 406, + 494, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 494, + 418 + ], + "score": 1.0, + "content": "the tiny training set, with MMD score and 1-NN accuracy significantly worse (higher) on the validation set.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + } + ], + "index": 8.75 + }, + { + "type": "text", + "bbox": [ + 108, + 440, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "score": 1.0, + "content": "metrics, especially for the Wasserstein distance. However, this may be simply due to the fact that the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 452, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 505, + 462 + ], + "score": 1.0, + "content": "feature dimensionality of the ResNet trained on CIFAR-10 is much smaller than that of the ResNet-34", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 462, + 246, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 246, + 475 + ], + "score": 1.0, + "content": "trained on ImageNet (64 v.s. 512).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 107, + 488, + 389, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 390, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 390, + 502 + ], + "score": 1.0, + "content": "C ARE GANS OVERFITTING TO THE TRAINING DATA?", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "We trained two representative GAN models, DCGAN (Radford et al., 2015) and WGAN (Arjovsky", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 520, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 291, + 534 + ], + "score": 1.0, + "content": "et al., 2017) on the CelebA dataset. Out of the", + "type": "text" + }, + { + "bbox": [ + 292, + 521, + 334, + 532 + ], + "score": 0.8, + "content": "{ \\sim } 2 0 0 { , } 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 520, + 505, + 534 + ], + "score": 1.0, + "content": "images in total, we holdout 20,000 images", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 530, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 546 + ], + "score": 1.0, + "content": "for validation, and the rest for training. As the training set is sufficiently large, which makes overfitting", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "unlikely to occur, we also create a small training set and a tiny training set respectively with only", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 553, + 327, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 327, + 567 + ], + "score": 1.0, + "content": "2000 and 10 images sampled from the full training set.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "The training setting for DCGAN and WGAN strictly follow their original implementation, except that", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "we change the default number of training iterations such that both models are sufficiently updated.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 592, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 506, + 606 + ], + "score": 1.0, + "content": "For each metric, we compute their score on 2000 real samples and 2000 generated samples, where", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 603, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 617 + ], + "score": 1.0, + "content": "the real samples are drawn from either the training set or the validation set, giving rise to training and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 613, + 507, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 507, + 628 + ], + "score": 1.0, + "content": "validation scores. The results are shown in Figure 11, from which we can make several observations:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 110, + 631, + 505, + 708 + ], + "lines": [ + { + "bbox": [ + 111, + 630, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 111, + 630, + 448, + 646 + ], + "score": 1.0, + "content": "• The training and validation scores almost overlap with each other with 2000 or", + "type": "text" + }, + { + "bbox": [ + 448, + 632, + 470, + 642 + ], + "score": 0.43, + "content": "1 8 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 630, + 506, + 646 + ], + "score": 1.0, + "content": "training", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 118, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 118, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "samples, showing that both DCGAN and WGAN do not overfit to the training data under of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 118, + 653, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 118, + 653, + 505, + 666 + ], + "score": 1.0, + "content": "these metrics. Even when using only 2000 training samples, there is still no significant difference", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 118, + 664, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 118, + 664, + 505, + 677 + ], + "score": 1.0, + "content": "between the training score and validation score. This shows that the training process of GANs", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 119, + 676, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 119, + 676, + 505, + 688 + ], + "score": 1.0, + "content": "behaves quite differently from those of supervised deep learning models, where a model can easily", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 118, + 686, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 118, + 686, + 507, + 700 + ], + "score": 1.0, + "content": "achieve 0 training error while behaving like random guess on the validation set (Zhang et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 696, + 157, + 710 + ], + "spans": [ + { + "bbox": [ + 117, + 696, + 157, + 710 + ], + "score": 1.0, + "content": "2016). 5", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 721, + 393, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 718, + 394, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 718, + 394, + 734 + ], + "score": 1.0, + "content": "5We observed that even memorizing 50 images is difficult for GAN models.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 147, + 80, + 462, + 170 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 147, + 80, + 462, + 170 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 80, + 462, + 170 + ], + "spans": [ + { + "bbox": [ + 147, + 80, + 462, + 170 + ], + "score": 0.961, + "type": "image", + "image_path": "2aec4fabff4f4efa259561ca43ac79e2d4419b14e514a7fa7e0dfdcdc789922b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 80, + 462, + 110.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 110.0, + 462, + 140.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 140.0, + 462, + 170.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 180, + 506, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 191 + ], + "score": 1.0, + "content": "Figure 10: Using features extracted from a ResNet trained on CIFAR-10 (right plot) to evaluate a GAN model", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 191, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 505, + 201 + ], + "score": 1.0, + "content": "trained on the same dataset. Compared to using an extractor trained on ImageNet, the metrics appear to have", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 199, + 460, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 460, + 211 + ], + "score": 1.0, + "content": "lower variance. However, this may due to the feature dimensionality being smaller for CIFAR-10.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 112, + 225, + 498, + 365 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 225, + 498, + 365 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 225, + 498, + 365 + ], + "spans": [ + { + "bbox": [ + 112, + 225, + 498, + 365 + ], + "score": 0.973, + "type": "image", + "image_path": "2272bebca854ed5b1c95d0a22f8cf047e84ae9c0cb02aceed35de9a3d7a44221.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 112, + 225, + 498, + 271.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 271.6666666666667, + 498, + 318.33333333333337 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 318.33333333333337, + 498, + 365.00000000000006 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 376, + 505, + 417 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "Figure 11: Training curves of DCGAN and WGAN on a large (left two panels), small (middle two panels) and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "score": 1.0, + "content": "tiny (right two panels) subsets of CelebA. Note that for the first four plots, blue (yellow) curves almost overlap", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "with the red (green) curves, indicating no overfitting detected by the two metrics. Overfitting only observed on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 406, + 494, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 494, + 418 + ], + "score": 1.0, + "content": "the tiny training set, with MMD score and 1-NN accuracy significantly worse (higher) on the validation set.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + } + ], + "index": 8.75 + }, + { + "type": "text", + "bbox": [ + 108, + 440, + 505, + 474 + ], + "lines": [], + "index": 14, + "bbox_fs": [ + 106, + 441, + 505, + 475 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 488, + 389, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 390, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 390, + 502 + ], + "score": 1.0, + "content": "C ARE GANS OVERFITTING TO THE TRAINING DATA?", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "We trained two representative GAN models, DCGAN (Radford et al., 2015) and WGAN (Arjovsky", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 520, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 291, + 534 + ], + "score": 1.0, + "content": "et al., 2017) on the CelebA dataset. Out of the", + "type": "text" + }, + { + "bbox": [ + 292, + 521, + 334, + 532 + ], + "score": 0.8, + "content": "{ \\sim } 2 0 0 { , } 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 520, + 505, + 534 + ], + "score": 1.0, + "content": "images in total, we holdout 20,000 images", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 530, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 546 + ], + "score": 1.0, + "content": "for validation, and the rest for training. As the training set is sufficiently large, which makes overfitting", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "unlikely to occur, we also create a small training set and a tiny training set respectively with only", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 553, + 327, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 327, + 567 + ], + "score": 1.0, + "content": "2000 and 10 images sampled from the full training set.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 509, + 506, + 567 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "The training setting for DCGAN and WGAN strictly follow their original implementation, except that", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "we change the default number of training iterations such that both models are sufficiently updated.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 592, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 506, + 606 + ], + "score": 1.0, + "content": "For each metric, we compute their score on 2000 real samples and 2000 generated samples, where", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 603, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 617 + ], + "score": 1.0, + "content": "the real samples are drawn from either the training set or the validation set, giving rise to training and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 613, + 507, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 507, + 628 + ], + "score": 1.0, + "content": "validation scores. The results are shown in Figure 11, from which we can make several observations:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 571, + 507, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 110, + 631, + 505, + 708 + ], + "lines": [ + { + "bbox": [ + 111, + 630, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 111, + 630, + 448, + 646 + ], + "score": 1.0, + "content": "• The training and validation scores almost overlap with each other with 2000 or", + "type": "text" + }, + { + "bbox": [ + 448, + 632, + 470, + 642 + ], + "score": 0.43, + "content": "1 8 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 630, + 506, + 646 + ], + "score": 1.0, + "content": "training", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 118, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 118, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "samples, showing that both DCGAN and WGAN do not overfit to the training data under of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 118, + 653, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 118, + 653, + 505, + 666 + ], + "score": 1.0, + "content": "these metrics. Even when using only 2000 training samples, there is still no significant difference", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 118, + 664, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 118, + 664, + 505, + 677 + ], + "score": 1.0, + "content": "between the training score and validation score. This shows that the training process of GANs", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 119, + 676, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 119, + 676, + 505, + 688 + ], + "score": 1.0, + "content": "behaves quite differently from those of supervised deep learning models, where a model can easily", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 118, + 686, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 118, + 686, + 507, + 700 + ], + "score": 1.0, + "content": "achieve 0 training error while behaving like random guess on the validation set (Zhang et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 696, + 157, + 710 + ], + "spans": [ + { + "bbox": [ + 117, + 696, + 157, + 710 + ], + "score": 1.0, + "content": "2016). 5", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 111, + 630, + 507, + 710 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 124, + 99, + 485, + 153 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 80, + 426, + 91 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 183, + 80, + 427, + 92 + ], + "spans": [ + { + "bbox": [ + 183, + 80, + 427, + 92 + ], + "score": 1.0, + "content": "Table 1: Comparison of several GAN models on the LSUN dataset", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 124, + 99, + 485, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 99, + 485, + 153 + ], + "spans": [ + { + "bbox": [ + 124, + 99, + 485, + 153 + ], + "score": 0.98, + "html": "
RealDCGANWGANWGAN-GPLSGAN
Conv SpaceMMD0.0190.2050.2700.1940.232
1-NN Accuracy0.4990.8250.9200.8120.871
1-NN Accuracy (real)0.4950.7590.8800.7650.804
1-NN Accuracy (fake)0.5030.8920.9610.8600.938
", + "type": "table", + "image_path": "ab92d3d0286e87eafef72ed0918c3d6c9537d5ee1f44d2019ef0f4a7e4ed705e.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 124, + 99, + 485, + 117.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 117.0, + 485, + 135.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 124, + 135.0, + 485, + 153.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 110, + 165, + 506, + 187 + ], + "lines": [ + { + "bbox": [ + 109, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 109, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "• DCGAN outperforms WGAN on the full training set under both metrics, and converges faster.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 118, + 174, + 507, + 189 + ], + "spans": [ + { + "bbox": [ + 118, + 174, + 507, + 189 + ], + "score": 1.0, + "content": "However, WGAN is much more stable on the small training set, and converges to better positions.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 109, + 203, + 475, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 478, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 478, + 219 + ], + "score": 1.0, + "content": "D COMPARISON OF POPULAR GAN MODELS BASED ON QUANTITATIVE", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 219, + 243, + 231 + ], + "spans": [ + { + "bbox": [ + 127, + 219, + 243, + 231 + ], + "score": 1.0, + "content": "EVALUATION METRICS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 106, + 238, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 505, + 250 + ], + "score": 1.0, + "content": "Based on our analysis, we chose MMD and 1-NN accuracy in the feature space of a 34-layer ResNet", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "trained on ImageNet to compare several state-of-the-art GAN models. All scores are computed", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "using 2000 samples from the holdout set and 2000 generated samples. The GAN models evaluated", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 270, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 284 + ], + "score": 1.0, + "content": "include DCGAN (Radford et al., 2015), WGAN (Arjovsky et al., 2017), WGAN with gradient penalty", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "(WGAN-GP ) (Gulrajani et al., 2017), and LSGAN (Mao et al., 2016) , all trained on the CelebA", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 292, + 460, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 460, + 306 + ], + "score": 1.0, + "content": "dataset. The results are reported in Table 1, from which we highlight three observations:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 132, + 313, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 131, + 313, + 371, + 326 + ], + "spans": [ + { + "bbox": [ + 131, + 313, + 371, + 326 + ], + "score": 1.0, + "content": "• WGAN-GP performs the best under most of the metrics.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 137, + 327, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 137, + 327, + 506, + 342 + ], + "score": 1.0, + "content": "DCGAN achieves 0.759 overall 1-NN accuracy on real samples, slightly better than 0.765", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 142, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "achieved by WGAN-GP; while the 1-NN accuracy on generated (fake) samples achieved by", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "DCGAN is higher than that by WGAN-GP (0.892 v.s. 0.860). This seems to suggest that", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "DCGAN is better at capturing modes in the training data distribution, while its generated", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 373, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 505, + 385 + ], + "score": 1.0, + "content": "samples are more collapsed compared to WGAN-GP. Such subtle difference is unlikely to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 384, + 378, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 378, + 396 + ], + "score": 1.0, + "content": "be discovered by the Inception Score or human evaluation.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 132, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 132, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "• The 1-NN accuracy for all evaluated GAN models are higher than 0.8 , far above the ground", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 409, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 409, + 506, + 421 + ], + "score": 1.0, + "content": "truth of 0.5. The MMD score of the four GAN models are also much larger than that of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "ground truth (0.019). This indicates that even state-of-the-art GAN models are far from", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 142, + 432, + 260, + 444 + ], + "spans": [ + { + "bbox": [ + 142, + 432, + 260, + 444 + ], + "score": 1.0, + "content": "learning the true distribution.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 124, + 99, + 485, + 153 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 80, + 426, + 91 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 183, + 80, + 427, + 92 + ], + "spans": [ + { + "bbox": [ + 183, + 80, + 427, + 92 + ], + "score": 1.0, + "content": "Table 1: Comparison of several GAN models on the LSUN dataset", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 124, + 99, + 485, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 99, + 485, + 153 + ], + "spans": [ + { + "bbox": [ + 124, + 99, + 485, + 153 + ], + "score": 0.98, + "html": "
RealDCGANWGANWGAN-GPLSGAN
Conv SpaceMMD0.0190.2050.2700.1940.232
1-NN Accuracy0.4990.8250.9200.8120.871
1-NN Accuracy (real)0.4950.7590.8800.7650.804
1-NN Accuracy (fake)0.5030.8920.9610.8600.938
", + "type": "table", + "image_path": "ab92d3d0286e87eafef72ed0918c3d6c9537d5ee1f44d2019ef0f4a7e4ed705e.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 124, + 99, + 485, + 117.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 117.0, + 485, + 135.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 124, + 135.0, + 485, + 153.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 110, + 165, + 506, + 187 + ], + "lines": [ + { + "bbox": [ + 109, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 109, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "• DCGAN outperforms WGAN on the full training set under both metrics, and converges faster.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 118, + 174, + 507, + 189 + ], + "spans": [ + { + "bbox": [ + 118, + 174, + 507, + 189 + ], + "score": 1.0, + "content": "However, WGAN is much more stable on the small training set, and converges to better positions.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 109, + 164, + 507, + 189 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 203, + 475, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 478, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 478, + 219 + ], + "score": 1.0, + "content": "D COMPARISON OF POPULAR GAN MODELS BASED ON QUANTITATIVE", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 219, + 243, + 231 + ], + "spans": [ + { + "bbox": [ + 127, + 219, + 243, + 231 + ], + "score": 1.0, + "content": "EVALUATION METRICS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 106, + 238, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 505, + 250 + ], + "score": 1.0, + "content": "Based on our analysis, we chose MMD and 1-NN accuracy in the feature space of a 34-layer ResNet", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "trained on ImageNet to compare several state-of-the-art GAN models. All scores are computed", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "using 2000 samples from the holdout set and 2000 generated samples. The GAN models evaluated", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 270, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 284 + ], + "score": 1.0, + "content": "include DCGAN (Radford et al., 2015), WGAN (Arjovsky et al., 2017), WGAN with gradient penalty", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "(WGAN-GP ) (Gulrajani et al., 2017), and LSGAN (Mao et al., 2016) , all trained on the CelebA", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 292, + 460, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 460, + 306 + ], + "score": 1.0, + "content": "dataset. The results are reported in Table 1, from which we highlight three observations:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 238, + 506, + 306 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 313, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 131, + 313, + 371, + 326 + ], + "spans": [ + { + "bbox": [ + 131, + 313, + 371, + 326 + ], + "score": 1.0, + "content": "• WGAN-GP performs the best under most of the metrics.", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 137, + 327, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 137, + 327, + 506, + 342 + ], + "score": 1.0, + "content": "DCGAN achieves 0.759 overall 1-NN accuracy on real samples, slightly better than 0.765", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 142, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "achieved by WGAN-GP; while the 1-NN accuracy on generated (fake) samples achieved by", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "DCGAN is higher than that by WGAN-GP (0.892 v.s. 0.860). This seems to suggest that", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "DCGAN is better at capturing modes in the training data distribution, while its generated", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 373, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 505, + 385 + ], + "score": 1.0, + "content": "samples are more collapsed compared to WGAN-GP. Such subtle difference is unlikely to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 384, + 378, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 378, + 396 + ], + "score": 1.0, + "content": "be discovered by the Inception Score or human evaluation.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 132, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "• The 1-NN accuracy for all evaluated GAN models are higher than 0.8 , far above the ground", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 409, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 409, + 506, + 421 + ], + "score": 1.0, + "content": "truth of 0.5. The MMD score of the four GAN models are also much larger than that of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "ground truth (0.019). This indicates that even state-of-the-art GAN models are far from", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 142, + 432, + 260, + 444 + ], + "spans": [ + { + "bbox": [ + 142, + 432, + 260, + 444 + ], + "score": 1.0, + "content": "learning the true distribution.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + } + ], + "index": 19, + "bbox_fs": [ + 131, + 313, + 506, + 444 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/Sy1f0e-R-/Sy1f0e-R-_model.json b/parse/train/Sy1f0e-R-/Sy1f0e-R-_model.json new file mode 100644 index 0000000000000000000000000000000000000000..37d1b913c55964210316f87df94a52ae3eaa1838 --- /dev/null +++ b/parse/train/Sy1f0e-R-/Sy1f0e-R-_model.json @@ -0,0 +1,24842 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 592, + 1304, + 592, + 1304, + 1048, + 398, + 1048 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1177, + 1404, + 1177, + 1404, + 1513, + 298, + 1513 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1529, + 1403, + 1529, + 1403, + 1803, + 298, + 1803 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1820, + 1404, + 1820, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.979 + }, + { + "category_id": 0, + "poly": [ + 300, + 223, + 1401, + 223, + 1401, + 323, + 300, + 323 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 313, + 377, + 680, + 377, + 680, + 438, + 313, + 438 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 302, + 1117, + 573, + 1117, + 573, + 1152, + 302, + 1152 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.898 + }, + { + "category_id": 0, + "poly": [ + 773, + 518, + 927, + 518, + 927, + 553, + 773, + 553 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 856, + 2089, + 856, + 2112, + 841, + 2112 + ], + "score": 0.716 + }, + { + "category_id": 15, + "poly": [ + 293.0, + 222.0, + 1409.0, + 222.0, + 1409.0, + 278.0, + 293.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 285.0, + 1114.0, + 285.0, + 1114.0, + 328.0, + 296.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1114.0, + 579.0, + 1114.0, + 579.0, + 1161.0, + 294.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 515.0, + 934.0, + 515.0, + 934.0, + 558.0, + 768.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 859.0, + 2088.0, + 859.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 591.0, + 1305.0, + 591.0, + 1305.0, + 628.0, + 393.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 624.0, + 1304.0, + 624.0, + 1304.0, + 655.0, + 395.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 653.0, + 1306.0, + 653.0, + 1306.0, + 688.0, + 394.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 681.0, + 1309.0, + 681.0, + 1309.0, + 719.0, + 393.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 715.0, + 1308.0, + 715.0, + 1308.0, + 748.0, + 394.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 745.0, + 1306.0, + 745.0, + 1306.0, + 780.0, + 394.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 775.0, + 1306.0, + 775.0, + 1306.0, + 810.0, + 393.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 807.0, + 1306.0, + 807.0, + 1306.0, + 840.0, + 393.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 836.0, + 1306.0, + 836.0, + 1306.0, + 869.0, + 392.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 867.0, + 1306.0, + 867.0, + 1306.0, + 902.0, + 394.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 897.0, + 1306.0, + 897.0, + 1306.0, + 931.0, + 394.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 929.0, + 1304.0, + 929.0, + 1304.0, + 961.0, + 395.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 960.0, + 1305.0, + 960.0, + 1305.0, + 992.0, + 394.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 986.0, + 1306.0, + 986.0, + 1306.0, + 1025.0, + 393.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1018.0, + 1236.0, + 1018.0, + 1236.0, + 1053.0, + 393.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1180.0, + 1404.0, + 1180.0, + 1404.0, + 1211.0, + 297.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1208.0, + 1407.0, + 1208.0, + 1407.0, + 1245.0, + 292.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1240.0, + 1408.0, + 1240.0, + 1408.0, + 1274.0, + 294.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1268.0, + 1408.0, + 1268.0, + 1408.0, + 1305.0, + 293.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 1408.0, + 1298.0, + 1408.0, + 1336.0, + 293.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1329.0, + 1406.0, + 1329.0, + 1406.0, + 1366.0, + 293.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1361.0, + 1405.0, + 1361.0, + 1405.0, + 1396.0, + 294.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1391.0, + 1408.0, + 1391.0, + 1408.0, + 1429.0, + 293.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1421.0, + 1407.0, + 1421.0, + 1407.0, + 1458.0, + 292.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1450.0, + 1405.0, + 1450.0, + 1405.0, + 1489.0, + 293.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1481.0, + 1282.0, + 1481.0, + 1282.0, + 1517.0, + 293.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1527.0, + 1405.0, + 1527.0, + 1405.0, + 1565.0, + 293.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1559.0, + 1405.0, + 1559.0, + 1405.0, + 1597.0, + 293.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1592.0, + 1404.0, + 1592.0, + 1404.0, + 1624.0, + 296.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1618.0, + 1405.0, + 1618.0, + 1405.0, + 1657.0, + 293.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1650.0, + 1404.0, + 1650.0, + 1404.0, + 1687.0, + 294.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1682.0, + 1405.0, + 1682.0, + 1405.0, + 1718.0, + 294.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1714.0, + 1405.0, + 1714.0, + 1405.0, + 1747.0, + 296.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1743.0, + 1407.0, + 1743.0, + 1407.0, + 1779.0, + 293.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1775.0, + 507.0, + 1775.0, + 507.0, + 1809.0, + 296.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1850.0, + 1405.0, + 1850.0, + 1405.0, + 1885.0, + 293.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1907.0, + 1406.0, + 1907.0, + 1406.0, + 1952.0, + 292.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 1409.0, + 1939.0, + 1409.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1974.0, + 1405.0, + 1974.0, + 1405.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1999.0, + 1409.0, + 1999.0, + 1409.0, + 2041.0, + 292.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 377.0, + 560.0, + 377.0, + 560.0, + 410.0, + 315.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 407.0, + 682.0, + 407.0, + 682.0, + 440.0, + 311.0, + 440.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1487, + 1404, + 1487, + 1404, + 1703, + 297, + 1703 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1350, + 1406, + 1350, + 1406, + 1473, + 298, + 1473 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1156, + 1403, + 1156, + 1403, + 1281, + 297, + 1281 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1772, + 1404, + 1772, + 1404, + 1957, + 298, + 1957 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 707, + 1404, + 707, + 1404, + 831, + 298, + 831 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1048, + 1405, + 1048, + 1405, + 1142, + 298, + 1142 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 392, + 225, + 1314, + 225, + 1314, + 423, + 392, + 423 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 298, + 917, + 1402, + 917, + 1402, + 981, + 298, + 981 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 297, + 507, + 1405, + 507, + 1405, + 694, + 297, + 694 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 458, + 1290, + 1242, + 1290, + 1242, + 1333, + 458, + 1333 + ], + "score": 0.939 + }, + { + "category_id": 2, + "poly": [ + 297, + 1978, + 1404, + 1978, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.936 + }, + { + "category_id": 4, + "poly": [ + 363, + 448, + 1334, + 448, + 1334, + 479, + 363, + 479 + ], + "score": 0.915 + }, + { + "category_id": 0, + "poly": [ + 299, + 862, + 558, + 862, + 558, + 898, + 299, + 898 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 299, + 1728, + 665, + 1728, + 665, + 1758, + 299, + 1758 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 298, + 1005, + 841, + 1005, + 841, + 1036, + 298, + 1036 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1295, + 1399, + 1295, + 1399, + 1324, + 1366, + 1324 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.753 + }, + { + "category_id": 13, + "poly": [ + 1137, + 1219, + 1211, + 1219, + 1211, + 1252, + 1137, + 1252 + ], + "score": 0.94, + "latex": "G ( \\mathbb { P } _ { z } )" + }, + { + "category_id": 13, + "poly": [ + 1031, + 1050, + 1249, + 1050, + 1249, + 1083, + 1031, + 1083 + ], + "score": 0.93, + "latex": "S _ { r } = \\{ \\mathbf { x } _ { 1 } ^ { r } , \\ldots , \\mathbf { x } _ { n } ^ { r } \\}" + }, + { + "category_id": 13, + "poly": [ + 967, + 1352, + 1045, + 1352, + 1045, + 1380, + 967, + 1380 + ], + "score": 0.93, + "latex": "\\mathbf { x } \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1189, + 471, + 1189, + 471, + 1221, + 297, + 1221 + ], + "score": 0.92, + "latex": "D : \\mathcal { X } [ 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 340, + 1046, + 465, + 1046, + 465, + 1077, + 340, + 1077 + ], + "score": 0.92, + "latex": "\\mathcal { X } = \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 13, + "poly": [ + 680, + 1189, + 817, + 1189, + 817, + 1217, + 680, + 1217 + ], + "score": 0.91, + "latex": "G : { \\mathcal { Z } } \\to { \\mathcal { X } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1640, + 598, + 1640, + 598, + 1675, + 297, + 1675 + ], + "score": 0.91, + "latex": "S _ { g } = \\{ \\mathbf { x } _ { 1 } ^ { g } , . . . , \\mathbf { x } _ { m } ^ { \\bar { g } } \\} \\sim \\mathbb { P } _ { q } ^ { m }" + }, + { + "category_id": 14, + "poly": [ + 454, + 1289, + 1243, + 1289, + 1243, + 1340, + 454, + 1340 + ], + "score": 0.9, + "latex": "\\operatorname* { m i n } _ { G } \\operatorname* { m a x } _ { D } L ( D , G ) = \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { r } } \\log { [ D ( \\mathbf { x } ) ] } + \\mathbb { E } _ { \\mathbf { z } \\sim \\mathbb { P } _ { z } } \\left[ \\log ( 1 - D ( G ( \\mathbf { z } ) ) ) \\right] ." + }, + { + "category_id": 13, + "poly": [ + 954, + 1219, + 987, + 1219, + 987, + 1253, + 954, + 1253 + ], + "score": 0.9, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1281, + 1413, + 1313, + 1413, + 1313, + 1442, + 1281, + 1442 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 400, + 1520, + 433, + 1520, + 433, + 1553, + 400, + 1553 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 543, + 2006, + 574, + 2006, + 574, + 2035, + 543, + 2035 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 534, + 1382, + 566, + 1382, + 566, + 1412, + 534, + 1412 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1364, + 1413, + 1396, + 1413, + 1396, + 1446, + 1364, + 1446 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1039, + 1612, + 1072, + 1612, + 1072, + 1644, + 1039, + 1644 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 477, + 1082, + 509, + 1082, + 509, + 1111, + 477, + 1111 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 408, + 1674, + 628, + 1674, + 628, + 1703, + 408, + 1703 + ], + "score": 0.88, + "latex": "\\hat { \\rho } : \\mathcal X ^ { n } \\times \\mathcal X ^ { m } \\stackrel { \\smile } { } \\mathbb R" + }, + { + "category_id": 13, + "poly": [ + 472, + 1113, + 503, + 1113, + 503, + 1141, + 472, + 1141 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 549, + 1582, + 580, + 1582, + 580, + 1611, + 549, + 1611 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1166, + 1083, + 1199, + 1083, + 1199, + 1114, + 1166, + 1114 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1059, + 1582, + 1091, + 1582, + 1091, + 1611, + 1059, + 1611 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 469, + 1520, + 502, + 1520, + 502, + 1550, + 469, + 1550 + ], + "score": 0.87, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 619, + 2006, + 649, + 2006, + 649, + 2032, + 619, + 2032 + ], + "score": 0.86, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1220, + 330, + 1220, + 330, + 1249, + 297, + 1249 + ], + "score": 0.86, + "latex": "\\mathbb { P } _ { z }" + }, + { + "category_id": 13, + "poly": [ + 904, + 1190, + 929, + 1190, + 929, + 1216, + 904, + 1216 + ], + "score": 0.85, + "latex": "\\mathcal { Z }" + }, + { + "category_id": 13, + "poly": [ + 756, + 1251, + 782, + 1251, + 782, + 1277, + 756, + 1277 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 745, + 1443, + 772, + 1443, + 772, + 1469, + 745, + 1469 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 609, + 1352, + 636, + 1352, + 636, + 1378, + 609, + 1378 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 823, + 1443, + 847, + 1443, + 847, + 1469, + 823, + 1469 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1104, + 1382, + 1129, + 1382, + 1129, + 1409, + 1104, + 1409 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 567, + 1082, + 594, + 1082, + 594, + 1108, + 567, + 1108 + ], + "score": 0.82, + "latex": "\\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 390, + 1221, + 415, + 1221, + 415, + 1247, + 390, + 1247 + ], + "score": 0.82, + "latex": "\\mathcal { Z }" + }, + { + "category_id": 13, + "poly": [ + 833, + 1250, + 859, + 1250, + 859, + 1277, + 833, + 1277 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1525, + 1266, + 1525, + 1266, + 1552, + 1247, + 1552 + ], + "score": 0.81, + "latex": "\\rho" + }, + { + "category_id": 13, + "poly": [ + 997, + 1494, + 1016, + 1494, + 1016, + 1521, + 997, + 1521 + ], + "score": 0.81, + "latex": "\\rho" + }, + { + "category_id": 13, + "poly": [ + 441, + 1983, + 459, + 1983, + 459, + 2007, + 441, + 2007 + ], + "score": 0.75, + "latex": "\\rho" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 231.0, + 1000.0, + 231.0, + 1000.0, + 257.0, + 879.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 243.0, + 546.0, + 243.0, + 546.0, + 373.0, + 410.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 269.0, + 767.0, + 269.0, + 767.0, + 361.0, + 625.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 250.0, + 1016.0, + 250.0, + 1016.0, + 382.0, + 863.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 260.0, + 1304.0, + 260.0, + 1304.0, + 370.0, + 1105.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 365.0, + 528.0, + 365.0, + 528.0, + 401.0, + 423.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 375.0, + 1015.0, + 375.0, + 1015.0, + 397.0, + 866.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 376.0, + 1210.0, + 376.0, + 1210.0, + 383.0, + 1199.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 392.0, + 998.0, + 392.0, + 998.0, + 416.0, + 882.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 331.0, + 1004.0, + 331.0, + 1004.0, + 364.5, + 877.0, + 364.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 1973.0, + 440.0, + 1973.0, + 440.0, + 2011.0, + 333.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1973.0, + 1406.0, + 1973.0, + 1406.0, + 2011.0, + 460.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2002.0, + 542.0, + 2002.0, + 542.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 2002.0, + 618.0, + 2002.0, + 618.0, + 2038.0, + 575.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 2002.0, + 1258.0, + 2002.0, + 1258.0, + 2038.0, + 650.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 447.0, + 1337.0, + 447.0, + 1337.0, + 480.0, + 361.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 856.0, + 562.0, + 856.0, + 562.0, + 907.0, + 290.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1726.0, + 667.0, + 1726.0, + 667.0, + 1761.0, + 295.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1004.0, + 844.0, + 1004.0, + 844.0, + 1040.0, + 294.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1486.0, + 996.0, + 1486.0, + 996.0, + 1525.0, + 294.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1486.0, + 1402.0, + 1486.0, + 1402.0, + 1525.0, + 1017.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1515.0, + 399.0, + 1515.0, + 399.0, + 1555.0, + 291.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1515.0, + 468.0, + 1515.0, + 468.0, + 1555.0, + 434.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1515.0, + 1246.0, + 1515.0, + 1246.0, + 1555.0, + 503.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 1515.0, + 1407.0, + 1515.0, + 1407.0, + 1555.0, + 1267.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1548.0, + 1409.0, + 1548.0, + 1409.0, + 1585.0, + 294.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 548.0, + 1580.0, + 548.0, + 1615.0, + 294.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1580.0, + 1058.0, + 1580.0, + 1058.0, + 1615.0, + 581.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1580.0, + 1409.0, + 1580.0, + 1409.0, + 1615.0, + 1092.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1607.0, + 1038.0, + 1607.0, + 1038.0, + 1648.0, + 290.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1607.0, + 1406.0, + 1607.0, + 1406.0, + 1648.0, + 1073.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1624.0, + 296.0, + 1624.0, + 296.0, + 1709.0, + 284.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1624.0, + 1416.0, + 1624.0, + 1416.0, + 1709.0, + 629.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1351.0, + 608.0, + 1351.0, + 608.0, + 1384.0, + 295.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1351.0, + 966.0, + 1351.0, + 966.0, + 1384.0, + 637.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1351.0, + 1403.0, + 1351.0, + 1403.0, + 1384.0, + 1046.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1382.0, + 533.0, + 1382.0, + 533.0, + 1415.0, + 296.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1382.0, + 1103.0, + 1382.0, + 1103.0, + 1415.0, + 567.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1382.0, + 1406.0, + 1382.0, + 1406.0, + 1415.0, + 1130.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1406.0, + 1280.0, + 1406.0, + 1280.0, + 1452.0, + 291.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1406.0, + 1363.0, + 1406.0, + 1363.0, + 1452.0, + 1314.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1406.0, + 1409.0, + 1406.0, + 1409.0, + 1452.0, + 1397.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1442.0, + 744.0, + 1442.0, + 744.0, + 1477.0, + 293.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1442.0, + 822.0, + 1442.0, + 822.0, + 1477.0, + 773.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1442.0, + 1112.0, + 1442.0, + 1112.0, + 1477.0, + 848.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1154.0, + 1408.0, + 1154.0, + 1408.0, + 1193.0, + 292.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1184.0, + 679.0, + 1184.0, + 679.0, + 1223.0, + 472.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1184.0, + 903.0, + 1184.0, + 903.0, + 1223.0, + 818.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1184.0, + 1407.0, + 1184.0, + 1407.0, + 1223.0, + 930.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1216.0, + 296.0, + 1216.0, + 296.0, + 1254.0, + 292.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1216.0, + 389.0, + 1216.0, + 389.0, + 1254.0, + 331.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1216.0, + 953.0, + 1216.0, + 953.0, + 1254.0, + 416.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1216.0, + 1136.0, + 1216.0, + 1136.0, + 1254.0, + 988.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1216.0, + 1407.0, + 1216.0, + 1407.0, + 1254.0, + 1212.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1250.0, + 755.0, + 1250.0, + 755.0, + 1283.0, + 292.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1250.0, + 832.0, + 1250.0, + 832.0, + 1283.0, + 783.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1772.0, + 1405.0, + 1772.0, + 1405.0, + 1807.0, + 294.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 1405.0, + 1802.0, + 1405.0, + 1838.0, + 294.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1833.0, + 1408.0, + 1833.0, + 1408.0, + 1868.0, + 293.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1863.0, + 1406.0, + 1863.0, + 1406.0, + 1898.0, + 293.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1894.0, + 1405.0, + 1894.0, + 1405.0, + 1928.0, + 292.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1924.0, + 965.0, + 1924.0, + 965.0, + 1961.0, + 292.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 706.0, + 1405.0, + 706.0, + 1405.0, + 745.0, + 293.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 740.0, + 1404.0, + 740.0, + 1404.0, + 773.0, + 294.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 768.0, + 1405.0, + 768.0, + 1405.0, + 806.0, + 292.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 801.0, + 699.0, + 801.0, + 699.0, + 833.0, + 296.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1042.0, + 339.0, + 1042.0, + 339.0, + 1087.0, + 291.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1042.0, + 1030.0, + 1042.0, + 1030.0, + 1087.0, + 466.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1042.0, + 1406.0, + 1042.0, + 1406.0, + 1087.0, + 1250.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1077.0, + 476.0, + 1077.0, + 476.0, + 1117.0, + 292.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1077.0, + 566.0, + 1077.0, + 566.0, + 1117.0, + 510.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1077.0, + 1165.0, + 1077.0, + 1165.0, + 1117.0, + 595.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1077.0, + 1405.0, + 1077.0, + 1405.0, + 1117.0, + 1200.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1111.0, + 471.0, + 1111.0, + 471.0, + 1142.0, + 296.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1111.0, + 513.0, + 1111.0, + 513.0, + 1142.0, + 504.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 917.0, + 1406.0, + 917.0, + 1406.0, + 953.0, + 296.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 950.0, + 1113.0, + 950.0, + 1113.0, + 982.0, + 296.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 510.0, + 1405.0, + 510.0, + 1405.0, + 542.0, + 295.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 538.0, + 1405.0, + 538.0, + 1405.0, + 575.0, + 292.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 571.0, + 1407.0, + 571.0, + 1407.0, + 603.0, + 295.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 599.0, + 1407.0, + 599.0, + 1407.0, + 635.0, + 294.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 627.0, + 1406.0, + 627.0, + 1406.0, + 666.0, + 292.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 661.0, + 1254.0, + 661.0, + 1254.0, + 697.0, + 294.0, + 697.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 388, + 1406, + 388, + 1406, + 701, + 295, + 701 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 1340, + 1405, + 1340, + 1405, + 1464, + 298, + 1464 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1085, + 1405, + 1085, + 1405, + 1210, + 297, + 1210 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1802, + 1406, + 1802, + 1406, + 1896, + 298, + 1896 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1910, + 1403, + 1910, + 1403, + 2036, + 297, + 2036 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 813, + 1405, + 813, + 1405, + 910, + 298, + 910 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 295, + 228, + 1402, + 228, + 1402, + 323, + 295, + 323 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 428, + 972, + 1270, + 972, + 1270, + 1074, + 428, + 1074 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 295, + 1671, + 1398, + 1671, + 1398, + 1736, + 295, + 1736 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 566, + 1268, + 1133, + 1268, + 1133, + 1325, + 566, + 1325 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 636, + 335, + 1062, + 335, + 1062, + 378, + 636, + 378 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 298, + 1594, + 1403, + 1594, + 1403, + 1659, + 298, + 1659 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 509, + 1750, + 1186, + 1750, + 1186, + 1793, + 509, + 1793 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 514, + 759, + 1183, + 759, + 1183, + 801, + 514, + 801 + ], + "score": 0.934 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 105, + 297, + 105 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 293, + 713, + 1277, + 713, + 1277, + 748, + 293, + 748 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 297, + 921, + 995, + 921, + 995, + 956, + 297, + 956 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 298, + 1223, + 963, + 1223, + 963, + 1257, + 298, + 1257 + ], + "score": 0.91 + }, + { + "category_id": 8, + "poly": [ + 299, + 1478, + 1402, + 1478, + 1402, + 1569, + 299, + 1569 + ], + "score": 0.906 + }, + { + "category_id": 9, + "poly": [ + 1366, + 768, + 1400, + 768, + 1400, + 798, + 1366, + 798 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1756, + 1400, + 1756, + 1400, + 1787, + 1366, + 1787 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1366, + 344, + 1400, + 344, + 1400, + 374, + 1366, + 374 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1273, + 1400, + 1273, + 1400, + 1303, + 1366, + 1303 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1009, + 1400, + 1009, + 1400, + 1038, + 1366, + 1038 + ], + "score": 0.869 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1568, + 1400, + 1568, + 1400, + 1594, + 1366, + 1594 + ], + "score": 0.85 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.635 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.539 + }, + { + "category_id": 13, + "poly": [ + 1178, + 456, + 1282, + 456, + 1282, + 488, + 1178, + 488 + ], + "score": 0.95, + "latex": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 657, + 1372, + 765, + 1372, + 765, + 1404, + 657, + 1404 + ], + "score": 0.93, + "latex": "d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } )" + }, + { + "category_id": 14, + "poly": [ + 428, + 968, + 1267, + 968, + 1267, + 1075, + 428, + 1075 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathrm { M M D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\biggl ( \\mathbb { E } _ { \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } \\sim \\mathbb { P } _ { r } , } \\biggl [ k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { r } ^ { \\prime } ) - 2 k ( \\mathbf { x } _ { r } , \\mathbf { x } _ { g } ) + k ( \\mathbf { x } _ { g } , \\mathbf { x } _ { g } ^ { \\prime } ) \\biggr ] \\biggr ) ^ { \\frac { 1 } { 2 } } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 373, + 814, + 689, + 814, + 689, + 852, + 373, + 852 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { p _ { \\mathcal M } ( y ^ { \\ast } ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { r } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 368, + 1341, + 480, + 1341, + 480, + 1375, + 368, + 1375 + ], + "score": 0.93, + "latex": "\\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )" + }, + { + "category_id": 13, + "poly": [ + 520, + 424, + 625, + 424, + 625, + 458, + 520, + 458 + ], + "score": 0.93, + "latex": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 368, + 391, + 472, + 391, + 472, + 425, + 368, + 425 + ], + "score": 0.93, + "latex": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 784, + 516, + 865, + 516, + 865, + 550, + 784, + 550 + ], + "score": 0.92, + "latex": "p _ { \\mathcal { M } } ( y )" + }, + { + "category_id": 13, + "poly": [ + 1096, + 390, + 1397, + 390, + 1397, + 429, + 1096, + 429 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { p _ { \\mathcal M } ( y ) = \\int _ { \\mathbf x } p _ { \\mathcal M } ( y | \\mathbf x ) d \\mathbb P _ { g } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 634, + 336, + 1063, + 336, + 1063, + 379, + 634, + 379 + ], + "score": 0.92, + "latex": "\\mathrm { I S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] } ," + }, + { + "category_id": 13, + "poly": [ + 297, + 455, + 402, + 455, + 402, + 489, + 297, + 489 + ], + "score": 0.92, + "latex": "p _ { \\mathcal { M } } ( y | \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 795, + 1595, + 938, + 1595, + 938, + 1630, + 795, + 1630 + ], + "score": 0.92, + "latex": "\\mathrm { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } )" + }, + { + "category_id": 13, + "poly": [ + 1072, + 1942, + 1199, + 1942, + 1199, + 1975, + 1072, + 1975 + ], + "score": 0.91, + "latex": "| S _ { r } | = | S _ { g } |" + }, + { + "category_id": 13, + "poly": [ + 992, + 876, + 1242, + 876, + 1242, + 911, + 992, + 911 + ], + "score": 0.91, + "latex": "K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( \\bar { y } ^ { * } ) )" + }, + { + "category_id": 14, + "poly": [ + 310, + 1475, + 1405, + 1475, + 1405, + 1572, + 310, + 1572 + ], + "score": 0.91, + "latex": "N \\mathrm { D } ( p _ { r } , p _ { g } ) = \\operatorname* { m i n } _ { w \\in \\mathbb { R } ^ { n \\times m } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { m } w _ { i j } d ( \\mathbf { x } _ { i } ^ { r } , \\mathbf { x } _ { j } ^ { g } ) \\quad \\mathrm { s . t . } \\quad \\sum _ { j = 1 } ^ { m } w _ { i , j } = p _ { r } ( \\mathbf { x } _ { i } ^ { r } ) \\ \\forall i , \\sum _ { i = 1 } ^ { n } w _ { i , j } = p _ { g } ( \\mathbf { x } _ { j } ^ { g } ) \\ \\forall j ." + }, + { + "category_id": 13, + "poly": [ + 744, + 1944, + 844, + 1944, + 844, + 1975, + 744, + 1975 + ], + "score": 0.91, + "latex": "S _ { r } \\sim \\mathbb { P } _ { r } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 567, + 1270, + 1131, + 1270, + 1131, + 1325, + 567, + 1325 + ], + "score": 0.91, + "latex": "\\operatorname { W D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\operatorname* { i n f } _ { \\substack { \\gamma \\in \\Gamma ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) } } \\mathbb { E } _ { ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\sim \\gamma } \\left[ d ( \\mathbf { x } ^ { r } , \\mathbf { x } ^ { g } ) \\right] ," + }, + { + "category_id": 14, + "poly": [ + 515, + 759, + 1179, + 759, + 1179, + 803, + 515, + 803 + ], + "score": 0.91, + "latex": "\\mathrm { M S } ( \\mathbb { P } _ { g } ) = e ^ { \\mathbb { E } _ { \\mathbf { x } \\sim \\mathbb { P } _ { g } } [ K L ( p _ { \\mathcal { M } } ( y | \\mathbf { x } ) | | p _ { \\mathcal { M } } ( y ) ) ] - K L ( p _ { \\mathcal { M } } ( y ) | | p _ { \\mathcal { M } } ( y ^ { \\ast } ) ) } ," + }, + { + "category_id": 13, + "poly": [ + 896, + 1943, + 1003, + 1943, + 1003, + 1978, + 896, + 1978 + ], + "score": 0.9, + "latex": "S _ { g } \\sim \\mathbb { P } _ { g } ^ { m }" + }, + { + "category_id": 13, + "poly": [ + 783, + 1226, + 816, + 1226, + 816, + 1260, + 783, + 1260 + ], + "score": 0.9, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 700, + 1226, + 733, + 1226, + 733, + 1256, + 700, + 1256 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 993, + 1835, + 1026, + 1835, + 1026, + 1868, + 993, + 1868 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1085, + 640, + 1117, + 640, + 1117, + 669, + 1085, + 669 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 784, + 1087, + 816, + 1087, + 816, + 1120, + 784, + 1120 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 632, + 2004, + 664, + 2004, + 664, + 2038, + 632, + 2038 + ], + "score": 0.89, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 700, + 1088, + 732, + 1088, + 732, + 1117, + 700, + 1117 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 564, + 1373, + 597, + 1373, + 597, + 1406, + 564, + 1406 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 480, + 1374, + 513, + 1374, + 513, + 1402, + 480, + 1402 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 912, + 1835, + 944, + 1835, + 944, + 1865, + 912, + 1865 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1142, + 1148, + 1175, + 1148, + 1175, + 1182, + 1142, + 1182 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 763, + 878, + 796, + 878, + 796, + 911, + 763, + 911 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 337, + 2003, + 369, + 2003, + 369, + 2034, + 337, + 2034 + ], + "score": 0.89, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 539, + 1118, + 572, + 1118, + 572, + 1151, + 539, + 1151 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 431, + 878, + 463, + 878, + 463, + 907, + 431, + 907 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 14, + "poly": [ + 512, + 1748, + 1185, + 1748, + 1185, + 1792, + 512, + 1792 + ], + "score": 0.88, + "latex": "\\mathrm { F I D } ( \\mathbb { P } _ { r } , \\mathbb { P } _ { g } ) = \\| \\mu _ { r } - \\mu _ { g } \\| + \\operatorname { T r } ( \\mathbf { C } _ { r } + \\mathbf { C } _ { g } - 2 ( \\mathbf { C } _ { r } \\mathbf { C } _ { g } ) ^ { 1 / 2 } ) ," + }, + { + "category_id": 13, + "poly": [ + 1139, + 1975, + 1171, + 1975, + 1171, + 2007, + 1139, + 2007 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 455, + 1118, + 488, + 1118, + 488, + 1148, + 455, + 1148 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 938, + 458, + 971, + 458, + 971, + 490, + 938, + 490 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 959, + 425, + 992, + 425, + 992, + 459, + 959, + 459 + ], + "score": 0.87, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1053, + 1976, + 1084, + 1976, + 1084, + 2003, + 1053, + 2003 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 787, + 1406, + 816, + 1406, + 816, + 1437, + 787, + 1437 + ], + "score": 0.87, + "latex": "p _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1308, + 1148, + 1340, + 1148, + 1340, + 1178, + 1308, + 1178 + ], + "score": 0.86, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 705, + 1408, + 734, + 1408, + 734, + 1435, + 705, + 1435 + ], + "score": 0.86, + "latex": "p _ { r }" + }, + { + "category_id": 13, + "poly": [ + 612, + 640, + 647, + 640, + 647, + 667, + 612, + 667 + ], + "score": 0.82, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 1007, + 392, + 1042, + 392, + 1042, + 419, + 1007, + 419 + ], + "score": 0.82, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 532, + 262, + 567, + 262, + 567, + 290, + 532, + 290 + ], + "score": 0.81, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 1169, + 1088, + 1187, + 1088, + 1187, + 1114, + 1169, + 1114 + ], + "score": 0.8, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 818, + 397, + 838, + 397, + 838, + 419, + 818, + 419 + ], + "score": 0.74, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 508, + 1803, + 546, + 1803, + 546, + 1836, + 508, + 1836 + ], + "score": 0.7, + "latex": "\\mathbf { C } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 370, + 1807, + 402, + 1807, + 402, + 1836, + 370, + 1836 + ], + "score": 0.66, + "latex": "\\mu _ { r }" + }, + { + "category_id": 13, + "poly": [ + 553, + 1804, + 605, + 1804, + 605, + 1838, + 553, + 1838 + ], + "score": 0.62, + "latex": "( \\mathbf { C } _ { g } )" + }, + { + "category_id": 13, + "poly": [ + 370, + 1805, + 457, + 1805, + 457, + 1838, + 370, + 1838 + ], + "score": 0.47, + "latex": "\\mu _ { r } \\left( \\mu _ { g } \\right)" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 859.0, + 72.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 382.0, + 367.0, + 382.0, + 367.0, + 435.0, + 292.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 382.0, + 817.0, + 382.0, + 817.0, + 435.0, + 473.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 382.0, + 1006.0, + 382.0, + 1006.0, + 435.0, + 839.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 382.0, + 1095.0, + 382.0, + 1095.0, + 435.0, + 1043.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 382.0, + 1409.0, + 382.0, + 1409.0, + 435.0, + 1398.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 422.0, + 519.0, + 422.0, + 519.0, + 462.0, + 293.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 422.0, + 958.0, + 422.0, + 958.0, + 462.0, + 626.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 422.0, + 1407.0, + 422.0, + 1407.0, + 462.0, + 993.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 453.0, + 296.0, + 453.0, + 296.0, + 493.0, + 292.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 453.0, + 937.0, + 453.0, + 937.0, + 493.0, + 403.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 453.0, + 1177.0, + 453.0, + 1177.0, + 493.0, + 972.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 453.0, + 1408.0, + 453.0, + 1408.0, + 493.0, + 1283.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 483.0, + 1407.0, + 483.0, + 1407.0, + 525.0, + 292.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 516.0, + 783.0, + 516.0, + 783.0, + 552.0, + 293.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 516.0, + 1404.0, + 516.0, + 1404.0, + 552.0, + 866.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 546.0, + 1404.0, + 546.0, + 1404.0, + 582.0, + 293.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 577.0, + 1406.0, + 577.0, + 1406.0, + 612.0, + 293.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 608.0, + 1405.0, + 608.0, + 1405.0, + 641.0, + 293.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 637.0, + 611.0, + 637.0, + 611.0, + 674.0, + 294.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 637.0, + 1084.0, + 637.0, + 1084.0, + 674.0, + 648.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 637.0, + 1408.0, + 637.0, + 1408.0, + 674.0, + 1118.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 670.0, + 559.0, + 670.0, + 559.0, + 703.0, + 292.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1341.0, + 367.0, + 1341.0, + 367.0, + 1374.0, + 296.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1341.0, + 1405.0, + 1341.0, + 1405.0, + 1374.0, + 481.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1370.0, + 479.0, + 1370.0, + 479.0, + 1407.0, + 295.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1370.0, + 563.0, + 1370.0, + 563.0, + 1407.0, + 514.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1370.0, + 656.0, + 1370.0, + 656.0, + 1407.0, + 598.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1370.0, + 1407.0, + 1370.0, + 1407.0, + 1407.0, + 766.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1401.0, + 704.0, + 1401.0, + 704.0, + 1440.0, + 293.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1401.0, + 786.0, + 1401.0, + 786.0, + 1440.0, + 735.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1401.0, + 1407.0, + 1401.0, + 1407.0, + 1440.0, + 817.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1430.0, + 1378.0, + 1430.0, + 1378.0, + 1469.0, + 292.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1085.0, + 699.0, + 1085.0, + 699.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1085.0, + 783.0, + 1085.0, + 783.0, + 1122.0, + 733.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1085.0, + 1168.0, + 1085.0, + 1168.0, + 1122.0, + 817.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1085.0, + 1407.0, + 1085.0, + 1407.0, + 1122.0, + 1188.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1117.0, + 454.0, + 1117.0, + 454.0, + 1150.0, + 295.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1117.0, + 538.0, + 1117.0, + 538.0, + 1150.0, + 489.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1117.0, + 1403.0, + 1117.0, + 1403.0, + 1150.0, + 573.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1147.0, + 1141.0, + 1147.0, + 1141.0, + 1184.0, + 295.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1147.0, + 1307.0, + 1147.0, + 1307.0, + 1184.0, + 1176.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 1147.0, + 1406.0, + 1147.0, + 1406.0, + 1184.0, + 1341.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1177.0, + 1381.0, + 1177.0, + 1381.0, + 1213.0, + 294.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1802.0, + 369.0, + 1802.0, + 369.0, + 1839.0, + 293.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1802.0, + 507.0, + 1802.0, + 507.0, + 1839.0, + 458.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1802.0, + 552.0, + 1802.0, + 552.0, + 1839.0, + 547.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1802.0, + 1408.0, + 1802.0, + 1408.0, + 1839.0, + 606.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1833.0, + 911.0, + 1833.0, + 911.0, + 1871.0, + 293.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1833.0, + 992.0, + 1833.0, + 992.0, + 1871.0, + 945.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1833.0, + 1407.0, + 1833.0, + 1407.0, + 1871.0, + 1027.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1860.0, + 634.0, + 1860.0, + 634.0, + 1899.0, + 294.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1908.0, + 1405.0, + 1908.0, + 1405.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1935.0, + 743.0, + 1935.0, + 743.0, + 1979.0, + 291.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1935.0, + 895.0, + 1935.0, + 895.0, + 1979.0, + 845.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1935.0, + 1071.0, + 1935.0, + 1071.0, + 1979.0, + 1004.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1935.0, + 1408.0, + 1935.0, + 1408.0, + 1979.0, + 1200.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1052.0, + 1972.0, + 1052.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1972.0, + 1138.0, + 1972.0, + 1138.0, + 2008.0, + 1085.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1972.0, + 1407.0, + 1972.0, + 1407.0, + 2008.0, + 1172.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2001.0, + 336.0, + 2001.0, + 336.0, + 2038.0, + 292.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 2001.0, + 631.0, + 2001.0, + 631.0, + 2038.0, + 370.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 2001.0, + 1404.0, + 2001.0, + 1404.0, + 2038.0, + 665.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 812.0, + 372.0, + 812.0, + 372.0, + 849.0, + 293.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 812.0, + 1406.0, + 812.0, + 1406.0, + 849.0, + 690.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 845.0, + 1405.0, + 845.0, + 1405.0, + 880.0, + 296.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 874.0, + 430.0, + 874.0, + 430.0, + 912.0, + 295.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 874.0, + 762.0, + 874.0, + 762.0, + 912.0, + 464.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 874.0, + 991.0, + 874.0, + 991.0, + 912.0, + 797.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 874.0, + 1253.0, + 874.0, + 1253.0, + 912.0, + 1243.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 225.0, + 1406.0, + 225.0, + 1406.0, + 268.0, + 293.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 262.0, + 531.0, + 262.0, + 531.0, + 296.0, + 297.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 262.0, + 1405.0, + 262.0, + 1405.0, + 296.0, + 568.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 293.0, + 850.0, + 293.0, + 850.0, + 327.0, + 297.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1673.0, + 1403.0, + 1673.0, + 1403.0, + 1709.0, + 295.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1699.0, + 645.0, + 1699.0, + 645.0, + 1741.0, + 295.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1593.0, + 794.0, + 1593.0, + 794.0, + 1629.0, + 293.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1593.0, + 1403.0, + 1593.0, + 1403.0, + 1629.0, + 939.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1626.0, + 1079.0, + 1626.0, + 1079.0, + 1659.0, + 297.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 707.0, + 1280.0, + 707.0, + 1280.0, + 756.0, + 292.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 915.0, + 1000.0, + 915.0, + 1000.0, + 966.0, + 291.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1218.0, + 699.0, + 1218.0, + 699.0, + 1262.0, + 293.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1218.0, + 782.0, + 1218.0, + 782.0, + 1262.0, + 734.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1218.0, + 964.0, + 1218.0, + 964.0, + 1262.0, + 817.0, + 1262.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 459, + 1404, + 459, + 1404, + 766, + 298, + 766 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1031, + 1404, + 1031, + 1404, + 1247, + 298, + 1247 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1788, + 1404, + 1788, + 1404, + 2035, + 298, + 2035 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 864, + 1403, + 864, + 1403, + 1016, + 299, + 1016 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1403, + 228, + 1403, + 445, + 298, + 445 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1619, + 1404, + 1619, + 1404, + 1774, + 299, + 1774 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1261, + 1405, + 1261, + 1405, + 1448, + 298, + 1448 + ], + "score": 0.966 + }, + { + "category_id": 0, + "poly": [ + 298, + 1498, + 1051, + 1498, + 1051, + 1533, + 298, + 1533 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 300, + 811, + 563, + 811, + 563, + 842, + 300, + 842 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 300, + 1568, + 558, + 1568, + 558, + 1599, + 300, + 1599 + ], + "score": 0.832 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 856, + 76, + 856, + 104, + 299, + 104 + ], + "score": 0.791 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 857, + 2089, + 857, + 2111, + 841, + 2111 + ], + "score": 0.777 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 855, + 76, + 855, + 104, + 300, + 104 + ], + "score": 0.333 + }, + { + "category_id": 0, + "poly": [ + 300, + 1568, + 558, + 1568, + 558, + 1599, + 300, + 1599 + ], + "score": 0.192 + }, + { + "category_id": 13, + "poly": [ + 557, + 1324, + 633, + 1324, + 633, + 1356, + 557, + 1356 + ], + "score": 0.93, + "latex": "p ( \\mathbf { x } | \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 883, + 230, + 1012, + 230, + 1012, + 264, + 883, + 264 + ], + "score": 0.92, + "latex": "| S _ { r } | = | S _ { g } |" + }, + { + "category_id": 13, + "poly": [ + 906, + 2005, + 969, + 2005, + 969, + 2036, + 906, + 2036 + ], + "score": 0.92, + "latex": "\\phi ( \\mathcal { X } )" + }, + { + "category_id": 13, + "poly": [ + 717, + 322, + 821, + 322, + 821, + 355, + 717, + 355 + ], + "score": 0.91, + "latex": "S _ { g } = S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 437, + 1294, + 493, + 1294, + 493, + 1326, + 437, + 1326 + ], + "score": 0.91, + "latex": "p ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 1316, + 1093, + 1403, + 1093, + 1403, + 1127, + 1316, + 1127 + ], + "score": 0.91, + "latex": "p ( \\mathbf { x } ) \\approx" + }, + { + "category_id": 13, + "poly": [ + 677, + 1943, + 992, + 1943, + 992, + 1976, + 677, + 1976 + ], + "score": 0.91, + "latex": "d ( \\mathbf { x } , \\mathbf { x } ^ { \\prime } ) = \\| \\boldsymbol { \\phi } ( \\mathbf { x } ) - \\boldsymbol { \\phi } ( \\mathbf { x } ^ { \\prime } ) \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1076, + 1942, + 1124, + 1942, + 1124, + 1975, + 1076, + 1975 + ], + "score": 0.9, + "latex": "\\phi ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 711, + 1098, + 837, + 1098, + 837, + 1125, + 711, + 1125 + ], + "score": 0.9, + "latex": "\\mathbf { x } _ { 1 } , \\ldots , \\mathbf { x } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 841, + 1186, + 875, + 1186, + 875, + 1219, + 841, + 1219 + ], + "score": 0.9, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 299, + 1122, + 517, + 1122, + 517, + 1160, + 299, + 1160 + ], + "score": 0.89, + "latex": "\\textstyle { \\frac { 1 } { z } } \\sum _ { i = 1 } ^ { n ^ { \\cdot } } K ( \\mathbf { x } - \\mathbf { x } _ { i } )" + }, + { + "category_id": 13, + "poly": [ + 564, + 230, + 643, + 230, + 643, + 260, + 564, + 260 + ], + "score": 0.89, + "latex": "\\sim 5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 388, + 293, + 422, + 293, + 422, + 325, + 388, + 325 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 773, + 1186, + 805, + 1186, + 805, + 1216, + 773, + 1216 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1100, + 1416, + 1133, + 1416, + 1133, + 1450, + 1100, + 1450 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 328, + 323, + 360, + 323, + 360, + 353, + 328, + 353 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 452, + 643, + 485, + 643, + 485, + 677, + 452, + 677 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 772, + 895, + 805, + 895, + 805, + 929, + 772, + 929 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 994, + 262, + 1047, + 262, + 1047, + 291, + 994, + 291 + ], + "score": 0.88, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 758, + 355, + 792, + 355, + 792, + 386, + 758, + 386 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1099, + 321, + 1139, + 321, + 1139, + 351, + 1099, + 351 + ], + "score": 0.88, + "latex": "0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1053, + 614, + 1085, + 614, + 1085, + 643, + 1053, + 643 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 455, + 292, + 487, + 292, + 487, + 323, + 455, + 323 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 298, + 354, + 329, + 354, + 329, + 383, + 298, + 383 + ], + "score": 0.86, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 639, + 1714, + 658, + 1714, + 658, + 1740, + 639, + 1740 + ], + "score": 0.77, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 601, + 1130, + 620, + 1130, + 620, + 1152, + 601, + 1152 + ], + "score": 0.76, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1194, + 1064, + 1224, + 1064, + 1224, + 1092, + 1194, + 1092 + ], + "score": 0.75, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 398, + 1683, + 417, + 1683, + 417, + 1709, + 398, + 1709 + ], + "score": 0.74, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 709, + 1329, + 728, + 1329, + 728, + 1351, + 709, + 1351 + ], + "score": 0.66, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 1262, + 1098, + 1284, + 1098, + 1284, + 1122, + 1262, + 1122 + ], + "score": 0.48, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 517, + 1657, + 540, + 1657, + 540, + 1680, + 517, + 1680 + ], + "score": 0.39, + "latex": "\\mathbf { x }" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1495.0, + 1056.0, + 1495.0, + 1056.0, + 1541.0, + 291.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 807.0, + 568.0, + 807.0, + 568.0, + 846.0, + 294.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1568.0, + 559.0, + 1568.0, + 559.0, + 1601.0, + 296.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 837.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1568.0, + 559.0, + 1568.0, + 559.0, + 1601.0, + 296.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 460.0, + 1404.0, + 460.0, + 1404.0, + 495.0, + 294.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 488.0, + 1408.0, + 488.0, + 1408.0, + 529.0, + 292.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 520.0, + 1407.0, + 520.0, + 1407.0, + 558.0, + 292.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 551.0, + 1404.0, + 551.0, + 1404.0, + 586.0, + 293.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 582.0, + 1405.0, + 582.0, + 1405.0, + 616.0, + 293.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 610.0, + 1052.0, + 610.0, + 1052.0, + 649.0, + 293.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 610.0, + 1407.0, + 610.0, + 1407.0, + 649.0, + 1086.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 643.0, + 451.0, + 643.0, + 451.0, + 679.0, + 294.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 643.0, + 1406.0, + 643.0, + 1406.0, + 679.0, + 486.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 674.0, + 1405.0, + 674.0, + 1405.0, + 709.0, + 294.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 702.0, + 1409.0, + 702.0, + 1409.0, + 742.0, + 292.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 733.0, + 1213.0, + 733.0, + 1213.0, + 769.0, + 294.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1029.0, + 1405.0, + 1029.0, + 1405.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1062.0, + 1193.0, + 1062.0, + 1193.0, + 1098.0, + 293.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1062.0, + 1406.0, + 1062.0, + 1406.0, + 1098.0, + 1225.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1092.0, + 710.0, + 1092.0, + 710.0, + 1130.0, + 293.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1092.0, + 1261.0, + 1092.0, + 1261.0, + 1130.0, + 838.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1092.0, + 1315.0, + 1092.0, + 1315.0, + 1130.0, + 1285.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1124.0, + 600.0, + 1124.0, + 600.0, + 1158.0, + 518.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1124.0, + 1405.0, + 1124.0, + 1405.0, + 1158.0, + 621.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1153.0, + 1409.0, + 1153.0, + 1409.0, + 1191.0, + 292.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1185.0, + 772.0, + 1185.0, + 772.0, + 1223.0, + 292.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1185.0, + 840.0, + 1185.0, + 840.0, + 1223.0, + 806.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1185.0, + 1410.0, + 1185.0, + 1410.0, + 1223.0, + 876.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1217.0, + 1263.0, + 1217.0, + 1263.0, + 1251.0, + 294.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1129.0, + 405.0, + 1129.0, + 405.0, + 1162.5, + 312.0, + 1162.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 1406.0, + 1790.0, + 1406.0, + 1824.0, + 294.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1850.0, + 1404.0, + 1850.0, + 1404.0, + 1886.0, + 295.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1880.0, + 1408.0, + 1880.0, + 1408.0, + 1915.0, + 295.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 1404.0, + 1910.0, + 1404.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1937.0, + 676.0, + 1937.0, + 676.0, + 1983.0, + 291.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1937.0, + 1075.0, + 1937.0, + 1075.0, + 1983.0, + 993.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 1937.0, + 1406.0, + 1937.0, + 1406.0, + 1983.0, + 1125.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 905.0, + 2002.0, + 905.0, + 2039.0, + 293.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 2002.0, + 982.0, + 2002.0, + 982.0, + 2039.0, + 970.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 863.0, + 1407.0, + 863.0, + 1407.0, + 899.0, + 294.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 891.0, + 771.0, + 891.0, + 771.0, + 931.0, + 294.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 891.0, + 1405.0, + 891.0, + 1405.0, + 931.0, + 806.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 924.0, + 1407.0, + 924.0, + 1407.0, + 960.0, + 293.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 957.0, + 1405.0, + 957.0, + 1405.0, + 987.0, + 295.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 987.0, + 1382.0, + 987.0, + 1382.0, + 1020.0, + 297.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 228.0, + 563.0, + 228.0, + 563.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 228.0, + 882.0, + 228.0, + 882.0, + 270.0, + 644.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 228.0, + 1408.0, + 228.0, + 1408.0, + 270.0, + 1013.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 993.0, + 261.0, + 993.0, + 296.0, + 293.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 296.0, + 1048.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 387.0, + 292.0, + 387.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 292.0, + 454.0, + 292.0, + 454.0, + 327.0, + 423.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 292.0, + 1403.0, + 292.0, + 1403.0, + 327.0, + 488.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 322.0, + 327.0, + 322.0, + 327.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 322.0, + 716.0, + 322.0, + 716.0, + 356.0, + 361.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 322.0, + 1098.0, + 322.0, + 1098.0, + 356.0, + 822.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 322.0, + 1405.0, + 322.0, + 1405.0, + 356.0, + 1140.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 297.0, + 350.0, + 297.0, + 389.0, + 293.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 350.0, + 757.0, + 350.0, + 757.0, + 389.0, + 330.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 350.0, + 1407.0, + 350.0, + 1407.0, + 389.0, + 793.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 381.0, + 1407.0, + 381.0, + 1407.0, + 419.0, + 293.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 412.0, + 1407.0, + 412.0, + 1407.0, + 450.0, + 293.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1619.0, + 1409.0, + 1619.0, + 1409.0, + 1657.0, + 294.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1650.0, + 516.0, + 1650.0, + 516.0, + 1686.0, + 293.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1650.0, + 1406.0, + 1650.0, + 1406.0, + 1686.0, + 541.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 397.0, + 1681.0, + 397.0, + 1718.0, + 294.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1681.0, + 1406.0, + 1681.0, + 1406.0, + 1718.0, + 418.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 638.0, + 1711.0, + 638.0, + 1748.0, + 294.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1711.0, + 1406.0, + 1711.0, + 1406.0, + 1748.0, + 659.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 403.0, + 1740.0, + 403.0, + 1779.0, + 294.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1261.0, + 1405.0, + 1261.0, + 1405.0, + 1297.0, + 295.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1290.0, + 436.0, + 1290.0, + 436.0, + 1328.0, + 293.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1290.0, + 1408.0, + 1290.0, + 1408.0, + 1328.0, + 494.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1322.0, + 556.0, + 1322.0, + 556.0, + 1358.0, + 295.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1322.0, + 708.0, + 1322.0, + 708.0, + 1358.0, + 634.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1322.0, + 1408.0, + 1322.0, + 1408.0, + 1358.0, + 729.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1351.0, + 1408.0, + 1351.0, + 1408.0, + 1388.0, + 293.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1385.0, + 1405.0, + 1385.0, + 1405.0, + 1417.0, + 296.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1410.0, + 1099.0, + 1410.0, + 1099.0, + 1456.0, + 291.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1410.0, + 1144.0, + 1410.0, + 1144.0, + 1456.0, + 1134.0, + 1456.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1186, + 1405, + 1186, + 1405, + 1523, + 298, + 1523 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1597, + 1403, + 1597, + 1403, + 1782, + 298, + 1782 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 883, + 1404, + 883, + 1404, + 1005, + 298, + 1005 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 713, + 1404, + 713, + 1404, + 867, + 298, + 867 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 301, + 219, + 1401, + 219, + 1401, + 544, + 301, + 544 + ], + "score": 0.961 + }, + { + "category_id": 2, + "poly": [ + 300, + 1915, + 1404, + 1915, + 1404, + 2033, + 300, + 2033 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 303, + 1079, + 1401, + 1079, + 1401, + 1172, + 303, + 1172 + ], + "score": 0.957 + }, + { + "category_id": 4, + "poly": [ + 297, + 564, + 1404, + 564, + 1404, + 677, + 297, + 677 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 302, + 1795, + 1401, + 1795, + 1401, + 1889, + 302, + 1889 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 300, + 1551, + 595, + 1551, + 595, + 1582, + 300, + 1582 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 299, + 1034, + 445, + 1034, + 445, + 1064, + 299, + 1064 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 857, + 76, + 857, + 104, + 298, + 104 + ], + "score": 0.826 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.751 + }, + { + "category_id": 2, + "poly": [ + 300, + 77, + 855, + 77, + 855, + 103, + 300, + 103 + ], + "score": 0.283 + }, + { + "category_id": 13, + "poly": [ + 669, + 1370, + 908, + 1370, + 908, + 1403, + 669, + 1403 + ], + "score": 0.93, + "latex": "R I S = \\left( 1 - { \\mathrm { I S } } / { \\mathrm { I S } } _ { 0 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 937, + 1796, + 1001, + 1796, + 1001, + 1831, + 937, + 1831 + ], + "score": 0.93, + "latex": "S _ { g } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1045, + 1796, + 1147, + 1796, + 1147, + 1830, + 1045, + 1830 + ], + "score": 0.92, + "latex": "t \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 579, + 1689, + 643, + 1689, + 643, + 1722, + 579, + 1722 + ], + "score": 0.92, + "latex": "S _ { g } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1339, + 621, + 1370, + 621, + 1370, + 650, + 1339, + 650 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 853, + 1797, + 885, + 1797, + 885, + 1827, + 853, + 1827 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 764, + 1310, + 796, + 1310, + 796, + 1340, + 764, + 1340 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1235, + 1310, + 1276, + 1310, + 1276, + 1341, + 1235, + 1341 + ], + "score": 0.87, + "latex": "\\mathrm { I S } _ { \\mathrm { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 443, + 1659, + 476, + 1659, + 476, + 1689, + 443, + 1689 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 367, + 1828, + 402, + 1828, + 402, + 1858, + 367, + 1858 + ], + "score": 0.87, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 665, + 1659, + 780, + 1659, + 780, + 1688, + 665, + 1688 + ], + "score": 0.87, + "latex": "n = 2 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 1298, + 1829, + 1318, + 1829, + 1318, + 1860, + 1298, + 1860 + ], + "score": 0.84, + "latex": "\\hat { \\rho }" + }, + { + "category_id": 13, + "poly": [ + 1217, + 1600, + 1236, + 1600, + 1236, + 1629, + 1217, + 1629 + ], + "score": 0.84, + "latex": "\\hat { \\rho }" + }, + { + "category_id": 13, + "poly": [ + 1254, + 1722, + 1269, + 1722, + 1269, + 1747, + 1254, + 1747 + ], + "score": 0.72, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 831, + 1694, + 852, + 1694, + 852, + 1716, + 831, + 1716 + ], + "score": 0.72, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 632, + 1663, + 652, + 1663, + 652, + 1686, + 632, + 1686 + ], + "score": 0.61, + "latex": "n" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 223.0, + 572.0, + 223.0, + 572.0, + 254.0, + 374.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 223.0, + 839.0, + 223.0, + 839.0, + 254.0, + 636.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 223.0, + 1094.0, + 223.0, + 1094.0, + 254.0, + 906.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 223.0, + 1358.0, + 223.0, + 1358.0, + 254.0, + 1166.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 254.0, + 358.0, + 254.0, + 358.0, + 286.0, + 317.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 300.0, + 330.0, + 300.0, + 330.0, + 364.0, + 302.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 305.0, + 358.0, + 305.0, + 358.0, + 338.0, + 317.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 359.0, + 358.0, + 359.0, + 358.0, + 390.0, + 317.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 412.0, + 376.0, + 412.0, + 376.0, + 439.0, + 336.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 412.0, + 436.0, + 412.0, + 436.0, + 439.0, + 395.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 410.0, + 496.0, + 410.0, + 496.0, + 442.0, + 452.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 410.0, + 553.0, + 410.0, + 553.0, + 442.0, + 511.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 412.0, + 640.0, + 412.0, + 640.0, + 441.0, + 572.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 410.0, + 701.0, + 410.0, + 701.0, + 442.0, + 657.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 410.0, + 759.0, + 410.0, + 759.0, + 442.0, + 715.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 410.0, + 817.0, + 410.0, + 817.0, + 442.0, + 775.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 412.0, + 903.0, + 412.0, + 903.0, + 441.0, + 835.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 412.0, + 963.0, + 412.0, + 963.0, + 439.0, + 922.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 410.0, + 1024.0, + 410.0, + 1024.0, + 442.0, + 978.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 410.0, + 1082.0, + 410.0, + 1082.0, + 441.0, + 1037.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 412.0, + 1167.0, + 412.0, + 1167.0, + 441.0, + 1098.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 412.0, + 1227.0, + 412.0, + 1227.0, + 439.0, + 1186.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 410.0, + 1287.0, + 410.0, + 1287.0, + 442.0, + 1242.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 410.0, + 1345.0, + 410.0, + 1345.0, + 442.0, + 1302.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 412.0, + 1402.0, + 412.0, + 1402.0, + 439.0, + 1363.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 432.0, + 581.0, + 432.0, + 581.0, + 462.0, + 365.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 432.0, + 846.0, + 432.0, + 846.0, + 462.0, + 628.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 432.0, + 1109.0, + 432.0, + 1109.0, + 462.0, + 892.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 432.0, + 1372.0, + 432.0, + 1372.0, + 462.0, + 1155.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 484.0, + 612.0, + 484.0, + 612.0, + 511.0, + 377.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 483.0, + 855.0, + 483.0, + 855.0, + 512.0, + 680.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 481.0, + 1046.0, + 481.0, + 1046.0, + 513.0, + 922.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 483.0, + 1327.0, + 483.0, + 1327.0, + 513.0, + 1156.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 509.0, + 595.0, + 509.0, + 595.0, + 536.0, + 376.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 507.0, + 734.0, + 507.0, + 734.0, + 536.0, + 677.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 507.0, + 1090.0, + 507.0, + 1090.0, + 537.0, + 921.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 509.0, + 1367.0, + 509.0, + 1367.0, + 540.0, + 1155.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1910.0, + 1376.0, + 1910.0, + 1376.0, + 1954.0, + 329.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1940.0, + 1371.0, + 1940.0, + 1371.0, + 1984.0, + 328.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1967.0, + 1405.0, + 1967.0, + 1405.0, + 2016.0, + 325.0, + 2016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 594.0, + 2000.0, + 594.0, + 2040.0, + 295.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 561.0, + 1405.0, + 561.0, + 1405.0, + 601.0, + 292.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 590.0, + 1407.0, + 590.0, + 1407.0, + 627.0, + 291.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 616.0, + 1338.0, + 616.0, + 1338.0, + 655.0, + 292.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 616.0, + 1406.0, + 616.0, + 1406.0, + 655.0, + 1371.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 644.0, + 1180.0, + 644.0, + 1180.0, + 680.0, + 295.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1548.0, + 600.0, + 1548.0, + 600.0, + 1586.0, + 293.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1030.0, + 450.0, + 1030.0, + 450.0, + 1069.0, + 293.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 70.0, + 860.0, + 70.0, + 860.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1187.0, + 1408.0, + 1187.0, + 1408.0, + 1222.0, + 295.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1215.0, + 1406.0, + 1215.0, + 1406.0, + 1254.0, + 293.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1247.0, + 1406.0, + 1247.0, + 1406.0, + 1284.0, + 292.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1278.0, + 1405.0, + 1278.0, + 1405.0, + 1314.0, + 292.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1305.0, + 763.0, + 1305.0, + 763.0, + 1347.0, + 292.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 1305.0, + 1234.0, + 1305.0, + 1234.0, + 1347.0, + 797.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1305.0, + 1409.0, + 1305.0, + 1409.0, + 1347.0, + 1277.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1341.0, + 1406.0, + 1341.0, + 1406.0, + 1376.0, + 295.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1371.0, + 668.0, + 1371.0, + 668.0, + 1405.0, + 295.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1371.0, + 1405.0, + 1371.0, + 1405.0, + 1405.0, + 909.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1400.0, + 1406.0, + 1400.0, + 1406.0, + 1435.0, + 295.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1431.0, + 1405.0, + 1431.0, + 1405.0, + 1466.0, + 295.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1461.0, + 1406.0, + 1461.0, + 1406.0, + 1496.0, + 293.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1492.0, + 1313.0, + 1492.0, + 1313.0, + 1527.0, + 293.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1594.0, + 1216.0, + 1594.0, + 1216.0, + 1636.0, + 292.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1594.0, + 1405.0, + 1594.0, + 1405.0, + 1636.0, + 1237.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1626.0, + 1405.0, + 1626.0, + 1405.0, + 1664.0, + 293.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1660.0, + 442.0, + 1660.0, + 442.0, + 1692.0, + 296.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1660.0, + 631.0, + 1660.0, + 631.0, + 1692.0, + 477.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1660.0, + 664.0, + 1660.0, + 664.0, + 1692.0, + 653.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1660.0, + 1404.0, + 1660.0, + 1404.0, + 1692.0, + 781.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1688.0, + 578.0, + 1688.0, + 578.0, + 1724.0, + 294.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1688.0, + 830.0, + 1688.0, + 830.0, + 1724.0, + 644.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1688.0, + 1405.0, + 1688.0, + 1405.0, + 1724.0, + 853.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1719.0, + 1253.0, + 1719.0, + 1253.0, + 1754.0, + 294.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1719.0, + 1404.0, + 1719.0, + 1404.0, + 1754.0, + 1270.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1747.0, + 592.0, + 1747.0, + 592.0, + 1786.0, + 293.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 883.0, + 1405.0, + 883.0, + 1405.0, + 919.0, + 293.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 912.0, + 1405.0, + 912.0, + 1405.0, + 950.0, + 292.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 945.0, + 1406.0, + 945.0, + 1406.0, + 978.0, + 296.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 975.0, + 1287.0, + 975.0, + 1287.0, + 1008.0, + 296.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 713.0, + 1406.0, + 713.0, + 1406.0, + 746.0, + 296.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 741.0, + 1406.0, + 741.0, + 1406.0, + 782.0, + 293.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 774.0, + 1404.0, + 774.0, + 1404.0, + 810.0, + 293.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 799.0, + 1405.0, + 799.0, + 1405.0, + 846.0, + 291.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 836.0, + 467.0, + 836.0, + 467.0, + 869.0, + 294.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1078.0, + 1406.0, + 1078.0, + 1406.0, + 1116.0, + 295.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1111.0, + 1404.0, + 1111.0, + 1404.0, + 1145.0, + 297.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1140.0, + 1406.0, + 1140.0, + 1406.0, + 1176.0, + 298.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1795.0, + 852.0, + 1795.0, + 852.0, + 1832.0, + 297.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1795.0, + 936.0, + 1795.0, + 936.0, + 1832.0, + 886.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1795.0, + 1044.0, + 1795.0, + 1044.0, + 1832.0, + 1002.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1795.0, + 1407.0, + 1795.0, + 1407.0, + 1832.0, + 1148.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1826.0, + 366.0, + 1826.0, + 366.0, + 1864.0, + 297.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1826.0, + 1297.0, + 1826.0, + 1297.0, + 1864.0, + 403.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1826.0, + 1406.0, + 1826.0, + 1406.0, + 1864.0, + 1319.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1855.0, + 1406.0, + 1855.0, + 1406.0, + 1893.0, + 296.0, + 1893.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 840, + 1404, + 840, + 1404, + 1145, + 298, + 1145 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1529, + 1405, + 1529, + 1405, + 1895, + 298, + 1895 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1299, + 1404, + 1299, + 1404, + 1513, + 298, + 1513 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1161, + 1402, + 1161, + 1402, + 1284, + 298, + 1284 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1911, + 1403, + 1911, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 702, + 1403, + 702, + 1403, + 824, + 298, + 824 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 302, + 220, + 1400, + 220, + 1400, + 525, + 302, + 525 + ], + "score": 0.963 + }, + { + "category_id": 4, + "poly": [ + 297, + 545, + 1404, + 545, + 1404, + 658, + 297, + 658 + ], + "score": 0.962 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 857, + 76, + 857, + 104, + 298, + 104 + ], + "score": 0.868 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 858, + 2089, + 858, + 2112, + 840, + 2112 + ], + "score": 0.779 + }, + { + "category_id": 13, + "poly": [ + 405, + 1620, + 514, + 1620, + 514, + 1654, + 405, + 1654 + ], + "score": 0.94, + "latex": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 550, + 1084, + 613, + 1084, + 613, + 1119, + 550, + 1119 + ], + "score": 0.92, + "latex": "S _ { g } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1285, + 1192, + 1348, + 1192, + 1348, + 1226, + 1285, + 1226 + ], + "score": 0.92, + "latex": "S _ { g } ( t )" + }, + { + "category_id": 13, + "poly": [ + 512, + 1943, + 544, + 1943, + 544, + 1975, + 512, + 1975 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1091, + 1743, + 1159, + 1743, + 1159, + 1772, + 1091, + 1772 + ], + "score": 0.89, + "latex": "C = 0" + }, + { + "category_id": 13, + "poly": [ + 1020, + 1529, + 1051, + 1529, + 1051, + 1562, + 1020, + 1562 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 965, + 1423, + 999, + 1423, + 999, + 1455, + 965, + 1455 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 403, + 1453, + 436, + 1453, + 436, + 1483, + 403, + 1483 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1331, + 1069, + 1331, + 1069, + 1364, + 1036, + 1364 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1005, + 1912, + 1036, + 1912, + 1036, + 1945, + 1005, + 1945 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 662, + 1423, + 694, + 1423, + 694, + 1452, + 662, + 1452 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 639, + 1454, + 673, + 1454, + 673, + 1486, + 639, + 1486 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1270, + 1055, + 1302, + 1055, + 1302, + 1084, + 1270, + 1084 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1326, + 1713, + 1395, + 1713, + 1395, + 1742, + 1326, + 1742 + ], + "score": 0.88, + "latex": "C > 0" + }, + { + "category_id": 13, + "poly": [ + 966, + 1301, + 999, + 1301, + 999, + 1331, + 966, + 1331 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 938, + 1530, + 970, + 1530, + 970, + 1560, + 938, + 1560 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 703, + 1913, + 734, + 1913, + 734, + 1942, + 703, + 1942 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1221, + 1331, + 1255, + 1331, + 1255, + 1361, + 1221, + 1361 + ], + "score": 0.87, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 693, + 2008, + 716, + 2008, + 716, + 2030, + 693, + 2030 + ], + "score": 0.83, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 1010, + 1563, + 1029, + 1563, + 1029, + 1588, + 1010, + 1588 + ], + "score": 0.82, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1171, + 1622, + 1196, + 1622, + 1196, + 1649, + 1171, + 1649 + ], + "score": 0.81, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1224, + 1149, + 1224, + 1149, + 1250, + 1122, + 1250 + ], + "score": 0.8, + "latex": "\\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 666, + 1652, + 691, + 1652, + 691, + 1679, + 666, + 1679 + ], + "score": 0.77, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 490, + 797, + 505, + 797, + 505, + 821, + 490, + 821 + ], + "score": 0.55, + "latex": "t" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 223.0, + 568.0, + 223.0, + 568.0, + 255.0, + 373.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 219.0, + 835.0, + 219.0, + 835.0, + 255.0, + 635.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 219.0, + 1092.0, + 219.0, + 1092.0, + 255.0, + 909.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 219.0, + 1358.0, + 219.0, + 1358.0, + 255.0, + 1172.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 245.0, + 358.0, + 245.0, + 358.0, + 274.0, + 318.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 298.0, + 327.0, + 298.0, + 327.0, + 357.0, + 301.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 314.0, + 356.0, + 314.0, + 356.0, + 343.0, + 317.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 384.0, + 358.0, + 384.0, + 358.0, + 413.0, + 317.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 403.0, + 373.0, + 403.0, + 373.0, + 432.0, + 333.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 403.0, + 433.0, + 403.0, + 433.0, + 431.0, + 392.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 403.0, + 492.0, + 403.0, + 492.0, + 431.0, + 450.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 403.0, + 549.0, + 403.0, + 549.0, + 431.0, + 509.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 402.0, + 639.0, + 402.0, + 639.0, + 432.0, + 570.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 403.0, + 698.0, + 403.0, + 698.0, + 431.0, + 658.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 402.0, + 756.0, + 402.0, + 756.0, + 431.0, + 716.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 403.0, + 814.0, + 403.0, + 814.0, + 431.0, + 774.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 403.0, + 905.0, + 403.0, + 905.0, + 432.0, + 834.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 403.0, + 963.0, + 403.0, + 963.0, + 431.0, + 923.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 403.0, + 1022.0, + 403.0, + 1022.0, + 431.0, + 981.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 403.0, + 1079.0, + 403.0, + 1079.0, + 431.0, + 1039.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 403.0, + 1169.0, + 403.0, + 1169.0, + 432.0, + 1098.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 403.0, + 1228.0, + 403.0, + 1228.0, + 431.0, + 1188.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 403.0, + 1287.0, + 403.0, + 1287.0, + 431.0, + 1245.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 403.0, + 1344.0, + 403.0, + 1344.0, + 431.0, + 1304.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1364.0, + 403.0, + 1403.0, + 403.0, + 1403.0, + 431.0, + 1364.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 423.0, + 581.0, + 423.0, + 581.0, + 451.0, + 361.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 423.0, + 847.0, + 423.0, + 847.0, + 451.0, + 625.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 423.0, + 1112.0, + 423.0, + 1112.0, + 451.0, + 891.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 423.0, + 1376.0, + 423.0, + 1376.0, + 451.0, + 1155.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 462.0, + 625.0, + 462.0, + 625.0, + 494.0, + 371.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 461.0, + 884.0, + 461.0, + 884.0, + 493.0, + 698.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 457.0, + 1178.0, + 457.0, + 1178.0, + 494.0, + 995.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 457.0, + 1388.0, + 457.0, + 1388.0, + 494.0, + 1252.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 488.0, + 606.0, + 488.0, + 606.0, + 519.0, + 369.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 488.0, + 925.0, + 488.0, + 925.0, + 523.0, + 695.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 484.0, + 1181.0, + 484.0, + 1181.0, + 522.0, + 994.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 486.0, + 1310.0, + 486.0, + 1310.0, + 518.0, + 1250.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 544.0, + 1405.0, + 544.0, + 1405.0, + 578.0, + 294.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 574.0, + 1404.0, + 574.0, + 1404.0, + 603.0, + 296.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 596.0, + 1406.0, + 596.0, + 1406.0, + 636.0, + 292.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 628.0, + 954.0, + 628.0, + 954.0, + 661.0, + 295.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 838.0, + 1405.0, + 838.0, + 1405.0, + 876.0, + 293.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 871.0, + 1405.0, + 871.0, + 1405.0, + 908.0, + 295.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 901.0, + 1405.0, + 901.0, + 1405.0, + 937.0, + 294.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 932.0, + 1405.0, + 932.0, + 1405.0, + 967.0, + 294.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 962.0, + 1404.0, + 962.0, + 1404.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 989.0, + 1405.0, + 989.0, + 1405.0, + 1030.0, + 293.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1022.0, + 1406.0, + 1022.0, + 1406.0, + 1058.0, + 293.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1053.0, + 1269.0, + 1053.0, + 1269.0, + 1089.0, + 294.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1053.0, + 1404.0, + 1053.0, + 1404.0, + 1089.0, + 1303.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1085.0, + 549.0, + 1085.0, + 549.0, + 1120.0, + 294.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1120.0, + 614.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1118.0, + 489.0, + 1118.0, + 489.0, + 1148.0, + 294.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1526.0, + 937.0, + 1526.0, + 937.0, + 1566.0, + 292.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1526.0, + 1019.0, + 1526.0, + 1019.0, + 1566.0, + 971.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1526.0, + 1407.0, + 1526.0, + 1407.0, + 1566.0, + 1052.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1559.0, + 1009.0, + 1559.0, + 1009.0, + 1596.0, + 293.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1559.0, + 1405.0, + 1559.0, + 1405.0, + 1596.0, + 1030.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1591.0, + 1405.0, + 1591.0, + 1405.0, + 1625.0, + 295.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1621.0, + 404.0, + 1621.0, + 404.0, + 1655.0, + 295.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1621.0, + 1170.0, + 1621.0, + 1170.0, + 1655.0, + 515.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1621.0, + 1405.0, + 1621.0, + 1405.0, + 1655.0, + 1197.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 665.0, + 1653.0, + 665.0, + 1687.0, + 295.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1653.0, + 1405.0, + 1653.0, + 1405.0, + 1687.0, + 692.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1683.0, + 1406.0, + 1683.0, + 1406.0, + 1717.0, + 295.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1713.0, + 1325.0, + 1713.0, + 1325.0, + 1747.0, + 295.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 1713.0, + 1405.0, + 1713.0, + 1405.0, + 1747.0, + 1396.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1740.0, + 1090.0, + 1740.0, + 1090.0, + 1779.0, + 292.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1779.0, + 1160.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1808.0, + 295.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1804.0, + 1403.0, + 1804.0, + 1403.0, + 1838.0, + 295.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1831.0, + 1405.0, + 1831.0, + 1405.0, + 1873.0, + 292.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1866.0, + 629.0, + 1866.0, + 629.0, + 1898.0, + 293.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 965.0, + 1298.0, + 965.0, + 1336.0, + 293.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1298.0, + 1406.0, + 1298.0, + 1406.0, + 1336.0, + 1000.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1327.0, + 1035.0, + 1327.0, + 1035.0, + 1368.0, + 292.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1327.0, + 1220.0, + 1327.0, + 1220.0, + 1368.0, + 1070.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 1327.0, + 1405.0, + 1327.0, + 1405.0, + 1368.0, + 1256.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1361.0, + 1406.0, + 1361.0, + 1406.0, + 1395.0, + 294.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1390.0, + 1406.0, + 1390.0, + 1406.0, + 1427.0, + 293.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1421.0, + 661.0, + 1421.0, + 661.0, + 1459.0, + 293.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1421.0, + 964.0, + 1421.0, + 964.0, + 1459.0, + 695.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1421.0, + 1406.0, + 1421.0, + 1406.0, + 1459.0, + 1000.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1451.0, + 402.0, + 1451.0, + 402.0, + 1489.0, + 293.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1451.0, + 638.0, + 1451.0, + 638.0, + 1489.0, + 437.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1451.0, + 1406.0, + 1451.0, + 1406.0, + 1489.0, + 674.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1481.0, + 796.0, + 1481.0, + 796.0, + 1519.0, + 293.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1159.0, + 1406.0, + 1159.0, + 1406.0, + 1197.0, + 292.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1191.0, + 1284.0, + 1191.0, + 1284.0, + 1227.0, + 294.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1191.0, + 1406.0, + 1191.0, + 1406.0, + 1227.0, + 1349.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1221.0, + 1121.0, + 1221.0, + 1121.0, + 1257.0, + 293.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1221.0, + 1407.0, + 1221.0, + 1407.0, + 1257.0, + 1150.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1254.0, + 1220.0, + 1254.0, + 1220.0, + 1286.0, + 296.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1911.0, + 702.0, + 1911.0, + 702.0, + 1947.0, + 293.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1911.0, + 1004.0, + 1911.0, + 1004.0, + 1947.0, + 735.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1911.0, + 1409.0, + 1911.0, + 1409.0, + 1947.0, + 1037.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 511.0, + 1941.0, + 511.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1941.0, + 1407.0, + 1941.0, + 1407.0, + 1977.0, + 545.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1974.0, + 1407.0, + 1974.0, + 1407.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 692.0, + 2000.0, + 692.0, + 2040.0, + 293.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2040.0, + 717.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 703.0, + 1402.0, + 703.0, + 1402.0, + 735.0, + 294.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 734.0, + 1405.0, + 734.0, + 1405.0, + 766.0, + 294.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 764.0, + 1406.0, + 764.0, + 1406.0, + 799.0, + 293.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 793.0, + 489.0, + 793.0, + 489.0, + 829.0, + 295.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 793.0, + 587.0, + 793.0, + 587.0, + 829.0, + 506.0, + 829.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1407, + 958, + 1407, + 958, + 1622, + 298, + 1622 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1230, + 1405, + 1230, + 1405, + 1324, + 299, + 1324 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 300, + 644, + 1403, + 644, + 1403, + 1062, + 300, + 1062 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 1637, + 957, + 1637, + 957, + 1943, + 298, + 1943 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 978, + 1765, + 1402, + 1765, + 1402, + 1932, + 978, + 1932 + ], + "score": 0.966 + }, + { + "category_id": 3, + "poly": [ + 301, + 221, + 1400, + 221, + 1400, + 523, + 301, + 523 + ], + "score": 0.966 + }, + { + "category_id": 3, + "poly": [ + 988, + 1417, + 1389, + 1417, + 1389, + 1743, + 988, + 1743 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 299, + 1944, + 1403, + 1944, + 1403, + 2035, + 299, + 2035 + ], + "score": 0.962 + }, + { + "category_id": 4, + "poly": [ + 297, + 1073, + 1406, + 1073, + 1406, + 1187, + 297, + 1187 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 298, + 532, + 1400, + 532, + 1400, + 618, + 298, + 618 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 302, + 1360, + 804, + 1360, + 804, + 1389, + 302, + 1389 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.756 + }, + { + "category_id": 13, + "poly": [ + 298, + 1699, + 329, + 1699, + 329, + 1732, + 298, + 1732 + ], + "score": 0.87, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1759, + 329, + 1759, + 329, + 1793, + 298, + 1793 + ], + "score": 0.87, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 872, + 1669, + 905, + 1669, + 905, + 1699, + 872, + 1699 + ], + "score": 0.87, + "latex": "S _ { r }" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 644.0, + 561.0, + 644.0, + 561.0, + 674.0, + 383.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 645.0, + 826.0, + 645.0, + 826.0, + 672.0, + 647.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 645.0, + 1083.0, + 645.0, + 1083.0, + 672.0, + 917.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 642.0, + 1348.0, + 642.0, + 1348.0, + 674.0, + 1178.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 672.0, + 355.0, + 672.0, + 355.0, + 700.0, + 320.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 674.0, + 330.0, + 674.0, + 330.0, + 796.0, + 297.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 718.0, + 356.0, + 718.0, + 356.0, + 748.0, + 318.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 765.0, + 357.0, + 765.0, + 357.0, + 797.0, + 318.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 797.0, + 373.0, + 797.0, + 373.0, + 824.0, + 335.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 796.0, + 433.0, + 796.0, + 433.0, + 823.0, + 394.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 796.0, + 493.0, + 796.0, + 493.0, + 823.0, + 453.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 797.0, + 552.0, + 797.0, + 552.0, + 824.0, + 513.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 797.0, + 636.0, + 797.0, + 636.0, + 823.0, + 576.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 798.0, + 695.0, + 798.0, + 695.0, + 822.0, + 660.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 798.0, + 755.0, + 798.0, + 755.0, + 822.0, + 720.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 797.0, + 816.0, + 797.0, + 816.0, + 823.0, + 776.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 796.0, + 902.0, + 796.0, + 902.0, + 824.0, + 838.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 796.0, + 961.0, + 796.0, + 961.0, + 823.0, + 922.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 798.0, + 1019.0, + 798.0, + 1019.0, + 822.0, + 983.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 798.0, + 1078.0, + 798.0, + 1078.0, + 822.0, + 1042.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 796.0, + 1166.0, + 796.0, + 1166.0, + 824.0, + 1102.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 797.0, + 1225.0, + 797.0, + 1225.0, + 823.0, + 1185.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 798.0, + 1283.0, + 798.0, + 1283.0, + 822.0, + 1246.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 797.0, + 1344.0, + 797.0, + 1344.0, + 823.0, + 1304.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 799.0, + 1402.0, + 799.0, + 1402.0, + 822.0, + 1368.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 822.0, + 333.0, + 822.0, + 333.0, + 948.0, + 300.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 822.0, + 355.0, + 822.0, + 355.0, + 849.0, + 319.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 869.0, + 357.0, + 869.0, + 357.0, + 897.0, + 319.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 917.0, + 355.0, + 917.0, + 355.0, + 945.0, + 318.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 946.0, + 373.0, + 946.0, + 373.0, + 974.0, + 335.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 946.0, + 433.0, + 946.0, + 433.0, + 974.0, + 394.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 946.0, + 493.0, + 946.0, + 493.0, + 974.0, + 453.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 946.0, + 552.0, + 946.0, + 552.0, + 974.0, + 513.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 948.0, + 636.0, + 948.0, + 636.0, + 974.0, + 576.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 946.0, + 697.0, + 946.0, + 697.0, + 974.0, + 659.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 946.0, + 756.0, + 946.0, + 756.0, + 974.0, + 719.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 946.0, + 816.0, + 946.0, + 816.0, + 974.0, + 778.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 948.0, + 899.0, + 948.0, + 899.0, + 974.0, + 840.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 946.0, + 961.0, + 946.0, + 961.0, + 974.0, + 922.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 946.0, + 1021.0, + 946.0, + 1021.0, + 974.0, + 982.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 946.0, + 1080.0, + 946.0, + 1080.0, + 974.0, + 1041.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 948.0, + 1164.0, + 948.0, + 1164.0, + 974.0, + 1105.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 946.0, + 1225.0, + 946.0, + 1225.0, + 974.0, + 1186.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 946.0, + 1284.0, + 946.0, + 1284.0, + 974.0, + 1245.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 946.0, + 1344.0, + 946.0, + 1344.0, + 974.0, + 1305.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 946.0, + 1403.0, + 946.0, + 1403.0, + 974.0, + 1367.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 965.0, + 588.0, + 965.0, + 588.0, + 992.0, + 359.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 965.0, + 852.0, + 965.0, + 852.0, + 992.0, + 622.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 965.0, + 1116.0, + 965.0, + 1116.0, + 992.0, + 887.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 965.0, + 1379.0, + 965.0, + 1379.0, + 992.0, + 1150.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1005.0, + 530.0, + 1005.0, + 530.0, + 1035.0, + 352.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1003.0, + 724.0, + 1003.0, + 724.0, + 1035.0, + 598.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1005.0, + 1012.0, + 1005.0, + 1012.0, + 1035.0, + 838.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1002.0, + 1373.0, + 1002.0, + 1373.0, + 1036.0, + 1146.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1029.0, + 408.0, + 1029.0, + 408.0, + 1058.0, + 350.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1027.0, + 769.0, + 1027.0, + 769.0, + 1062.0, + 596.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 1030.0, + 1080.0, + 1030.0, + 1080.0, + 1060.0, + 837.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1032.0, + 1364.0, + 1032.0, + 1364.0, + 1064.0, + 1146.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1764.0, + 1404.0, + 1764.0, + 1404.0, + 1795.0, + 976.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1793.0, + 1403.0, + 1793.0, + 1403.0, + 1822.0, + 977.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1820.0, + 1404.0, + 1820.0, + 1404.0, + 1850.0, + 977.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1849.0, + 1404.0, + 1849.0, + 1404.0, + 1879.0, + 975.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1875.0, + 1405.0, + 1875.0, + 1405.0, + 1906.0, + 977.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1904.0, + 1253.0, + 1904.0, + 1253.0, + 1933.0, + 978.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 221.0, + 566.0, + 221.0, + 566.0, + 253.0, + 374.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 221.0, + 833.0, + 221.0, + 833.0, + 253.0, + 637.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 222.0, + 1092.0, + 222.0, + 1092.0, + 253.0, + 909.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 220.0, + 1359.0, + 220.0, + 1359.0, + 254.0, + 1173.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 260.0, + 356.0, + 260.0, + 356.0, + 289.0, + 318.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 298.0, + 326.0, + 298.0, + 326.0, + 356.0, + 301.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 322.0, + 355.0, + 322.0, + 355.0, + 351.0, + 317.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 386.0, + 357.0, + 386.0, + 357.0, + 412.0, + 317.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 403.0, + 373.0, + 403.0, + 373.0, + 431.0, + 335.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 403.0, + 433.0, + 403.0, + 433.0, + 431.0, + 393.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 403.0, + 491.0, + 403.0, + 491.0, + 431.0, + 451.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 403.0, + 548.0, + 403.0, + 548.0, + 431.0, + 508.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 402.0, + 639.0, + 402.0, + 639.0, + 432.0, + 569.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 403.0, + 698.0, + 403.0, + 698.0, + 431.0, + 658.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 403.0, + 757.0, + 403.0, + 757.0, + 431.0, + 716.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 403.0, + 814.0, + 403.0, + 814.0, + 431.0, + 773.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 403.0, + 904.0, + 403.0, + 904.0, + 432.0, + 834.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 403.0, + 963.0, + 403.0, + 963.0, + 431.0, + 923.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 403.0, + 1022.0, + 403.0, + 1022.0, + 431.0, + 982.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 403.0, + 1079.0, + 403.0, + 1079.0, + 431.0, + 1038.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 403.0, + 1169.0, + 403.0, + 1169.0, + 432.0, + 1099.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 403.0, + 1228.0, + 403.0, + 1228.0, + 431.0, + 1188.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 403.0, + 1286.0, + 403.0, + 1286.0, + 431.0, + 1246.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 403.0, + 1344.0, + 403.0, + 1344.0, + 431.0, + 1304.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 403.0, + 1403.0, + 403.0, + 1403.0, + 431.0, + 1365.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 424.0, + 577.0, + 424.0, + 577.0, + 450.0, + 365.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 420.0, + 842.0, + 420.0, + 842.0, + 454.0, + 630.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 424.0, + 1107.0, + 424.0, + 1107.0, + 451.0, + 895.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 420.0, + 1371.0, + 420.0, + 1371.0, + 454.0, + 1159.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 464.0, + 623.0, + 464.0, + 623.0, + 491.0, + 372.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 461.0, + 884.0, + 461.0, + 884.0, + 491.0, + 697.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 461.0, + 1177.0, + 461.0, + 1177.0, + 493.0, + 997.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 460.0, + 1387.0, + 460.0, + 1387.0, + 493.0, + 1253.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 488.0, + 607.0, + 488.0, + 607.0, + 519.0, + 371.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 487.0, + 926.0, + 487.0, + 926.0, + 523.0, + 696.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 485.0, + 1179.0, + 485.0, + 1179.0, + 520.0, + 996.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 485.0, + 1311.0, + 485.0, + 1311.0, + 520.0, + 1251.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1417.0, + 1293.0, + 1417.0, + 1293.0, + 1450.0, + 1144.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1481.0, + 1047.0, + 1481.0, + 1047.0, + 1510.0, + 1008.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1528.0, + 1014.0, + 1528.0, + 1014.0, + 1614.0, + 986.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1579.0, + 1049.0, + 1579.0, + 1049.0, + 1609.0, + 1008.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1593.0, + 1330.0, + 1593.0, + 1330.0, + 1619.0, + 1204.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1615.0, + 1376.0, + 1615.0, + 1376.0, + 1640.0, + 1204.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1634.0, + 1254.0, + 1634.0, + 1254.0, + 1662.0, + 1204.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1657.0, + 1326.0, + 1657.0, + 1326.0, + 1686.0, + 1205.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1677.0, + 1048.0, + 1677.0, + 1048.0, + 1702.0, + 1007.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1695.0, + 1390.0, + 1695.0, + 1390.0, + 1720.0, + 1074.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1717.0, + 1350.0, + 1717.0, + 1350.0, + 1746.0, + 1087.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1072.0, + 1404.0, + 1072.0, + 1404.0, + 1106.0, + 296.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1099.0, + 1403.0, + 1099.0, + 1403.0, + 1133.0, + 295.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1127.0, + 1403.0, + 1127.0, + 1403.0, + 1162.0, + 294.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1157.0, + 1159.0, + 1157.0, + 1159.0, + 1190.0, + 295.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 1404.0, + 532.0, + 1404.0, + 565.0, + 294.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 560.0, + 1405.0, + 560.0, + 1405.0, + 592.0, + 294.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 587.0, + 1339.0, + 587.0, + 1339.0, + 620.0, + 293.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1357.0, + 809.0, + 1357.0, + 809.0, + 1394.0, + 294.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1407.0, + 958.0, + 1407.0, + 958.0, + 1441.0, + 296.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1438.0, + 960.0, + 1438.0, + 960.0, + 1472.0, + 293.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1469.0, + 961.0, + 1469.0, + 961.0, + 1500.0, + 295.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1499.0, + 962.0, + 1499.0, + 962.0, + 1531.0, + 295.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1530.0, + 960.0, + 1530.0, + 960.0, + 1561.0, + 294.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1561.0, + 959.0, + 1561.0, + 959.0, + 1594.0, + 295.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1591.0, + 744.0, + 1591.0, + 744.0, + 1626.0, + 295.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1230.0, + 1405.0, + 1230.0, + 1405.0, + 1265.0, + 293.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1262.0, + 1405.0, + 1262.0, + 1405.0, + 1296.0, + 294.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1290.0, + 1409.0, + 1290.0, + 1409.0, + 1327.0, + 294.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1637.0, + 959.0, + 1637.0, + 959.0, + 1669.0, + 295.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1668.0, + 871.0, + 1668.0, + 871.0, + 1702.0, + 295.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1668.0, + 959.0, + 1668.0, + 959.0, + 1702.0, + 906.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1695.0, + 297.0, + 1695.0, + 297.0, + 1735.0, + 294.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1695.0, + 962.0, + 1695.0, + 962.0, + 1735.0, + 330.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1728.0, + 960.0, + 1728.0, + 960.0, + 1763.0, + 294.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1757.0, + 297.0, + 1757.0, + 297.0, + 1794.0, + 294.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1757.0, + 961.0, + 1757.0, + 961.0, + 1794.0, + 330.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 962.0, + 1790.0, + 962.0, + 1823.0, + 294.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1820.0, + 961.0, + 1820.0, + 961.0, + 1854.0, + 293.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 960.0, + 1849.0, + 960.0, + 1884.0, + 293.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 961.0, + 1881.0, + 961.0, + 1914.0, + 294.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1908.0, + 961.0, + 1908.0, + 961.0, + 1948.0, + 294.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1943.0, + 1408.0, + 1943.0, + 1408.0, + 1976.0, + 297.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1407.0, + 1972.0, + 1407.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1403.0, + 2001.0, + 1403.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1418, + 1404, + 1418, + 1404, + 1633, + 298, + 1633 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1126, + 1404, + 1126, + 1404, + 1403, + 298, + 1403 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1711, + 1404, + 1711, + 1404, + 1928, + 297, + 1928 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 896, + 1404, + 896, + 1404, + 1113, + 297, + 1113 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 789, + 1403, + 789, + 1403, + 882, + 298, + 882 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1403, + 1942, + 1403, + 2035, + 300, + 2035 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 298, + 221, + 1400, + 221, + 1400, + 539, + 298, + 539 + ], + "score": 0.969 + }, + { + "category_id": 4, + "poly": [ + 296, + 548, + 1405, + 548, + 1405, + 689, + 296, + 689 + ], + "score": 0.965 + }, + { + "category_id": 0, + "poly": [ + 300, + 1663, + 674, + 1663, + 674, + 1695, + 300, + 1695 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 299, + 741, + 509, + 741, + 509, + 772, + 299, + 772 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.791 + }, + { + "category_id": 13, + "poly": [ + 354, + 1865, + 472, + 1865, + 472, + 1898, + 354, + 1898 + ], + "score": 0.95, + "latex": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { t r } )" + }, + { + "category_id": 13, + "poly": [ + 938, + 1833, + 1068, + 1833, + 1068, + 1868, + 938, + 1868 + ], + "score": 0.93, + "latex": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1340, + 406, + 1340, + 406, + 1374, + 298, + 1374 + ], + "score": 0.92, + "latex": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1050, + 406, + 1050, + 406, + 1083, + 298, + 1083 + ], + "score": 0.92, + "latex": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 538, + 1053, + 646, + 1053, + 646, + 1085, + 538, + 1085 + ], + "score": 0.92, + "latex": "\\hat { \\rho } ( S _ { r } , S _ { g } )" + }, + { + "category_id": 13, + "poly": [ + 714, + 1976, + 754, + 1976, + 754, + 2006, + 714, + 2006 + ], + "score": 0.91, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 675, + 576, + 782, + 576, + 782, + 607, + 675, + 607 + ], + "score": 0.91, + "latex": "\\hat { \\rho } ( S _ { r } , S _ { r } ^ { \\prime } ) )" + }, + { + "category_id": 13, + "poly": [ + 991, + 576, + 1097, + 576, + 1097, + 607, + 991, + 607 + ], + "score": 0.91, + "latex": "\\hat { \\rho } ( \\bar { S } _ { r } , S _ { g } ) \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1050, + 1742, + 1091, + 1742, + 1091, + 1776, + 1050, + 1776 + ], + "score": 0.9, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 806, + 1975, + 858, + 1975, + 858, + 2006, + 806, + 2006 + ], + "score": 0.9, + "latex": "S _ { r } ^ { v a l }" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1971, + 1404, + 1971, + 1404, + 2007, + 1127, + 2007 + ], + "score": 0.9, + "latex": "\\hat { \\rho } ( S _ { g } , S _ { r } ^ { v a l } ) - \\hat { \\rho } ( \\mathbf { \\dot { } { } } S _ { g } , S _ { r } ^ { t r } )" + }, + { + "category_id": 13, + "poly": [ + 903, + 1774, + 944, + 1774, + 944, + 1807, + 903, + 1807 + ], + "score": 0.9, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 510, + 1020, + 774, + 1020, + 774, + 1053, + 510, + 1053 + ], + "score": 0.9, + "latex": "| S _ { r } | = | S _ { r } ^ { \\prime } | = | S _ { g } | = n" + }, + { + "category_id": 13, + "poly": [ + 819, + 1834, + 871, + 1834, + 871, + 1867, + 819, + 1867 + ], + "score": 0.89, + "latex": "S _ { r } ^ { v a l }" + }, + { + "category_id": 13, + "poly": [ + 813, + 1946, + 842, + 1946, + 842, + 1975, + 813, + 1975 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 733, + 1190, + 764, + 1190, + 764, + 1222, + 733, + 1222 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1052, + 2004, + 1082, + 2004, + 1082, + 2036, + 1052, + 2036 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 698, + 1081, + 728, + 1081, + 728, + 1114, + 698, + 1114 + ], + "score": 0.89, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 544, + 1866, + 577, + 1866, + 577, + 1898, + 544, + 1898 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1069, + 990, + 1100, + 990, + 1100, + 1022, + 1069, + 1022 + ], + "score": 0.88, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 434, + 1743, + 468, + 1743, + 468, + 1777, + 434, + 1777 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 781, + 1081, + 813, + 1081, + 813, + 1115, + 781, + 1115 + ], + "score": 0.88, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1261, + 959, + 1292, + 959, + 1292, + 989, + 1261, + 989 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 813, + 1867, + 853, + 1867, + 853, + 1898, + 813, + 1898 + ], + "score": 0.88, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 775, + 960, + 806, + 960, + 806, + 992, + 775, + 992 + ], + "score": 0.88, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1134, + 2005, + 1175, + 2005, + 1175, + 2036, + 1134, + 2036 + ], + "score": 0.87, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 1295, + 1941, + 1335, + 1941, + 1335, + 1973, + 1295, + 1973 + ], + "score": 0.87, + "latex": "S _ { r } ^ { t r }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1189, + 330, + 1189, + 330, + 1222, + 298, + 1222 + ], + "score": 0.87, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1775, + 330, + 1775, + 330, + 1807, + 298, + 1807 + ], + "score": 0.87, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1367, + 929, + 1401, + 929, + 1401, + 963, + 1367, + 963 + ], + "score": 0.86, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1175, + 1776, + 1194, + 1776, + 1194, + 1806, + 1175, + 1806 + ], + "score": 0.85, + "latex": "\\hat { \\rho }" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1024, + 1148, + 1024, + 1148, + 1052, + 1129, + 1052 + ], + "score": 0.83, + "latex": "\\rho" + }, + { + "category_id": 13, + "poly": [ + 721, + 1779, + 741, + 1779, + 741, + 1801, + 721, + 1801 + ], + "score": 0.77, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 909, + 1059, + 927, + 1059, + 927, + 1077, + 909, + 1077 + ], + "score": 0.73, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1381, + 1056, + 1401, + 1056, + 1401, + 1077, + 1381, + 1077 + ], + "score": 0.72, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 955, + 1134, + 974, + 1134, + 974, + 1155, + 955, + 1155 + ], + "score": 0.71, + "latex": "n" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 225.0, + 557.0, + 225.0, + 557.0, + 249.0, + 386.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 220.0, + 825.0, + 220.0, + 825.0, + 252.0, + 649.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 220.0, + 1085.0, + 220.0, + 1085.0, + 252.0, + 921.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 220.0, + 1351.0, + 220.0, + 1351.0, + 252.0, + 1187.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 268.0, + 352.0, + 268.0, + 352.0, + 296.0, + 316.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 300.0, + 324.0, + 300.0, + 324.0, + 355.0, + 299.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 309.0, + 351.0, + 309.0, + 351.0, + 341.0, + 313.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 355.0, + 351.0, + 355.0, + 351.0, + 384.0, + 314.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 402.0, + 485.0, + 402.0, + 485.0, + 439.0, + 446.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 404.0, + 634.0, + 404.0, + 634.0, + 437.0, + 561.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 401.0, + 900.0, + 401.0, + 900.0, + 438.0, + 825.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 402.0, + 1017.0, + 402.0, + 1017.0, + 439.0, + 978.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 401.0, + 1167.0, + 401.0, + 1167.0, + 438.0, + 1091.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 402.0, + 1400.0, + 402.0, + 1400.0, + 439.0, + 1361.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 430.0, + 528.0, + 430.0, + 528.0, + 453.0, + 413.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 427.0, + 795.0, + 427.0, + 795.0, + 456.0, + 680.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 430.0, + 1061.0, + 430.0, + 1061.0, + 454.0, + 947.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 430.0, + 1327.0, + 430.0, + 1327.0, + 453.0, + 1212.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 474.0, + 522.0, + 474.0, + 522.0, + 495.0, + 344.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 474.0, + 808.0, + 474.0, + 808.0, + 495.0, + 631.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 473.0, + 998.0, + 473.0, + 998.0, + 494.0, + 919.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 473.0, + 1263.0, + 473.0, + 1263.0, + 494.0, + 1185.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 490.0, + 575.0, + 490.0, + 575.0, + 517.0, + 342.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 490.0, + 859.0, + 490.0, + 859.0, + 517.0, + 629.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 487.0, + 1058.0, + 487.0, + 1058.0, + 516.0, + 917.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 487.0, + 1324.0, + 487.0, + 1324.0, + 516.0, + 1182.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 512.0, + 560.0, + 512.0, + 560.0, + 537.0, + 343.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 509.0, + 848.0, + 509.0, + 848.0, + 537.0, + 629.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 512.0, + 1130.0, + 512.0, + 1130.0, + 538.0, + 918.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 512.0, + 1395.0, + 512.0, + 1395.0, + 539.0, + 1184.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 406.0, + 374.0, + 406.0, + 374.0, + 436.5, + 322.0, + 436.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 405.0, + 756.0, + 405.0, + 756.0, + 437.0, + 706.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 405.0, + 1288.0, + 405.0, + 1288.0, + 436.5, + 1239.0, + 436.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 548.0, + 1406.0, + 548.0, + 1406.0, + 580.0, + 295.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 575.0, + 674.0, + 575.0, + 674.0, + 610.0, + 294.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 575.0, + 990.0, + 575.0, + 990.0, + 610.0, + 783.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 575.0, + 1405.0, + 575.0, + 1405.0, + 610.0, + 1098.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 600.0, + 1408.0, + 600.0, + 1408.0, + 637.0, + 293.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 630.0, + 1405.0, + 630.0, + 1405.0, + 665.0, + 294.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 657.0, + 1184.0, + 657.0, + 1184.0, + 694.0, + 293.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1663.0, + 678.0, + 1663.0, + 678.0, + 1699.0, + 294.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 737.0, + 512.0, + 737.0, + 512.0, + 779.0, + 293.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1418.0, + 1408.0, + 1418.0, + 1408.0, + 1453.0, + 297.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1449.0, + 1404.0, + 1449.0, + 1404.0, + 1484.0, + 294.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1477.0, + 1406.0, + 1477.0, + 1406.0, + 1517.0, + 292.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1510.0, + 1408.0, + 1510.0, + 1408.0, + 1544.0, + 294.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1539.0, + 1407.0, + 1539.0, + 1407.0, + 1576.0, + 292.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1571.0, + 1406.0, + 1571.0, + 1406.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1601.0, + 1286.0, + 1601.0, + 1286.0, + 1635.0, + 294.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1128.0, + 954.0, + 1128.0, + 954.0, + 1161.0, + 296.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1128.0, + 1404.0, + 1128.0, + 1404.0, + 1161.0, + 975.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1155.0, + 1405.0, + 1155.0, + 1405.0, + 1194.0, + 292.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1187.0, + 297.0, + 1187.0, + 297.0, + 1224.0, + 294.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1187.0, + 732.0, + 1187.0, + 732.0, + 1224.0, + 331.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1187.0, + 1406.0, + 1187.0, + 1406.0, + 1224.0, + 765.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1217.0, + 1405.0, + 1217.0, + 1405.0, + 1254.0, + 294.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1249.0, + 1406.0, + 1249.0, + 1406.0, + 1285.0, + 294.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1276.0, + 1404.0, + 1276.0, + 1404.0, + 1316.0, + 293.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1309.0, + 1405.0, + 1309.0, + 1405.0, + 1346.0, + 294.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1338.0, + 297.0, + 1338.0, + 297.0, + 1380.0, + 293.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1338.0, + 1405.0, + 1338.0, + 1405.0, + 1380.0, + 407.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1371.0, + 843.0, + 1371.0, + 843.0, + 1407.0, + 293.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1750.0, + 294.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1738.0, + 433.0, + 1738.0, + 433.0, + 1780.0, + 292.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1738.0, + 1049.0, + 1738.0, + 1049.0, + 1780.0, + 469.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1738.0, + 1406.0, + 1738.0, + 1406.0, + 1780.0, + 1092.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1766.0, + 297.0, + 1766.0, + 297.0, + 1813.0, + 291.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1766.0, + 720.0, + 1766.0, + 720.0, + 1813.0, + 331.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1766.0, + 902.0, + 1766.0, + 902.0, + 1813.0, + 742.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1766.0, + 1174.0, + 1766.0, + 1174.0, + 1813.0, + 945.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1766.0, + 1409.0, + 1766.0, + 1409.0, + 1813.0, + 1195.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1801.0, + 1405.0, + 1801.0, + 1405.0, + 1838.0, + 294.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1819.0, + 818.0, + 1819.0, + 818.0, + 1879.0, + 287.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1819.0, + 937.0, + 1819.0, + 937.0, + 1879.0, + 872.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1819.0, + 1413.0, + 1819.0, + 1413.0, + 1879.0, + 1069.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1859.0, + 353.0, + 1859.0, + 353.0, + 1901.0, + 291.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1859.0, + 543.0, + 1859.0, + 543.0, + 1901.0, + 473.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1859.0, + 812.0, + 1859.0, + 812.0, + 1901.0, + 578.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1859.0, + 1407.0, + 1859.0, + 1407.0, + 1901.0, + 854.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1889.0, + 765.0, + 1889.0, + 765.0, + 1932.0, + 292.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 898.0, + 1404.0, + 898.0, + 1404.0, + 933.0, + 295.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 922.0, + 1366.0, + 922.0, + 1366.0, + 970.0, + 291.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 922.0, + 1405.0, + 922.0, + 1405.0, + 970.0, + 1402.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 956.0, + 774.0, + 956.0, + 774.0, + 993.0, + 294.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 956.0, + 1260.0, + 956.0, + 1260.0, + 993.0, + 807.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 956.0, + 1405.0, + 956.0, + 1405.0, + 993.0, + 1293.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 987.0, + 1068.0, + 987.0, + 1068.0, + 1026.0, + 294.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 987.0, + 1405.0, + 987.0, + 1405.0, + 1026.0, + 1101.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1018.0, + 509.0, + 1018.0, + 509.0, + 1057.0, + 294.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1018.0, + 1128.0, + 1018.0, + 1128.0, + 1057.0, + 775.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1018.0, + 1405.0, + 1018.0, + 1405.0, + 1057.0, + 1149.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1049.0, + 537.0, + 1049.0, + 537.0, + 1087.0, + 407.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1049.0, + 908.0, + 1049.0, + 908.0, + 1087.0, + 647.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1049.0, + 1380.0, + 1049.0, + 1380.0, + 1087.0, + 928.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1049.0, + 1407.0, + 1049.0, + 1407.0, + 1087.0, + 1402.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1080.0, + 697.0, + 1080.0, + 697.0, + 1118.0, + 292.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1080.0, + 780.0, + 1080.0, + 780.0, + 1118.0, + 729.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 1080.0, + 1267.0, + 1080.0, + 1267.0, + 1118.0, + 814.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 790.0, + 1403.0, + 790.0, + 1403.0, + 823.0, + 294.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 821.0, + 1405.0, + 821.0, + 1405.0, + 855.0, + 294.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 852.0, + 1040.0, + 852.0, + 1040.0, + 886.0, + 296.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1938.0, + 812.0, + 1938.0, + 812.0, + 1978.0, + 293.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1938.0, + 1294.0, + 1938.0, + 1294.0, + 1978.0, + 843.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1938.0, + 1409.0, + 1938.0, + 1409.0, + 1978.0, + 1336.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1966.0, + 713.0, + 1966.0, + 713.0, + 2011.0, + 291.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1966.0, + 805.0, + 1966.0, + 805.0, + 2011.0, + 755.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1966.0, + 1126.0, + 1966.0, + 1126.0, + 2011.0, + 859.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1966.0, + 1409.0, + 1966.0, + 1409.0, + 2011.0, + 1405.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1996.0, + 1051.0, + 1996.0, + 1051.0, + 2040.0, + 293.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1996.0, + 1133.0, + 1996.0, + 1133.0, + 2040.0, + 1083.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1996.0, + 1408.0, + 1996.0, + 1408.0, + 2040.0, + 1176.0, + 2040.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 655, + 1405, + 655, + 1405, + 904, + 297, + 904 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1175, + 1403, + 1175, + 1403, + 1390, + 298, + 1390 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1819, + 1404, + 1819, + 1404, + 2034, + 297, + 2034 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1543, + 1403, + 1543, + 1403, + 1697, + 298, + 1697 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1005, + 1403, + 1005, + 1403, + 1160, + 298, + 1160 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1405, + 1404, + 1405, + 1404, + 1527, + 299, + 1527 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 300, + 220, + 1401, + 220, + 1401, + 509, + 300, + 509 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 300, + 1712, + 1397, + 1712, + 1397, + 1804, + 300, + 1804 + ], + "score": 0.963 + }, + { + "category_id": 4, + "poly": [ + 298, + 516, + 1406, + 516, + 1406, + 627, + 298, + 627 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 302, + 944, + 816, + 944, + 816, + 979, + 302, + 979 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.794 + }, + { + "category_id": 13, + "poly": [ + 754, + 656, + 883, + 656, + 883, + 692, + 754, + 692 + ], + "score": 0.93, + "latex": "\\hat { \\rho } ( S _ { r } ^ { \\prime } , S _ { r } ^ { v a l } )" + }, + { + "category_id": 13, + "poly": [ + 459, + 1633, + 533, + 1633, + 533, + 1668, + 459, + 1668 + ], + "score": 0.92, + "latex": "O ( n ^ { 3 } )" + }, + { + "category_id": 13, + "poly": [ + 1009, + 750, + 1040, + 750, + 1040, + 783, + 1009, + 783 + ], + "score": 0.9, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 408, + 1237, + 443, + 1237, + 443, + 1270, + 408, + 1270 + ], + "score": 0.89, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 511, + 1911, + 567, + 1911, + 567, + 1942, + 511, + 1942 + ], + "score": 0.89, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 680, + 720, + 712, + 720, + 712, + 750, + 680, + 750 + ], + "score": 0.89, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 596, + 719, + 628, + 719, + 628, + 752, + 596, + 752 + ], + "score": 0.88, + "latex": "S _ { r } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 1237, + 1146, + 1237, + 1146, + 1267, + 1113, + 1267 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 679, + 1329, + 711, + 1329, + 711, + 1358, + 679, + 1358 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 1229, + 750, + 1260, + 750, + 1260, + 780, + 1229, + 780 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 222.0, + 559.0, + 222.0, + 559.0, + 253.0, + 373.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 221.0, + 828.0, + 221.0, + 828.0, + 254.0, + 639.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 222.0, + 1088.0, + 222.0, + 1088.0, + 253.0, + 912.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 221.0, + 1357.0, + 221.0, + 1357.0, + 254.0, + 1177.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 247.0, + 355.0, + 247.0, + 355.0, + 275.0, + 318.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 292.0, + 361.0, + 292.0, + 361.0, + 351.0, + 293.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 364.0, + 356.0, + 364.0, + 356.0, + 396.0, + 315.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 391.0, + 373.0, + 391.0, + 373.0, + 419.0, + 333.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 391.0, + 430.0, + 391.0, + 430.0, + 418.0, + 389.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 391.0, + 488.0, + 391.0, + 488.0, + 418.0, + 448.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 391.0, + 545.0, + 391.0, + 545.0, + 418.0, + 504.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 391.0, + 639.0, + 391.0, + 639.0, + 419.0, + 563.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 391.0, + 699.0, + 391.0, + 699.0, + 418.0, + 655.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 391.0, + 754.0, + 391.0, + 754.0, + 418.0, + 714.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 391.0, + 812.0, + 391.0, + 812.0, + 418.0, + 772.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 390.0, + 907.0, + 390.0, + 907.0, + 419.0, + 830.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 391.0, + 964.0, + 391.0, + 964.0, + 418.0, + 923.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 391.0, + 1022.0, + 391.0, + 1022.0, + 418.0, + 981.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 391.0, + 1078.0, + 391.0, + 1078.0, + 418.0, + 1037.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 390.0, + 1175.0, + 390.0, + 1175.0, + 419.0, + 1097.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 391.0, + 1232.0, + 391.0, + 1232.0, + 418.0, + 1191.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 391.0, + 1288.0, + 391.0, + 1288.0, + 418.0, + 1248.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 391.0, + 1346.0, + 391.0, + 1346.0, + 418.0, + 1305.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 391.0, + 1402.0, + 391.0, + 1402.0, + 418.0, + 1365.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 409.0, + 553.0, + 409.0, + 553.0, + 440.0, + 381.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 409.0, + 820.0, + 409.0, + 820.0, + 440.0, + 648.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 412.0, + 1085.0, + 412.0, + 1085.0, + 437.0, + 915.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 409.0, + 1355.0, + 409.0, + 1355.0, + 440.0, + 1183.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 452.0, + 604.0, + 452.0, + 604.0, + 482.0, + 358.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 451.0, + 853.0, + 451.0, + 853.0, + 481.0, + 673.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 449.0, + 1098.0, + 449.0, + 1098.0, + 484.0, + 922.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 451.0, + 1300.0, + 451.0, + 1300.0, + 483.0, + 1170.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 479.0, + 585.0, + 479.0, + 585.0, + 506.0, + 358.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 476.0, + 729.0, + 476.0, + 729.0, + 506.0, + 670.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 475.0, + 1102.0, + 475.0, + 1102.0, + 508.0, + 921.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 477.0, + 1391.0, + 477.0, + 1391.0, + 511.0, + 1167.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 515.0, + 1403.0, + 515.0, + 1403.0, + 548.0, + 296.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 540.0, + 1408.0, + 540.0, + 1408.0, + 579.0, + 292.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 570.0, + 1406.0, + 570.0, + 1406.0, + 603.0, + 295.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 599.0, + 613.0, + 599.0, + 613.0, + 632.0, + 295.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 940.0, + 822.0, + 940.0, + 822.0, + 987.0, + 293.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 652.0, + 753.0, + 652.0, + 753.0, + 696.0, + 291.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 652.0, + 1407.0, + 652.0, + 1407.0, + 696.0, + 884.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 689.0, + 1406.0, + 689.0, + 1406.0, + 723.0, + 294.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 718.0, + 595.0, + 718.0, + 595.0, + 756.0, + 294.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 718.0, + 679.0, + 718.0, + 679.0, + 756.0, + 629.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 718.0, + 1405.0, + 718.0, + 1405.0, + 756.0, + 713.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 748.0, + 1008.0, + 748.0, + 1008.0, + 786.0, + 294.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 748.0, + 1228.0, + 748.0, + 1228.0, + 786.0, + 1041.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 748.0, + 1406.0, + 748.0, + 1406.0, + 786.0, + 1261.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 780.0, + 1408.0, + 780.0, + 1408.0, + 813.0, + 294.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 809.0, + 1406.0, + 809.0, + 1406.0, + 845.0, + 294.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 840.0, + 1407.0, + 840.0, + 1407.0, + 875.0, + 294.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 873.0, + 1408.0, + 873.0, + 1408.0, + 907.0, + 295.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1173.0, + 1405.0, + 1173.0, + 1405.0, + 1210.0, + 293.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1205.0, + 1405.0, + 1205.0, + 1405.0, + 1243.0, + 293.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1236.0, + 407.0, + 1236.0, + 407.0, + 1271.0, + 292.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1236.0, + 1112.0, + 1236.0, + 1112.0, + 1271.0, + 444.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1236.0, + 1405.0, + 1236.0, + 1405.0, + 1271.0, + 1147.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1268.0, + 1404.0, + 1268.0, + 1404.0, + 1301.0, + 293.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1297.0, + 1405.0, + 1297.0, + 1405.0, + 1332.0, + 294.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1328.0, + 678.0, + 1328.0, + 678.0, + 1362.0, + 294.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1328.0, + 1405.0, + 1328.0, + 1405.0, + 1362.0, + 712.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1358.0, + 1277.0, + 1358.0, + 1277.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1855.0, + 292.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1886.0, + 295.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1407.0, + 1882.0, + 1407.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1911.0, + 510.0, + 1911.0, + 510.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1911.0, + 1407.0, + 1911.0, + 1407.0, + 1947.0, + 568.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1970.0, + 1407.0, + 1970.0, + 1407.0, + 2009.0, + 291.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 1405.0, + 2000.0, + 1405.0, + 2040.0, + 292.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1545.0, + 1409.0, + 1545.0, + 1409.0, + 1578.0, + 297.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1572.0, + 1407.0, + 1572.0, + 1407.0, + 1609.0, + 294.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1605.0, + 1404.0, + 1605.0, + 1404.0, + 1639.0, + 294.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1632.0, + 458.0, + 1632.0, + 458.0, + 1672.0, + 292.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1632.0, + 1408.0, + 1632.0, + 1408.0, + 1672.0, + 534.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1664.0, + 1192.0, + 1664.0, + 1192.0, + 1702.0, + 293.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1007.0, + 1404.0, + 1007.0, + 1404.0, + 1040.0, + 296.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1035.0, + 1404.0, + 1035.0, + 1404.0, + 1072.0, + 294.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1067.0, + 1404.0, + 1067.0, + 1404.0, + 1102.0, + 294.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1098.0, + 1404.0, + 1098.0, + 1404.0, + 1132.0, + 296.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1128.0, + 763.0, + 1128.0, + 763.0, + 1165.0, + 294.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1405.0, + 1408.0, + 1405.0, + 1408.0, + 1441.0, + 295.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1435.0, + 1404.0, + 1435.0, + 1404.0, + 1471.0, + 293.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1466.0, + 1406.0, + 1466.0, + 1406.0, + 1505.0, + 293.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1498.0, + 642.0, + 1498.0, + 642.0, + 1531.0, + 295.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1712.0, + 1403.0, + 1712.0, + 1403.0, + 1746.0, + 295.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1744.0, + 1403.0, + 1744.0, + 1403.0, + 1776.0, + 295.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1775.0, + 693.0, + 1775.0, + 693.0, + 1805.0, + 296.0, + 1805.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 399, + 1404, + 399, + 1404, + 613, + 297, + 613 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 230, + 1403, + 230, + 1403, + 383, + 299, + 383 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 629, + 1404, + 629, + 1404, + 782, + 299, + 782 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 857, + 76, + 857, + 104, + 298, + 104 + ], + "score": 0.878 + }, + { + "category_id": 0, + "poly": [ + 299, + 831, + 488, + 831, + 488, + 863, + 299, + 863 + ], + "score": 0.843 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 298, + 882, + 1401, + 882, + 1401, + 944, + 298, + 944 + ], + "score": 0.558 + }, + { + "category_id": 1, + "poly": [ + 299, + 1971, + 1401, + 1971, + 1401, + 2033, + 299, + 2033 + ], + "score": 0.536 + }, + { + "category_id": 1, + "poly": [ + 298, + 1452, + 1404, + 1452, + 1404, + 1547, + 298, + 1547 + ], + "score": 0.51 + }, + { + "category_id": 1, + "poly": [ + 297, + 965, + 1400, + 965, + 1400, + 1030, + 297, + 1030 + ], + "score": 0.505 + }, + { + "category_id": 1, + "poly": [ + 299, + 1568, + 1404, + 1568, + 1404, + 1662, + 299, + 1662 + ], + "score": 0.502 + }, + { + "category_id": 1, + "poly": [ + 301, + 1051, + 1399, + 1051, + 1399, + 1144, + 301, + 1144 + ], + "score": 0.495 + }, + { + "category_id": 1, + "poly": [ + 298, + 1368, + 1403, + 1368, + 1403, + 1432, + 298, + 1432 + ], + "score": 0.472 + }, + { + "category_id": 1, + "poly": [ + 290, + 1684, + 1404, + 1684, + 1404, + 1749, + 290, + 1749 + ], + "score": 0.459 + }, + { + "category_id": 1, + "poly": [ + 298, + 1250, + 1401, + 1250, + 1401, + 1347, + 298, + 1347 + ], + "score": 0.448 + }, + { + "category_id": 1, + "poly": [ + 291, + 1886, + 1402, + 1886, + 1402, + 1949, + 291, + 1949 + ], + "score": 0.439 + }, + { + "category_id": 1, + "poly": [ + 292, + 1166, + 1402, + 1166, + 1402, + 1231, + 292, + 1231 + ], + "score": 0.426 + }, + { + "category_id": 1, + "poly": [ + 301, + 1770, + 1403, + 1770, + 1403, + 1864, + 301, + 1864 + ], + "score": 0.408 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 828.0, + 491.0, + 828.0, + 491.0, + 868.0, + 296.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 399.0, + 1404.0, + 399.0, + 1404.0, + 434.0, + 296.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 429.0, + 1404.0, + 429.0, + 1404.0, + 463.0, + 294.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 457.0, + 1409.0, + 457.0, + 1409.0, + 499.0, + 291.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 489.0, + 1406.0, + 489.0, + 1406.0, + 526.0, + 294.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 523.0, + 1404.0, + 523.0, + 1404.0, + 554.0, + 296.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 550.0, + 1404.0, + 550.0, + 1404.0, + 587.0, + 294.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 581.0, + 432.0, + 581.0, + 432.0, + 620.0, + 294.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 231.0, + 1403.0, + 231.0, + 1403.0, + 264.0, + 297.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 258.0, + 1404.0, + 258.0, + 1404.0, + 299.0, + 294.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 291.0, + 1404.0, + 291.0, + 1404.0, + 326.0, + 293.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 318.0, + 1405.0, + 318.0, + 1405.0, + 358.0, + 294.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 351.0, + 695.0, + 351.0, + 695.0, + 388.0, + 295.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 626.0, + 1404.0, + 626.0, + 1404.0, + 667.0, + 293.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 657.0, + 1405.0, + 657.0, + 1405.0, + 696.0, + 293.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 684.0, + 1405.0, + 684.0, + 1405.0, + 729.0, + 292.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 720.0, + 1404.0, + 720.0, + 1404.0, + 757.0, + 295.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 749.0, + 1357.0, + 749.0, + 1357.0, + 786.0, + 295.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 878.0, + 1406.0, + 878.0, + 1406.0, + 918.0, + 292.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 912.0, + 570.0, + 912.0, + 570.0, + 942.0, + 323.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1969.0, + 1405.0, + 1969.0, + 1405.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2002.0, + 541.0, + 2002.0, + 541.0, + 2035.0, + 323.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1451.0, + 1409.0, + 1451.0, + 1409.0, + 1491.0, + 293.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1485.0, + 1406.0, + 1485.0, + 1406.0, + 1519.0, + 323.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1515.0, + 848.0, + 1515.0, + 848.0, + 1549.0, + 323.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 967.0, + 1402.0, + 967.0, + 1402.0, + 1003.0, + 296.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 998.0, + 1192.0, + 998.0, + 1192.0, + 1031.0, + 321.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1568.0, + 1409.0, + 1568.0, + 1409.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1596.0, + 1409.0, + 1596.0, + 1409.0, + 1639.0, + 319.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1631.0, + 481.0, + 1631.0, + 481.0, + 1662.0, + 321.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1052.0, + 1403.0, + 1052.0, + 1403.0, + 1085.0, + 297.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1084.0, + 1404.0, + 1084.0, + 1404.0, + 1118.0, + 323.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1114.0, + 610.0, + 1114.0, + 610.0, + 1143.0, + 322.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1367.0, + 1408.0, + 1367.0, + 1408.0, + 1406.0, + 293.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1400.0, + 657.0, + 1400.0, + 657.0, + 1432.0, + 322.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1684.0, + 1405.0, + 1684.0, + 1405.0, + 1722.0, + 294.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1715.0, + 1100.0, + 1715.0, + 1100.0, + 1750.0, + 322.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1253.0, + 1403.0, + 1253.0, + 1403.0, + 1287.0, + 296.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1284.0, + 1405.0, + 1284.0, + 1405.0, + 1318.0, + 324.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1315.0, + 861.0, + 1315.0, + 861.0, + 1349.0, + 324.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 1403.0, + 1884.0, + 1403.0, + 1921.0, + 294.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1916.0, + 1152.0, + 1916.0, + 1152.0, + 1951.0, + 322.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1168.0, + 1404.0, + 1168.0, + 1404.0, + 1204.0, + 295.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1197.0, + 1022.0, + 1197.0, + 1022.0, + 1232.0, + 321.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1770.0, + 1405.0, + 1770.0, + 1405.0, + 1805.0, + 295.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1801.0, + 1405.0, + 1801.0, + 1405.0, + 1835.0, + 324.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1832.0, + 777.0, + 1832.0, + 777.0, + 1865.0, + 322.0, + 1865.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 862, + 2088, + 862, + 2113, + 835, + 2113 + ], + "score": 0.832 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1404, + 229, + 1404, + 324, + 299, + 324 + ], + "score": 0.7 + }, + { + "category_id": 1, + "poly": [ + 293, + 757, + 1402, + 757, + 1402, + 822, + 293, + 822 + ], + "score": 0.667 + }, + { + "category_id": 1, + "poly": [ + 294, + 674, + 1402, + 674, + 1402, + 739, + 294, + 739 + ], + "score": 0.649 + }, + { + "category_id": 1, + "poly": [ + 299, + 508, + 1403, + 508, + 1403, + 573, + 299, + 573 + ], + "score": 0.634 + }, + { + "category_id": 1, + "poly": [ + 293, + 591, + 1402, + 591, + 1402, + 656, + 293, + 656 + ], + "score": 0.63 + }, + { + "category_id": 1, + "poly": [ + 297, + 342, + 1403, + 342, + 1403, + 406, + 297, + 406 + ], + "score": 0.622 + }, + { + "category_id": 1, + "poly": [ + 298, + 840, + 1400, + 840, + 1400, + 905, + 298, + 905 + ], + "score": 0.613 + }, + { + "category_id": 1, + "poly": [ + 292, + 1573, + 1401, + 1573, + 1401, + 1639, + 292, + 1639 + ], + "score": 0.611 + }, + { + "category_id": 1, + "poly": [ + 292, + 1823, + 1402, + 1823, + 1402, + 1889, + 292, + 1889 + ], + "score": 0.595 + }, + { + "category_id": 1, + "poly": [ + 296, + 426, + 1403, + 426, + 1403, + 490, + 296, + 490 + ], + "score": 0.594 + }, + { + "category_id": 1, + "poly": [ + 292, + 1656, + 1401, + 1656, + 1401, + 1722, + 292, + 1722 + ], + "score": 0.582 + }, + { + "category_id": 1, + "poly": [ + 295, + 1036, + 1403, + 1036, + 1403, + 1132, + 295, + 1132 + ], + "score": 0.578 + }, + { + "category_id": 1, + "poly": [ + 298, + 923, + 1403, + 923, + 1403, + 1018, + 298, + 1018 + ], + "score": 0.578 + }, + { + "category_id": 1, + "poly": [ + 298, + 1150, + 1403, + 1150, + 1403, + 1244, + 298, + 1244 + ], + "score": 0.57 + }, + { + "category_id": 1, + "poly": [ + 293, + 1740, + 1403, + 1740, + 1403, + 1806, + 293, + 1806 + ], + "score": 0.568 + }, + { + "category_id": 1, + "poly": [ + 299, + 1460, + 1403, + 1460, + 1403, + 1556, + 299, + 1556 + ], + "score": 0.554 + }, + { + "category_id": 1, + "poly": [ + 298, + 1263, + 1401, + 1263, + 1401, + 1360, + 298, + 1360 + ], + "score": 0.551 + }, + { + "category_id": 1, + "poly": [ + 297, + 1377, + 1402, + 1377, + 1402, + 1442, + 297, + 1442 + ], + "score": 0.536 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 1408.0, + 227.0, + 1408.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 262.0, + 1408.0, + 262.0, + 1408.0, + 296.0, + 323.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 286.0, + 399.0, + 286.0, + 399.0, + 327.0, + 321.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 755.0, + 1405.0, + 755.0, + 1405.0, + 796.0, + 293.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 788.0, + 612.0, + 788.0, + 612.0, + 819.0, + 322.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 673.0, + 1405.0, + 673.0, + 1405.0, + 713.0, + 294.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 707.0, + 610.0, + 707.0, + 610.0, + 737.0, + 322.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 508.0, + 1407.0, + 508.0, + 1407.0, + 544.0, + 295.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 539.0, + 519.0, + 539.0, + 519.0, + 573.0, + 323.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 593.0, + 1404.0, + 593.0, + 1404.0, + 629.0, + 295.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 624.0, + 1235.0, + 624.0, + 1235.0, + 657.0, + 322.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 343.0, + 1407.0, + 343.0, + 1407.0, + 379.0, + 295.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 374.0, + 567.0, + 374.0, + 567.0, + 405.0, + 322.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 839.0, + 1404.0, + 839.0, + 1404.0, + 879.0, + 294.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 873.0, + 1300.0, + 873.0, + 1300.0, + 906.0, + 323.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1572.0, + 1406.0, + 1572.0, + 1406.0, + 1611.0, + 296.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1606.0, + 1160.0, + 1606.0, + 1160.0, + 1638.0, + 324.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1822.0, + 1405.0, + 1822.0, + 1405.0, + 1860.0, + 294.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1855.0, + 1274.0, + 1855.0, + 1274.0, + 1890.0, + 323.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 426.0, + 1404.0, + 426.0, + 1404.0, + 462.0, + 296.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 458.0, + 939.0, + 458.0, + 939.0, + 490.0, + 323.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1654.0, + 1406.0, + 1654.0, + 1406.0, + 1697.0, + 295.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1686.0, + 1321.0, + 1686.0, + 1321.0, + 1723.0, + 323.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1037.0, + 1406.0, + 1037.0, + 1406.0, + 1072.0, + 297.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1068.0, + 1405.0, + 1068.0, + 1405.0, + 1105.0, + 322.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1100.0, + 610.0, + 1100.0, + 610.0, + 1129.0, + 323.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 923.0, + 1407.0, + 923.0, + 1407.0, + 960.0, + 294.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 953.0, + 1408.0, + 953.0, + 1408.0, + 991.0, + 321.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 985.0, + 575.0, + 985.0, + 575.0, + 1019.0, + 319.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1148.0, + 1408.0, + 1148.0, + 1408.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1181.0, + 1408.0, + 1181.0, + 1408.0, + 1216.0, + 322.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1213.0, + 760.0, + 1213.0, + 760.0, + 1247.0, + 325.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1739.0, + 1406.0, + 1739.0, + 1406.0, + 1776.0, + 295.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1772.0, + 775.0, + 1772.0, + 775.0, + 1808.0, + 324.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1457.0, + 1409.0, + 1457.0, + 1409.0, + 1501.0, + 293.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1490.0, + 1405.0, + 1490.0, + 1405.0, + 1529.0, + 322.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1519.0, + 1040.0, + 1519.0, + 1040.0, + 1559.0, + 323.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1263.0, + 1405.0, + 1263.0, + 1405.0, + 1302.0, + 296.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1297.0, + 1405.0, + 1297.0, + 1405.0, + 1331.0, + 323.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1323.0, + 932.0, + 1323.0, + 932.0, + 1362.0, + 322.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1377.0, + 1404.0, + 1377.0, + 1404.0, + 1415.0, + 293.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1410.0, + 873.0, + 1410.0, + 873.0, + 1443.0, + 322.0, + 1443.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 952, + 1405, + 952, + 1405, + 1258, + 297, + 1258 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1849, + 1403, + 1849, + 1403, + 2034, + 297, + 2034 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 528, + 1403, + 528, + 1403, + 650, + 299, + 650 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 358, + 1403, + 358, + 1403, + 513, + 298, + 513 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 298, + 1285, + 1403, + 1285, + 1403, + 1708, + 298, + 1708 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 774, + 1404, + 774, + 1404, + 867, + 298, + 867 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 666, + 1397, + 666, + 1397, + 759, + 298, + 759 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 297, + 1734, + 1404, + 1734, + 1404, + 1819, + 297, + 1819 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 291, + 281, + 1401, + 281, + 1401, + 343, + 291, + 343 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 304, + 899, + 812, + 899, + 812, + 933, + 304, + 933 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.851 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 995, + 227, + 995, + 261, + 299, + 261 + ], + "score": 0.848 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 104, + 298, + 104 + ], + "score": 0.737 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.159 + }, + { + "category_id": 13, + "poly": [ + 686, + 452, + 781, + 452, + 781, + 481, + 686, + 481 + ], + "score": 0.9, + "latex": "6 4 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 591, + 590, + 623, + 590, + 623, + 623, + 591, + 623 + ], + "score": 0.89, + "latex": "S _ { g }" + }, + { + "category_id": 13, + "poly": [ + 1120, + 837, + 1154, + 837, + 1154, + 871, + 1120, + 871 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { g }" + }, + { + "category_id": 13, + "poly": [ + 783, + 837, + 815, + 837, + 815, + 867, + 783, + 867 + ], + "score": 0.88, + "latex": "\\mathbb { P } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 502, + 590, + 535, + 590, + 535, + 620, + 502, + 620 + ], + "score": 0.88, + "latex": "S _ { r }" + }, + { + "category_id": 13, + "poly": [ + 298, + 835, + 330, + 835, + 330, + 868, + 298, + 868 + ], + "score": 0.87, + "latex": "\\chi ^ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1283.0, + 518.0, + 1283.0, + 518.0, + 1312.0, + 425.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 1282.0, + 763.0, + 1282.0, + 763.0, + 1313.0, + 709.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1283.0, + 1053.0, + 1283.0, + 1053.0, + 1312.0, + 946.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1284.0, + 1345.0, + 1284.0, + 1345.0, + 1312.0, + 1179.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1314.0, + 327.0, + 1314.0, + 327.0, + 1440.0, + 298.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1317.0, + 350.0, + 1317.0, + 350.0, + 1341.0, + 317.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1354.0, + 351.0, + 1354.0, + 351.0, + 1383.0, + 316.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1366.0, + 974.0, + 1366.0, + 974.0, + 1375.0, + 953.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1396.0, + 351.0, + 1396.0, + 351.0, + 1424.0, + 316.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1437.0, + 369.0, + 1437.0, + 369.0, + 1465.0, + 332.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1439.0, + 429.0, + 1439.0, + 429.0, + 1463.0, + 394.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1440.0, + 490.0, + 1440.0, + 490.0, + 1463.0, + 455.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1440.0, + 551.0, + 1440.0, + 551.0, + 1463.0, + 517.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1439.0, + 632.0, + 1439.0, + 632.0, + 1464.0, + 578.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1441.0, + 690.0, + 1441.0, + 690.0, + 1460.0, + 660.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1440.0, + 753.0, + 1440.0, + 753.0, + 1463.0, + 719.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1440.0, + 814.0, + 1440.0, + 814.0, + 1463.0, + 779.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 1439.0, + 893.0, + 1439.0, + 893.0, + 1464.0, + 842.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1439.0, + 953.0, + 1439.0, + 953.0, + 1460.0, + 919.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1440.0, + 1016.0, + 1440.0, + 1016.0, + 1463.0, + 981.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1440.0, + 1078.0, + 1440.0, + 1078.0, + 1463.0, + 1043.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1439.0, + 1158.0, + 1439.0, + 1158.0, + 1464.0, + 1105.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 1441.0, + 1217.0, + 1441.0, + 1217.0, + 1460.0, + 1186.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1440.0, + 1280.0, + 1440.0, + 1280.0, + 1463.0, + 1245.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1441.0, + 1339.0, + 1441.0, + 1339.0, + 1460.0, + 1307.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 1440.0, + 1402.0, + 1440.0, + 1402.0, + 1463.0, + 1369.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1465.0, + 351.0, + 1465.0, + 351.0, + 1489.0, + 317.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1466.0, + 328.0, + 1466.0, + 328.0, + 1586.0, + 296.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1504.0, + 352.0, + 1504.0, + 352.0, + 1532.0, + 316.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1513.0, + 720.0, + 1513.0, + 720.0, + 1522.0, + 695.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1544.0, + 352.0, + 1544.0, + 352.0, + 1573.0, + 316.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1587.0, + 370.0, + 1587.0, + 370.0, + 1615.0, + 331.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1587.0, + 431.0, + 1587.0, + 431.0, + 1615.0, + 392.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1587.0, + 492.0, + 1587.0, + 492.0, + 1615.0, + 453.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1588.0, + 551.0, + 1588.0, + 551.0, + 1612.0, + 515.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1588.0, + 632.0, + 1588.0, + 632.0, + 1613.0, + 577.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1587.0, + 694.0, + 1587.0, + 694.0, + 1615.0, + 655.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1586.0, + 755.0, + 1586.0, + 755.0, + 1615.0, + 716.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1587.0, + 817.0, + 1587.0, + 817.0, + 1615.0, + 778.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 1588.0, + 894.0, + 1588.0, + 894.0, + 1613.0, + 840.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1588.0, + 956.0, + 1588.0, + 956.0, + 1612.0, + 920.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1587.0, + 1019.0, + 1587.0, + 1019.0, + 1615.0, + 979.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1588.0, + 1078.0, + 1588.0, + 1078.0, + 1612.0, + 1041.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1588.0, + 1158.0, + 1588.0, + 1158.0, + 1613.0, + 1104.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1588.0, + 1220.0, + 1588.0, + 1220.0, + 1612.0, + 1183.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 1587.0, + 1282.0, + 1587.0, + 1282.0, + 1615.0, + 1242.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1588.0, + 1341.0, + 1588.0, + 1341.0, + 1612.0, + 1305.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 1588.0, + 1402.0, + 1588.0, + 1402.0, + 1612.0, + 1368.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1605.0, + 558.0, + 1605.0, + 558.0, + 1632.0, + 386.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1605.0, + 820.0, + 1605.0, + 820.0, + 1632.0, + 651.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1605.0, + 1085.0, + 1605.0, + 1085.0, + 1632.0, + 913.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 1605.0, + 1348.0, + 1605.0, + 1348.0, + 1632.0, + 1177.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1649.0, + 620.0, + 1649.0, + 620.0, + 1676.0, + 377.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1650.0, + 866.0, + 1650.0, + 866.0, + 1674.0, + 691.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1649.0, + 1026.0, + 1649.0, + 1026.0, + 1675.0, + 974.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1645.0, + 1343.0, + 1645.0, + 1343.0, + 1676.0, + 1171.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1669.0, + 601.0, + 1669.0, + 601.0, + 1702.0, + 376.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1671.0, + 905.0, + 1671.0, + 905.0, + 1707.0, + 688.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1668.0, + 1101.0, + 1668.0, + 1101.0, + 1704.0, + 973.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1668.0, + 1347.0, + 1668.0, + 1347.0, + 1704.0, + 1169.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1734.0, + 1405.0, + 1734.0, + 1405.0, + 1765.0, + 294.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1764.0, + 1405.0, + 1764.0, + 1405.0, + 1793.0, + 295.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1790.0, + 1075.0, + 1790.0, + 1075.0, + 1820.0, + 292.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 896.0, + 817.0, + 896.0, + 817.0, + 938.0, + 294.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 224.0, + 999.0, + 224.0, + 999.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 859.0, + 72.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 953.0, + 1405.0, + 953.0, + 1405.0, + 987.0, + 295.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 982.0, + 1405.0, + 982.0, + 1405.0, + 1020.0, + 294.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1014.0, + 1406.0, + 1014.0, + 1406.0, + 1050.0, + 294.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1047.0, + 1407.0, + 1047.0, + 1407.0, + 1078.0, + 295.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1075.0, + 1403.0, + 1075.0, + 1403.0, + 1109.0, + 295.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1104.0, + 1408.0, + 1104.0, + 1408.0, + 1143.0, + 292.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1134.0, + 1405.0, + 1134.0, + 1405.0, + 1172.0, + 292.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1167.0, + 1406.0, + 1167.0, + 1406.0, + 1201.0, + 295.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1198.0, + 1406.0, + 1198.0, + 1406.0, + 1233.0, + 295.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1229.0, + 699.0, + 1229.0, + 699.0, + 1261.0, + 296.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1407.0, + 1849.0, + 1407.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1880.0, + 1407.0, + 1880.0, + 1407.0, + 1916.0, + 295.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1912.0, + 1404.0, + 1912.0, + 1404.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1940.0, + 1407.0, + 1940.0, + 1407.0, + 1977.0, + 292.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2005.0, + 295.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 529.0, + 1404.0, + 529.0, + 1404.0, + 561.0, + 295.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 560.0, + 1404.0, + 560.0, + 1404.0, + 592.0, + 295.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 588.0, + 501.0, + 588.0, + 501.0, + 627.0, + 293.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 588.0, + 590.0, + 588.0, + 590.0, + 627.0, + 536.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 588.0, + 1405.0, + 588.0, + 1405.0, + 627.0, + 624.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 620.0, + 894.0, + 620.0, + 894.0, + 655.0, + 293.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 358.0, + 1404.0, + 358.0, + 1404.0, + 392.0, + 296.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 388.0, + 1405.0, + 388.0, + 1405.0, + 426.0, + 294.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 420.0, + 1407.0, + 420.0, + 1407.0, + 457.0, + 293.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 449.0, + 685.0, + 449.0, + 685.0, + 490.0, + 293.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 449.0, + 1405.0, + 449.0, + 1405.0, + 490.0, + 782.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 482.0, + 1051.0, + 482.0, + 1051.0, + 519.0, + 294.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 773.0, + 1405.0, + 773.0, + 1405.0, + 811.0, + 294.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 806.0, + 1404.0, + 806.0, + 1404.0, + 840.0, + 293.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 831.0, + 297.0, + 831.0, + 297.0, + 875.0, + 292.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 831.0, + 782.0, + 831.0, + 782.0, + 875.0, + 331.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 831.0, + 1119.0, + 831.0, + 1119.0, + 875.0, + 816.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 831.0, + 1165.0, + 831.0, + 1165.0, + 875.0, + 1155.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 663.0, + 1403.0, + 663.0, + 1403.0, + 705.0, + 290.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 697.0, + 1402.0, + 697.0, + 1402.0, + 733.0, + 294.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 728.0, + 964.0, + 728.0, + 964.0, + 763.0, + 293.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 279.0, + 1404.0, + 279.0, + 1404.0, + 317.0, + 294.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 312.0, + 663.0, + 312.0, + 663.0, + 347.0, + 294.0, + 347.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1417, + 1403, + 1417, + 1403, + 1571, + 298, + 1571 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 312, + 626, + 1385, + 626, + 1385, + 1016, + 312, + 1016 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1585, + 1405, + 1585, + 1405, + 1740, + 298, + 1740 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 300, + 1224, + 1404, + 1224, + 1404, + 1317, + 300, + 1317 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 308, + 1755, + 1405, + 1755, + 1405, + 1969, + 308, + 1969 + ], + "score": 0.964 + }, + { + "category_id": 4, + "poly": [ + 296, + 1046, + 1405, + 1046, + 1405, + 1159, + 296, + 1159 + ], + "score": 0.964 + }, + { + "category_id": 3, + "poly": [ + 410, + 223, + 1286, + 223, + 1286, + 474, + 410, + 474 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 295, + 501, + 1406, + 501, + 1406, + 587, + 295, + 587 + ], + "score": 0.954 + }, + { + "category_id": 2, + "poly": [ + 330, + 2004, + 1093, + 2004, + 1093, + 2034, + 330, + 2034 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 857, + 76, + 857, + 104, + 298, + 104 + ], + "score": 0.867 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.857 + }, + { + "category_id": 0, + "poly": [ + 298, + 1357, + 1081, + 1357, + 1081, + 1392, + 298, + 1392 + ], + "score": 0.748 + }, + { + "category_id": 13, + "poly": [ + 812, + 1449, + 928, + 1449, + 928, + 1478, + 812, + 1478 + ], + "score": 0.8, + "latex": "{ \\sim } 2 0 0 { , } 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 1246, + 1757, + 1306, + 1757, + 1306, + 1786, + 1246, + 1786 + ], + "score": 0.43, + "latex": "1 8 0 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 627.0, + 620.0, + 627.0, + 620.0, + 656.0, + 400.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 627.0, + 971.0, + 627.0, + 971.0, + 656.0, + 774.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 629.0, + 1335.0, + 629.0, + 1335.0, + 655.0, + 1139.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 644.0, + 371.0, + 644.0, + 371.0, + 671.0, + 334.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 644.0, + 734.0, + 644.0, + 734.0, + 671.0, + 697.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 644.0, + 1098.0, + 644.0, + 1098.0, + 671.0, + 1060.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 666.0, + 371.0, + 666.0, + 371.0, + 694.0, + 334.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 666.0, + 734.0, + 666.0, + 734.0, + 693.0, + 697.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 684.0, + 372.0, + 684.0, + 372.0, + 759.0, + 311.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 677.0, + 745.0, + 677.0, + 745.0, + 759.0, + 664.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 682.0, + 1098.0, + 682.0, + 1098.0, + 743.0, + 1038.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 729.0, + 1127.0, + 729.0, + 1127.0, + 741.0, + 1103.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 753.0, + 371.0, + 753.0, + 371.0, + 779.0, + 334.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 753.0, + 735.0, + 753.0, + 735.0, + 780.0, + 697.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 753.0, + 1097.0, + 753.0, + 1097.0, + 781.0, + 1060.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 769.0, + 378.0, + 769.0, + 378.0, + 789.0, + 360.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 769.0, + 418.0, + 769.0, + 418.0, + 790.0, + 400.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 768.0, + 460.0, + 768.0, + 460.0, + 790.0, + 439.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 768.0, + 500.0, + 768.0, + 500.0, + 789.0, + 479.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 769.0, + 539.0, + 769.0, + 539.0, + 788.0, + 519.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 768.0, + 580.0, + 768.0, + 580.0, + 789.0, + 560.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 768.0, + 620.0, + 768.0, + 620.0, + 789.0, + 600.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 770.0, + 657.0, + 770.0, + 657.0, + 787.0, + 643.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 768.0, + 741.0, + 768.0, + 741.0, + 789.0, + 721.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 769.0, + 783.0, + 769.0, + 783.0, + 790.0, + 763.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 768.0, + 823.0, + 768.0, + 823.0, + 790.0, + 802.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 768.0, + 863.0, + 768.0, + 863.0, + 789.0, + 842.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 769.0, + 903.0, + 769.0, + 903.0, + 788.0, + 883.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 768.0, + 943.0, + 768.0, + 943.0, + 790.0, + 923.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 768.0, + 984.0, + 768.0, + 984.0, + 789.0, + 963.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 770.0, + 1021.0, + 770.0, + 1021.0, + 787.0, + 1006.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 769.0, + 1103.0, + 769.0, + 1103.0, + 788.0, + 1088.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 770.0, + 1143.0, + 770.0, + 1143.0, + 788.0, + 1128.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 768.0, + 1185.0, + 768.0, + 1185.0, + 789.0, + 1166.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 768.0, + 1225.0, + 768.0, + 1225.0, + 789.0, + 1207.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 769.0, + 1266.0, + 769.0, + 1266.0, + 788.0, + 1246.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 768.0, + 1305.0, + 768.0, + 1305.0, + 789.0, + 1287.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 768.0, + 1346.0, + 768.0, + 1346.0, + 789.0, + 1327.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 770.0, + 1384.0, + 770.0, + 1384.0, + 787.0, + 1369.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 790.0, + 394.0, + 790.0, + 394.0, + 970.0, + 297.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 784.0, + 552.0, + 784.0, + 552.0, + 814.0, + 468.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 784.0, + 657.0, + 784.0, + 657.0, + 810.0, + 613.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 811.0, + 735.0, + 811.0, + 735.0, + 950.0, + 674.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 784.0, + 915.0, + 784.0, + 915.0, + 814.0, + 831.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 784.0, + 1020.0, + 784.0, + 1020.0, + 810.0, + 977.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 812.0, + 1097.0, + 812.0, + 1097.0, + 907.0, + 1039.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 788.0, + 1276.0, + 788.0, + 1276.0, + 812.0, + 1196.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 784.0, + 1384.0, + 784.0, + 1384.0, + 809.0, + 1340.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 850.0, + 371.0, + 850.0, + 371.0, + 876.0, + 334.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 849.0, + 734.0, + 849.0, + 734.0, + 875.0, + 696.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 840.0, + 1098.0, + 840.0, + 1098.0, + 867.0, + 1061.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 867.0, + 1098.0, + 867.0, + 1098.0, + 894.0, + 1060.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 884.0, + 373.0, + 884.0, + 373.0, + 914.0, + 335.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 913.0, + 409.0, + 913.0, + 409.0, + 921.0, + 382.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 885.0, + 735.0, + 885.0, + 735.0, + 912.0, + 697.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 912.0, + 1000.0, + 912.0, + 1000.0, + 922.0, + 913.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 892.0, + 1097.0, + 892.0, + 1097.0, + 948.0, + 1038.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 923.0, + 370.0, + 923.0, + 370.0, + 946.0, + 334.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 923.0, + 732.0, + 923.0, + 732.0, + 946.0, + 698.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 922.0, + 1097.0, + 922.0, + 1097.0, + 949.0, + 1060.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 939.0, + 377.0, + 939.0, + 377.0, + 955.0, + 361.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 938.0, + 418.0, + 938.0, + 418.0, + 958.0, + 400.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 936.0, + 460.0, + 936.0, + 460.0, + 959.0, + 439.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 936.0, + 500.0, + 936.0, + 500.0, + 958.0, + 479.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 938.0, + 539.0, + 938.0, + 539.0, + 958.0, + 520.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 936.0, + 580.0, + 936.0, + 580.0, + 958.0, + 560.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 936.0, + 621.0, + 936.0, + 621.0, + 959.0, + 600.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 941.0, + 655.0, + 941.0, + 655.0, + 954.0, + 646.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 938.0, + 741.0, + 938.0, + 741.0, + 958.0, + 723.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 940.0, + 780.0, + 940.0, + 780.0, + 956.0, + 764.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 936.0, + 823.0, + 936.0, + 823.0, + 958.0, + 802.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 936.0, + 863.0, + 936.0, + 863.0, + 958.0, + 844.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 938.0, + 903.0, + 938.0, + 903.0, + 958.0, + 883.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 936.0, + 944.0, + 936.0, + 944.0, + 959.0, + 923.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 936.0, + 984.0, + 936.0, + 984.0, + 958.0, + 963.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 940.0, + 1020.0, + 940.0, + 1020.0, + 955.0, + 1006.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 938.0, + 1104.0, + 938.0, + 1104.0, + 958.0, + 1086.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 940.0, + 1143.0, + 940.0, + 1143.0, + 956.0, + 1128.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 938.0, + 1185.0, + 938.0, + 1185.0, + 958.0, + 1166.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 938.0, + 1226.0, + 938.0, + 1226.0, + 958.0, + 1207.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 938.0, + 1266.0, + 938.0, + 1266.0, + 958.0, + 1246.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 936.0, + 1307.0, + 936.0, + 1307.0, + 958.0, + 1286.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 938.0, + 1346.0, + 938.0, + 1346.0, + 958.0, + 1326.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 940.0, + 1384.0, + 940.0, + 1384.0, + 955.0, + 1369.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 955.0, + 551.0, + 955.0, + 551.0, + 983.0, + 469.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 953.0, + 657.0, + 953.0, + 657.0, + 979.0, + 613.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 956.0, + 913.0, + 956.0, + 913.0, + 980.0, + 834.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 953.0, + 1020.0, + 953.0, + 1020.0, + 979.0, + 976.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 956.0, + 1276.0, + 956.0, + 1276.0, + 980.0, + 1196.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 953.0, + 1384.0, + 953.0, + 1384.0, + 979.0, + 1338.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 988.0, + 603.0, + 988.0, + 603.0, + 1011.0, + 462.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 986.0, + 788.0, + 986.0, + 788.0, + 1013.0, + 662.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 989.0, + 979.0, + 989.0, + 979.0, + 1008.0, + 850.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 986.0, + 1159.0, + 986.0, + 1159.0, + 1013.0, + 1041.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 995.0, + 1207.0, + 995.0, + 1207.0, + 1004.0, + 1189.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 988.0, + 1325.0, + 988.0, + 1325.0, + 1011.0, + 1219.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1047.0, + 1403.0, + 1047.0, + 1403.0, + 1080.0, + 295.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1109.0, + 293.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1101.0, + 1406.0, + 1101.0, + 1406.0, + 1134.0, + 295.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1129.0, + 1374.0, + 1129.0, + 1374.0, + 1162.0, + 295.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 225.0, + 665.0, + 225.0, + 665.0, + 251.0, + 492.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 225.0, + 929.0, + 225.0, + 929.0, + 251.0, + 755.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 246.0, + 660.0, + 246.0, + 660.0, + 271.0, + 496.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 245.0, + 915.0, + 245.0, + 915.0, + 271.0, + 770.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 244.0, + 1284.0, + 244.0, + 1284.0, + 272.0, + 1032.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 267.0, + 466.0, + 267.0, + 466.0, + 297.0, + 427.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 269.0, + 1267.0, + 269.0, + 1267.0, + 299.0, + 1031.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 293.0, + 1216.0, + 293.0, + 1216.0, + 325.0, + 1032.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 310.0, + 474.0, + 310.0, + 474.0, + 389.0, + 400.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 321.0, + 1087.0, + 321.0, + 1087.0, + 348.0, + 1031.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 343.0, + 1165.0, + 343.0, + 1165.0, + 377.0, + 1031.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 369.0, + 465.0, + 369.0, + 465.0, + 399.0, + 426.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 371.0, + 1210.0, + 371.0, + 1210.0, + 402.0, + 1031.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 398.0, + 1214.0, + 398.0, + 1214.0, + 429.0, + 1031.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 430.0, + 481.0, + 430.0, + 481.0, + 456.0, + 444.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 430.0, + 598.0, + 430.0, + 598.0, + 456.0, + 500.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 430.0, + 655.0, + 430.0, + 655.0, + 455.0, + 615.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 430.0, + 746.0, + 430.0, + 746.0, + 456.0, + 676.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 430.0, + 805.0, + 430.0, + 805.0, + 456.0, + 765.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 429.0, + 863.0, + 429.0, + 863.0, + 457.0, + 806.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 430.0, + 919.0, + 430.0, + 919.0, + 456.0, + 881.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 430.0, + 977.0, + 430.0, + 977.0, + 457.0, + 940.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 429.0, + 1256.0, + 429.0, + 1256.0, + 457.0, + 1033.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 450.0, + 666.0, + 450.0, + 666.0, + 475.0, + 490.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 450.0, + 930.0, + 450.0, + 930.0, + 475.0, + 754.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 501.0, + 1404.0, + 501.0, + 1404.0, + 531.0, + 294.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 531.0, + 1404.0, + 531.0, + 1404.0, + 560.0, + 295.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 554.0, + 1278.0, + 554.0, + 1278.0, + 588.0, + 292.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1997.0, + 1097.0, + 1997.0, + 1097.0, + 2041.0, + 328.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 867.0, + 2084.0, + 867.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1353.0, + 1086.0, + 1353.0, + 1086.0, + 1397.0, + 296.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1454.0, + 294.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1445.0, + 811.0, + 1445.0, + 811.0, + 1485.0, + 292.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 1445.0, + 1405.0, + 1445.0, + 1405.0, + 1485.0, + 929.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1473.0, + 1407.0, + 1473.0, + 1407.0, + 1518.0, + 292.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1507.0, + 1403.0, + 1507.0, + 1403.0, + 1547.0, + 293.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1538.0, + 910.0, + 1538.0, + 910.0, + 1577.0, + 293.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1587.0, + 1406.0, + 1587.0, + 1406.0, + 1620.0, + 297.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1616.0, + 1408.0, + 1616.0, + 1408.0, + 1653.0, + 295.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1647.0, + 1407.0, + 1647.0, + 1407.0, + 1684.0, + 295.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1676.0, + 1406.0, + 1676.0, + 1406.0, + 1716.0, + 293.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1705.0, + 1409.0, + 1705.0, + 1409.0, + 1745.0, + 292.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1225.0, + 1404.0, + 1225.0, + 1404.0, + 1258.0, + 296.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1256.0, + 1403.0, + 1256.0, + 1403.0, + 1286.0, + 295.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1286.0, + 684.0, + 1286.0, + 684.0, + 1320.0, + 296.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1752.0, + 1245.0, + 1752.0, + 1245.0, + 1795.0, + 309.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1752.0, + 1406.0, + 1752.0, + 1406.0, + 1795.0, + 1307.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1786.0, + 1405.0, + 1786.0, + 1405.0, + 1820.0, + 329.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1815.0, + 1405.0, + 1815.0, + 1405.0, + 1852.0, + 330.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1846.0, + 1405.0, + 1846.0, + 1405.0, + 1883.0, + 330.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 1878.0, + 1405.0, + 1878.0, + 1405.0, + 1912.0, + 333.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1906.0, + 1409.0, + 1906.0, + 1409.0, + 1945.0, + 329.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1935.0, + 437.0, + 1935.0, + 437.0, + 1973.0, + 326.0, + 1973.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 346, + 277, + 1349, + 277, + 1349, + 426, + 346, + 426 + ], + "score": 0.98, + "html": "
RealDCGANWGANWGAN-GPLSGAN
Conv SpaceMMD0.0190.2050.2700.1940.232
1-NN Accuracy0.4990.8250.9200.8120.871
1-NN Accuracy (real)0.4950.7590.8800.7650.804
1-NN Accuracy (fake)0.5030.8920.9610.8600.938
" + }, + { + "category_id": 1, + "poly": [ + 298, + 661, + 1405, + 661, + 1405, + 847, + 298, + 847 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 307, + 459, + 1408, + 459, + 1408, + 521, + 307, + 521 + ], + "score": 0.942 + }, + { + "category_id": 6, + "poly": [ + 513, + 224, + 1186, + 224, + 1186, + 254, + 513, + 254 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 305, + 565, + 1321, + 565, + 1321, + 640, + 305, + 640 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 367, + 871, + 1405, + 871, + 1405, + 1231, + 367, + 1231 + ], + "score": 0.864 + }, + { + "category_id": 2, + "poly": [ + 835, + 2089, + 863, + 2089, + 863, + 2112, + 835, + 2112 + ], + "score": 0.841 + }, + { + "category_id": 15, + "poly": [ + 511.0, + 223.0, + 1188.0, + 223.0, + 1188.0, + 257.0, + 511.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 564.0, + 1328.0, + 564.0, + 1328.0, + 610.0, + 293.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 611.0, + 677.0, + 611.0, + 677.0, + 643.0, + 355.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 663.0, + 1405.0, + 663.0, + 1405.0, + 695.0, + 296.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 691.0, + 1406.0, + 691.0, + 1406.0, + 727.0, + 295.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 722.0, + 1406.0, + 722.0, + 1406.0, + 758.0, + 293.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 751.0, + 1405.0, + 751.0, + 1405.0, + 790.0, + 293.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 782.0, + 1406.0, + 782.0, + 1406.0, + 818.0, + 293.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 813.0, + 1279.0, + 813.0, + 1279.0, + 850.0, + 293.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 456.0, + 1408.0, + 456.0, + 1408.0, + 495.0, + 305.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 486.0, + 1411.0, + 486.0, + 1411.0, + 525.0, + 330.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 871.0, + 1031.0, + 871.0, + 1031.0, + 907.0, + 366.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 911.0, + 1406.0, + 911.0, + 1406.0, + 952.0, + 382.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 946.0, + 1405.0, + 946.0, + 1405.0, + 979.0, + 395.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 974.0, + 1406.0, + 974.0, + 1406.0, + 1011.0, + 393.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1003.0, + 1406.0, + 1003.0, + 1406.0, + 1041.0, + 392.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1037.0, + 1405.0, + 1037.0, + 1405.0, + 1070.0, + 394.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1068.0, + 1051.0, + 1068.0, + 1051.0, + 1101.0, + 395.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1105.0, + 1406.0, + 1105.0, + 1406.0, + 1143.0, + 367.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1138.0, + 1406.0, + 1138.0, + 1406.0, + 1172.0, + 394.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1167.0, + 1408.0, + 1167.0, + 1408.0, + 1203.0, + 393.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1201.0, + 724.0, + 1201.0, + 724.0, + 1234.0, + 395.0, + 1234.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/SybqeKgA-/SybqeKgA-_layout.pdf b/parse/train/SybqeKgA-/SybqeKgA-_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0b51d9739f2c30c41892bb6503562a8c8a5a9d7a --- /dev/null +++ b/parse/train/SybqeKgA-/SybqeKgA-_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8da9e420fdbe7126b862659f359bf689c16beefa4ead74fa484bbaf5a0ab404 +size 1663619 diff --git a/parse/train/SybqeKgA-/SybqeKgA-_origin.pdf b/parse/train/SybqeKgA-/SybqeKgA-_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7129f49f6b0b2c10ad17fb62acccfe3ba1cc78d7 --- /dev/null +++ b/parse/train/SybqeKgA-/SybqeKgA-_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94bcbffb063035c80c57b438ec708c0f889872cc6425a6902167f5eed1071133 +size 1533784 diff --git a/parse/train/SybqeKgA-/SybqeKgA-_span.pdf b/parse/train/SybqeKgA-/SybqeKgA-_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0b73f2f68d2dd538e8ffa55612b444e3d31d3fb6 --- /dev/null +++ b/parse/train/SybqeKgA-/SybqeKgA-_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6bcb069bef89a0e2506dc769a0c42c941734c1f14179074295a23027a7319ee +size 1665198 diff --git a/parse/train/SygwwGbRW/SygwwGbRW_layout.pdf b/parse/train/SygwwGbRW/SygwwGbRW_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..52300ed66368f44cd569ab36cbcb230325acd2cb --- /dev/null +++ b/parse/train/SygwwGbRW/SygwwGbRW_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b37a4c93b79fcb359fe800bc995f05f1d34f982eff4bfc6d39edcc13a4d2a92 +size 3970827 diff --git a/parse/train/SygwwGbRW/SygwwGbRW_origin.pdf b/parse/train/SygwwGbRW/SygwwGbRW_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9867212fda3aab82179d0f1138ae41b4c0cfc1a6 --- /dev/null +++ b/parse/train/SygwwGbRW/SygwwGbRW_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24aad1cff2d4ebd59f067a1adcf2910360a67620ff21f2035cc62f092aedd771 +size 3831710 diff --git a/parse/train/SygwwGbRW/SygwwGbRW_span.pdf b/parse/train/SygwwGbRW/SygwwGbRW_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..516e9026985ec5334d7102792fe500647b9eb124 --- /dev/null +++ b/parse/train/SygwwGbRW/SygwwGbRW_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34b474c0be87e4c9bbd9fc3cc9049599dcfdcea45b79d562ea819ba0bd5c5143 +size 3975768 diff --git a/parse/train/SylVJTNKDr/SylVJTNKDr_layout.pdf b/parse/train/SylVJTNKDr/SylVJTNKDr_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..88d863edc255573a7f0a0f6ccfcc9a0444912f73 --- /dev/null +++ b/parse/train/SylVJTNKDr/SylVJTNKDr_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b8c1090745bfc7d230bad8e33619209b4dcd1afd79bd487099e5aed69a5a3ed +size 1204972 diff --git a/parse/train/SylVJTNKDr/SylVJTNKDr_origin.pdf b/parse/train/SylVJTNKDr/SylVJTNKDr_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..aa9eb59355753fb8ad1631faf16a3524113a0f9e --- /dev/null +++ b/parse/train/SylVJTNKDr/SylVJTNKDr_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d44bbfc4f977324bfa6fb97ba48cfce1763a115801a9028b68c8069b1161bbb1 +size 1040379 diff --git a/parse/train/SylVJTNKDr/SylVJTNKDr_span.pdf b/parse/train/SylVJTNKDr/SylVJTNKDr_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3a05654d53f5875e04a7c0f86516fdd455843dad --- /dev/null +++ b/parse/train/SylVJTNKDr/SylVJTNKDr_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9992159fa74621776d47c87006e8bce48c17e4325d2216a67647445e7be0701 +size 1208865 diff --git a/parse/train/SyqShMZRb/SyqShMZRb_layout.pdf b/parse/train/SyqShMZRb/SyqShMZRb_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..053697437de1f3cfb1a232687361a2e4226b03f3 --- /dev/null +++ b/parse/train/SyqShMZRb/SyqShMZRb_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9920df419cb63091658fbf134c506eaf2e45f6d8967aa53912317fbc8af7171 +size 2997512 diff --git a/parse/train/SyqShMZRb/SyqShMZRb_origin.pdf b/parse/train/SyqShMZRb/SyqShMZRb_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4c3e8133366c756b7b11e868729f72b5e11223f7 --- /dev/null +++ b/parse/train/SyqShMZRb/SyqShMZRb_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3d5d080ea3b4ac58fd90625aad124ae2690404b1401f743af2c383126ea098f +size 2817526 diff --git a/parse/train/SyqShMZRb/SyqShMZRb_span.pdf b/parse/train/SyqShMZRb/SyqShMZRb_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..dd117108b21d8af706b5c07e13d3e39e476b5231 --- /dev/null +++ b/parse/train/SyqShMZRb/SyqShMZRb_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db6bd8ba54b984c2dc946b4ecd4d1dd0ce2471ebfdf54890e421013ef51a12c9 +size 3002805 diff --git a/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_layout.pdf b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8d8cebab58aa5778238e93710f7d23a24e914de3 --- /dev/null +++ b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1880f56135211c13c3420176049f07af930d726648eeb90de1e84a7f35ac119d +size 428151 diff --git a/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_origin.pdf b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5e352be87147dc19db2e85827566ef2661f4d2fb --- /dev/null +++ b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:829aa10d86e0d335d69214f86adfb061a9ca4173a78c4d9ceed33a5b48800625 +size 313096 diff --git a/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_span.pdf b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..409be3a5d450fcc677776a9f29e9f175509b20a3 --- /dev/null +++ b/parse/train/SyxvSiCcFQ/SyxvSiCcFQ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:003bdae3229aca53fdff555caa2cf68a5738143fe30a3b7b06823c5b500f986d +size 431875 diff --git a/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_layout.pdf b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a50fc7e10b56c015938108a2f8535db262d8771f --- /dev/null +++ b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b676c2acef0ef8a011e986ec66b68505eb38e23f1aea1a42009852d7f9ca4f6f +size 1189593 diff --git a/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_origin.pdf b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..83464bf09d416481fe217ed61e1decd1a15f2ac2 --- /dev/null +++ b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffd150678d5bf3edb152567dc65054c670c0c208aac04c105aa5fd77040c6b2c +size 643996 diff --git a/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_span.pdf b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6672ba7fcc4cdca5effa24039168a5a500333661 --- /dev/null +++ b/parse/train/TMUR2ovJfjE/TMUR2ovJfjE_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8615ba9fd39e69dc1fb50f23d7211c2613688c1d826992b27c0b4ed813efc75 +size 1212601 diff --git a/parse/train/TV9INIrmtWN/TV9INIrmtWN_layout.pdf b/parse/train/TV9INIrmtWN/TV9INIrmtWN_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e89e1b5ffb38565ed9c402dc69330ee3a2a81953 --- /dev/null +++ b/parse/train/TV9INIrmtWN/TV9INIrmtWN_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40adf4fe5d21f25770a0d46bc309166a7794a83c0a021d2c3809a72b8b85b3c8 +size 8531650 diff --git a/parse/train/TV9INIrmtWN/TV9INIrmtWN_origin.pdf b/parse/train/TV9INIrmtWN/TV9INIrmtWN_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9e3d1a76e176d6f6a034503e18f6f0879e9e4f47 --- /dev/null +++ b/parse/train/TV9INIrmtWN/TV9INIrmtWN_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:115c37dda8e03f3388424df2795b332b8fc08a0e39c1a3bc1f12c1cd12321082 +size 8339966 diff --git a/parse/train/TV9INIrmtWN/TV9INIrmtWN_span.pdf b/parse/train/TV9INIrmtWN/TV9INIrmtWN_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8355ed1e84cff8dcff0eac1b58772dd3b87c3cd8 --- /dev/null +++ b/parse/train/TV9INIrmtWN/TV9INIrmtWN_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a874028e227b53a452f29bdea7684aace695f1ffd6d943f1b842623550e258a +size 8533527 diff --git a/parse/train/Twf_XYunk5j/Twf_XYunk5j_layout.pdf b/parse/train/Twf_XYunk5j/Twf_XYunk5j_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..718f0af8de2c4a53fbe794e5e8490a8ae1024fe1 --- /dev/null +++ b/parse/train/Twf_XYunk5j/Twf_XYunk5j_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f25f675d5f44e9023fac34ea5b9437bb6d404e99d05a3919c4dbffcae8810cc +size 988225 diff --git a/parse/train/Twf_XYunk5j/Twf_XYunk5j_origin.pdf b/parse/train/Twf_XYunk5j/Twf_XYunk5j_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..acd8848f90a991c1ac7cc7f68221abf33f4c19f1 --- /dev/null +++ b/parse/train/Twf_XYunk5j/Twf_XYunk5j_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38bd34ac0f9ec8fc7a81c802c71193c92519cd5fa1a959d2cb8bcc42d2919d13 +size 804908 diff --git a/parse/train/Twf_XYunk5j/Twf_XYunk5j_span.pdf b/parse/train/Twf_XYunk5j/Twf_XYunk5j_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0da058f281c15d7325db3b35e6af99d294523f66 --- /dev/null +++ b/parse/train/Twf_XYunk5j/Twf_XYunk5j_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cd8ac659f511e1cdacb4b2bd4d9ddb1808eca727e0d0741123f923ef3a8d5cb +size 1003184 diff --git a/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_layout.pdf b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b377a0a30f3786dfbf38e70daf4bfb340ba5a3b6 --- /dev/null +++ b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:017798bcd3b168d86956bee13cc29fa675920b8191b4d94b79faa85a621940b7 +size 546091 diff --git a/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_origin.pdf b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..39518887e09c8e450e8590fd965d88be1420e196 --- /dev/null +++ b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40e1d862b63555f0558cc2a4cdafff70739f2ab74980c54c767e257c7cdca4bb +size 321383 diff --git a/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_span.pdf b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a72b20e0afef19c3b7dcd3161e4e0a60aea1eb17 --- /dev/null +++ b/parse/train/UFWnZn2v0bV/UFWnZn2v0bV_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ce733b10bfdbe15b02ec2b2fcbcdf3ebd98c2acd74a5550f84be3884a2dc3d6 +size 549936 diff --git a/parse/train/VRgITLy0l2/VRgITLy0l2.md b/parse/train/VRgITLy0l2/VRgITLy0l2.md new file mode 100644 index 0000000000000000000000000000000000000000..c1d4a2fbf4055b6da77216dd066b344051ada0fc --- /dev/null +++ b/parse/train/VRgITLy0l2/VRgITLy0l2.md @@ -0,0 +1,349 @@ +# A PRIORI GUARANTEES OF FINITE-TIME CONVERGENCE FOR DEEP NEURAL NETWORKS + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +In this paper, we perform Lyapunov based analysis of the loss function to derive an a priori upper bound on the settling time of deep neural networks. While previous studies have attempted to understand deep learning using control theory framework, there is limited work on a priori finite time convergence analysis. Drawing from the advances in analysis of finite-time control of non-linear systems, we provide a priori guarantees of finite-time convergence in a deterministic control theoretic setting. We formulate the supervised learning framework as a control problem where weights of the network are control inputs and learning translates into a tracking problem. An analytical formula for finite-time upper bound on settling time is provided a priori under the assumptions of boundedness of input. Finally, we prove that our loss function is robust against input perturbations. + +# 1 INTRODUCTION + +Over the past decade, Deep neural networks have achieved human-like performance in various machine learning tasks, such as classification, natural language processing and speech recognition. Despite the popularity of deep learning, the underlying theoretical understanding remains relatively less explored. While attempts have been made to develop deep learning theory by drawing inspiration from other related fields such as statistical learning and information theory, a comprehensive theoretical framework is still in an early developmental stage. It is difficult to perform mathematical analysis on Deep neural networks due to the large number of parameters involved. Other problems in deep neural networks revolve around the stability and desired convergence rate of the training. Since the performance of the network depends highly on the training data and the choice of the optimization algorithm, there is no guarantee that the training will converge. Our work attempts to give finite-time convergence guarantees for training of a deep neural network by utilizing an established stabilization framework from control theory. + +Existing works in deep learning theory have attempted to bridge the gap in understanding deep learning dynamics by focusing on simple models of neural networks [Saxe et al. (2013), Li & Yuan (2017), Arora et al. (2018), Jacot et al. (2018)]. This could be attributed to the fact that current state-of-the-art deep learning models are highly complex structures to analyze. Jacot et al. (2018) proved that a multilayer fully-connected network with infinite width converges to a deterministic limit at initialization and the rate of change of weights goes to zero. Saxe et al. (2013) analyzed deep linear networks and proved that these networks, surprisingly, have a rich non-linear structure. The study shows that given the right initial conditions, deep linear networks are a finite amount slower than shallow networks. Following this work, Arora et al. (2018) proved the convergence of gradient descent to global minima for networks with dimensions of every layer being full rank in dimensions. While these studies give important insights into the design of neural network architecture and the behavior of training, their results may need to be modified in order to provide convergence guarantees for the conventional deep neural networks. Du et al. (2018) extended the work of Jacot et al. (2018) further by proving convergence for gradient descent to achieve zero training loss in deep neural networks with residual connections. + +When it comes to convergence of certain state variables of a dynamical system, control theory provides a rich mathematical framework which can be utilized for analyzing the non-linear dynamics of deep learning [Liu & Theodorou (2019)]. One of the early works relating deep learning to control theory was of LeCun et al. (1988), which used the concept of optimal control and formulated back-propagation as an optimization problem with non-linear constraints. Non-linear control has gained increasing attention over the past few years in the context of neural networks, especially for recurrent neural networks [Allen-Zhu et al. (2019), Xiao (2017)] and reinforcement learning [Xu et al. (2013), Gupta et al. (2019), Wang et al. (2019), Kaledin et al. (2020)]. A new class of recurrent neural networks, called Zhang Neural Networks (ZNN), was developed that expressed dynamics of the network as a set of ordinary differential equations and used non-linear control to prove global or exponential stability for time-varying Sylvester equation [Zhang et al. (2002), Guo et al. (2011)]. Li et al. (2013) introduced the sign bi-power activation function for Zhang Neural Networks (ZNN) which helps to prove the existence of finite-time convergence property. Haber & Ruthotto (2017) presents deep learning as a parameter estimation problem of non-linear dynamical systems to tackle the exploding and vanishing gradients. + +The focus of this paper is on deriving a priori guarantee of attaining finite-time convergence of training in a supervised learning framework under some assumptions on inputs. The novelty lies in the fact that the weight update is cast as a finite-time control synthesis such that the loss function is proven to be a valid Lyapunov function. The resulting training update is derived as a function of time such that it ensures the convergence of Lyapunov function in finite time. The only assumption used is that the magnitude of atleast one input is greater than zero. Thus, the learning problem is converted into a finite time stabilization problem as studied rigorously in Bhat & Bernstein (2000). The contributions of the proposed study are twofold. First, we propose a Lyapunov candidate function to be used as loss function. Second, we modify the weight update of the neural network in such a way that the supervised training is converted into a dynamical control system. This allows us to use results from Bhat & Bernstein (2000) to a priori guarantee finite time convergence on the training. To the best of our knowledge, a guarantee of finite-time convergence is being studied for the first time in context of training a general multi-layer neural network. The proposed results will enable time bound training that will be useful in real-time applications. + +The paper is organized as follows. Section 2 starts with introducing the Lyapunov function from the control theory perspective. Section 2.1 derives the weight update and lyapunov loss function for a single neuron case and proves that it satisfies the conditions required for finite-time stability theorems developed in Bhat & Bernstein (2000) to be applicable. Section 2.2 then proves that a similar result extends to a multi-layer perceptron network under reasonable assumptions on the input. In Section 2.3, we state the equations to compute upper bounds on the convergence time for training neural networks. Section 2.4 provides an extension to the case when bounded perturbations are admitted at the input and convergence guarantees are shown to hold true. In Section 3, some numerical simulations are presented for both single neuron and multi-layer perceptron cases for regression. Section 4 collects conclusions and discusses future scope. + +# 2 PROPOSED METHOD TO CONVERT SUPERVISED LEARNING INTO DYNAMICAL CONTROL SYSTEM + +This section motivates the development of a priori bounds on settling time with certain assumptions on the input. The weight update problem for supervised learning in neural networks is similar to the tracking problem of non-linear control systems. The idea is to synthesize a feedback control law based on certain Lyapunov function $V ( x )$ . A Lyapunov function is a positive definite function, i.e. $V ( x ) > 0$ , and its time derivative is negative definite along the given system dynamics, i.e. $\dot { V } < 0$ . We cast the supervised learning problem as a dynamical system which admits a valid Lyapunov function as the loss function with the weight update is designed as a function of time. + +# 2.1 SINGLE NEURON CASE + +We start with a simplistic single neuron case. Let $x \in \mathbb { R } ^ { n }$ be the input to the network where $x = [ x _ { 1 } \quad x _ { 2 } \quad \ldots \quad x _ { n } ] ^ { \top } , | x _ { i } | < c , i = 1 , 2 , \cdots n$ $c \in \mathsf { \Gamma } ( 0 , \infty )$ . Let with i $y ^ { \star }$ beuts e target oand bias tput and . The de $\textstyle z = \sum _ { i = 1 } ^ { n } w _ { i } x _ { i } + b$ holds true for some a priori but arbitrary scalar be the linear comction used here is $w _ { i }$ $x _ { i }$ $b$ $\mathrm { s i g n } ( x ) =$ $1 , \forall x > 0 , { \mathrm { s i g n } } ( x ) = - 1 , \forall x < 0 , { \mathrm { s i g n } } ( x ) \in [ - 1 , 1 ] , x = 0$ . For our analysis, we choose sigmoid function as our activation function, i.e. $\begin{array} { r } { \sigma ( z ) \doteq \frac { 1 ^ { - } } { 1 + e ^ { - z } } } \end{array}$ . The output of the neural network is given by $y = \sigma ( z )$ . Let the error in output be defined as $\bar { e } = y - y ^ { * }$ . The first objective is to convert our loss function into a candidate Lyapunov function in order to apply the control theoretic principles. Consider a continuous function $E ( \bar { e } )$ to be a candidate Lyapunov function as follows: + +$$ +E = \frac { | \bar { e } | ^ { ( \alpha + 1 ) } } { ( \alpha + 1 ) } +$$ + +where $\alpha \in ( 0 , 1 )$ is a user-defined parameter. The second objective is to define the temporal rate of weight as the control input to enforce the stability of the origin $\bar { e } = 0$ as $t \to \infty$ . The Lyapunov function in (1) is used to show that it is indeed plausible to achieve this asymptotic stability goal. Taking the temporal derivative of the candidate Lyapunov function (1) produces (by chain rule of differentiation) + +$$ +\frac { \mathrm { d } E } { \mathrm { d } t } = \frac { \mathrm { d } E } { \mathrm { d } \bar { e } } \frac { \mathrm { d } \bar { e } } { \mathrm { d } y } \frac { \mathrm { d } y } { \mathrm { d } z } \frac { \mathrm { d } z } { \mathrm { d } w } \frac { \mathrm { d } w } { \mathrm { d } t } +$$ + +where $\frac { \mathrm { d } E } { \mathrm { d } \bar { e } } \frac { \mathrm { d } \bar { e } } { \mathrm { d } y } \frac { \mathrm { d } y } { \mathrm { d } z } \frac { \mathrm { d } z } { \mathrm { d } w }$ is the weight update for standard gradient descent algorithm and $\textstyle { \frac { \mathrm { d } w } { \mathrm { d } t } }$ is the additional term we introduce to the weight update equation. Using (1), we get: + +$$ +\frac { d E } { d t } = | \bar { e } | ^ { \alpha } \mathrm { s i g n } ( \bar { e } ) \left( \frac { e ^ { - z } } { ( 1 + e ^ { - z } ) ^ { 2 } } \right) \left( x _ { 1 } \dot { w } _ { 1 } + x _ { 2 } \dot { w } _ { 2 } + \cdot \cdot \cdot + x _ { n } \dot { w } _ { n } \right) +$$ + +Define, $u _ { 1 } \triangleq \dot { w } _ { 1 } , u _ { 2 } \triangleq \dot { w } _ { 2 } , \cdot \cdot \cdot , u _ { n } \triangleq \dot { w } _ { n }$ , and + +$$ +u _ { i } = - k _ { i } \mathrm { s i g n } ( x _ { i } ) \mathrm { s i g n } ( \bar { e } ) e ^ { z } ( 1 + e ^ { - z } ) ^ { 2 } , \quad i = 1 , 2 , \cdots , n , +$$ + +where $k _ { i } > 0$ , for all $i$ , are tuning parameters to be chosen by the user. It can be noted that all control inputs $u _ { 1 } , u _ { 2 } , \cdots , u _ { n }$ remain bounded due to boundedness assumption of all the inputs $x _ { i }$ and that of $e ^ { z }$ . Substituting (3) into (2) produces + +$$ +\frac { d E } { d t } = - | \bar { e } | ^ { \alpha } \left( \sum _ { i = 1 } ^ { n } k _ { i } | x _ { i } | \right) +$$ + +Assumption 1. At least one input of all $x _ { i } , i = 1 , 2 , \cdots , n$ is non-zero such that $| x _ { j } | > \gamma > 0$ where $\gamma$ is a priori known scalar for some integers $j \in [ 1 , n ]$ . + +It can be noted that Assumption 1 is reasonable for many practical applications in that some inputs will always be nonzero with a known lower bound on its magnitude. First main result of the paper is in order. + +Theorem 1. Assuming Assumption 1 holds true, let the output of the neural network be given by $y = \sigma ( z )$ . Let all the inputs $x _ { i } , i = 1 , 2 , \cdots , n$ be bounded by some a priori known scalar $a \in$ $( 0 , \infty )$ such that $| x _ { i } | < a$ holds true for all $i$ . Then, weight update (3) causes the error $\bar { e } = y - y ^ { * }$ to converge to zero in finite time. + +Proof. The proof of the theorem is furnished using standard Lyapunov analysis arguments. Consider $E$ defined by (1) as a candidate Lyapunov function. Observing (4), it can be concluded that the right hand side of the temporal derivative of remains negative definite since it involves power terms and norm of inputs. Furthermore, (4) can be rewritten under the assumption 1 as follows: + +$$ +\frac { d E } { d t } \leq - k _ { \operatorname* { m i n } } \gamma E ^ { \beta } +$$ + +where $\begin{array} { r } { k _ { \operatorname* { m i n } } = \operatorname* { m i n } ( k _ { i } ) , i = 1 , 2 , \cdots , n , \beta = \frac { \alpha } { \alpha + 1 } } \end{array}$ = αα+1 and |e¯|α = |e¯|α+1 αα+1 = Eβ has been utilized. Noting that $E$ is a positive definite function and scalars $k _ { \mathrm { m i n } }$ and $\gamma$ are always positive, the proof is complete by applying (Bhat & Bernstein, 2000, Theorem 4.2). □ + +# 2.2 MULTI NEURON CASE + +Consider a multi-layer perceptron with $N$ layers where the layers are connected in a feed-forward manner (Bishop, 1995, Chapter 4). Let $\boldsymbol { x } \in \mathbb { R } ^ { n }$ be the input to the network where $x \quad =$ $\left[ x _ { 1 } \quad x _ { 2 } \quad \cdots \quad x _ { n } \right] ^ { \top }$ and $| x _ { i } | < c , i = 1 , 2 , \cdot \cdot \cdot n$ holds true for some a priori but arbitrary scalar $c \in$ $( 0 , \infty )$ . Let $y \in \mathbb { R } ^ { m }$ define the multi-neuron output to the network where $y = [ y _ { 1 } \quad y _ { 2 } \quad \cdot \cdot \cdot \quad y _ { m } ]$ . Let $y ^ { \star } \in \mathbb { R } ^ { m }$ define the target output values of the network where $y ^ { \star } = [ y _ { 1 } ^ { \star } \quad y _ { 2 } ^ { \star } \quad \cdot \cdot \quad y _ { m } ^ { \star } ]$ . The error in the output layer can be expressed as $\bar { e } = [ \left| y _ { 1 } - y _ { 1 } ^ { \star } \right| \quad \left| y _ { 2 } - y _ { 2 } ^ { \star } \right| \quad \cdot \cdot \quad \left| y _ { m } - y _ { m } ^ { \star } \right| ]$ . Hence, the scalar candidate Lyapunov function can be written as follows: + +$$ +E = E _ { 1 } + \cdot \cdot \cdot + E _ { m } = { \frac { | { \bar { e } } _ { 1 } | ^ { ( \alpha + 1 ) } } { ( \alpha + 1 ) } } + \cdot \cdot \cdot + { \frac { | { \bar { e } } _ { m } | ^ { ( \alpha + 1 ) } } { ( \alpha + 1 ) } } +$$ + +As is usually done in the case of feed-forward networks, consider unit $j$ of layer $l$ that computes its output + +$$ +a _ { j } ^ { l } = \sum _ { i } w _ { j i } ^ { l } z _ { i } ^ { l } +$$ + +using its inputs $z _ { i } ^ { l }$ from layer $l$ where bias parameter has been embedded inside the linear combination and $z _ { j } ^ { l } = \sigma ( a _ { j } ^ { l - 1 } )$ , where $\sigma$ is non-linear activation function. The aim of this section is to extend the single neuron case to a multi-neuron one. The simplest way do achieve this is to find sensitivity of $E$ to the weight $w _ { j i } ^ { l }$ of a given layer $l$ , which is given by + +$$ +\frac { \partial E _ { m } } { \partial w _ { j i } ^ { l } } = \frac { \partial E _ { m } } { \partial a _ { j } ^ { l } } \frac { \partial a _ { j } ^ { l } } { \partial w _ { j i } ^ { l } } +$$ + +Using standard notation $\begin{array} { r } { \delta _ { j } ^ { l } \triangleq \frac { \partial E _ { m } } { \partial a _ { j } ^ { l } } } \end{array}$ with (7) results in: + +$$ +\frac { \partial E _ { m } } { \partial w _ { j i } ^ { l } } = \delta _ { j } ^ { l } z _ { i } ^ { l } +$$ + +It is straightforward to compute $\delta _ { m } ^ { L }$ that belongs to the output layer $L$ as shown below: + +$$ +\delta _ { m } ^ { L } = \frac { \partial E _ { m } } { \partial a _ { m } ^ { L } } = \sigma ^ { \prime } ( a _ { m } ^ { L } ) \frac { \partial E _ { m } } { \partial y _ { m } } +$$ + +where $z _ { m } ^ { L }$ is replaced by $y _ { m }$ as it is the output layer. Finally, computation of $\delta _ { j } ^ { l }$ for all hidden units is given by + +$$ +\delta _ { j } ^ { l } = \frac { \partial E _ { m } } { \partial a _ { j } ^ { l } } = \sum _ { k } \frac { \partial E _ { m } } { \partial a _ { k } ^ { l + 1 } } \frac { \partial a _ { k } ^ { l + 1 } } { \partial a _ { j } ^ { l } } +$$ + +where units with a label $k$ includes either hidden layer units or an output unit in layer $l + 1$ . Combining (7), $z _ { j } ^ { l } = \sigma ( a _ { j } ^ { l } )$ and $\begin{array} { r } { \delta _ { j } ^ { l } \triangleq \frac { \partial E _ { m } } { \partial a _ { j } ^ { l } } } \end{array}$ produces + +$$ +\delta _ { j } ^ { l } = \sigma ^ { \prime } ( a _ { j } ^ { l } ) \sum _ { k } w _ { k j } ^ { l + 1 } \delta _ { k } ^ { l + 1 } +$$ + +A slightly modified version of assumption 1 is required before the next result of the paper is presented. + +Assumption 2. At least one input of all $z _ { i } , i = 1 , 2 , \cdots , L$ is non-zero such that $\left| z _ { n } \right| > \gamma > 0$ where $\gamma$ is a priori known scalar for some integers $n \in [ 1 , L ]$ where $L$ is the number of units in layer $l$ . + +Theorem 2. Let the weight update for connecting unit $i$ of layer $l$ to unit $j$ of layer $l + 1$ of $a$ multi-layer neural network be given by + +$$ +\begin{array} { r } { \dot { w } _ { j i } ^ { l } = - k _ { j i } ^ { l } \mathrm { s i g n } ( \delta _ { j } ^ { l } z _ { i } ^ { l } ) | \delta _ { j } ^ { l } z _ { i } ^ { l } | ^ { \alpha } E ^ { \beta } } \end{array} +$$ + +with some scalar $\beta \in ( 0 , 1 )$ such that $\alpha + \beta < 1$ and $k _ { j i } > 0$ is a tuning parameter. Then the output vector y converges to $y ^ { * }$ in finite time. + +Proof. Consider the candidate Lyapunov function $E$ given by (6). The temporal derivative of the Lyapunov function is given by + +$$ +\dot { E } = \sum _ { m } \dot { E } _ { m } = \sum _ { m } \frac { \partial E _ { m } } { \partial w _ { j i ^ { l } } } \dot { w } _ { j i } ^ { l } +$$ + +which can be simplified using (13) and (9) as + +$$ +\dot { E } = - E ^ { \beta } \sum _ { m } k _ { j i } ^ { l } | \delta _ { j } ^ { l } z _ { i } ^ { l } | ^ { \alpha + 1 } +$$ + +Using Assumption 2 it is easy to conclude that for some $k _ { \mathrm { m i n } } = \operatorname* { m i n } _ { i , j , l } k _ { j i } ^ { l } > 0$ , the following inequality holds true: + +$$ +\dot { E } \le - k _ { \mathrm { m i n } } \gamma ^ { \alpha + 1 } E ^ { \beta } +$$ + +Noting that $E$ is a positive definite function and scalars $k _ { \mathrm { m i n } }$ , $\gamma$ are always positive, the proof is complete by applying (Bhat & Bernstein, 2000, Theorem 4.2). □ + +Remark 1. It can be seen that weight update $( I 3 )$ (or respectively (3)) is in a feedback control form where state $z _ { i }$ and $\delta _ { j }$ (respectively $x _ { i }$ and $\bar { e }$ ) are being used for influencing the learning process. + +# 2.3 SETTLING TIME FOR NEURAL NETWORK TRAINING + +Theorem 1 and 2 prove that using candidate Lyapunov function as the loss function and the temporal rate of change of the loss function as the weight update equation, we can convert supervised learning framework into a control problem. (Bhat & Bernstein, 2000, Theorem 4.2) states that if there is continuous function that is positive definite and it’s rate of change with respect to time is negative definite, then it’s finite time stable equilibrium is at origin. The theorem also gives the settling time when the above conditions are satisfied by the control system. For our case, this means that the Lyapunov based loss function will converge in finite time and time taken to converge is as given below: + +The settling time for Single Neuron case: + +$$ +T \leq \frac { 1 } { k _ { m i n } \gamma ( 1 - \beta ) } E _ { i n i } ^ { ( 1 - \beta ) } +$$ + +The settling time for Multi Neuron case: + +$$ +T \leq \frac { 1 } { k _ { m i n } \gamma ^ { \alpha + 1 } ( 1 - \beta ) } E _ { i n i } ^ { ( 1 - \beta ) } +$$ + +where $\gamma$ is the lower bound on the inputs, $k _ { m i n }$ is the minimum value of tuning parameter $k , \alpha$ is the scalar used in the Lyapunov loss function and $E _ { i n i }$ is the initial value of loss, i.e. loss at $\mathrm { { t } } = 0$ . + +# 2.4 SENSITIVITY TO PERTURBATIONS + +This section considers the robustness of training a neural network based on our proposed algorithm. In control theoretic terms, perturbation can be understood either as a disturbance to the process or as modelling uncertainty. When supervised learning is viewed as a control process, perturbation in each neuron of the input layer can be seen as external noise. This perturbation could cause the control system to diverge. In this section, we develop theoretical claims that even with perturbed inputs, training a neural network with the proposed algorithm will converge. The following assumption of the upper bound on the perturbation of inputs is invoked. + +Assumption 3. There exists an a priori known constant $M \ > \ 0$ such that all inputs $x _ { i } , i \ =$ $1 , 2 , \cdots , N$ admit additive perturbations $\Delta x _ { i }$ such that + +$$ +| \Delta x _ { i } | \leq M | x _ { i } | ^ { \alpha } +$$ + +for all i where $N$ is the number of inputs. + +The following result is in order. + +Theorem 3. Let assumptions 2 and 3 hold true. Let the weight update for connecting unit i of layer $l$ to unit $j$ of layer $l + 1$ of a multi-layer neural network be given by (13). Then the output vector $y$ converges to $y ^ { * }$ in finite time in the presence of additive perturbations $\Delta x _ { n } , n \in \left[ 1 , L \right] i \bar { f } k _ { \operatorname* { m i n } } > M$ . + +Proof. It can be seen that the perturbation is considered only in inputs. Hence all hidden layer weights are updated as done in the proof of Theorem 2. Hence, the dynamics of learning results in the following revised temporal derivative of Lyapunov function: + +$$ +\dot { E } = - E ^ { \beta } \sum _ { m } k _ { j i } | \delta _ { j } z _ { i } | ^ { \alpha + 1 } + \sum _ { p = 0 } ^ { n } k _ { 1 p } | \delta _ { p } x _ { p } | ^ { \alpha } \mathrm { s i g n } ( \delta _ { p } x _ { p } ) \delta _ { p } \Delta x _ { p } +$$ + +where $k _ { 1 p }$ is the gain parameter for training of all the input layer weights. The expression in (20) can be simplified using Assumption 3 as follows: + +$$ +\begin{array} { l } { \displaystyle \dot { E } \leq - E ^ { \beta } \sum _ { m } k _ { j i } | \delta _ { j } z _ { i } | ^ { \alpha + 1 } + E ^ { \beta } \sum _ { p = 0 } ^ { n } { k _ { 1 p } | \delta _ { p } x _ { p } | ^ { \alpha + 1 } M } , } \\ { \displaystyle \leq - E ^ { \beta } \sum _ { m } ( k _ { j i } - M ) | \delta _ { j } z _ { i } | ^ { \alpha + 1 } } \end{array} +$$ + +where inputs $z _ { i }$ now collect inputs $x _ { i }$ as well. Similar to the proof of Theorem 2, (21) can be re-written by applying Assumption 2 as follows: + +$$ +\dot { E } \le - ( k _ { \operatorname* { m i n } } - M ) \gamma E ^ { \beta } +$$ + +Since $k _ { \operatorname* { m i n } { } } > M$ , the proof is complete by applying (Bhat & Bernstein, 2000, Theorem 4.2). + +# 3 EXPERIMENTS + +This section presents empirical evidence that the theoretical results derived in the above sections hold for real-life applications. + +Convergence Rate for training. To analyze the training convergence rate of the proposed method, we train a multi-layer perceptron (covered in Section 2.2) to perform regression on the Boston Housing dataset (Harrison Jr & Rubinfeld (1978)). Experiment on single neuron case is covered in Appendix A.2.2 + +Training details. We compare our proposed method (Lyapunov based loss and modified weight update equation) with traditional $L _ { 1 }$ , $L _ { 2 }$ loss functions and standard SGD weight update equation. The modifications to the weight update equation for Lyapunov loss are as stated in (15) for multi-neuron case. We plot the training and test loss with respect to time in Figure 1. + +The details about the training examples, epochs, learning rate etc. are specified under the description of the figures. We observe that the Lyapunov Loss function converges faster than $L _ { 1 }$ and $L _ { 2 }$ loss functions, demonstrating the results proven extensively in Section 2. This is attributed mainly to the non-Lipschitz weight updates given by (3) and (13). + +The settling time as well as metrics of the trained networks are reported in Table 1. The time taken by the proposed algorithm to converge falls within the a priori upper bound. We also observe that the metrics (Accuracy for single neuron case and rmse (Root mean square error) for multi-neuron case) achieved by the proposed method is better than the baselines. Admittedly, the theoretical upper bound on settling time for single neuron and multi-layer perceptron produced by (5) and (16) are very conservative, yet it is an a priori deterministic upper bound and the aggressive learning proves to be faster than traditional loss functions in the scenarios considered here. + +![](images/94677af0d4160d1d0f013e93407e9bd174fcde65f7757d87a59d03e10d863a68.jpg) +Figure 1: Comparison of convergence with respect to time for the multi-layer perceptron trained on the Boston Housing dataset. All networks are trained for 4000 epochs with learning rate $= 0 . 0 2$ and $\alpha = 0 . 6 8$ . + +
ExperimentTheoretical Upper BoundExp. Convergence Time (in seconds)Metric
(in seconds)L1L2Lyap.L1L2Lyap.
SingleNeuron (acc)~3.4e30.0650.0640.0250.950.951.0
MLP (rmse)~ 1.4e104.3e33.4e33.1e30.0910.1790.085
+ +Table 1: Settling time in seconds for each experiment conducted. We compare the time taken for convergence by three different loss functions, $L _ { 1 }$ , $L _ { 2 }$ and Lyapunov Loss function. The training conditions were similar for individual cases in the experiment. The single neuron case is trained on Iris dataset and the metric used is Accuracy (higher is better). For Multi-layer perceptron, we use Boston Housing dataset and the metric used for this is rmse (lower is better). + +This can be attributed to the fact that we operate in the realm of non-smooth functions which tend to be more aggressive than the smooth $L _ { 2 }$ function we usually encounter in most learning problems. We can clearly observe that the loss value does not converge to zero. According to (Bhat & Bernstein, 2000, Theorem 5.2), a control system will converge to a non-zero value at steady state, if there is constant perturbation in the input. Usually in real life datasets, there is an inherent bias or constant perturbation, which prevents the loss to converge to zero. Of course, by setting $\alpha = 0$ , we can reject all persisting disturbances (Orlov (2005)) and force the loss to converge to zero, but this may result in discontinuity in back propagation (discussed in Appendix A.1) and result into large numerical errors. + +Hyperparameter analysis and its effect on training. In this section, we attempt to give an empirical analysis as to how changes in the hyperparameters like $k$ and $\alpha$ affect the training of the neural network. In order to study the effect of tuning $k$ , we work with the single neuron case for simplicity and ease of understanding. We use Iris dataset for this experiment Dua & Graff (2017). Existing literature in control theory suggests that a monotonic increase in the tuning parameter, $k$ should result in a monotonic decrease in the corresponding loss. In neural network terminology, $k$ can be considered as the ’learning rate’. We can clearly see in Figure 2 that $k$ behaves like learning rate parameter where increasing $k$ makes the learning more aggressive. + +![](images/a553b1180d6f11315ea77c381a836a1152ae4f595414a80947015fee38d20db9.jpg) +Figure 2: Comparison of training and test convergence with respect to time for different values of tuning parameter, $k$ in a single neuron trained on the Iris dataset. We convert the problem into a binary classification problem by considering only two output classes. All networks are trained for 600 epochs with 80 training examples and 20 test examples. The learning rate for all $L _ { 1 }$ , $L _ { 2 }$ and Lyapunov is set to the tuning parameter value $k$ and $\alpha = 0 . 8$ + +From Table 2, we observe that the settling time for the training loss of our Lyapunov loss function is always less than $L _ { 1 }$ and $L _ { 2 }$ . We also observe that the Accuracy of our method is either better or similar to the baselines, which shows that the optimized solution achieved by our method is either better or on par with the baselines. For the multi-neuron case, we have defined $k _ { i j } ^ { l }$ for every weight. Similar to adaptive learning rate, we can either tune $k _ { i j } ^ { l }$ for every weight (which can be quite cumbersome for larger neural networks) or we can set it arbitrarily to the same value for all weights. It must be noted that $k _ { i j } ^ { l }$ can be set to any value greater than an a priori known positive constant $M$ . + +
Experiment for single neuron case on Iris datasetTheoretical Upper Bound (in seconds)Exp. Convergence Time (in seconds)Accuracy on Test Set
L1L2Lyap.L1L2Lyap.
k=0.001~ 37213.8480.3120.3100.09180.60.950.95
k = 0.005~7442.7690.09430.09420.03440.951.01.0
k= 0.01~ 3721.3850.05660.05730.02210.951.01.0
+ +Table 2: Settling time in seconds for different values of tuning parameter, $k$ for the single neuron case on Iris dataset. We compare the time taken for convergence by three different loss functions, $L _ { 1 }$ , $L _ { 2 }$ and Lyapunov Loss function. The training conditions were similar for individual cases in the experiment. + +We will discuss the impact of different values of $\alpha$ on the training process of our neural network. For the given experiments, we tested the network for various values of $\alpha$ ranging between $( 0 , 1 )$ . We observed that the training loss follows equations proved in above sections as long as $\alpha \in ( 0 . 5 , 0 . 9 )$ . If $\alpha$ happens to be too close to zero, the resulting control law becomes discontinuous thereby resulting in numerical instabilities owing to the fact that the current ODE solvers are unable to handle functions that are discontinuous. More details about the case $\alpha = 0$ is given in Appendix A.1 (Bhat & Bernstein, 2000, Theorem 5.2). + +![](images/eb0cb9a2a0c72346b1743ee9223cfec4cf4ee23758e8756964bd84b532fb7c69.jpg) +Stability of learning in the presence of amplitude-bounded perturbations. +Figure 3: Comparison of training loss convergence with respect to time for different values of input perturbations, $\Delta x$ for a multi-layer perceptron trained on the IMDB Wiki Faces dataset.The a priori upper bound on input perturbations are as follows: (a) $\Delta x = 0 . 1$ , (b) $\Delta x = 0 . 2$ , $( \mathrm { c } ) \Delta x = 0 . 3$ Learning rate $k = 0 . 0 0 0 9$ , $\alpha = 0 . 8$ , epochs $_ { \mathrm { - } 1 0 0 }$ + +In this section, we present results that demonstrate the robustness of proposed algorithm to bounded perturbations in the input while maintaining convergence of training dynamics. For this case, we train a multi-layer perceptron on a regression task of predicting the age given the image of a face. We use IMDB Wiki Faces Dataset Rothe et al. (2015) which has over 0.5 million face images of celebrities with their age and gender labels. Only part of the dataset is used for this experiment, i.e. 20,000 images for training and 4,000 images each for validation and test. + +For adding input perturbations, we asssume that an a priori upper bound, $M$ is known on the amplitude of possible input perturbations following Assumption 3. We add input perturbations to each pixel in the image using a randomized uniform distribution ranging from $( - \Delta x , \Delta x )$ . Hence, our additive noise remains in the abovementioned a priori bounds. We vary the values of $M$ from (0.1, 0.3) with a 0.1 increment, giving us three training cases. Figure 3 shows that the proposed training loss still manages to achieve steady state error and does not diverge due to perturbations introduced in the input dataset. Since we are dealing with noisy data, the loss values converge to a non-zero value in the steady-state + +Experiments on larger datasets. In this section, we present results to demonstrate the performance of our proposed algorithm on a larger dataset. We use the entire IMDB Wiki Faces Dataset Rothe et al. (2015) with 0.5 million images for this experiment. Training dataset consists of ${ \sim } 0 . 2$ million images and the test and validation set consists of ${ \sim } 0 . 1$ million images each. As described in the previous section, a multi-layer perceptron is trained to predict the age from the image of a face. The MLP is trained for 100 epochs with learning rate 0.0005 for all three loss functions and with $\alpha = 0 . 7$ for Lyapunov loss function. From Figure 4 and Table 3, we can see that our proposed algorithm achieves similar generalization / testing rmse (Root mean square error) as $L _ { 1 }$ and $L _ { 2 }$ baselines while providing finite time convergence guarantees even on large datasets. The convergence time is also within the theoretically derived upper bound. + +![](images/5b5329f6c05dd7e9ed3e8faf2625740921023501de47d20852c27b587256edbe.jpg) +Figure 4: Comparison of training convergence with respect to time for 0.5 million IMDB Wiki dataset + +
ExperimentTheoretical Upper BoundExp. Convergence Time (in seconds)Metric
(in seconds)L1L2Lyap.L1rmse L2Lyap.
IMDBWiki8.3e6980.40712.99467.960.4140.4150.416
+ +Table 3: rmse on IMDB Wiki test dataset for $L _ { 1 }$ , $L _ { 2 }$ and Lyapunov Loss function. + +# 4 CONCLUSION AND FUTURE WORK + +This paper studies the training of a deep neural network from control theory perspective. We pose the supervised learning problem as a control problem by jointly designing loss function as Lyapunov function and weight update as temporal derivative of the Lyapunov function. Control theory principles are then applied to provide guarantees on finite time convergence and settling time of the neural network. Through experiments on benchmark datasets, our proposed method converges within the a priori bounds derived from theory. It is also observed that in some cases our method enforces faster convergence as compared to standard $L _ { 1 }$ and $L _ { 2 }$ loss functions. We also prove that our method is robust to any perturbations in the input and convergence guarantees still hold true. The given a priori guarantees for the convergence time is a desirable result for training networks that are extremely difficult to converge, specifically in Reinforcement Learning. A future scope of this work may be to convert the continuous time analysis framework to discrete time. This study introduces a novel perspective of viewing neural networks as control systems and opens up the field of machine learning research to a plethora of new results that can be derived from control theory. + +# REFERENCES + +Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. On the convergence rate of training recurrent neural networks. In Advances in Neural Information Processing Systems, pp. 6673–6685, 2019. + +Sanjeev Arora, Nadav Cohen, Noah Golowich, and Wei Hu. A convergence analysis of gradient descent for deep linear neural networks. arXiv preprint arXiv:1810.02281, 2018. + +S.P. Bhat and D.S. Bernstein. Finite-Time Stability of Continuous Autonomous Systems. SIAM Journal of Control and Optimization, 38(3):751–766, 2000. + +Christopher M. Bishop. Neural Networks for Pattern Recognition. OXFORD University Press, 1995. + +Simon S Du, Jason D Lee, Haochuan Li, Liwei Wang, and Xiyu Zhai. Gradient descent finds global minima of deep neural networks. arXiv preprint arXiv:1811.03804, 2018. + +Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive. ics.uci.edu/ml. + +Dongsheng Guo, Chenfu Yi, and Yunong Zhang. Zhang neural network versus gradient-based neural network for time-varying linear matrix equation solving. Neurocomputing, 74(17):3708–3712, 2011. + +Harsh Gupta, R Srikant, and Lei Ying. Finite-time performance bounds and adaptive learning rate selection for two time-scale reinforcement learning. In Advances in Neural Information Processing Systems, pp. 4706–4715, 2019. + +Eldad Haber and Lars Ruthotto. Stable architectures for deep neural networks. Inverse Problems, 34(1):014004, 2017. + +David Harrison Jr and Daniel L Rubinfeld. Hedonic housing prices and the demand for clean air. 1978. + +Arthur Jacot, Franck Gabriel, and Clement Hongler. Neural tangent kernel: Convergence and gen- ´ eralization in neural networks. In Advances in neural information processing systems, pp. 8571– 8580, 2018. + +Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, and Hoi-To Wai. Finite time analysis of linear two-timescale stochastic approximation with markovian noise. arXiv preprint arXiv:2002.01268, 2020. + +Yann LeCun, D Touresky, G Hinton, and T Sejnowski. A theoretical framework for backpropagation. In Proceedings of the 1988 connectionist models summer school, volume 1, pp. 21–28. CMU, Pittsburgh, Pa: Morgan Kaufmann, 1988. + +Shuai Li, Sanfeng Chen, and Bo Liu. Accelerating a recurrent neural network to finite-time convergence for solving time-varying sylvester equation by using a sign-bi-power activation function. Neural processing letters, 37(2):189–205, 2013. + +Yuanzhi Li and Yang Yuan. Convergence analysis of two-layer neural networks with relu activation. In Advances in neural information processing systems, pp. 597–607, 2017. + +Guan-Horng Liu and Evangelos A Theodorou. Deep learning theory review: An optimal control and dynamical systems perspective. arXiv preprint arXiv:1908.10920, 2019. + +Yury Orlov. Finite-Time Stability and Robust Control Synthesis of Uncertain Switched Systems. SIAM Journal of Control and Optimization, 43(4):1253–1271, 2005. + +Rasmus Rothe, Radu Timofte, and Luc Van Gool. Dex: Deep expectation of apparent age from a single image. In IEEE International Conference on Computer Vision Workshops (ICCVW), December 2015. + +Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. arXiv preprint arXiv:1312.6120, 2013. + +Gang Wang, Bingcong Li, and Georgios B Giannakis. A multistep lyapunov approach for finite-time analysis of biased stochastic approximation. arXiv preprint arXiv:1909.04299, 2019. + +Lin Xiao. Accelerating a recurrent neural network to finite-time convergence using a new design formula and its application to time-varying matrix square root. Journal of the Franklin Institute, 354(13):5667–5677, 2017. + +Bin Xu, Chenguang Yang, and Zhongke Shi. Reinforcement learning output feedback nn control using deterministic learning technique. IEEE Transactions on Neural Networks and Learning Systems, 25(3):635–641, 2013. + +Yunong Zhang, Danchi Jiang, and Jun Wang. A recurrent neural network for solving sylvester equation with time-varying coefficients. IEEE Transactions on Neural Networks, 13(5):1053– 1063, 2002. + +# A APPENDIX + +# A.1 IMPACT OF $\alpha$ ON TRAINING + +The case when $\alpha = 0$ , as depicted in Figure 5, raises continuity issues. + +![](images/dba33dd8e19f76befd2fd50e3232f785f2f61abf9d083b943e646c81f5a3789d.jpg) +Figure 5: Depiction of the numerical instability observed when we take $\alpha = 0$ . Effectively, at this point, the loss function contains a discontinuous signum function. + +Consider the left and right limits of the function $f ( \varrho ) = | \varrho | ^ { \alpha } \mathrm { s i g n } ( \varrho )$ , where $\varrho = \delta _ { j } z _ { i }$ as appearing in (15). It can be seen that $\begin{array} { r } { \operatorname* { l i m } _ { \varrho \to 0 ^ { - } } f ( \varrho ) = \operatorname* { l i m } _ { \varrho \to 0 ^ { + } } f ( \varrho ) = 0 } \end{array}$ for $\alpha \in ( 0 , 1 )$ . However, for $\alpha = 0$ , $\begin{array} { r } { \operatorname* { l i m } _ { \varrho \to 0 ^ { - } } f ( \varrho ) = - 1 } \end{array}$ and $\begin{array} { r } { \operatorname* { l i m } _ { \varrho \to 0 ^ { + } } f ( \varrho ) = 1 } \end{array}$ . It should be noted that function $f ( \varrho )$ is non-Lipschitz since $\partial f / \partial \varrho$ tends to infinity in the limit $\varrho \to 0$ . We agree that $\alpha = 0$ reduces the loss function to L1 loss, but the control update becomes purely discontinuous due to the presence of $f ( \varrho )$ in (15). We do not deal with this case for the reasons of continuity as mentioned in the main paper. + +# A.2 EXPERIMENTS + +# A.2.1 EXPERIMENTAL SETUP + +For our experiments, we use the NVIDIA GEFORCE GTX1080Ti GPU card with 12GB RAM, 4- core CPU with 32GB of RAM for training. The training code is in Python. All networks are trained on data with an $8 0 \% - 2 0 \%$ train data-test data split. + +# A.2.2 CONVERGENCE EXPERIMENTS ON SINGLE NEURON + +In this experiment, we perform binary classification on the Iris dataset (Dua & Graff (2017)) as a representative example of single neuron case presented in Section 2.1. + +![](images/484caabc86462e795402029c89b573f150b284c3ff0f88ef33e68e42ffbeaa24.jpg) +Figure 6: Comparison of convergence with respect to time for a single neuron trained on the Iris dataset. We convert the problem into a binary classification problem by considering only two of the three output classes. All networks are trained for 2100 epochs with 80 training examples and 20 test examples, $\alpha = 0 . 8$ , learning rate $/ \operatorname { k } = 0 . 0 1$ + +From the Figure 6, we can clearly see that Lyapunov loss function with control weight update converges much faster towards zero as compared to $L _ { 1 }$ and $L _ { 2 }$ loss with standard gradient descent weight update. This shows that explicitly adding a control weight update (15) that drives the loss towards zero is helpful in reaching convergence faster. + +# A.2.3 BOSTON HOUSING DATASET EXPERIMENT WITH PERTURBATIONS + +The Boston Housing Dataset predicts the price of houses in various areas of Boston Mass, given 14 different relevant attributes, like per capita crime rate and pupil-teacher ratio by town. The 506 examples in the data are divided into a 80-20 training-test split to give us 505 training examples and 101 test examples respectively. We consider three cases here, $M = 0 . 1$ , $M = 0 . 2$ ,and $M = 0 . 3$ . Table 4 presents the theoretical upper bounds for the proposed Lyapunov function’s settling time and experimental results for the training conducted for different upper bounds assumed for the input perturbation. We observe that the settling time for the proposed Lyapunov function is similar to the one observed for $L _ { 2 }$ loss function whereas it performs way better than the $L _ { 1 }$ loss function. + +
Experiment for MLP case on BostonTheoretical Upper Bound (in seconds)Exp. Convergence Time to within 10e-9 (in seconds)Metric rmse
L1L2Lyap.L1L2Lyap.
dataset M=0.1~1.97e111277.66906.61755.590.0950.1450.092
M= 0.2~ 5.67e101297.111097.62914.780.0960.1460.093
M= 0.3~2.73e101290.481197.53998.050.0930.1470.101
+ +Table 4: Settling time in seconds for different values of the upper bound on additive input perturbations, $M$ for the multi layer perceptron case on Boston Housing dataset. We compare the time taken for convergence by three different loss functions, $L _ { 1 }$ , $L _ { 2 }$ and Lyapunov Loss function. The training conditions were similar for individual cases in the experiment. + +![](images/11d49f7bb289d1b3734f3bf88fc7219a72fb3e0346b5a0100b0d547b4a55cde2.jpg) +Figure 7: Pictorial representation of the effect of random additive noise with an upper bound of 0.2 on IMDB Wiki Faces Dataset. In this experiment, we work with monochrome images for the sake of simplicity. + +Figure 7 shows us how the noise affects the training images. We take the specific case where $M =$ 0.2. We can clearly observe that the images obtained after adding input perturbations happens to be quite noisy. Figure 8 and Figure 9 show that even when inputs are perturbed, our proposed method converges in finite time. + +![](images/97c396dd6181b32b466eaaa868e746b184b69235d7d298d2c9dd76c686d88565.jpg) +Figure 8: Comparison of training loss convergence with respect to time for different values of input perturbations, $\Delta x$ for a multi-layer perceptron trained on the IMDB Wiki Faces dataset. The a priori upper bound on input perturbations are as follows: (a) $\Delta x = 0 . 1$ , (b) $\Delta x = 0 . 2$ , $( \mathrm { c } ) \Delta x = 0 . 3$ , (d) $\Delta x = 0 . 4$ , (e) $\Delta x = 0 . 5$ , and (f) $\Delta x = 1 . 2$ . + +![](images/73d5133f391dd03ab63be601efd6eb5e468d1b8b4a9c16d108dbdd1ddc4e42f2.jpg) +Figure 9: Comparison of test loss convergence with respect to time for different values of input perturbations, $\Delta x$ for a multi-layer perceptron trained on the IMDB Wiki Faces dataset.The a priori upper bound on input perturbations are as follows: (a) $\Delta x = 0 . 1$ , (b) $\Delta x = 0 . 2$ , $( \mathrm { c } ) \Delta x = 0 . 3$ , (d) $\Delta x = 0 . 4$ , (e) $\Delta x = 0 . 5$ , and (f) $\Delta x = 1 . 2$ . \ No newline at end of file diff --git a/parse/train/VRgITLy0l2/VRgITLy0l2_model.json b/parse/train/VRgITLy0l2/VRgITLy0l2_model.json new file mode 100644 index 0000000000000000000000000000000000000000..380df3dd0493ede76c252e8e3b14ca8fe52c6ae3 --- /dev/null +++ b/parse/train/VRgITLy0l2/VRgITLy0l2_model.json @@ -0,0 +1,24832 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 589, + 1302, + 589, + 1302, + 926, + 398, + 926 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1440, + 1404, + 1440, + 1404, + 1894, + 298, + 1894 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1056, + 1403, + 1056, + 1403, + 1422, + 298, + 1422 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 299, + 1911, + 1400, + 1911, + 1400, + 2034, + 299, + 2034 + ], + "score": 0.973 + }, + { + "category_id": 0, + "poly": [ + 301, + 223, + 1398, + 223, + 1398, + 324, + 301, + 324 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 313, + 378, + 680, + 378, + 680, + 438, + 313, + 438 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 302, + 986, + 573, + 986, + 573, + 1021, + 302, + 1021 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 104, + 298, + 104 + ], + "score": 0.881 + }, + { + "category_id": 0, + "poly": [ + 773, + 520, + 926, + 520, + 926, + 553, + 773, + 553 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 842, + 2089, + 857, + 2089, + 857, + 2112, + 842, + 2112 + ], + "score": 0.64 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 221.0, + 1403.0, + 221.0, + 1403.0, + 271.0, + 295.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 276.0, + 1100.0, + 276.0, + 1100.0, + 328.0, + 295.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 983.0, + 579.0, + 983.0, + 579.0, + 1030.0, + 294.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 519.0, + 932.0, + 519.0, + 932.0, + 556.0, + 769.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 860.0, + 2088.0, + 860.0, + 2118.0, + 841.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 589.0, + 1304.0, + 589.0, + 1304.0, + 625.0, + 394.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 619.0, + 1305.0, + 619.0, + 1305.0, + 658.0, + 392.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 649.0, + 1304.0, + 649.0, + 1304.0, + 686.0, + 392.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 679.0, + 1305.0, + 679.0, + 1305.0, + 719.0, + 393.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 712.0, + 1306.0, + 712.0, + 1306.0, + 745.0, + 394.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 744.0, + 1306.0, + 744.0, + 1306.0, + 776.0, + 393.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 773.0, + 1306.0, + 773.0, + 1306.0, + 806.0, + 394.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 804.0, + 1305.0, + 804.0, + 1305.0, + 836.0, + 394.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 834.0, + 1305.0, + 834.0, + 1305.0, + 867.0, + 394.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 865.0, + 1304.0, + 865.0, + 1304.0, + 897.0, + 394.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 894.0, + 1248.0, + 894.0, + 1248.0, + 930.0, + 393.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1437.0, + 1404.0, + 1437.0, + 1404.0, + 1474.0, + 293.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1469.0, + 1405.0, + 1469.0, + 1405.0, + 1504.0, + 293.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1500.0, + 1406.0, + 1500.0, + 1406.0, + 1533.0, + 293.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1530.0, + 1402.0, + 1530.0, + 1402.0, + 1564.0, + 294.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1559.0, + 1406.0, + 1559.0, + 1406.0, + 1595.0, + 292.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1589.0, + 1405.0, + 1589.0, + 1405.0, + 1626.0, + 293.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1619.0, + 1405.0, + 1619.0, + 1405.0, + 1656.0, + 293.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1650.0, + 1406.0, + 1650.0, + 1406.0, + 1687.0, + 293.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1682.0, + 1406.0, + 1682.0, + 1406.0, + 1716.0, + 294.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1712.0, + 1405.0, + 1712.0, + 1405.0, + 1747.0, + 292.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1742.0, + 1404.0, + 1742.0, + 1404.0, + 1777.0, + 294.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1809.0, + 293.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1804.0, + 1406.0, + 1804.0, + 1406.0, + 1836.0, + 293.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1835.0, + 1404.0, + 1835.0, + 1404.0, + 1870.0, + 293.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1864.0, + 838.0, + 1864.0, + 838.0, + 1900.0, + 293.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1053.0, + 1404.0, + 1053.0, + 1404.0, + 1093.0, + 295.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1086.0, + 1404.0, + 1086.0, + 1404.0, + 1123.0, + 293.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1116.0, + 1404.0, + 1116.0, + 1404.0, + 1154.0, + 293.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1146.0, + 1403.0, + 1146.0, + 1403.0, + 1183.0, + 293.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1177.0, + 1405.0, + 1177.0, + 1405.0, + 1214.0, + 293.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1209.0, + 1404.0, + 1209.0, + 1404.0, + 1243.0, + 294.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1243.0, + 1402.0, + 1243.0, + 1402.0, + 1273.0, + 296.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1266.0, + 1404.0, + 1266.0, + 1404.0, + 1307.0, + 292.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 1403.0, + 1298.0, + 1403.0, + 1335.0, + 293.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1330.0, + 1405.0, + 1330.0, + 1405.0, + 1366.0, + 293.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1362.0, + 1405.0, + 1362.0, + 1405.0, + 1396.0, + 294.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1389.0, + 795.0, + 1389.0, + 795.0, + 1428.0, + 293.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1402.0, + 1910.0, + 1402.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 1404.0, + 1941.0, + 1404.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2002.0, + 1405.0, + 2002.0, + 1405.0, + 2036.0, + 294.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 379.0, + 560.0, + 379.0, + 560.0, + 411.0, + 315.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 407.0, + 681.0, + 407.0, + 681.0, + 442.0, + 312.0, + 442.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 582, + 1404, + 582, + 1404, + 1009, + 298, + 1009 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1023, + 1404, + 1023, + 1404, + 1329, + 298, + 1329 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1404, + 229, + 1404, + 565, + 298, + 565 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1486, + 1403, + 1486, + 1403, + 1707, + 298, + 1707 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1804, + 1404, + 1804, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.979 + }, + { + "category_id": 0, + "poly": [ + 302, + 1380, + 1244, + 1380, + 1244, + 1451, + 302, + 1451 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 300, + 1746, + 640, + 1746, + 640, + 1778, + 300, + 1778 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 852, + 76, + 852, + 104, + 300, + 104 + ], + "score": 0.832 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.695 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.216 + }, + { + "category_id": 13, + "poly": [ + 863, + 2004, + 993, + 2004, + 993, + 2035, + 863, + 2035 + ], + "score": 0.93, + "latex": "\\bar { e } = y - y ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 332, + 2001, + 440, + 2001, + 440, + 2036, + 332, + 2036 + ], + "score": 0.92, + "latex": "y = \\sigma ( z )" + }, + { + "category_id": 13, + "poly": [ + 824, + 1874, + 1066, + 1874, + 1066, + 1912, + 824, + 1912 + ], + "score": 0.92, + "latex": "\\textstyle z = \\sum _ { i = 1 } ^ { n } w _ { i } x _ { i } + b" + }, + { + "category_id": 13, + "poly": [ + 704, + 1579, + 766, + 1579, + 766, + 1613, + 704, + 1613 + ], + "score": 0.91, + "latex": "V ( x )" + }, + { + "category_id": 13, + "poly": [ + 746, + 1968, + 913, + 1968, + 913, + 2005, + 746, + 2005 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\sigma ( z ) \\doteq \\frac { 1 ^ { - } } { 1 + e ^ { - z } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1613, + 412, + 1613, + 412, + 1647, + 297, + 1647 + ], + "score": 0.91, + "latex": "V ( x ) > 0" + }, + { + "category_id": 13, + "poly": [ + 892, + 1807, + 994, + 1807, + 994, + 1835, + 892, + 1835 + ], + "score": 0.91, + "latex": "x \\in \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1877, + 432, + 1877, + 432, + 1910, + 298, + 1910 + ], + "score": 0.9, + "latex": "c \\in \\mathsf { \\Gamma } ( 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 1316, + 1610, + 1394, + 1610, + 1394, + 1644, + 1316, + 1644 + ], + "score": 0.9, + "latex": "\\dot { V } < 0" + }, + { + "category_id": 13, + "poly": [ + 1278, + 1906, + 1403, + 1906, + 1403, + 1940, + 1278, + 1940 + ], + "score": 0.89, + "latex": "\\mathrm { s i g n } ( x ) =" + }, + { + "category_id": 13, + "poly": [ + 502, + 1878, + 533, + 1878, + 533, + 1909, + 502, + 1909 + ], + "score": 0.87, + "latex": "y ^ { \\star }" + }, + { + "category_id": 13, + "poly": [ + 393, + 1913, + 425, + 1913, + 425, + 1937, + 393, + 1937 + ], + "score": 0.85, + "latex": "w _ { i }" + }, + { + "category_id": 13, + "poly": [ + 567, + 1912, + 595, + 1912, + 595, + 1937, + 567, + 1937 + ], + "score": 0.83, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1938, + 971, + 1938, + 971, + 1971, + 297, + 1971 + ], + "score": 0.82, + "latex": "1 , \\forall x > 0 , { \\mathrm { s i g n } } ( x ) = - 1 , \\forall x < 0 , { \\mathrm { s i g n } } ( x ) \\in [ - 1 , 1 ] , x = 0" + }, + { + "category_id": 13, + "poly": [ + 705, + 1909, + 720, + 1909, + 720, + 1934, + 705, + 1934 + ], + "score": 0.75, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 608, + 1838, + 871, + 1838, + 871, + 1875, + 608, + 1875 + ], + "score": 0.74, + "latex": "| x _ { i } | < c , i = 1 , 2 , \\cdot \\cdot \\cdot n" + }, + { + "category_id": 13, + "poly": [ + 296, + 1835, + 872, + 1835, + 872, + 1876, + 296, + 1876 + ], + "score": 0.67, + "latex": "x = [ x _ { 1 } \\quad x _ { 2 } \\quad \\ldots \\quad x _ { n } ] ^ { \\top } , | x _ { i } | < c , i = 1 , 2 , \\cdots n" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1374.0, + 1250.0, + 1374.0, + 1250.0, + 1420.0, + 289.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1423.0, + 797.0, + 1423.0, + 797.0, + 1456.0, + 348.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1746.0, + 644.0, + 1746.0, + 644.0, + 1782.0, + 295.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 579.0, + 1405.0, + 579.0, + 1405.0, + 619.0, + 293.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 613.0, + 1406.0, + 613.0, + 1406.0, + 649.0, + 294.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 642.0, + 1406.0, + 642.0, + 1406.0, + 678.0, + 293.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 673.0, + 1407.0, + 673.0, + 1407.0, + 708.0, + 292.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 704.0, + 1406.0, + 704.0, + 1406.0, + 740.0, + 294.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 735.0, + 1406.0, + 735.0, + 1406.0, + 770.0, + 294.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 766.0, + 1402.0, + 766.0, + 1402.0, + 798.0, + 294.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 793.0, + 1404.0, + 793.0, + 1404.0, + 831.0, + 293.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 827.0, + 1404.0, + 827.0, + 1404.0, + 859.0, + 296.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 858.0, + 1405.0, + 858.0, + 1405.0, + 889.0, + 296.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 887.0, + 1405.0, + 887.0, + 1405.0, + 922.0, + 293.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 916.0, + 1405.0, + 916.0, + 1405.0, + 950.0, + 293.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 948.0, + 1404.0, + 948.0, + 1404.0, + 983.0, + 294.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 977.0, + 1091.0, + 977.0, + 1091.0, + 1013.0, + 293.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1024.0, + 1404.0, + 1024.0, + 1404.0, + 1059.0, + 296.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1055.0, + 1405.0, + 1055.0, + 1405.0, + 1091.0, + 294.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1087.0, + 1402.0, + 1087.0, + 1402.0, + 1119.0, + 296.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1116.0, + 1406.0, + 1116.0, + 1406.0, + 1152.0, + 294.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1147.0, + 1406.0, + 1147.0, + 1406.0, + 1182.0, + 292.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1176.0, + 1406.0, + 1176.0, + 1406.0, + 1214.0, + 293.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1206.0, + 1405.0, + 1206.0, + 1405.0, + 1243.0, + 293.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1238.0, + 1404.0, + 1238.0, + 1404.0, + 1274.0, + 294.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1267.0, + 1406.0, + 1267.0, + 1406.0, + 1304.0, + 293.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 1072.0, + 1298.0, + 1072.0, + 1336.0, + 293.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 231.0, + 1405.0, + 231.0, + 1405.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 263.0, + 1405.0, + 263.0, + 1405.0, + 295.0, + 293.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 292.0, + 1405.0, + 292.0, + 1405.0, + 326.0, + 292.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 321.0, + 1406.0, + 321.0, + 1406.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 351.0, + 1405.0, + 351.0, + 1405.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 381.0, + 1407.0, + 381.0, + 1407.0, + 418.0, + 292.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 414.0, + 1404.0, + 414.0, + 1404.0, + 448.0, + 292.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 444.0, + 1405.0, + 444.0, + 1405.0, + 479.0, + 294.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 475.0, + 1405.0, + 475.0, + 1405.0, + 510.0, + 294.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 505.0, + 1406.0, + 505.0, + 1406.0, + 541.0, + 292.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 534.0, + 732.0, + 534.0, + 732.0, + 569.0, + 294.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1487.0, + 1405.0, + 1487.0, + 1405.0, + 1523.0, + 296.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1519.0, + 1405.0, + 1519.0, + 1405.0, + 1554.0, + 294.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1550.0, + 1403.0, + 1550.0, + 1403.0, + 1581.0, + 296.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1576.0, + 703.0, + 1576.0, + 703.0, + 1617.0, + 293.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1576.0, + 1404.0, + 1576.0, + 1404.0, + 1617.0, + 767.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1613.0, + 1315.0, + 1613.0, + 1315.0, + 1652.0, + 413.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1613.0, + 1405.0, + 1613.0, + 1405.0, + 1652.0, + 1395.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1642.0, + 1405.0, + 1642.0, + 1405.0, + 1682.0, + 293.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1673.0, + 1247.0, + 1673.0, + 1247.0, + 1711.0, + 293.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1803.0, + 891.0, + 1803.0, + 891.0, + 1839.0, + 296.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1803.0, + 1405.0, + 1803.0, + 1405.0, + 1839.0, + 995.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1839.0, + 295.0, + 1839.0, + 295.0, + 1879.0, + 292.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1839.0, + 1406.0, + 1839.0, + 1406.0, + 1879.0, + 873.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 1855.0, + 297.0, + 1855.0, + 297.0, + 1941.0, + 282.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1855.0, + 501.0, + 1855.0, + 501.0, + 1941.0, + 433.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1855.0, + 566.0, + 1855.0, + 566.0, + 1941.0, + 534.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1855.0, + 704.0, + 1855.0, + 704.0, + 1941.0, + 596.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1855.0, + 823.0, + 1855.0, + 823.0, + 1941.0, + 721.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1855.0, + 1277.0, + 1855.0, + 1277.0, + 1941.0, + 1067.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1855.0, + 1418.0, + 1855.0, + 1418.0, + 1941.0, + 1404.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1932.0, + 296.0, + 1932.0, + 296.0, + 1976.0, + 292.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1932.0, + 1409.0, + 1932.0, + 1409.0, + 1976.0, + 972.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1957.0, + 745.0, + 1957.0, + 745.0, + 2019.0, + 286.0, + 2019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1957.0, + 1413.0, + 1957.0, + 1413.0, + 2019.0, + 914.0, + 2019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 331.0, + 2000.0, + 331.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 2000.0, + 862.0, + 2000.0, + 862.0, + 2040.0, + 441.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 2000.0, + 1406.0, + 2000.0, + 1406.0, + 2040.0, + 994.0, + 2040.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1872, + 1405, + 1872, + 1405, + 2036, + 296, + 2036 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1457, + 1404, + 1457, + 1404, + 1582, + 297, + 1582 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 396, + 1405, + 396, + 1405, + 550, + 296, + 550 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1678, + 1404, + 1678, + 1404, + 1781, + 298, + 1781 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 898, + 1403, + 898, + 1403, + 992, + 297, + 992 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1299, + 1403, + 1299, + 1403, + 1424, + 297, + 1424 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1197, + 1405, + 1197, + 1405, + 1291, + 297, + 1291 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 696, + 1005, + 1001, + 1005, + 1001, + 1097, + 696, + 1097 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 294, + 1111, + 1405, + 1111, + 1405, + 1176, + 294, + 1176 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 294, + 228, + 1401, + 228, + 1401, + 294, + 294, + 294 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 705, + 547, + 994, + 547, + 994, + 618, + 705, + 618 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 743, + 1595, + 955, + 1595, + 955, + 1662, + 743, + 1662 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 626, + 1399, + 626, + 1399, + 696, + 299, + 696 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 770, + 308, + 929, + 308, + 929, + 385, + 770, + 385 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 477, + 709, + 1220, + 709, + 1220, + 786, + 477, + 786 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 514, + 846, + 1183, + 846, + 1183, + 887, + 514, + 887 + ], + "score": 0.933 + }, + { + "category_id": 0, + "poly": [ + 299, + 1816, + 632, + 1816, + 632, + 1849, + 299, + 1849 + ], + "score": 0.925 + }, + { + "category_id": 9, + "poly": [ + 1366, + 852, + 1400, + 852, + 1400, + 882, + 1366, + 882 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1366, + 332, + 1400, + 332, + 1400, + 363, + 1366, + 363 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1615, + 1400, + 1615, + 1400, + 1645, + 1366, + 1645 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1035, + 1400, + 1035, + 1400, + 1065, + 1366, + 1065 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1366, + 732, + 1400, + 732, + 1400, + 762, + 1366, + 762 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.85 + }, + { + "category_id": 1, + "poly": [ + 298, + 800, + 819, + 800, + 819, + 837, + 298, + 837 + ], + "score": 0.715 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.663 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.438 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 854, + 74, + 854, + 106, + 298, + 106 + ], + "score": 0.119 + }, + { + "category_id": 14, + "poly": [ + 768, + 306, + 930, + 306, + 930, + 386, + 768, + 386 + ], + "score": 0.96, + "latex": "E = \\frac { | \\bar { e } | ^ { ( \\alpha + 1 ) } } { ( \\alpha + 1 ) }" + }, + { + "category_id": 14, + "poly": [ + 695, + 1003, + 1005, + 1003, + 1005, + 1099, + 695, + 1099 + ], + "score": 0.95, + "latex": "\\frac { d E } { d t } = - | \\bar { e } | ^ { \\alpha } \\left( \\sum _ { i = 1 } ^ { n } k _ { i } | x _ { i } | \\right)" + }, + { + "category_id": 14, + "poly": [ + 704, + 544, + 997, + 544, + 997, + 619, + 704, + 619 + ], + "score": 0.94, + "latex": "\\frac { \\mathrm { d } E } { \\mathrm { d } t } = \\frac { \\mathrm { d } E } { \\mathrm { d } \\bar { e } } \\frac { \\mathrm { d } \\bar { e } } { \\mathrm { d } y } \\frac { \\mathrm { d } y } { \\mathrm { d } z } \\frac { \\mathrm { d } z } { \\mathrm { d } w } \\frac { \\mathrm { d } w } { \\mathrm { d } t }" + }, + { + "category_id": 14, + "poly": [ + 742, + 1594, + 957, + 1594, + 957, + 1661, + 742, + 1661 + ], + "score": 0.94, + "latex": "\\frac { d E } { d t } \\leq - k _ { \\operatorname* { m i n } } \\gamma E ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 904, + 1675, + 1201, + 1675, + 1201, + 1720, + 904, + 1720 + ], + "score": 0.94, + "latex": "| \\bar { e } | ^ { \\alpha } = \\left( | \\bar { e } | ^ { \\alpha + 1 } \\right) ^ { \\frac { \\alpha } { \\alpha + 1 } } = E ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 374, + 624, + 507, + 624, + 507, + 669, + 374, + 669 + ], + "score": 0.93, + "latex": "\\frac { \\mathrm { d } E } { \\mathrm { d } \\bar { e } } \\frac { \\mathrm { d } \\bar { e } } { \\mathrm { d } y } \\frac { \\mathrm { d } y } { \\mathrm { d } z } \\frac { \\mathrm { d } z } { \\mathrm { d } w }" + }, + { + "category_id": 13, + "poly": [ + 372, + 398, + 489, + 398, + 489, + 432, + 372, + 432 + ], + "score": 0.93, + "latex": "\\alpha \\in ( 0 , 1 )" + }, + { + "category_id": 14, + "poly": [ + 478, + 708, + 1220, + 708, + 1220, + 786, + 478, + 786 + ], + "score": 0.92, + "latex": "\\frac { d E } { d t } = | \\bar { e } | ^ { \\alpha } \\mathrm { s i g n } ( \\bar { e } ) \\left( \\frac { e ^ { - z } } { ( 1 + e ^ { - z } ) ^ { 2 } } \\right) \\left( x _ { 1 } \\dot { w } _ { 1 } + x _ { 2 } \\dot { w } _ { 2 } + \\cdot \\cdot \\cdot + x _ { n } \\dot { w } _ { n } \\right)" + }, + { + "category_id": 13, + "poly": [ + 652, + 262, + 710, + 262, + 710, + 295, + 652, + 295 + ], + "score": 0.92, + "latex": "E ( \\bar { e } )" + }, + { + "category_id": 13, + "poly": [ + 342, + 2005, + 450, + 2005, + 450, + 2034, + 342, + 2034 + ], + "score": 0.92, + "latex": "y ^ { \\star } \\in \\mathbb { R } ^ { m }" + }, + { + "category_id": 13, + "poly": [ + 489, + 1362, + 585, + 1362, + 585, + 1395, + 489, + 1395 + ], + "score": 0.92, + "latex": "| x _ { i } | < a" + }, + { + "category_id": 13, + "poly": [ + 1234, + 1111, + 1402, + 1111, + 1402, + 1146, + 1234, + 1146 + ], + "score": 0.92, + "latex": "| x _ { j } | > \\gamma > 0" + }, + { + "category_id": 13, + "poly": [ + 1270, + 1364, + 1399, + 1364, + 1399, + 1394, + 1270, + 1394 + ], + "score": 0.91, + "latex": "\\bar { e } = y - y ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 371, + 901, + 448, + 901, + 448, + 930, + 371, + 930 + ], + "score": 0.91, + "latex": "k _ { i } > 0" + }, + { + "category_id": 13, + "poly": [ + 864, + 1143, + 971, + 1143, + 971, + 1175, + 864, + 1175 + ], + "score": 0.91, + "latex": "j \\in [ 1 , n ]" + }, + { + "category_id": 13, + "poly": [ + 764, + 1112, + 982, + 1112, + 982, + 1143, + 764, + 1143 + ], + "score": 0.91, + "latex": "x _ { i } , i = 1 , 2 , \\cdots , n" + }, + { + "category_id": 13, + "poly": [ + 1233, + 627, + 1270, + 627, + 1270, + 665, + 1233, + 665 + ], + "score": 0.9, + "latex": "\\textstyle { \\frac { \\mathrm { d } w } { \\mathrm { d } t } }" + }, + { + "category_id": 13, + "poly": [ + 385, + 799, + 765, + 799, + 765, + 836, + 385, + 836 + ], + "score": 0.9, + "latex": "u _ { 1 } \\triangleq \\dot { w } _ { 1 } , u _ { 2 } \\triangleq \\dot { w } _ { 2 } , \\cdot \\cdot \\cdot , u _ { n } \\triangleq \\dot { w } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 633, + 1332, + 850, + 1332, + 850, + 1363, + 633, + 1363 + ], + "score": 0.9, + "latex": "x _ { i } , i = 1 , 2 , \\cdots , n" + }, + { + "category_id": 13, + "poly": [ + 790, + 1905, + 903, + 1905, + 903, + 1933, + 790, + 1933 + ], + "score": 0.9, + "latex": "\\boldsymbol { x } \\in \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 430, + 1092, + 430, + 1092, + 458, + 1018, + 458 + ], + "score": 0.9, + "latex": "\\bar { e } = 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 1331, + 411, + 1331, + 411, + 1364, + 298, + 1364 + ], + "score": 0.9, + "latex": "y = \\sigma ( z )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1682, + 852, + 1682, + 852, + 1721, + 371, + 1721 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { k _ { \\operatorname* { m i n } } = \\operatorname* { m i n } ( k _ { i } ) , i = 1 , 2 , \\cdots , n , \\beta = \\frac { \\alpha } { \\alpha + 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 299, + 1364, + 375, + 1364, + 375, + 1395, + 299, + 1395 + ], + "score": 0.89, + "latex": "( 0 , \\infty )" + }, + { + "category_id": 14, + "poly": [ + 514, + 845, + 1182, + 845, + 1182, + 887, + 514, + 887 + ], + "score": 0.89, + "latex": "u _ { i } = - k _ { i } \\mathrm { s i g n } ( x _ { i } ) \\mathrm { s i g n } ( \\bar { e } ) e ^ { z } ( 1 + e ^ { - z } ) ^ { 2 } , \\quad i = 1 , 2 , \\cdots , n ," + }, + { + "category_id": 13, + "poly": [ + 917, + 1720, + 972, + 1720, + 972, + 1748, + 917, + 1748 + ], + "score": 0.89, + "latex": "k _ { \\mathrm { m i n } }" + }, + { + "category_id": 13, + "poly": [ + 1128, + 432, + 1218, + 432, + 1218, + 458, + 1128, + 458 + ], + "score": 0.89, + "latex": "t \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 432, + 1975, + 525, + 1975, + 525, + 2003, + 432, + 2003 + ], + "score": 0.88, + "latex": "y \\in \\mathbb { R } ^ { m }" + }, + { + "category_id": 13, + "poly": [ + 590, + 1939, + 846, + 1939, + 846, + 1975, + 590, + 1975 + ], + "score": 0.88, + "latex": "| x _ { i } | < c , i = 1 , 2 , \\cdot \\cdot \\cdot n" + }, + { + "category_id": 13, + "poly": [ + 374, + 936, + 541, + 936, + 541, + 961, + 374, + 961 + ], + "score": 0.87, + "latex": "u _ { 1 } , u _ { 2 } , \\cdots , u _ { n }" + }, + { + "category_id": 13, + "poly": [ + 299, + 1974, + 375, + 1974, + 375, + 2005, + 299, + 2005 + ], + "score": 0.87, + "latex": "( 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 1339, + 1907, + 1403, + 1907, + 1403, + 1937, + 1339, + 1937 + ], + "score": 0.87, + "latex": "x \\quad =" + }, + { + "category_id": 13, + "poly": [ + 328, + 962, + 355, + 962, + 355, + 988, + 328, + 988 + ], + "score": 0.86, + "latex": "e ^ { z }" + }, + { + "category_id": 13, + "poly": [ + 1276, + 934, + 1304, + 934, + 1304, + 960, + 1276, + 960 + ], + "score": 0.85, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1360, + 1945, + 1404, + 1945, + 1404, + 1972, + 1360, + 1972 + ], + "score": 0.84, + "latex": "c \\in" + }, + { + "category_id": 13, + "poly": [ + 430, + 1719, + 455, + 1719, + 455, + 1745, + 430, + 1745 + ], + "score": 0.83, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 741, + 1876, + 769, + 1876, + 769, + 1902, + 741, + 1902 + ], + "score": 0.83, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 1353, + 1334, + 1403, + 1334, + 1403, + 1361, + 1353, + 1361 + ], + "score": 0.83, + "latex": "a \\in" + }, + { + "category_id": 13, + "poly": [ + 1023, + 1724, + 1044, + 1724, + 1044, + 1750, + 1023, + 1750 + ], + "score": 0.82, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 530, + 903, + 542, + 903, + 542, + 927, + 530, + 927 + ], + "score": 0.79, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1116, + 1972, + 1394, + 1972, + 1394, + 2004, + 1116, + 2004 + ], + "score": 0.78, + "latex": "y = [ y _ { 1 } \\quad y _ { 2 } \\quad \\cdot \\cdot \\cdot \\quad y _ { m } ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1490, + 323, + 1490, + 323, + 1516, + 298, + 1516 + ], + "score": 0.77, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 298, + 1935, + 539, + 1935, + 539, + 1973, + 298, + 1973 + ], + "score": 0.77, + "latex": "\\left[ x _ { 1 } \\quad x _ { 2 } \\quad \\cdots \\quad x _ { n } \\right] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1046, + 2004, + 1336, + 2004, + 1336, + 2037, + 1046, + 2037 + ], + "score": 0.75, + "latex": "y ^ { \\star } = [ y _ { 1 } ^ { \\star } \\quad y _ { 2 } ^ { \\star } \\quad \\cdot \\cdot \\quad y _ { m } ^ { \\star } ]" + }, + { + "category_id": 13, + "poly": [ + 370, + 1146, + 391, + 1146, + 391, + 1174, + 370, + 1174 + ], + "score": 0.74, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 1292, + 2004, + 1338, + 2004, + 1338, + 2036, + 1292, + 2036 + ], + "score": 0.55, + "latex": "y _ { m } ^ { \\star } ]" + }, + { + "category_id": 13, + "poly": [ + 480, + 1935, + 539, + 1935, + 539, + 1974, + 480, + 1974 + ], + "score": 0.5, + "latex": "x _ { n } \\big ] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1116, + 1972, + 1205, + 1972, + 1205, + 2004, + 1116, + 2004 + ], + "score": 0.45, + "latex": "y = [ y _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 784, + 1365, + 796, + 1365, + 796, + 1389, + 784, + 1389 + ], + "score": 0.39, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 747, + 1683, + 852, + 1683, + 852, + 1722, + 747, + 1722 + ], + "score": 0.3, + "latex": "\\begin{array} { r } { \\beta = \\frac { \\alpha } { \\alpha + 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1347, + 1974, + 1392, + 1974, + 1392, + 2005, + 1347, + 2005 + ], + "score": 0.28, + "latex": "y _ { m } ]" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1814.0, + 634.0, + 1814.0, + 634.0, + 1853.0, + 294.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1871.0, + 740.0, + 1871.0, + 740.0, + 1907.0, + 295.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1871.0, + 1405.0, + 1871.0, + 1405.0, + 1907.0, + 770.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1902.0, + 789.0, + 1902.0, + 789.0, + 1939.0, + 294.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1902.0, + 1338.0, + 1902.0, + 1338.0, + 1939.0, + 904.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1936.0, + 589.0, + 1936.0, + 589.0, + 1981.0, + 540.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1936.0, + 1359.0, + 1936.0, + 1359.0, + 1981.0, + 847.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1936.0, + 1408.0, + 1936.0, + 1408.0, + 1981.0, + 1405.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1967.0, + 298.0, + 1967.0, + 298.0, + 2013.0, + 293.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1967.0, + 431.0, + 1967.0, + 431.0, + 2013.0, + 376.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1967.0, + 1115.0, + 1967.0, + 1115.0, + 2013.0, + 526.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1967.0, + 1410.0, + 1967.0, + 1410.0, + 2013.0, + 1395.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1998.0, + 341.0, + 1998.0, + 341.0, + 2041.0, + 290.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1998.0, + 1045.0, + 1998.0, + 1045.0, + 2041.0, + 451.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1998.0, + 1408.0, + 1998.0, + 1408.0, + 2041.0, + 1339.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1456.0, + 1405.0, + 1456.0, + 1405.0, + 1494.0, + 294.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1488.0, + 297.0, + 1488.0, + 297.0, + 1524.0, + 294.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1488.0, + 1405.0, + 1488.0, + 1405.0, + 1524.0, + 324.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1521.0, + 1405.0, + 1521.0, + 1405.0, + 1553.0, + 295.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1550.0, + 1244.0, + 1550.0, + 1244.0, + 1585.0, + 292.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 395.0, + 371.0, + 395.0, + 371.0, + 433.0, + 294.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 395.0, + 1407.0, + 395.0, + 1407.0, + 433.0, + 490.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 427.0, + 1017.0, + 427.0, + 1017.0, + 465.0, + 294.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 427.0, + 1127.0, + 427.0, + 1127.0, + 465.0, + 1093.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 427.0, + 1405.0, + 427.0, + 1405.0, + 465.0, + 1219.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 458.0, + 1403.0, + 458.0, + 1403.0, + 495.0, + 294.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 486.0, + 1408.0, + 486.0, + 1408.0, + 527.0, + 293.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 515.0, + 473.0, + 515.0, + 473.0, + 558.0, + 293.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 370.0, + 1681.0, + 370.0, + 1721.0, + 294.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1716.0, + 429.0, + 1716.0, + 429.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1716.0, + 916.0, + 1716.0, + 916.0, + 1751.0, + 456.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1716.0, + 1022.0, + 1716.0, + 1022.0, + 1751.0, + 973.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1716.0, + 1404.0, + 1716.0, + 1404.0, + 1751.0, + 1045.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1747.0, + 994.0, + 1747.0, + 994.0, + 1783.0, + 296.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1751.0, + 1402.0, + 1751.0, + 1402.0, + 1778.0, + 1377.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.25, + 1668.5, + 1410.25, + 1668.5, + 1410.25, + 1728.0, + 785.25, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 899.0, + 370.0, + 899.0, + 370.0, + 933.0, + 296.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 899.0, + 529.0, + 899.0, + 529.0, + 933.0, + 449.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 899.0, + 1403.0, + 899.0, + 1403.0, + 933.0, + 543.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 928.0, + 373.0, + 928.0, + 373.0, + 964.0, + 292.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 928.0, + 1275.0, + 928.0, + 1275.0, + 964.0, + 542.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 928.0, + 1406.0, + 928.0, + 1406.0, + 964.0, + 1305.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 958.0, + 327.0, + 958.0, + 327.0, + 995.0, + 294.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 958.0, + 743.0, + 958.0, + 743.0, + 995.0, + 356.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1297.0, + 1407.0, + 1297.0, + 1407.0, + 1336.0, + 295.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1331.0, + 297.0, + 1331.0, + 297.0, + 1367.0, + 293.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1331.0, + 632.0, + 1331.0, + 632.0, + 1367.0, + 412.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1331.0, + 1352.0, + 1331.0, + 1352.0, + 1367.0, + 851.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1331.0, + 1407.0, + 1331.0, + 1407.0, + 1367.0, + 1404.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1359.0, + 298.0, + 1359.0, + 298.0, + 1398.0, + 293.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1359.0, + 488.0, + 1359.0, + 488.0, + 1398.0, + 376.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1359.0, + 783.0, + 1359.0, + 783.0, + 1398.0, + 586.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 1359.0, + 1269.0, + 1359.0, + 1269.0, + 1398.0, + 797.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1359.0, + 1405.0, + 1359.0, + 1405.0, + 1398.0, + 1400.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1394.0, + 664.0, + 1394.0, + 664.0, + 1427.0, + 295.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1198.0, + 1403.0, + 1198.0, + 1403.0, + 1234.0, + 292.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1227.0, + 1405.0, + 1227.0, + 1405.0, + 1265.0, + 294.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1259.0, + 422.0, + 1259.0, + 422.0, + 1294.0, + 291.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1110.0, + 763.0, + 1110.0, + 763.0, + 1146.0, + 297.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1110.0, + 1233.0, + 1110.0, + 1233.0, + 1146.0, + 983.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1138.0, + 369.0, + 1138.0, + 369.0, + 1178.0, + 294.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1138.0, + 863.0, + 1138.0, + 863.0, + 1178.0, + 392.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1138.0, + 984.0, + 1138.0, + 984.0, + 1178.0, + 972.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 230.0, + 1403.0, + 230.0, + 1403.0, + 266.0, + 296.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 258.0, + 651.0, + 258.0, + 651.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 258.0, + 1251.0, + 258.0, + 1251.0, + 297.0, + 711.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 617.0, + 373.0, + 617.0, + 373.0, + 668.0, + 296.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 617.0, + 1232.0, + 617.0, + 1232.0, + 668.0, + 508.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 617.0, + 1406.0, + 617.0, + 1406.0, + 668.0, + 1271.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 642.0, + 511.0, + 642.0, + 511.0, + 672.0, + 508.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 660.0, + 1124.0, + 660.0, + 1124.0, + 701.0, + 293.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 795.0, + 384.0, + 795.0, + 384.0, + 842.0, + 294.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 795.0, + 824.0, + 795.0, + 824.0, + 842.0, + 766.0, + 842.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 561, + 1405, + 561, + 1405, + 694, + 297, + 694 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1537, + 1406, + 1537, + 1406, + 1630, + 297, + 1630 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 680, + 1963, + 1019, + 1963, + 1019, + 2044, + 680, + 2044 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 669, + 1185, + 1029, + 1185, + 1029, + 1270, + 669, + 1270 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 688, + 1028, + 1009, + 1028, + 1009, + 1104, + 688, + 1104 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 300, + 1279, + 1399, + 1279, + 1399, + 1358, + 300, + 1358 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 701, + 1375, + 995, + 1375, + 995, + 1448, + 701, + 1448 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 732, + 714, + 965, + 714, + 965, + 800, + 732, + 800 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 296, + 1780, + 1406, + 1780, + 1406, + 1846, + 296, + 1846 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 291, + 226, + 1403, + 226, + 1403, + 295, + 291, + 295 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 549, + 331, + 1149, + 331, + 1149, + 409, + 549, + 409 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 291, + 1880, + 1402, + 1880, + 1402, + 1948, + 291, + 1948 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 296, + 416, + 1402, + 416, + 1402, + 478, + 296, + 478 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 773, + 882, + 927, + 882, + 927, + 961, + 773, + 961 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 1463, + 1401, + 1463, + 1401, + 1529, + 296, + 1529 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 297, + 1121, + 1402, + 1121, + 1402, + 1183, + 297, + 1183 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 662, + 1719, + 1036, + 1719, + 1036, + 1766, + 662, + 1766 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 297, + 1640, + 1400, + 1640, + 1400, + 1705, + 297, + 1705 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 760, + 480, + 938, + 480, + 938, + 549, + 760, + 549 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 815, + 900, + 815, + 900, + 864, + 297, + 864 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 294, + 978, + 1259, + 978, + 1259, + 1014, + 294, + 1014 + ], + "score": 0.928 + }, + { + "category_id": 2, + "poly": [ + 296, + 73, + 855, + 73, + 855, + 106, + 296, + 106 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1387, + 1400, + 1387, + 1400, + 1419, + 1352, + 1419 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1727, + 1400, + 1727, + 1400, + 1758, + 1352, + 1758 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1049, + 1400, + 1049, + 1400, + 1081, + 1352, + 1081 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1209, + 1400, + 1209, + 1400, + 1240, + 1352, + 1240 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1365, + 903, + 1401, + 903, + 1401, + 934, + 1365, + 934 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1365, + 741, + 1400, + 741, + 1400, + 772, + 1365, + 772 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1983, + 1401, + 1983, + 1401, + 2014, + 1351, + 2014 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1366, + 357, + 1400, + 357, + 1400, + 388, + 1366, + 388 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1365, + 489, + 1400, + 489, + 1400, + 520, + 1365, + 520 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.8 + }, + { + "category_id": 14, + "poly": [ + 678, + 1961, + 1022, + 1961, + 1022, + 2043, + 678, + 2043 + ], + "score": 0.95, + "latex": "\\dot { E } = \\sum _ { m } \\dot { E } _ { m } = \\sum _ { m } \\frac { \\partial E _ { m } } { \\partial w _ { j i ^ { l } } } \\dot { w } _ { j i } ^ { l }" + }, + { + "category_id": 14, + "poly": [ + 762, + 477, + 939, + 477, + 939, + 551, + 762, + 551 + ], + "score": 0.95, + "latex": "a _ { j } ^ { l } = \\sum _ { i } w _ { j i } ^ { l } z _ { i } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 570, + 816, + 685, + 816, + 685, + 867, + 570, + 867 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { \\delta _ { j } ^ { l } \\triangleq \\frac { \\partial E _ { m } } { \\partial a _ { j } ^ { l } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 668, + 1181, + 1032, + 1181, + 1032, + 1271, + 668, + 1271 + ], + "score": 0.95, + "latex": "\\delta _ { j } ^ { l } = \\frac { \\partial E _ { m } } { \\partial a _ { j } ^ { l } } = \\sum _ { k } \\frac { \\partial E _ { m } } { \\partial a _ { k } ^ { l + 1 } } \\frac { \\partial a _ { k } ^ { l + 1 } } { \\partial a _ { j } ^ { l } }" + }, + { + "category_id": 14, + "poly": [ + 687, + 1028, + 1013, + 1028, + 1013, + 1104, + 687, + 1104 + ], + "score": 0.94, + "latex": "\\delta _ { m } ^ { L } = \\frac { \\partial E _ { m } } { \\partial a _ { m } ^ { L } } = \\sigma ^ { \\prime } ( a _ { m } ^ { L } ) \\frac { \\partial E _ { m } } { \\partial y _ { m } }" + }, + { + "category_id": 14, + "poly": [ + 701, + 1375, + 998, + 1375, + 998, + 1450, + 701, + 1450 + ], + "score": 0.94, + "latex": "\\delta _ { j } ^ { l } = \\sigma ^ { \\prime } ( a _ { j } ^ { l } ) \\sum _ { k } w _ { k j } ^ { l + 1 } \\delta _ { k } ^ { l + 1 }" + }, + { + "category_id": 14, + "poly": [ + 733, + 712, + 968, + 712, + 968, + 800, + 733, + 800 + ], + "score": 0.94, + "latex": "\\frac { \\partial E _ { m } } { \\partial w _ { j i } ^ { l } } = \\frac { \\partial E _ { m } } { \\partial a _ { j } ^ { l } } \\frac { \\partial a _ { j } ^ { l } } { \\partial w _ { j i } ^ { l } }" + }, + { + "category_id": 13, + "poly": [ + 601, + 1310, + 718, + 1310, + 718, + 1361, + 601, + 1361 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\delta _ { j } ^ { l } \\triangleq \\frac { \\partial E _ { m } } { \\partial a _ { j } ^ { l } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 773, + 880, + 928, + 880, + 928, + 962, + 773, + 962 + ], + "score": 0.93, + "latex": "\\frac { \\partial E _ { m } } { \\partial w _ { j i } ^ { l } } = \\delta _ { j } ^ { l } z _ { i } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 425, + 596, + 584, + 596, + 584, + 638, + 425, + 638 + ], + "score": 0.93, + "latex": "z _ { j } ^ { l } = \\sigma ( a _ { j } ^ { l - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 421, + 1311, + 550, + 1311, + 550, + 1352, + 421, + 1352 + ], + "score": 0.93, + "latex": "z _ { j } ^ { l } = \\sigma ( a _ { j } ^ { l } )" + }, + { + "category_id": 14, + "poly": [ + 662, + 1719, + 1038, + 1719, + 1038, + 1765, + 662, + 1765 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\dot { w } _ { j i } ^ { l } = - k _ { j i } ^ { l } \\mathrm { s i g n } ( \\delta _ { j } ^ { l } z _ { i } ^ { l } ) | \\delta _ { j } ^ { l } z _ { i } ^ { l } | ^ { \\alpha } E ^ { \\beta } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 549, + 330, + 1150, + 330, + 1150, + 410, + 549, + 410 + ], + "score": 0.92, + "latex": "E = E _ { 1 } + \\cdot \\cdot \\cdot + E _ { m } = { \\frac { | { \\bar { e } } _ { 1 } | ^ { ( \\alpha + 1 ) } } { ( \\alpha + 1 ) } } + \\cdot \\cdot \\cdot + { \\frac { | { \\bar { e } } _ { m } | ^ { ( \\alpha + 1 ) } } { ( \\alpha + 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 716, + 1782, + 833, + 1782, + 833, + 1813, + 716, + 1813 + ], + "score": 0.92, + "latex": "\\alpha + \\beta < 1" + }, + { + "category_id": 13, + "poly": [ + 885, + 1782, + 973, + 1782, + 973, + 1816, + 885, + 1816 + ], + "score": 0.92, + "latex": "k _ { j i } > 0" + }, + { + "category_id": 13, + "poly": [ + 628, + 659, + 670, + 659, + 670, + 698, + 628, + 698 + ], + "score": 0.92, + "latex": "w _ { j i } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 491, + 1782, + 605, + 1782, + 605, + 1814, + 491, + 1814 + ], + "score": 0.91, + "latex": "\\beta \\in ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 1234, + 1538, + 1402, + 1538, + 1402, + 1572, + 1234, + 1572 + ], + "score": 0.91, + "latex": "\\left| z _ { n } \\right| > \\gamma > 0" + }, + { + "category_id": 13, + "poly": [ + 655, + 979, + 691, + 979, + 691, + 1014, + 655, + 1014 + ], + "score": 0.91, + "latex": "\\delta _ { m } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 764, + 1540, + 981, + 1540, + 981, + 1570, + 764, + 1570 + ], + "score": 0.91, + "latex": "z _ { i } , i = 1 , 2 , \\cdots , L" + }, + { + "category_id": 13, + "poly": [ + 372, + 1120, + 409, + 1120, + 409, + 1156, + 372, + 1156 + ], + "score": 0.91, + "latex": "z _ { m } ^ { L }" + }, + { + "category_id": 13, + "poly": [ + 856, + 1571, + 969, + 1571, + 969, + 1603, + 856, + 1603 + ], + "score": 0.9, + "latex": "n \\in [ 1 , L ]" + }, + { + "category_id": 13, + "poly": [ + 1155, + 1120, + 1181, + 1120, + 1181, + 1160, + 1155, + 1160 + ], + "score": 0.9, + "latex": "\\delta _ { j } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1280, + 1642, + 1345, + 1642, + 1345, + 1671, + 1280, + 1671 + ], + "score": 0.88, + "latex": "l + 1" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1281, + 1320, + 1281, + 1320, + 1310, + 1260, + 1310 + ], + "score": 0.88, + "latex": "l + 1" + }, + { + "category_id": 13, + "poly": [ + 478, + 562, + 503, + 562, + 503, + 597, + 478, + 597 + ], + "score": 0.86, + "latex": "z _ { i } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 540, + 1816, + 569, + 1816, + 569, + 1845, + 540, + 1845 + ], + "score": 0.86, + "latex": "y ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 576, + 1128, + 613, + 1128, + 613, + 1155, + 576, + 1155 + ], + "score": 0.85, + "latex": "y _ { m }" + }, + { + "category_id": 13, + "poly": [ + 1046, + 983, + 1067, + 983, + 1067, + 1007, + 1046, + 1007 + ], + "score": 0.85, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 1081, + 420, + 1097, + 420, + 1097, + 450, + 1081, + 450 + ], + "score": 0.85, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 572, + 1283, + 591, + 1283, + 591, + 1308, + 572, + 1308 + ], + "score": 0.84, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 448, + 662, + 472, + 662, + 472, + 688, + 448, + 688 + ], + "score": 0.84, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 867, + 1885, + 892, + 1885, + 892, + 1911, + 867, + 1911 + ], + "score": 0.84, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 1159, + 1644, + 1176, + 1644, + 1176, + 1674, + 1159, + 1674 + ], + "score": 0.8, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 1045, + 1571, + 1067, + 1571, + 1067, + 1597, + 1045, + 1597 + ], + "score": 0.79, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 794, + 228, + 1307, + 228, + 1307, + 265, + 794, + 265 + ], + "score": 0.78, + "latex": "\\bar { e } = [ \\left| y _ { 1 } - y _ { 1 } ^ { \\star } \\right| \\quad \\left| y _ { 2 } - y _ { 2 } ^ { \\star } \\right| \\quad \\cdot \\cdot \\quad \\left| y _ { m } - y _ { m } ^ { \\star } \\right| ]" + }, + { + "category_id": 13, + "poly": [ + 635, + 565, + 647, + 565, + 647, + 591, + 635, + 591 + ], + "score": 0.77, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 674, + 605, + 694, + 605, + 694, + 627, + 674, + 627 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 1193, + 419, + 1206, + 419, + 1206, + 445, + 1193, + 445 + ], + "score": 0.7, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 369, + 1574, + 390, + 1574, + 390, + 1601, + 369, + 1601 + ], + "score": 0.68, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1643, + 1067, + 1643, + 1067, + 1669, + 1055, + 1669 + ], + "score": 0.61, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 1384, + 1647, + 1402, + 1647, + 1402, + 1669, + 1384, + 1669 + ], + "score": 0.6, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1178, + 229, + 1306, + 229, + 1306, + 264, + 1178, + 264 + ], + "score": 0.59, + "latex": "| y _ { m } - y _ { m } ^ { \\star } | ]" + }, + { + "category_id": 13, + "poly": [ + 937, + 1644, + 951, + 1644, + 951, + 1669, + 937, + 1669 + ], + "score": 0.57, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 853, + 662, + 864, + 662, + 864, + 688, + 853, + 688 + ], + "score": 0.56, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 297, + 1601, + 309, + 1601, + 309, + 1627, + 297, + 1627 + ], + "score": 0.39, + "latex": "l" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 863.0, + 2086.0, + 863.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 562.0, + 477.0, + 562.0, + 477.0, + 600.0, + 296.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 562.0, + 634.0, + 562.0, + 634.0, + 600.0, + 504.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 562.0, + 1405.0, + 562.0, + 1405.0, + 600.0, + 648.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 595.0, + 424.0, + 595.0, + 424.0, + 640.0, + 291.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 595.0, + 673.0, + 595.0, + 673.0, + 640.0, + 585.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 595.0, + 1408.0, + 595.0, + 1408.0, + 640.0, + 695.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 629.0, + 1405.0, + 629.0, + 1405.0, + 663.0, + 294.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 659.0, + 447.0, + 659.0, + 447.0, + 698.0, + 292.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 659.0, + 627.0, + 659.0, + 627.0, + 698.0, + 473.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 659.0, + 852.0, + 659.0, + 852.0, + 698.0, + 671.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 659.0, + 1076.0, + 659.0, + 1076.0, + 698.0, + 865.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1536.0, + 763.0, + 1536.0, + 763.0, + 1574.0, + 294.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1536.0, + 1233.0, + 1536.0, + 1233.0, + 1574.0, + 982.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1536.0, + 1407.0, + 1536.0, + 1407.0, + 1574.0, + 1403.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1568.0, + 368.0, + 1568.0, + 368.0, + 1605.0, + 292.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1568.0, + 855.0, + 1568.0, + 855.0, + 1605.0, + 391.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1568.0, + 1044.0, + 1568.0, + 1044.0, + 1605.0, + 970.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1568.0, + 1407.0, + 1568.0, + 1407.0, + 1605.0, + 1068.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1598.0, + 296.0, + 1598.0, + 296.0, + 1633.0, + 290.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1598.0, + 325.0, + 1598.0, + 325.0, + 1633.0, + 310.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1277.0, + 571.0, + 1277.0, + 571.0, + 1316.0, + 295.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1277.0, + 1259.0, + 1277.0, + 1259.0, + 1316.0, + 592.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 1277.0, + 1402.0, + 1277.0, + 1402.0, + 1316.0, + 1321.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1309.0, + 420.0, + 1309.0, + 420.0, + 1349.0, + 295.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1309.0, + 600.0, + 1309.0, + 600.0, + 1349.0, + 551.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1313.0, + 831.0, + 1313.0, + 831.0, + 1348.0, + 719.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1302.0, + 719.0, + 1302.0, + 719.0, + 1341.0, + 661.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1777.0, + 490.0, + 1777.0, + 490.0, + 1818.0, + 293.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1777.0, + 715.0, + 1777.0, + 715.0, + 1818.0, + 606.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1777.0, + 884.0, + 1777.0, + 884.0, + 1818.0, + 834.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1777.0, + 1406.0, + 1777.0, + 1406.0, + 1818.0, + 974.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1814.0, + 539.0, + 1814.0, + 539.0, + 1847.0, + 296.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1814.0, + 724.0, + 1814.0, + 724.0, + 1847.0, + 570.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 224.0, + 793.0, + 224.0, + 793.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 224.0, + 1408.0, + 224.0, + 1408.0, + 270.0, + 1308.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 259.0, + 1030.0, + 259.0, + 1030.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 866.0, + 1879.0, + 866.0, + 1919.0, + 294.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1879.0, + 1404.0, + 1879.0, + 1404.0, + 1919.0, + 893.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1914.0, + 640.0, + 1914.0, + 640.0, + 1951.0, + 294.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 413.0, + 1080.0, + 413.0, + 1080.0, + 453.0, + 293.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 413.0, + 1192.0, + 413.0, + 1192.0, + 453.0, + 1098.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 453.0, + 1207.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 447.0, + 380.0, + 447.0, + 380.0, + 486.0, + 292.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1461.0, + 1403.0, + 1461.0, + 1403.0, + 1501.0, + 292.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1495.0, + 383.0, + 1495.0, + 383.0, + 1530.0, + 292.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1111.0, + 371.0, + 1111.0, + 371.0, + 1167.0, + 290.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1111.0, + 575.0, + 1111.0, + 575.0, + 1167.0, + 410.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1111.0, + 1154.0, + 1111.0, + 1154.0, + 1167.0, + 614.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1111.0, + 1411.0, + 1111.0, + 1411.0, + 1167.0, + 1182.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1148.0, + 426.0, + 1148.0, + 426.0, + 1192.0, + 292.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1636.0, + 936.0, + 1636.0, + 936.0, + 1678.0, + 293.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1636.0, + 1054.0, + 1636.0, + 1054.0, + 1678.0, + 952.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1636.0, + 1158.0, + 1636.0, + 1158.0, + 1678.0, + 1068.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 1636.0, + 1279.0, + 1636.0, + 1279.0, + 1678.0, + 1177.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1636.0, + 1383.0, + 1636.0, + 1383.0, + 1678.0, + 1346.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1672.0, + 736.0, + 1672.0, + 736.0, + 1709.0, + 295.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 812.0, + 569.0, + 812.0, + 569.0, + 854.0, + 295.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 812.0, + 902.0, + 812.0, + 902.0, + 854.0, + 686.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 835.5, + 678.0, + 835.5, + 678.0, + 865.0, + 637.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 976.0, + 654.0, + 976.0, + 654.0, + 1019.0, + 292.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 976.0, + 1045.0, + 976.0, + 1045.0, + 1019.0, + 692.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 976.0, + 1264.0, + 976.0, + 1264.0, + 1019.0, + 1068.0, + 1019.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 757, + 1405, + 757, + 1405, + 1005, + 296, + 1005 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1474, + 1405, + 1474, + 1405, + 1691, + 297, + 1691 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1404, + 1941, + 1404, + 2036, + 299, + 2036 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 672, + 1227, + 1023, + 1227, + 1023, + 1299, + 672, + 1299 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 695, + 1072, + 1002, + 1072, + 1002, + 1144, + 695, + 1144 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 300, + 1312, + 1406, + 1312, + 1406, + 1377, + 300, + 1377 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 292, + 506, + 1404, + 506, + 1404, + 570, + 292, + 570 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 695, + 280, + 1003, + 280, + 1003, + 352, + 695, + 352 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 596, + 1405, + 596, + 1405, + 661, + 297, + 661 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 1702, + 1403, + 1702, + 1403, + 1764, + 297, + 1764 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 732, + 453, + 968, + 453, + 968, + 495, + 732, + 495 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 753, + 1782, + 943, + 1782, + 943, + 1821, + 753, + 1821 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 299, + 1897, + 651, + 1897, + 651, + 1930, + 299, + 1930 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 300, + 229, + 802, + 229, + 802, + 263, + 300, + 263 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 1838, + 758, + 1838, + 758, + 1870, + 297, + 1870 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 301, + 1019, + 781, + 1019, + 781, + 1052, + 301, + 1052 + ], + "score": 0.913 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1243, + 1400, + 1243, + 1400, + 1275, + 1353, + 1275 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 301, + 1417, + 759, + 1417, + 759, + 1448, + 301, + 1448 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1352, + 291, + 1400, + 291, + 1400, + 322, + 1352, + 322 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1088, + 1400, + 1088, + 1400, + 1120, + 1352, + 1120 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1786, + 1400, + 1786, + 1400, + 1816, + 1352, + 1816 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1353, + 460, + 1399, + 460, + 1399, + 491, + 1353, + 491 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 300, + 1174, + 774, + 1174, + 774, + 1207, + 300, + 1207 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 297, + 702, + 966, + 702, + 966, + 734, + 297, + 734 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.754 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.734 + }, + { + "category_id": 1, + "poly": [ + 289, + 372, + 1404, + 372, + 1404, + 453, + 289, + 453 + ], + "score": 0.631 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.246 + }, + { + "category_id": 13, + "poly": [ + 916, + 371, + 1144, + 371, + 1144, + 424, + 916, + 424 + ], + "score": 0.95, + "latex": "k _ { \\mathrm { m i n } } = \\operatorname* { m i n } _ { i , j , l } k _ { j i } ^ { l } > 0" + }, + { + "category_id": 14, + "poly": [ + 671, + 1224, + 1027, + 1224, + 1027, + 1297, + 671, + 1297 + ], + "score": 0.95, + "latex": "T \\leq \\frac { 1 } { k _ { m i n } \\gamma ^ { \\alpha + 1 } ( 1 - \\beta ) } E _ { i n i } ^ { ( 1 - \\beta ) }" + }, + { + "category_id": 14, + "poly": [ + 692, + 1069, + 1005, + 1069, + 1005, + 1143, + 692, + 1143 + ], + "score": 0.94, + "latex": "T \\leq \\frac { 1 } { k _ { m i n } \\gamma ( 1 - \\beta ) } E _ { i n i } ^ { ( 1 - \\beta ) }" + }, + { + "category_id": 14, + "poly": [ + 692, + 279, + 1007, + 279, + 1007, + 353, + 692, + 353 + ], + "score": 0.94, + "latex": "\\dot { E } = - E ^ { \\beta } \\sum _ { m } k _ { j i } ^ { l } | \\delta _ { j } ^ { l } z _ { i } ^ { l } | ^ { \\alpha + 1 }" + }, + { + "category_id": 14, + "poly": [ + 753, + 1782, + 946, + 1782, + 946, + 1820, + 753, + 1820 + ], + "score": 0.93, + "latex": "| \\Delta x _ { i } | \\leq M | x _ { i } | ^ { \\alpha }" + }, + { + "category_id": 14, + "poly": [ + 731, + 452, + 969, + 452, + 969, + 494, + 731, + 494 + ], + "score": 0.92, + "latex": "\\dot { E } \\le - k _ { \\mathrm { m i n } } \\gamma ^ { \\alpha + 1 } E ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 850, + 1346, + 903, + 1346, + 903, + 1375, + 850, + 1375 + ], + "score": 0.91, + "latex": "E _ { i n i }" + }, + { + "category_id": 13, + "poly": [ + 776, + 1314, + 835, + 1314, + 835, + 1346, + 776, + 1346 + ], + "score": 0.91, + "latex": "k _ { m i n }" + }, + { + "category_id": 13, + "poly": [ + 760, + 1734, + 810, + 1734, + 810, + 1763, + 760, + 1763 + ], + "score": 0.9, + "latex": "\\Delta x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 966, + 1703, + 1068, + 1703, + 1068, + 1732, + 966, + 1732 + ], + "score": 0.9, + "latex": "M \\ > \\ 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 1733, + 432, + 1733, + 432, + 1764, + 298, + 1764 + ], + "score": 0.89, + "latex": "1 , 2 , \\cdots , N" + }, + { + "category_id": 13, + "poly": [ + 506, + 1974, + 567, + 1974, + 567, + 2002, + 506, + 2002 + ], + "score": 0.89, + "latex": "l + 1" + }, + { + "category_id": 13, + "poly": [ + 1317, + 1346, + 1372, + 1346, + 1372, + 1373, + 1317, + 1373 + ], + "score": 0.89, + "latex": "\\mathrm { { t } } = 0" + }, + { + "category_id": 13, + "poly": [ + 508, + 630, + 534, + 630, + 534, + 663, + 508, + 663 + ], + "score": 0.87, + "latex": "\\delta _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1704, + 1403, + 1704, + 1403, + 1735, + 1312, + 1735 + ], + "score": 0.87, + "latex": "x _ { i } , i \\ =" + }, + { + "category_id": 13, + "poly": [ + 939, + 507, + 994, + 507, + 994, + 538, + 939, + 538 + ], + "score": 0.86, + "latex": "k _ { \\mathrm { m i n } }" + }, + { + "category_id": 13, + "poly": [ + 685, + 632, + 713, + 632, + 713, + 659, + 685, + 659 + ], + "score": 0.85, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 435, + 508, + 460, + 508, + 460, + 534, + 435, + 534 + ], + "score": 0.83, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 430, + 633, + 455, + 633, + 455, + 659, + 430, + 659 + ], + "score": 0.83, + "latex": "z _ { i }" + }, + { + "category_id": 13, + "poly": [ + 372, + 1319, + 394, + 1319, + 394, + 1345, + 372, + 1345 + ], + "score": 0.82, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 441, + 2005, + 471, + 2005, + 471, + 2035, + 441, + 2035 + ], + "score": 0.75, + "latex": "y ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1009, + 511, + 1029, + 511, + 1029, + 539, + 1009, + 539 + ], + "score": 0.73, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 1384, + 1984, + 1401, + 1984, + 1401, + 2004, + 1384, + 2004 + ], + "score": 0.72, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1068, + 2002, + 1392, + 2002, + 1392, + 2036, + 1068, + 2036 + ], + "score": 0.72, + "latex": "\\Delta x _ { n } , n \\in \\left[ 1 , L \\right] i \\bar { f } k _ { \\operatorname* { m i n } } > M" + }, + { + "category_id": 13, + "poly": [ + 393, + 1975, + 410, + 1975, + 410, + 2005, + 393, + 2005 + ], + "score": 0.71, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1316, + 1340, + 1316, + 1340, + 1341, + 1322, + 1341 + ], + "score": 0.62, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1353, + 1319, + 1375, + 1319, + 1375, + 1341, + 1353, + 1341 + ], + "score": 0.62, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 767, + 632, + 782, + 632, + 782, + 656, + 767, + 656 + ], + "score": 0.56, + "latex": "\\bar { e }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1974, + 310, + 1974, + 310, + 2000, + 298, + 2000 + ], + "score": 0.49, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 1257, + 2003, + 1392, + 2003, + 1392, + 2036, + 1257, + 2036 + ], + "score": 0.42, + "latex": "\\mathrm { \\ddot { \\it f k } } _ { \\mathrm { m i n } } > M" + }, + { + "category_id": 13, + "poly": [ + 463, + 1839, + 492, + 1839, + 492, + 1866, + 463, + 1866 + ], + "score": 0.4, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1315, + 1375, + 1315, + 1375, + 1342, + 1322, + 1342 + ], + "score": 0.39, + "latex": "k , \\alpha" + }, + { + "category_id": 13, + "poly": [ + 805, + 600, + 854, + 600, + 854, + 630, + 805, + 630 + ], + "score": 0.33, + "latex": "( I 3 )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 764.0, + 1416.0, + 764.0, + 1452.0, + 295.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 702.0, + 968.0, + 702.0, + 968.0, + 737.0, + 294.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 758.0, + 1406.0, + 758.0, + 1406.0, + 796.0, + 294.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 788.0, + 1406.0, + 788.0, + 1406.0, + 829.0, + 291.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 822.0, + 1406.0, + 822.0, + 1406.0, + 856.0, + 294.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 851.0, + 1405.0, + 851.0, + 1405.0, + 886.0, + 294.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 917.0, + 293.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 911.0, + 1406.0, + 911.0, + 1406.0, + 949.0, + 294.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 943.0, + 1405.0, + 943.0, + 1405.0, + 979.0, + 291.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 971.0, + 381.0, + 971.0, + 381.0, + 1007.0, + 293.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1475.0, + 1402.0, + 1475.0, + 1402.0, + 1511.0, + 294.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1507.0, + 1405.0, + 1507.0, + 1405.0, + 1540.0, + 295.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1538.0, + 1405.0, + 1538.0, + 1405.0, + 1573.0, + 295.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1568.0, + 1406.0, + 1568.0, + 1406.0, + 1602.0, + 294.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1596.0, + 1405.0, + 1596.0, + 1405.0, + 1635.0, + 294.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1630.0, + 1406.0, + 1630.0, + 1406.0, + 1664.0, + 295.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1659.0, + 932.0, + 1659.0, + 932.0, + 1694.0, + 295.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1969.0, + 297.0, + 1969.0, + 297.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1969.0, + 392.0, + 1969.0, + 392.0, + 2010.0, + 311.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1969.0, + 505.0, + 1969.0, + 505.0, + 2010.0, + 411.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1969.0, + 1383.0, + 1969.0, + 1383.0, + 2010.0, + 568.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1969.0, + 1406.0, + 1969.0, + 1406.0, + 2010.0, + 1402.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 440.0, + 2003.0, + 440.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 2003.0, + 1067.0, + 2003.0, + 1067.0, + 2037.0, + 472.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 2003.0, + 1404.0, + 2003.0, + 1404.0, + 2037.0, + 1393.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1312.0, + 371.0, + 1312.0, + 371.0, + 1348.0, + 296.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1312.0, + 775.0, + 1312.0, + 775.0, + 1348.0, + 395.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1312.0, + 1321.0, + 1312.0, + 1321.0, + 1348.0, + 836.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1312.0, + 1406.0, + 1312.0, + 1406.0, + 1348.0, + 1376.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1343.0, + 849.0, + 1343.0, + 849.0, + 1378.0, + 295.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1343.0, + 1316.0, + 1343.0, + 1316.0, + 1378.0, + 904.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1343.0, + 1383.0, + 1343.0, + 1383.0, + 1378.0, + 1373.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 503.0, + 434.0, + 503.0, + 434.0, + 543.0, + 294.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 503.0, + 938.0, + 503.0, + 938.0, + 543.0, + 461.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 503.0, + 1008.0, + 503.0, + 1008.0, + 543.0, + 995.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 503.0, + 1406.0, + 503.0, + 1406.0, + 543.0, + 1030.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 537.0, + 993.0, + 537.0, + 993.0, + 573.0, + 295.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 543.0, + 1402.0, + 543.0, + 1402.0, + 564.0, + 1378.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 594.0, + 804.0, + 594.0, + 804.0, + 635.0, + 295.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 594.0, + 1403.0, + 594.0, + 1403.0, + 635.0, + 855.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 628.0, + 429.0, + 628.0, + 429.0, + 665.0, + 296.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 628.0, + 507.0, + 628.0, + 507.0, + 665.0, + 456.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 628.0, + 684.0, + 628.0, + 684.0, + 665.0, + 535.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 628.0, + 766.0, + 628.0, + 766.0, + 665.0, + 714.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 628.0, + 1371.0, + 628.0, + 1371.0, + 665.0, + 783.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1698.0, + 965.0, + 1698.0, + 965.0, + 1739.0, + 295.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1698.0, + 1311.0, + 1698.0, + 1311.0, + 1739.0, + 1069.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1698.0, + 1407.0, + 1698.0, + 1407.0, + 1739.0, + 1404.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1733.0, + 759.0, + 1733.0, + 759.0, + 1765.0, + 433.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1733.0, + 923.0, + 1733.0, + 923.0, + 1765.0, + 811.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1895.0, + 653.0, + 1895.0, + 653.0, + 1933.0, + 296.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 805.0, + 227.0, + 805.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1832.0, + 462.0, + 1832.0, + 462.0, + 1877.0, + 292.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 1832.0, + 761.0, + 1832.0, + 761.0, + 1877.0, + 493.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1015.0, + 785.0, + 1015.0, + 785.0, + 1059.0, + 294.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1175.0, + 777.0, + 1175.0, + 777.0, + 1211.0, + 297.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 359.0, + 915.0, + 359.0, + 915.0, + 429.0, + 287.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 359.0, + 1412.0, + 359.0, + 1412.0, + 429.0, + 1145.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 418.0, + 457.0, + 418.0, + 457.0, + 455.0, + 295.0, + 455.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1577, + 824, + 1577, + 824, + 2032, + 298, + 2032 + ], + "score": 0.978 + }, + { + "category_id": 8, + "poly": [ + 539, + 522, + 1161, + 522, + 1161, + 687, + 539, + 687 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1055, + 1404, + 1055, + 1404, + 1178, + 297, + 1178 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 851, + 1338, + 1395, + 1338, + 1395, + 1845, + 851, + 1845 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 1318, + 824, + 1318, + 824, + 1562, + 298, + 1562 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 845, + 1870, + 1403, + 1870, + 1403, + 2023, + 845, + 2023 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 295, + 1193, + 1404, + 1193, + 1404, + 1318, + 295, + 1318 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 296, + 228, + 1404, + 228, + 1404, + 324, + 296, + 324 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 491, + 340, + 1209, + 340, + 1209, + 428, + 491, + 428 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 297, + 977, + 1402, + 977, + 1402, + 1042, + 297, + 1042 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 293, + 700, + 1403, + 700, + 1403, + 763, + 293, + 763 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 295, + 442, + 1405, + 442, + 1405, + 507, + 295, + 507 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 711, + 778, + 988, + 778, + 988, + 821, + 711, + 821 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 296, + 831, + 1327, + 831, + 1327, + 867, + 296, + 867 + ], + "score": 0.926 + }, + { + "category_id": 0, + "poly": [ + 299, + 909, + 559, + 909, + 559, + 945, + 299, + 945 + ], + "score": 0.92 + }, + { + "category_id": 9, + "poly": [ + 1351, + 589, + 1400, + 589, + 1400, + 620, + 1351, + 620 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1352, + 784, + 1400, + 784, + 1400, + 815, + 1352, + 815 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1352, + 364, + 1400, + 364, + 1400, + 395, + 1352, + 395 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.82 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.792 + }, + { + "category_id": 2, + "poly": [ + 1375, + 834, + 1402, + 834, + 1402, + 862, + 1375, + 862 + ], + "score": 0.758 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 854, + 74, + 854, + 106, + 298, + 106 + ], + "score": 0.152 + }, + { + "category_id": 14, + "poly": [ + 489, + 335, + 1210, + 335, + 1210, + 430, + 489, + 430 + ], + "score": 0.95, + "latex": "\\dot { E } = - E ^ { \\beta } \\sum _ { m } k _ { j i } | \\delta _ { j } z _ { i } | ^ { \\alpha + 1 } + \\sum _ { p = 0 } ^ { n } k _ { 1 p } | \\delta _ { p } x _ { p } | ^ { \\alpha } \\mathrm { s i g n } ( \\delta _ { p } x _ { p } ) \\delta _ { p } \\Delta x _ { p }" + }, + { + "category_id": 14, + "poly": [ + 537, + 519, + 1158, + 519, + 1158, + 690, + 537, + 690 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { \\displaystyle \\dot { E } \\leq - E ^ { \\beta } \\sum _ { m } k _ { j i } | \\delta _ { j } z _ { i } | ^ { \\alpha + 1 } + E ^ { \\beta } \\sum _ { p = 0 } ^ { n } { k _ { 1 p } | \\delta _ { p } x _ { p } | ^ { \\alpha + 1 } M } , } \\\\ { \\displaystyle \\leq - E ^ { \\beta } \\sum _ { m } ( k _ { j i } - M ) | \\delta _ { j } z _ { i } | ^ { \\alpha + 1 } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 710, + 776, + 989, + 776, + 989, + 820, + 710, + 820 + ], + "score": 0.91, + "latex": "\\dot { E } \\le - ( k _ { \\operatorname* { m i n } } - M ) \\gamma E ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 365, + 834, + 487, + 834, + 487, + 864, + 365, + 864 + ], + "score": 0.91, + "latex": "k _ { \\operatorname* { m i n } { } } > M" + }, + { + "category_id": 13, + "poly": [ + 373, + 445, + 415, + 445, + 415, + 478, + 373, + 478 + ], + "score": 0.9, + "latex": "k _ { 1 p }" + }, + { + "category_id": 13, + "poly": [ + 894, + 1993, + 993, + 1993, + 993, + 2022, + 894, + 2022 + ], + "score": 0.89, + "latex": "\\alpha = 0 . 6 8" + }, + { + "category_id": 13, + "poly": [ + 355, + 1439, + 390, + 1439, + 390, + 1469, + 355, + 1469 + ], + "score": 0.88, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 446, + 1439, + 481, + 1439, + 481, + 1469, + 446, + 1469 + ], + "score": 0.87, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 706, + 708, + 735, + 708, + 735, + 732, + 706, + 732 + ], + "score": 0.85, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1962, + 1401, + 1962, + 1401, + 1991, + 1323, + 1991 + ], + "score": 0.83, + "latex": "= 0 . 0 2" + }, + { + "category_id": 13, + "poly": [ + 453, + 708, + 479, + 708, + 479, + 732, + 453, + 732 + ], + "score": 0.83, + "latex": "z _ { i }" + }, + { + "category_id": 13, + "poly": [ + 682, + 1226, + 716, + 1226, + 716, + 1256, + 682, + 1256 + ], + "score": 0.79, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 732, + 1226, + 767, + 1226, + 767, + 1256, + 732, + 1256 + ], + "score": 0.77, + "latex": "L _ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1339.0, + 893.0, + 1339.0, + 893.0, + 1357.0, + 865.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1347.0, + 908.0, + 1347.0, + 908.0, + 1354.0, + 897.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1379.0, + 894.0, + 1379.0, + 894.0, + 1401.0, + 862.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1405.0, + 876.0, + 1405.0, + 876.0, + 1497.0, + 850.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1419.0, + 894.0, + 1419.0, + 894.0, + 1443.0, + 863.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1461.0, + 894.0, + 1461.0, + 894.0, + 1485.0, + 863.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1496.0, + 978.0, + 1496.0, + 978.0, + 1516.0, + 924.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1500.0, + 930.0, + 1500.0, + 930.0, + 1511.0, + 914.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1502.0, + 895.0, + 1502.0, + 895.0, + 1528.0, + 861.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1510.0, + 980.0, + 1510.0, + 980.0, + 1534.0, + 924.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 1518.0, + 925.0, + 1518.0, + 925.0, + 1526.0, + 912.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1530.0, + 1020.0, + 1530.0, + 1020.0, + 1549.0, + 924.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1542.0, + 902.0, + 1542.0, + 902.0, + 1577.0, + 859.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1552.0, + 996.0, + 1552.0, + 996.0, + 1575.0, + 969.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1554.0, + 1088.0, + 1554.0, + 1088.0, + 1573.0, + 1064.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1554.0, + 1183.0, + 1554.0, + 1183.0, + 1573.0, + 1157.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1554.0, + 1275.0, + 1554.0, + 1275.0, + 1574.0, + 1251.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1554.0, + 1370.0, + 1554.0, + 1370.0, + 1574.0, + 1344.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1565.0, + 1194.0, + 1565.0, + 1194.0, + 1588.0, + 1084.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1587.0, + 910.0, + 1587.0, + 910.0, + 1608.0, + 863.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1579.0, + 1159.0, + 1579.0, + 1159.0, + 1601.0, + 1129.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1627.0, + 894.0, + 1627.0, + 894.0, + 1650.0, + 862.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1658.0, + 905.0, + 1658.0, + 905.0, + 1744.0, + 839.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 1743.0, + 979.0, + 1743.0, + 979.0, + 1766.0, + 923.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1752.0, + 895.0, + 1752.0, + 895.0, + 1775.0, + 862.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1757.0, + 979.0, + 1757.0, + 979.0, + 1785.0, + 924.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1764.0, + 928.0, + 1764.0, + 928.0, + 1778.0, + 908.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1778.0, + 1020.0, + 1778.0, + 1020.0, + 1801.0, + 925.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1789.0, + 903.0, + 1789.0, + 903.0, + 1827.0, + 860.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1801.0, + 996.0, + 1801.0, + 996.0, + 1824.0, + 969.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1803.0, + 1088.0, + 1803.0, + 1088.0, + 1823.0, + 1065.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 1803.0, + 1182.0, + 1803.0, + 1182.0, + 1822.0, + 1158.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1804.0, + 1274.0, + 1804.0, + 1274.0, + 1823.0, + 1251.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 1802.0, + 1371.0, + 1802.0, + 1371.0, + 1825.0, + 1343.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1814.0, + 1194.0, + 1814.0, + 1194.0, + 1837.0, + 1084.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1828.0, + 1157.0, + 1828.0, + 1157.0, + 1851.0, + 1129.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.25, + 1605.0, + 943.25, + 1605.0, + 943.25, + 1615.0, + 927.25, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1869.0, + 1403.0, + 1869.0, + 1403.0, + 1904.0, + 846.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1901.0, + 1403.0, + 1901.0, + 1403.0, + 1934.0, + 845.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1931.0, + 1405.0, + 1931.0, + 1405.0, + 1964.0, + 845.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1961.0, + 1322.0, + 1961.0, + 1322.0, + 1995.0, + 845.0, + 1995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1992.0, + 893.0, + 1992.0, + 893.0, + 2024.0, + 846.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1992.0, + 1004.0, + 1992.0, + 1004.0, + 2024.0, + 994.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 905.0, + 562.0, + 905.0, + 562.0, + 950.0, + 290.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 838.0, + 1402.0, + 838.0, + 1402.0, + 864.0, + 1378.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1577.0, + 825.0, + 1577.0, + 825.0, + 1609.0, + 297.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1606.0, + 824.0, + 1606.0, + 824.0, + 1640.0, + 295.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1636.0, + 825.0, + 1636.0, + 825.0, + 1673.0, + 296.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1670.0, + 826.0, + 1670.0, + 826.0, + 1701.0, + 296.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1697.0, + 825.0, + 1697.0, + 825.0, + 1732.0, + 295.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1729.0, + 827.0, + 1729.0, + 827.0, + 1764.0, + 295.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1761.0, + 826.0, + 1761.0, + 826.0, + 1792.0, + 295.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 826.0, + 1790.0, + 826.0, + 1821.0, + 294.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 827.0, + 1820.0, + 827.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1848.0, + 827.0, + 1848.0, + 827.0, + 1887.0, + 294.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 827.0, + 1883.0, + 827.0, + 1915.0, + 295.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 827.0, + 1912.0, + 827.0, + 1946.0, + 295.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 826.0, + 1942.0, + 826.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 826.0, + 1972.0, + 826.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 2004.0, + 557.0, + 2004.0, + 557.0, + 2034.0, + 297.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1054.0, + 1405.0, + 1054.0, + 1405.0, + 1091.0, + 294.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1086.0, + 1406.0, + 1086.0, + 1406.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1115.0, + 1406.0, + 1115.0, + 1406.0, + 1154.0, + 292.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1147.0, + 512.0, + 1147.0, + 512.0, + 1184.0, + 294.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1313.0, + 825.0, + 1313.0, + 825.0, + 1351.0, + 295.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1348.0, + 827.0, + 1348.0, + 827.0, + 1379.0, + 295.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1377.0, + 827.0, + 1377.0, + 827.0, + 1408.0, + 295.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1406.0, + 828.0, + 1406.0, + 828.0, + 1441.0, + 294.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 354.0, + 1436.0, + 354.0, + 1473.0, + 294.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1436.0, + 445.0, + 1436.0, + 445.0, + 1473.0, + 391.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1436.0, + 827.0, + 1436.0, + 827.0, + 1473.0, + 482.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1468.0, + 827.0, + 1468.0, + 827.0, + 1502.0, + 295.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 826.0, + 1498.0, + 826.0, + 1534.0, + 293.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1531.0, + 636.0, + 1531.0, + 636.0, + 1563.0, + 296.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1191.0, + 1406.0, + 1191.0, + 1406.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1226.0, + 681.0, + 1226.0, + 681.0, + 1259.0, + 297.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1226.0, + 731.0, + 1226.0, + 731.0, + 1259.0, + 717.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1226.0, + 1404.0, + 1226.0, + 1404.0, + 1259.0, + 768.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1253.0, + 1406.0, + 1253.0, + 1406.0, + 1291.0, + 293.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1284.0, + 1404.0, + 1284.0, + 1404.0, + 1321.0, + 293.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 1406.0, + 227.0, + 1406.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 263.0, + 1402.0, + 263.0, + 1402.0, + 293.0, + 296.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 292.0, + 1012.0, + 292.0, + 1012.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 973.0, + 1406.0, + 973.0, + 1406.0, + 1015.0, + 292.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1007.0, + 634.0, + 1007.0, + 634.0, + 1044.0, + 294.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 700.0, + 452.0, + 700.0, + 452.0, + 735.0, + 296.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 700.0, + 705.0, + 700.0, + 705.0, + 735.0, + 480.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 700.0, + 1404.0, + 700.0, + 1404.0, + 735.0, + 736.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 733.0, + 840.0, + 733.0, + 840.0, + 765.0, + 296.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 438.0, + 372.0, + 438.0, + 372.0, + 481.0, + 294.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 438.0, + 1403.0, + 438.0, + 1403.0, + 481.0, + 416.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 473.0, + 849.0, + 473.0, + 849.0, + 509.0, + 294.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 831.0, + 364.0, + 831.0, + 364.0, + 870.0, + 294.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 831.0, + 1329.0, + 831.0, + 1329.0, + 870.0, + 488.0, + 870.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 593, + 1404, + 593, + 1404, + 869, + 297, + 869 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 296, + 884, + 1405, + 884, + 1405, + 1130, + 296, + 1130 + ], + "score": 0.981 + }, + { + "category_id": 3, + "poly": [ + 355, + 1165, + 1350, + 1165, + 1350, + 1780, + 355, + 1780 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 297, + 1801, + 1405, + 1801, + 1405, + 1955, + 297, + 1955 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 297, + 1972, + 1401, + 1972, + 1401, + 2034, + 297, + 2034 + ], + "score": 0.936 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.906 + }, + { + "category_id": 1, + "poly": [ + 297, + 422, + 1404, + 422, + 1404, + 575, + 297, + 575 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 842, + 2087, + 858, + 2087, + 858, + 2111, + 842, + 2111 + ], + "score": 0.666 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.26 + }, + { + "category_id": 13, + "poly": [ + 1170, + 748, + 1242, + 748, + 1242, + 776, + 1170, + 776 + ], + "score": 0.9, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 518, + 2003, + 552, + 2003, + 552, + 2034, + 518, + 2034 + ], + "score": 0.89, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 608, + 2004, + 641, + 2004, + 641, + 2034, + 608, + 2034 + ], + "score": 0.89, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 725, + 626, + 760, + 626, + 760, + 656, + 725, + 656 + ], + "score": 0.89, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 877, + 1925, + 972, + 1925, + 972, + 1952, + 877, + 1952 + ], + "score": 0.87, + "latex": "\\alpha = 0 . 8" + }, + { + "category_id": 13, + "poly": [ + 808, + 1925, + 827, + 1925, + 827, + 1951, + 808, + 1951 + ], + "score": 0.82, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1140, + 1069, + 1160, + 1069, + 1160, + 1095, + 1140, + 1095 + ], + "score": 0.81, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 823, + 453, + 857, + 453, + 857, + 483, + 823, + 483 + ], + "score": 0.81, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1016, + 917, + 1036, + 917, + 1036, + 943, + 1016, + 943 + ], + "score": 0.8, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 503, + 1834, + 523, + 1834, + 523, + 1860, + 503, + 1860 + ], + "score": 0.8, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 871, + 453, + 906, + 453, + 906, + 484, + 871, + 484 + ], + "score": 0.8, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1091, + 920, + 1114, + 920, + 1114, + 944, + 1091, + 944 + ], + "score": 0.8, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1270, + 1894, + 1304, + 1894, + 1304, + 1924, + 1270, + 1924 + ], + "score": 0.78, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1382, + 1008, + 1401, + 1008, + 1401, + 1035, + 1382, + 1035 + ], + "score": 0.77, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1317, + 1894, + 1352, + 1894, + 1352, + 1924, + 1317, + 1924 + ], + "score": 0.75, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 658, + 1100, + 677, + 1100, + 677, + 1126, + 658, + 1126 + ], + "score": 0.74, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1381, + 1039, + 1402, + 1039, + 1402, + 1066, + 1381, + 1066 + ], + "score": 0.73, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 909, + 948, + 928, + 948, + 928, + 974, + 909, + 974 + ], + "score": 0.71, + "latex": "k" + }, + { + "category_id": 5, + "poly": [ + 297, + 223, + 1411, + 223, + 1411, + 401, + 297, + 401 + ], + "score": 0.728, + "html": "
ExperimentTheoretical Upper BoundExp. Convergence Time (in seconds)Metric
(in seconds)L1L2Lyap.L1L2Lyap.
SingleNeuron (acc)~3.4e30.0650.0640.0250.950.951.0
MLP (rmse)~ 1.4e104.3e33.4e33.1e30.0910.1790.085
" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1159.0, + 412.0, + 1159.0, + 412.0, + 1188.0, + 377.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1160.0, + 741.0, + 1160.0, + 741.0, + 1189.0, + 703.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1160.0, + 1069.0, + 1160.0, + 1069.0, + 1189.0, + 1032.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1180.0, + 626.0, + 1180.0, + 626.0, + 1205.0, + 562.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1175.0, + 956.0, + 1175.0, + 956.0, + 1208.0, + 889.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1179.0, + 1283.0, + 1179.0, + 1283.0, + 1205.0, + 1219.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1205.0, + 409.0, + 1205.0, + 409.0, + 1394.0, + 356.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1199.0, + 675.0, + 1199.0, + 675.0, + 1246.0, + 562.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1203.0, + 738.0, + 1203.0, + 738.0, + 1395.0, + 686.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1199.0, + 1004.0, + 1199.0, + 1004.0, + 1244.0, + 890.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1208.0, + 1067.0, + 1208.0, + 1067.0, + 1392.0, + 1012.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1199.0, + 1332.0, + 1199.0, + 1332.0, + 1244.0, + 1218.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1262.0, + 409.0, + 1262.0, + 409.0, + 1287.0, + 376.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1262.0, + 738.0, + 1262.0, + 738.0, + 1287.0, + 704.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1262.0, + 1067.0, + 1262.0, + 1067.0, + 1287.0, + 1033.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1274.0, + 1087.0, + 1274.0, + 1087.0, + 1287.0, + 1078.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1307.0, + 805.0, + 1307.0, + 805.0, + 1318.0, + 795.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1286.0, + 1094.0, + 1286.0, + 1094.0, + 1343.0, + 1063.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1315.0, + 408.0, + 1315.0, + 408.0, + 1336.0, + 377.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1313.0, + 737.0, + 1313.0, + 737.0, + 1337.0, + 705.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1315.0, + 1065.0, + 1315.0, + 1065.0, + 1335.0, + 1031.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1363.0, + 408.0, + 1363.0, + 408.0, + 1389.0, + 376.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1363.0, + 737.0, + 1363.0, + 737.0, + 1388.0, + 704.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1362.0, + 1065.0, + 1362.0, + 1065.0, + 1388.0, + 1033.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1408.0, + 437.0, + 1408.0, + 437.0, + 1454.0, + 373.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1423.0, + 483.0, + 1423.0, + 483.0, + 1449.0, + 438.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 1424.0, + 538.0, + 1424.0, + 538.0, + 1448.0, + 493.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1424.0, + 590.0, + 1424.0, + 590.0, + 1448.0, + 546.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1424.0, + 648.0, + 1424.0, + 648.0, + 1448.0, + 600.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1422.0, + 697.0, + 1422.0, + 697.0, + 1449.0, + 651.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1408.0, + 815.0, + 1408.0, + 815.0, + 1454.0, + 700.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1422.0, + 871.0, + 1422.0, + 871.0, + 1449.0, + 824.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1422.0, + 927.0, + 1422.0, + 927.0, + 1450.0, + 881.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1423.0, + 983.0, + 1423.0, + 983.0, + 1450.0, + 936.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1409.0, + 1140.0, + 1409.0, + 1140.0, + 1453.0, + 1030.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1424.0, + 1192.0, + 1424.0, + 1192.0, + 1448.0, + 1149.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1423.0, + 1247.0, + 1423.0, + 1247.0, + 1450.0, + 1200.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1424.0, + 1300.0, + 1424.0, + 1300.0, + 1448.0, + 1255.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1424.0, + 1351.0, + 1424.0, + 1351.0, + 1448.0, + 1306.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 1438.0, + 611.0, + 1438.0, + 611.0, + 1467.0, + 474.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1438.0, + 939.0, + 1438.0, + 939.0, + 1467.0, + 802.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1439.0, + 1268.0, + 1439.0, + 1268.0, + 1465.0, + 1132.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 1462.0, + 416.0, + 1462.0, + 416.0, + 1491.0, + 375.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1456.0, + 563.0, + 1456.0, + 563.0, + 1481.0, + 532.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1462.0, + 741.0, + 1462.0, + 741.0, + 1491.0, + 703.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1456.0, + 892.0, + 1456.0, + 892.0, + 1481.0, + 860.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1465.0, + 1068.0, + 1465.0, + 1068.0, + 1490.0, + 1033.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1456.0, + 1220.0, + 1456.0, + 1220.0, + 1481.0, + 1189.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1480.0, + 625.0, + 1480.0, + 625.0, + 1510.0, + 561.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1480.0, + 954.0, + 1480.0, + 954.0, + 1510.0, + 890.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1491.0, + 1215.0, + 1491.0, + 1215.0, + 1499.0, + 1199.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1480.0, + 1283.0, + 1480.0, + 1283.0, + 1510.0, + 1218.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1500.0, + 626.0, + 1500.0, + 626.0, + 1530.0, + 561.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1498.0, + 956.0, + 1498.0, + 956.0, + 1532.0, + 889.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1500.0, + 1283.0, + 1500.0, + 1283.0, + 1530.0, + 1218.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1515.0, + 409.0, + 1515.0, + 409.0, + 1688.0, + 358.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1525.0, + 675.0, + 1525.0, + 675.0, + 1550.0, + 563.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1516.0, + 738.0, + 1516.0, + 738.0, + 1687.0, + 686.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1525.0, + 1003.0, + 1525.0, + 1003.0, + 1550.0, + 891.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1517.0, + 1064.0, + 1517.0, + 1064.0, + 1684.0, + 1014.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1525.0, + 1333.0, + 1525.0, + 1333.0, + 1550.0, + 1220.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1565.0, + 410.0, + 1565.0, + 410.0, + 1591.0, + 376.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1561.0, + 572.0, + 1561.0, + 572.0, + 1574.0, + 557.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1566.0, + 738.0, + 1566.0, + 738.0, + 1591.0, + 704.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1566.0, + 1067.0, + 1566.0, + 1067.0, + 1591.0, + 1033.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1615.0, + 410.0, + 1615.0, + 410.0, + 1642.0, + 376.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1615.0, + 738.0, + 1615.0, + 738.0, + 1641.0, + 705.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1615.0, + 818.0, + 1615.0, + 818.0, + 1627.0, + 809.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1617.0, + 1067.0, + 1617.0, + 1067.0, + 1641.0, + 1033.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 1649.0, + 853.0, + 1649.0, + 853.0, + 1659.0, + 841.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1666.0, + 409.0, + 1666.0, + 409.0, + 1691.0, + 376.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1666.0, + 737.0, + 1666.0, + 737.0, + 1690.0, + 705.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1663.0, + 1066.0, + 1663.0, + 1066.0, + 1692.0, + 1033.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1711.0, + 437.0, + 1711.0, + 437.0, + 1758.0, + 372.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1727.0, + 483.0, + 1727.0, + 483.0, + 1753.0, + 438.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 1727.0, + 537.0, + 1727.0, + 537.0, + 1751.0, + 493.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1726.0, + 592.0, + 1726.0, + 592.0, + 1753.0, + 546.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1727.0, + 645.0, + 1727.0, + 645.0, + 1751.0, + 600.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1724.0, + 697.0, + 1724.0, + 697.0, + 1753.0, + 652.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1712.0, + 763.0, + 1712.0, + 763.0, + 1757.0, + 700.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1725.0, + 814.0, + 1725.0, + 814.0, + 1753.0, + 768.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1727.0, + 870.0, + 1727.0, + 870.0, + 1753.0, + 825.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1726.0, + 928.0, + 1726.0, + 928.0, + 1753.0, + 881.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1726.0, + 983.0, + 1726.0, + 983.0, + 1753.0, + 936.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1710.0, + 1139.0, + 1710.0, + 1139.0, + 1758.0, + 1030.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1727.0, + 1194.0, + 1727.0, + 1194.0, + 1753.0, + 1149.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1726.0, + 1248.0, + 1726.0, + 1248.0, + 1753.0, + 1200.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1727.0, + 1300.0, + 1727.0, + 1300.0, + 1751.0, + 1255.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 1727.0, + 1350.0, + 1727.0, + 1350.0, + 1751.0, + 1308.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1737.0, + 612.0, + 1737.0, + 612.0, + 1772.0, + 472.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 1741.0, + 939.0, + 1741.0, + 939.0, + 1770.0, + 803.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1741.0, + 1269.0, + 1741.0, + 1269.0, + 1770.0, + 1131.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1758.0, + 563.0, + 1758.0, + 563.0, + 1784.0, + 532.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1758.0, + 891.0, + 1758.0, + 891.0, + 1784.0, + 860.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1760.0, + 1217.0, + 1760.0, + 1217.0, + 1784.0, + 1192.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1177.0, + 416.0, + 1177.0, + 416.0, + 1185.0, + 411.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1230.0, + 513.0, + 1230.0, + 513.0, + 1245.0, + 479.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.75, + 1244.0, + 761.75, + 1244.0, + 761.75, + 1254.5, + 754.75, + 1254.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1248.5, + 806.0, + 1248.5, + 806.0, + 1309.5, + 743.0, + 1309.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.25, + 1353.5, + 1116.25, + 1353.5, + 1116.25, + 1367.5, + 1111.25, + 1367.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1368.5, + 882.0, + 1368.5, + 882.0, + 1388.5, + 846.0, + 1388.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1480.5, + 416.0, + 1480.5, + 416.0, + 1489.5, + 411.0, + 1489.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1573.5, + 795.0, + 1573.5, + 795.0, + 1593.5, + 764.0, + 1593.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.25, + 1624.5, + 1112.25, + 1624.5, + 1112.25, + 1639.0, + 1096.25, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.25, + 1659.5, + 892.25, + 1659.5, + 892.25, + 1681.5, + 850.25, + 1681.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1799.0, + 1406.0, + 1799.0, + 1406.0, + 1836.0, + 295.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1832.0, + 502.0, + 1832.0, + 502.0, + 1869.0, + 295.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1832.0, + 1407.0, + 1832.0, + 1407.0, + 1869.0, + 524.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1863.0, + 1405.0, + 1863.0, + 1405.0, + 1897.0, + 296.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1893.0, + 1269.0, + 1893.0, + 1269.0, + 1930.0, + 295.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1893.0, + 1316.0, + 1893.0, + 1316.0, + 1930.0, + 1305.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1893.0, + 1406.0, + 1893.0, + 1406.0, + 1930.0, + 1353.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1924.0, + 807.0, + 1924.0, + 807.0, + 1957.0, + 293.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1924.0, + 876.0, + 1924.0, + 876.0, + 1957.0, + 828.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1924.0, + 977.0, + 1924.0, + 977.0, + 1957.0, + 973.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 591.0, + 1405.0, + 591.0, + 1405.0, + 629.0, + 294.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 623.0, + 724.0, + 623.0, + 724.0, + 662.0, + 292.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 623.0, + 1406.0, + 623.0, + 1406.0, + 662.0, + 761.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 655.0, + 1404.0, + 655.0, + 1404.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 684.0, + 1405.0, + 684.0, + 1405.0, + 721.0, + 294.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 718.0, + 1405.0, + 718.0, + 1405.0, + 751.0, + 294.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 747.0, + 1169.0, + 747.0, + 1169.0, + 782.0, + 292.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 747.0, + 1406.0, + 747.0, + 1406.0, + 782.0, + 1243.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 777.0, + 1406.0, + 777.0, + 1406.0, + 812.0, + 294.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 807.0, + 1405.0, + 807.0, + 1405.0, + 843.0, + 294.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 842.0, + 374.0, + 842.0, + 374.0, + 871.0, + 295.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 884.0, + 1403.0, + 884.0, + 1403.0, + 921.0, + 294.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 916.0, + 1015.0, + 916.0, + 1015.0, + 949.0, + 295.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 916.0, + 1090.0, + 916.0, + 1090.0, + 949.0, + 1037.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 916.0, + 1405.0, + 916.0, + 1405.0, + 949.0, + 1115.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 947.0, + 908.0, + 947.0, + 908.0, + 981.0, + 294.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 947.0, + 1405.0, + 947.0, + 1405.0, + 981.0, + 929.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 978.0, + 1402.0, + 978.0, + 1402.0, + 1008.0, + 296.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1005.0, + 1381.0, + 1005.0, + 1381.0, + 1041.0, + 294.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1005.0, + 1406.0, + 1005.0, + 1406.0, + 1041.0, + 1402.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1036.0, + 1380.0, + 1036.0, + 1380.0, + 1072.0, + 293.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1036.0, + 1406.0, + 1036.0, + 1406.0, + 1072.0, + 1403.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1065.0, + 1139.0, + 1065.0, + 1139.0, + 1104.0, + 291.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1065.0, + 1406.0, + 1065.0, + 1406.0, + 1104.0, + 1161.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1098.0, + 657.0, + 1098.0, + 657.0, + 1134.0, + 293.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1098.0, + 1090.0, + 1098.0, + 1090.0, + 1134.0, + 678.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1970.0, + 1403.0, + 1970.0, + 1403.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2004.0, + 517.0, + 2004.0, + 517.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 2004.0, + 607.0, + 2004.0, + 607.0, + 2036.0, + 553.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 2004.0, + 1403.0, + 2004.0, + 1403.0, + 2036.0, + 642.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 421.0, + 1405.0, + 421.0, + 1405.0, + 455.0, + 296.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 451.0, + 822.0, + 451.0, + 822.0, + 489.0, + 294.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 451.0, + 870.0, + 451.0, + 870.0, + 489.0, + 858.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 451.0, + 1405.0, + 451.0, + 1405.0, + 489.0, + 907.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 484.0, + 1404.0, + 484.0, + 1404.0, + 517.0, + 296.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 510.0, + 1406.0, + 510.0, + 1406.0, + 550.0, + 294.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 545.0, + 1148.0, + 545.0, + 1148.0, + 578.0, + 296.0, + 578.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 320, + 419, + 1380, + 419, + 1380, + 656, + 320, + 656 + ], + "score": 0.981, + "html": "
Experiment for single neuron case on Iris datasetTheoretical Upper Bound (in seconds)Exp. Convergence Time (in seconds)Accuracy on Test Set
L1L2Lyap.L1L2Lyap.
k=0.001~ 37213.8480.3120.3100.09180.60.950.95
k = 0.005~7442.7690.09430.09420.03440.951.01.0
k= 0.01~ 3721.3850.05660.05730.02210.951.01.0
" + }, + { + "category_id": 1, + "poly": [ + 297, + 819, + 1404, + 819, + 1404, + 1034, + 297, + 1034 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1850, + 1405, + 1850, + 1405, + 2034, + 297, + 2034 + ], + "score": 0.979 + }, + { + "category_id": 3, + "poly": [ + 373, + 1120, + 1330, + 1120, + 1330, + 1687, + 373, + 1687 + ], + "score": 0.974 + }, + { + "category_id": 4, + "poly": [ + 296, + 1709, + 1404, + 1709, + 1404, + 1831, + 296, + 1831 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1403, + 228, + 1403, + 394, + 298, + 394 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 299, + 678, + 1403, + 678, + 1403, + 801, + 299, + 801 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 104, + 298, + 104 + ], + "score": 0.8 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.799 + }, + { + "category_id": 4, + "poly": [ + 297, + 1080, + 1165, + 1080, + 1165, + 1113, + 297, + 1113 + ], + "score": 0.641 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 852, + 76, + 852, + 104, + 299, + 104 + ], + "score": 0.294 + }, + { + "category_id": 0, + "poly": [ + 297, + 1080, + 1165, + 1080, + 1165, + 1113, + 297, + 1113 + ], + "score": 0.24 + }, + { + "category_id": 13, + "poly": [ + 529, + 357, + 566, + 357, + 566, + 397, + 529, + 397 + ], + "score": 0.91, + "latex": "k _ { i j } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1255, + 259, + 1292, + 259, + 1292, + 299, + 1255, + 299 + ], + "score": 0.91, + "latex": "k _ { i j } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 957, + 295, + 994, + 295, + 994, + 336, + 957, + 336 + ], + "score": 0.91, + "latex": "k _ { i j } ^ { l }" + }, + { + "category_id": 13, + "poly": [ + 902, + 1771, + 1015, + 1771, + 1015, + 1800, + 902, + 1800 + ], + "score": 0.9, + "latex": "\\Delta x = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1077, + 974, + 1157, + 974, + 1157, + 1001, + 1077, + 1001 + ], + "score": 0.9, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 1235, + 882, + 1392, + 882, + 1392, + 914, + 1235, + 914 + ], + "score": 0.89, + "latex": "\\alpha \\in ( 0 . 5 , 0 . 9 )" + }, + { + "category_id": 13, + "poly": [ + 458, + 1741, + 500, + 1741, + 500, + 1768, + 458, + 1768 + ], + "score": 0.89, + "latex": "\\Delta x" + }, + { + "category_id": 13, + "poly": [ + 1286, + 851, + 1348, + 851, + 1348, + 883, + 1286, + 883 + ], + "score": 0.89, + "latex": "( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 587, + 1802, + 682, + 1802, + 682, + 1829, + 587, + 1829 + ], + "score": 0.88, + "latex": "\\alpha = 0 . 8" + }, + { + "category_id": 13, + "poly": [ + 453, + 1801, + 577, + 1801, + 577, + 1830, + 453, + 1830 + ], + "score": 0.87, + "latex": "k = 0 . 0 0 0 9" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1770, + 1180, + 1770, + 1180, + 1800, + 1066, + 1800 + ], + "score": 0.86, + "latex": "\\Delta x = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 1191, + 1770, + 1337, + 1770, + 1337, + 1802, + 1191, + 1802 + ], + "score": 0.84, + "latex": "( \\mathrm { c } ) \\Delta x = 0 . 3" + }, + { + "category_id": 13, + "poly": [ + 769, + 1801, + 831, + 1801, + 831, + 1829, + 769, + 1829 + ], + "score": 0.82, + "latex": "_ { \\mathrm { - } 1 0 0 }" + }, + { + "category_id": 13, + "poly": [ + 1072, + 855, + 1093, + 855, + 1093, + 879, + 1072, + 879 + ], + "score": 0.8, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 297, + 740, + 331, + 740, + 331, + 770, + 297, + 770 + ], + "score": 0.79, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1138, + 680, + 1158, + 680, + 1158, + 706, + 1138, + 706 + ], + "score": 0.78, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 341, + 741, + 376, + 741, + 376, + 770, + 341, + 770 + ], + "score": 0.78, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 854, + 827, + 876, + 827, + 876, + 848, + 854, + 848 + ], + "score": 0.76, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1353, + 360, + 1385, + 360, + 1385, + 387, + 1353, + 387 + ], + "score": 0.74, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 327, + 916, + 350, + 916, + 350, + 940, + 327, + 940 + ], + "score": 0.73, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 924, + 491, + 960, + 491, + 960, + 522, + 924, + 522 + ], + "score": 0.35, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 491, + 1247, + 491, + 1247, + 523, + 1211, + 523 + ], + "score": 0.33, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 815, + 491, + 850, + 491, + 850, + 522, + 815, + 522 + ], + "score": 0.26, + "latex": "L _ { 1 }" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1122.0, + 422.0, + 1122.0, + 422.0, + 1141.0, + 374.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1120.0, + 742.0, + 1120.0, + 742.0, + 1142.0, + 694.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1121.0, + 1061.0, + 1121.0, + 1061.0, + 1141.0, + 1012.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1131.0, + 400.0, + 1131.0, + 400.0, + 1311.0, + 373.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 1131.0, + 1040.0, + 1131.0, + 1040.0, + 1311.0, + 1011.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1133.0, + 717.0, + 1133.0, + 717.0, + 1252.0, + 693.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1162.0, + 1064.0, + 1162.0, + 1064.0, + 1192.0, + 1028.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1164.0, + 423.0, + 1164.0, + 423.0, + 1190.0, + 391.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1164.0, + 742.0, + 1164.0, + 742.0, + 1191.0, + 711.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1212.0, + 423.0, + 1212.0, + 423.0, + 1236.0, + 391.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 1210.0, + 742.0, + 1210.0, + 742.0, + 1236.0, + 709.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1210.0, + 1063.0, + 1210.0, + 1063.0, + 1236.0, + 1029.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1259.0, + 422.0, + 1259.0, + 422.0, + 1280.0, + 391.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1241.0, + 742.0, + 1241.0, + 742.0, + 1308.0, + 695.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1255.0, + 1064.0, + 1255.0, + 1064.0, + 1284.0, + 1028.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1286.0, + 444.0, + 1286.0, + 444.0, + 1395.0, + 360.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1286.0, + 527.0, + 1286.0, + 527.0, + 1334.0, + 465.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1288.0, + 756.0, + 1288.0, + 756.0, + 1385.0, + 684.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1285.0, + 847.0, + 1285.0, + 847.0, + 1333.0, + 783.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1286.0, + 1077.0, + 1286.0, + 1077.0, + 1386.0, + 1003.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1285.0, + 1167.0, + 1285.0, + 1167.0, + 1333.0, + 1102.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1330.0, + 578.0, + 1330.0, + 578.0, + 1354.0, + 466.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1330.0, + 899.0, + 1330.0, + 899.0, + 1354.0, + 785.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1330.0, + 1218.0, + 1330.0, + 1218.0, + 1353.0, + 1103.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1356.0, + 479.0, + 1356.0, + 479.0, + 1382.0, + 449.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1356.0, + 526.0, + 1356.0, + 526.0, + 1380.0, + 494.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1357.0, + 569.0, + 1357.0, + 569.0, + 1379.0, + 542.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1359.0, + 618.0, + 1359.0, + 618.0, + 1378.0, + 581.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1357.0, + 663.0, + 1357.0, + 663.0, + 1380.0, + 625.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1356.0, + 800.0, + 1356.0, + 800.0, + 1382.0, + 770.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1356.0, + 847.0, + 1356.0, + 847.0, + 1380.0, + 815.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1357.0, + 892.0, + 1357.0, + 892.0, + 1379.0, + 866.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1357.0, + 944.0, + 1357.0, + 944.0, + 1380.0, + 904.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1357.0, + 990.0, + 1357.0, + 990.0, + 1380.0, + 951.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1356.0, + 1119.0, + 1356.0, + 1119.0, + 1382.0, + 1087.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1356.0, + 1165.0, + 1356.0, + 1165.0, + 1380.0, + 1132.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1359.0, + 1208.0, + 1359.0, + 1208.0, + 1379.0, + 1182.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1359.0, + 1258.0, + 1359.0, + 1258.0, + 1378.0, + 1220.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1357.0, + 1304.0, + 1357.0, + 1304.0, + 1380.0, + 1265.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1385.0, + 430.0, + 1385.0, + 430.0, + 1435.0, + 370.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1371.0, + 618.0, + 1371.0, + 618.0, + 1411.0, + 480.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1373.0, + 761.0, + 1373.0, + 761.0, + 1537.0, + 674.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1371.0, + 940.0, + 1371.0, + 940.0, + 1412.0, + 801.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1388.0, + 1063.0, + 1388.0, + 1063.0, + 1547.0, + 1010.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1369.0, + 1257.0, + 1369.0, + 1257.0, + 1410.0, + 1120.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 1416.0, + 423.0, + 1416.0, + 423.0, + 1545.0, + 375.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 1441.0, + 742.0, + 1441.0, + 742.0, + 1466.0, + 709.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1441.0, + 1063.0, + 1441.0, + 1063.0, + 1466.0, + 1028.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1487.0, + 423.0, + 1487.0, + 423.0, + 1512.0, + 388.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 1487.0, + 742.0, + 1487.0, + 742.0, + 1512.0, + 708.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1487.0, + 1063.0, + 1487.0, + 1063.0, + 1512.0, + 1028.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1518.0, + 440.0, + 1518.0, + 440.0, + 1671.0, + 350.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1562.0, + 526.0, + 1562.0, + 526.0, + 1588.0, + 466.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1512.0, + 750.0, + 1512.0, + 750.0, + 1655.0, + 687.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1562.0, + 845.0, + 1562.0, + 845.0, + 1588.0, + 785.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1516.0, + 1079.0, + 1516.0, + 1079.0, + 1670.0, + 996.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1562.0, + 1166.0, + 1562.0, + 1166.0, + 1588.0, + 1089.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1582.0, + 528.0, + 1582.0, + 528.0, + 1610.0, + 465.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 1579.0, + 741.0, + 1579.0, + 741.0, + 1604.0, + 708.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1582.0, + 847.0, + 1582.0, + 847.0, + 1610.0, + 784.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1582.0, + 1167.0, + 1582.0, + 1167.0, + 1610.0, + 1103.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1605.0, + 578.0, + 1605.0, + 578.0, + 1629.0, + 465.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1608.0, + 898.0, + 1608.0, + 898.0, + 1628.0, + 785.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1608.0, + 1217.0, + 1608.0, + 1217.0, + 1628.0, + 1106.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1632.0, + 479.0, + 1632.0, + 479.0, + 1657.0, + 449.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1632.0, + 526.0, + 1632.0, + 526.0, + 1656.0, + 494.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1634.0, + 569.0, + 1634.0, + 569.0, + 1656.0, + 541.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1633.0, + 663.0, + 1633.0, + 663.0, + 1658.0, + 577.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 1634.0, + 799.0, + 1634.0, + 799.0, + 1656.0, + 771.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1632.0, + 847.0, + 1632.0, + 847.0, + 1657.0, + 815.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1633.0, + 990.0, + 1633.0, + 990.0, + 1658.0, + 865.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1633.0, + 1119.0, + 1633.0, + 1119.0, + 1657.0, + 1088.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1632.0, + 1165.0, + 1632.0, + 1165.0, + 1656.0, + 1132.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1634.0, + 1208.0, + 1634.0, + 1208.0, + 1656.0, + 1181.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1633.0, + 1304.0, + 1633.0, + 1304.0, + 1658.0, + 1217.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1646.0, + 618.0, + 1646.0, + 618.0, + 1674.0, + 481.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1648.0, + 938.0, + 1648.0, + 938.0, + 1672.0, + 801.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1648.0, + 1257.0, + 1648.0, + 1257.0, + 1672.0, + 1121.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1666.0, + 570.0, + 1666.0, + 570.0, + 1687.0, + 541.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1666.0, + 889.0, + 1666.0, + 889.0, + 1687.0, + 860.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1665.0, + 1206.0, + 1665.0, + 1206.0, + 1688.0, + 1182.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1348.0, + 745.0, + 1348.0, + 745.0, + 1376.5, + 712.0, + 1376.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1348.0, + 1066.0, + 1348.0, + 1066.0, + 1377.0, + 1030.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.25, + 1622.5, + 747.25, + 1622.5, + 747.25, + 1654.0, + 709.25, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1622.5, + 1066.0, + 1622.5, + 1066.0, + 1654.0, + 1028.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1709.0, + 1405.0, + 1709.0, + 1405.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1740.0, + 457.0, + 1740.0, + 457.0, + 1773.0, + 291.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1773.0, + 501.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1771.0, + 901.0, + 1771.0, + 901.0, + 1803.0, + 293.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 1771.0, + 1065.0, + 1771.0, + 1065.0, + 1803.0, + 1016.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1771.0, + 1190.0, + 1771.0, + 1190.0, + 1803.0, + 1181.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1771.0, + 1343.0, + 1771.0, + 1343.0, + 1803.0, + 1338.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1801.0, + 452.0, + 1801.0, + 452.0, + 1834.0, + 294.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1801.0, + 586.0, + 1801.0, + 586.0, + 1834.0, + 578.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1801.0, + 768.0, + 1801.0, + 768.0, + 1834.0, + 683.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1078.0, + 1170.0, + 1078.0, + 1170.0, + 1118.0, + 294.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1078.0, + 1170.0, + 1078.0, + 1170.0, + 1118.0, + 294.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 818.0, + 853.0, + 818.0, + 853.0, + 855.0, + 295.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 818.0, + 1404.0, + 818.0, + 1404.0, + 855.0, + 877.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 850.0, + 1071.0, + 850.0, + 1071.0, + 885.0, + 295.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 850.0, + 1285.0, + 850.0, + 1285.0, + 885.0, + 1094.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 850.0, + 1405.0, + 850.0, + 1405.0, + 885.0, + 1349.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 878.0, + 1234.0, + 878.0, + 1234.0, + 915.0, + 294.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 878.0, + 1402.0, + 878.0, + 1402.0, + 915.0, + 1393.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 908.0, + 326.0, + 908.0, + 326.0, + 949.0, + 291.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 908.0, + 1405.0, + 908.0, + 1405.0, + 949.0, + 351.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 941.0, + 1406.0, + 941.0, + 1406.0, + 976.0, + 294.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 971.0, + 1076.0, + 971.0, + 1076.0, + 1008.0, + 292.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 971.0, + 1404.0, + 971.0, + 1404.0, + 1008.0, + 1158.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1000.0, + 794.0, + 1000.0, + 794.0, + 1037.0, + 294.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1882.0, + 295.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1918.0, + 292.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1947.0, + 295.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 1113.0, + 2000.0, + 1113.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1405.0, + 230.0, + 1405.0, + 263.0, + 294.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 255.0, + 1254.0, + 255.0, + 1254.0, + 300.0, + 291.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 255.0, + 1405.0, + 255.0, + 1405.0, + 300.0, + 1293.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 297.0, + 956.0, + 297.0, + 956.0, + 334.0, + 294.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 297.0, + 1405.0, + 297.0, + 1405.0, + 334.0, + 995.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 331.0, + 1402.0, + 331.0, + 1402.0, + 361.0, + 297.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 357.0, + 528.0, + 357.0, + 528.0, + 398.0, + 292.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 357.0, + 1352.0, + 357.0, + 1352.0, + 398.0, + 567.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 357.0, + 1400.0, + 357.0, + 1400.0, + 398.0, + 1386.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 677.0, + 1137.0, + 677.0, + 1137.0, + 713.0, + 294.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 677.0, + 1405.0, + 677.0, + 1405.0, + 713.0, + 1159.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 710.0, + 1405.0, + 710.0, + 1405.0, + 743.0, + 294.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 741.0, + 340.0, + 741.0, + 340.0, + 773.0, + 332.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 741.0, + 1404.0, + 741.0, + 1404.0, + 773.0, + 377.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 772.0, + 435.0, + 772.0, + 435.0, + 804.0, + 293.0, + 804.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1637, + 1404, + 1637, + 1404, + 2034, + 297, + 2034 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 490, + 1404, + 490, + 1404, + 796, + 297, + 796 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1404, + 229, + 1404, + 475, + 297, + 475 + ], + "score": 0.982 + }, + { + "category_id": 5, + "poly": [ + 323, + 1342, + 1378, + 1342, + 1378, + 1490, + 323, + 1490 + ], + "score": 0.969, + "html": "
ExperimentTheoretical Upper BoundExp. Convergence Time (in seconds)Metric
(in seconds)L1L2Lyap.L1rmse L2Lyap.
IMDBWiki8.3e6980.40712.99467.960.4140.4150.416
" + }, + { + "category_id": 3, + "poly": [ + 500, + 849, + 1186, + 849, + 1186, + 1222, + 500, + 1222 + ], + "score": 0.967 + }, + { + "category_id": 0, + "poly": [ + 298, + 1569, + 844, + 1569, + 844, + 1606, + 298, + 1606 + ], + "score": 0.921 + }, + { + "category_id": 4, + "poly": [ + 293, + 1252, + 1401, + 1252, + 1401, + 1315, + 293, + 1315 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 852, + 76, + 852, + 104, + 300, + 104 + ], + "score": 0.839 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 391, + 1510, + 1305, + 1510, + 1305, + 1543, + 391, + 1543 + ], + "score": 0.689 + }, + { + "category_id": 7, + "poly": [ + 391, + 1510, + 1305, + 1510, + 1305, + 1543, + 391, + 1543 + ], + "score": 0.147 + }, + { + "category_id": 13, + "poly": [ + 1124, + 292, + 1258, + 292, + 1258, + 324, + 1124, + 324 + ], + "score": 0.91, + "latex": "( - \\Delta x , \\Delta x )" + }, + { + "category_id": 13, + "poly": [ + 1253, + 704, + 1288, + 704, + 1288, + 734, + 1253, + 734 + ], + "score": 0.91, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1307, + 643, + 1402, + 643, + 1402, + 672, + 1307, + 672 + ], + "score": 0.9, + "latex": "\\alpha = 0 . 7" + }, + { + "category_id": 13, + "poly": [ + 724, + 1821, + 758, + 1821, + 758, + 1851, + 724, + 1851 + ], + "score": 0.89, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1160, + 705, + 1196, + 705, + 1196, + 734, + 1160, + 734 + ], + "score": 0.88, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 811, + 1821, + 846, + 1821, + 846, + 1851, + 811, + 1851 + ], + "score": 0.88, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 895, + 1512, + 929, + 1512, + 929, + 1541, + 895, + 1541 + ], + "score": 0.85, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 940, + 1511, + 975, + 1511, + 975, + 1542, + 940, + 1542 + ], + "score": 0.84, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1254, + 552, + 1314, + 552, + 1314, + 581, + 1254, + 581 + ], + "score": 0.84, + "latex": "{ \\sim } 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 862, + 583, + 922, + 583, + 922, + 612, + 862, + 612 + ], + "score": 0.83, + "latex": "{ \\sim } 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1127, + 231, + 1161, + 231, + 1161, + 259, + 1127, + 259 + ], + "score": 0.75, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1303, + 322, + 1337, + 322, + 1337, + 349, + 1303, + 349 + ], + "score": 0.75, + "latex": "M" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 850.0, + 543.0, + 850.0, + 543.0, + 872.0, + 511.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 852.0, + 894.0, + 852.0, + 894.0, + 871.0, + 866.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 914.0, + 542.0, + 914.0, + 542.0, + 937.0, + 511.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 914.0, + 894.0, + 914.0, + 894.0, + 937.0, + 863.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 959.0, + 907.0, + 959.0, + 907.0, + 1085.0, + 838.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 965.0, + 555.0, + 965.0, + 555.0, + 1079.0, + 485.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 977.0, + 894.0, + 977.0, + 894.0, + 1003.0, + 862.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1105.0, + 542.0, + 1105.0, + 542.0, + 1131.0, + 510.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1105.0, + 895.0, + 1105.0, + 895.0, + 1131.0, + 863.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1118.0, + 626.0, + 1118.0, + 626.0, + 1144.0, + 573.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 1118.0, + 979.0, + 1118.0, + 979.0, + 1144.0, + 923.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1137.0, + 627.0, + 1137.0, + 627.0, + 1161.0, + 573.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1137.0, + 979.0, + 1137.0, + 979.0, + 1161.0, + 926.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1155.0, + 668.0, + 1155.0, + 668.0, + 1177.0, + 574.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1155.0, + 1021.0, + 1155.0, + 1021.0, + 1177.0, + 926.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1176.0, + 612.0, + 1176.0, + 612.0, + 1200.0, + 570.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1178.0, + 664.0, + 1178.0, + 664.0, + 1199.0, + 624.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1178.0, + 718.0, + 1178.0, + 718.0, + 1199.0, + 677.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1178.0, + 826.0, + 1178.0, + 826.0, + 1198.0, + 729.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1167.0, + 902.0, + 1167.0, + 902.0, + 1201.0, + 862.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1176.0, + 964.0, + 1176.0, + 964.0, + 1200.0, + 922.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1178.0, + 1017.0, + 1178.0, + 1017.0, + 1199.0, + 977.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1177.0, + 1179.0, + 1177.0, + 1179.0, + 1199.0, + 1028.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 1190.0, + 738.0, + 1190.0, + 738.0, + 1212.0, + 626.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1190.0, + 1092.0, + 1190.0, + 1092.0, + 1212.0, + 978.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1203.0, + 700.0, + 1203.0, + 700.0, + 1226.0, + 672.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1203.0, + 1052.0, + 1203.0, + 1052.0, + 1226.0, + 1024.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.75, + 1168.5, + 544.75, + 1168.5, + 544.75, + 1197.5, + 512.75, + 1197.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1564.0, + 847.0, + 1564.0, + 847.0, + 1611.0, + 293.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1250.0, + 1406.0, + 1250.0, + 1406.0, + 1290.0, + 292.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1278.0, + 386.0, + 1278.0, + 386.0, + 1320.0, + 292.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1508.0, + 894.0, + 1508.0, + 894.0, + 1545.0, + 390.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1508.0, + 939.0, + 1508.0, + 939.0, + 1545.0, + 930.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1508.0, + 1306.0, + 1508.0, + 1306.0, + 1545.0, + 976.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1635.0, + 1406.0, + 1635.0, + 1406.0, + 1674.0, + 295.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1669.0, + 1404.0, + 1669.0, + 1404.0, + 1705.0, + 295.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1697.0, + 1402.0, + 1697.0, + 1402.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1728.0, + 1404.0, + 1728.0, + 1404.0, + 1764.0, + 295.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1759.0, + 1407.0, + 1759.0, + 1407.0, + 1795.0, + 294.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1788.0, + 1407.0, + 1788.0, + 1407.0, + 1826.0, + 292.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1820.0, + 723.0, + 1820.0, + 723.0, + 1855.0, + 292.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1820.0, + 810.0, + 1820.0, + 810.0, + 1855.0, + 759.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1820.0, + 1406.0, + 1820.0, + 1406.0, + 1855.0, + 847.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1850.0, + 1404.0, + 1850.0, + 1404.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1883.0, + 1401.0, + 1883.0, + 1401.0, + 1916.0, + 296.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1913.0, + 1404.0, + 1913.0, + 1404.0, + 1946.0, + 296.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1943.0, + 1401.0, + 1943.0, + 1401.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2012.0, + 294.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 1147.0, + 2000.0, + 1147.0, + 2040.0, + 292.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 488.0, + 1405.0, + 488.0, + 1405.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 521.0, + 1404.0, + 521.0, + 1404.0, + 557.0, + 295.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 550.0, + 1253.0, + 550.0, + 1253.0, + 586.0, + 295.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 550.0, + 1404.0, + 550.0, + 1404.0, + 586.0, + 1315.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 583.0, + 861.0, + 583.0, + 861.0, + 615.0, + 295.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 583.0, + 1404.0, + 583.0, + 1404.0, + 615.0, + 923.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 614.0, + 1405.0, + 614.0, + 1405.0, + 646.0, + 294.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 642.0, + 1306.0, + 642.0, + 1306.0, + 677.0, + 294.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 642.0, + 1406.0, + 642.0, + 1406.0, + 677.0, + 1403.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 675.0, + 1402.0, + 675.0, + 1402.0, + 706.0, + 296.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 705.0, + 1159.0, + 705.0, + 1159.0, + 737.0, + 295.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 705.0, + 1252.0, + 705.0, + 1252.0, + 737.0, + 1197.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 705.0, + 1402.0, + 705.0, + 1402.0, + 737.0, + 1289.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 732.0, + 1406.0, + 732.0, + 1406.0, + 772.0, + 292.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 766.0, + 853.0, + 766.0, + 853.0, + 798.0, + 296.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 231.0, + 1126.0, + 231.0, + 1126.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 231.0, + 1404.0, + 231.0, + 1404.0, + 264.0, + 1162.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 1123.0, + 292.0, + 1123.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 292.0, + 1404.0, + 292.0, + 1404.0, + 326.0, + 1259.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 322.0, + 1302.0, + 322.0, + 1302.0, + 356.0, + 295.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 322.0, + 1404.0, + 322.0, + 1404.0, + 356.0, + 1338.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 352.0, + 1406.0, + 352.0, + 1406.0, + 389.0, + 295.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 382.0, + 1405.0, + 382.0, + 1405.0, + 419.0, + 294.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 449.0, + 294.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 445.0, + 674.0, + 445.0, + 674.0, + 478.0, + 295.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1508.0, + 894.0, + 1508.0, + 894.0, + 1545.0, + 390.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1508.0, + 939.0, + 1508.0, + 939.0, + 1545.0, + 930.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1508.0, + 1306.0, + 1508.0, + 1306.0, + 1545.0, + 976.0, + 1545.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2088, + 866, + 2088, + 866, + 2113, + 836, + 2113 + ], + "score": 0.826 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 489, + 227, + 489, + 262, + 299, + 262 + ], + "score": 0.813 + }, + { + "category_id": 1, + "poly": [ + 291, + 996, + 1401, + 996, + 1401, + 1061, + 291, + 1061 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 294, + 1078, + 1401, + 1078, + 1401, + 1144, + 294, + 1144 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 295, + 1273, + 1401, + 1273, + 1401, + 1368, + 295, + 1368 + ], + "score": 0.782 + }, + { + "category_id": 1, + "poly": [ + 298, + 1160, + 1400, + 1160, + 1400, + 1254, + 298, + 1254 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 298, + 1499, + 1400, + 1499, + 1400, + 1594, + 298, + 1594 + ], + "score": 0.752 + }, + { + "category_id": 1, + "poly": [ + 295, + 688, + 1396, + 688, + 1396, + 753, + 295, + 753 + ], + "score": 0.747 + }, + { + "category_id": 1, + "poly": [ + 297, + 1612, + 1400, + 1612, + 1400, + 1677, + 297, + 1677 + ], + "score": 0.746 + }, + { + "category_id": 1, + "poly": [ + 300, + 1384, + 1401, + 1384, + 1401, + 1482, + 300, + 1482 + ], + "score": 0.744 + }, + { + "category_id": 1, + "poly": [ + 299, + 524, + 1402, + 524, + 1402, + 588, + 299, + 588 + ], + "score": 0.738 + }, + { + "category_id": 1, + "poly": [ + 296, + 883, + 1398, + 883, + 1398, + 979, + 296, + 979 + ], + "score": 0.73 + }, + { + "category_id": 1, + "poly": [ + 298, + 1693, + 1399, + 1693, + 1399, + 1759, + 298, + 1759 + ], + "score": 0.721 + }, + { + "category_id": 1, + "poly": [ + 296, + 606, + 1399, + 606, + 1399, + 672, + 296, + 672 + ], + "score": 0.711 + }, + { + "category_id": 1, + "poly": [ + 299, + 1775, + 1399, + 1775, + 1399, + 1842, + 299, + 1842 + ], + "score": 0.679 + }, + { + "category_id": 1, + "poly": [ + 297, + 1971, + 1399, + 1971, + 1399, + 2035, + 297, + 2035 + ], + "score": 0.67 + }, + { + "category_id": 1, + "poly": [ + 299, + 278, + 1400, + 278, + 1400, + 343, + 299, + 343 + ], + "score": 0.666 + }, + { + "category_id": 1, + "poly": [ + 298, + 441, + 1401, + 441, + 1401, + 507, + 298, + 507 + ], + "score": 0.654 + }, + { + "category_id": 1, + "poly": [ + 297, + 358, + 1399, + 358, + 1399, + 424, + 297, + 424 + ], + "score": 0.626 + }, + { + "category_id": 1, + "poly": [ + 296, + 1857, + 1401, + 1857, + 1401, + 1952, + 296, + 1952 + ], + "score": 0.606 + }, + { + "category_id": 1, + "poly": [ + 297, + 770, + 1401, + 770, + 1401, + 865, + 297, + 865 + ], + "score": 0.595 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.545 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.346 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 995.0, + 1404.0, + 995.0, + 1404.0, + 1034.0, + 293.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1026.0, + 561.0, + 1026.0, + 561.0, + 1062.0, + 323.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1076.0, + 1403.0, + 1076.0, + 1403.0, + 1115.0, + 293.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1107.0, + 397.0, + 1107.0, + 397.0, + 1143.0, + 322.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1274.0, + 1405.0, + 1274.0, + 1405.0, + 1308.0, + 295.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1306.0, + 1405.0, + 1306.0, + 1405.0, + 1340.0, + 323.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1336.0, + 609.0, + 1336.0, + 609.0, + 1366.0, + 323.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1160.0, + 1402.0, + 1160.0, + 1402.0, + 1197.0, + 294.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1191.0, + 1405.0, + 1191.0, + 1405.0, + 1229.0, + 321.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1223.0, + 462.0, + 1223.0, + 462.0, + 1254.0, + 322.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1499.0, + 1402.0, + 1499.0, + 1402.0, + 1536.0, + 294.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1531.0, + 1405.0, + 1531.0, + 1405.0, + 1566.0, + 321.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1560.0, + 864.0, + 1560.0, + 864.0, + 1598.0, + 322.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 687.0, + 1401.0, + 687.0, + 1401.0, + 727.0, + 294.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 722.0, + 571.0, + 722.0, + 571.0, + 753.0, + 324.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1612.0, + 1404.0, + 1612.0, + 1404.0, + 1648.0, + 298.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1641.0, + 1160.0, + 1641.0, + 1160.0, + 1679.0, + 321.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1386.0, + 1403.0, + 1386.0, + 1403.0, + 1421.0, + 298.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1416.0, + 1406.0, + 1416.0, + 1406.0, + 1457.0, + 320.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1448.0, + 948.0, + 1448.0, + 948.0, + 1482.0, + 323.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 520.0, + 1407.0, + 520.0, + 1407.0, + 564.0, + 293.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 553.0, + 396.0, + 553.0, + 396.0, + 589.0, + 322.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 881.0, + 1404.0, + 881.0, + 1404.0, + 922.0, + 292.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 914.0, + 1404.0, + 914.0, + 1404.0, + 949.0, + 322.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 947.0, + 718.0, + 947.0, + 718.0, + 980.0, + 324.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1694.0, + 1404.0, + 1694.0, + 1404.0, + 1730.0, + 297.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1725.0, + 1182.0, + 1725.0, + 1182.0, + 1760.0, + 323.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 607.0, + 1404.0, + 607.0, + 1404.0, + 643.0, + 296.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 638.0, + 1151.0, + 638.0, + 1151.0, + 673.0, + 321.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1773.0, + 1403.0, + 1773.0, + 1403.0, + 1815.0, + 297.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1807.0, + 1089.0, + 1807.0, + 1089.0, + 1842.0, + 321.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 1403.0, + 1971.0, + 1403.0, + 2007.0, + 296.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2003.0, + 1279.0, + 2003.0, + 1279.0, + 2036.0, + 322.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 276.0, + 1405.0, + 276.0, + 1405.0, + 317.0, + 293.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 309.0, + 1316.0, + 309.0, + 1316.0, + 345.0, + 322.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 440.0, + 1406.0, + 440.0, + 1406.0, + 480.0, + 293.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 473.0, + 992.0, + 473.0, + 992.0, + 509.0, + 323.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 360.0, + 1404.0, + 360.0, + 1404.0, + 396.0, + 296.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 390.0, + 1227.0, + 390.0, + 1227.0, + 425.0, + 322.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1856.0, + 1403.0, + 1856.0, + 1403.0, + 1895.0, + 295.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1891.0, + 1405.0, + 1891.0, + 1405.0, + 1925.0, + 320.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1920.0, + 516.0, + 1920.0, + 516.0, + 1951.0, + 324.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 772.0, + 1402.0, + 772.0, + 1402.0, + 807.0, + 296.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 803.0, + 1405.0, + 803.0, + 1405.0, + 841.0, + 321.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 828.0, + 399.0, + 828.0, + 399.0, + 868.0, + 320.0, + 868.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1592, + 1404, + 1592, + 1404, + 1779, + 297, + 1779 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1402, + 1941, + 1402, + 2034, + 299, + 2034 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 544, + 1014, + 1112, + 1014, + 1112, + 1434, + 544, + 1434 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 297, + 457, + 1404, + 457, + 1404, + 549, + 297, + 549 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 297, + 328, + 1404, + 328, + 1404, + 420, + 297, + 420 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 299, + 586, + 1403, + 586, + 1403, + 679, + 299, + 679 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1402, + 229, + 1402, + 293, + 295, + 293 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 296, + 887, + 1083, + 887, + 1083, + 921, + 296, + 921 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 300, + 1884, + 670, + 1884, + 670, + 1916, + 300, + 1916 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 300, + 1823, + 544, + 1823, + 544, + 1856, + 300, + 1856 + ], + "score": 0.898 + }, + { + "category_id": 0, + "poly": [ + 301, + 752, + 509, + 752, + 509, + 787, + 301, + 787 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 301, + 827, + 694, + 827, + 694, + 858, + 301, + 858 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.858 + }, + { + "category_id": 4, + "poly": [ + 294, + 1465, + 1401, + 1465, + 1401, + 1530, + 294, + 1530 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 862, + 2088, + 862, + 2112, + 836, + 2112 + ], + "score": 0.834 + }, + { + "category_id": 1, + "poly": [ + 294, + 1465, + 1401, + 1465, + 1401, + 1530, + 294, + 1530 + ], + "score": 0.103 + }, + { + "category_id": 13, + "poly": [ + 1152, + 1596, + 1253, + 1596, + 1253, + 1628, + 1152, + 1628 + ], + "score": 0.92, + "latex": "\\varrho = \\delta _ { j } z _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1254, + 1716, + 1309, + 1716, + 1309, + 1750, + 1254, + 1750 + ], + "score": 0.92, + "latex": "f ( \\varrho )" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1626, + 1157, + 1626, + 1157, + 1657, + 1043, + 1657 + ], + "score": 0.92, + "latex": "\\alpha \\in ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 1159, + 1656, + 1214, + 1656, + 1214, + 1689, + 1159, + 1689 + ], + "score": 0.91, + "latex": "f ( \\varrho )" + }, + { + "category_id": 13, + "poly": [ + 474, + 2002, + 614, + 2002, + 614, + 2033, + 474, + 2033 + ], + "score": 0.91, + "latex": "8 0 \\% - 2 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 840, + 1593, + 1063, + 1593, + 1063, + 1627, + 840, + 1627 + ], + "score": 0.91, + "latex": "f ( \\varrho ) = | \\varrho | ^ { \\alpha } \\mathrm { s i g n } ( \\varrho )" + }, + { + "category_id": 13, + "poly": [ + 761, + 1687, + 837, + 1687, + 837, + 1717, + 761, + 1717 + ], + "score": 0.91, + "latex": "\\varrho \\to 0" + }, + { + "category_id": 13, + "poly": [ + 1013, + 1687, + 1087, + 1687, + 1087, + 1714, + 1013, + 1714 + ], + "score": 0.9, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 362, + 1688, + 443, + 1688, + 443, + 1718, + 362, + 1718 + ], + "score": 0.9, + "latex": "\\partial f / \\partial \\varrho" + }, + { + "category_id": 13, + "poly": [ + 1105, + 1468, + 1180, + 1468, + 1180, + 1495, + 1105, + 1495 + ], + "score": 0.9, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 469, + 891, + 540, + 891, + 540, + 917, + 469, + 917 + ], + "score": 0.89, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1627, + 1393, + 1627, + 1393, + 1653, + 1322, + 1653 + ], + "score": 0.88, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 594, + 1625, + 1000, + 1625, + 1000, + 1659, + 594, + 1659 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\varrho \\to 0 ^ { - } } f ( \\varrho ) = \\operatorname* { l i m } _ { \\varrho \\to 0 ^ { + } } f ( \\varrho ) = 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 582, + 1658, + 794, + 1658, + 794, + 1690, + 582, + 1690 + ], + "score": 0.87, + "latex": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\varrho \\to 0 ^ { + } } f ( \\varrho ) = 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1655, + 531, + 1655, + 531, + 1688, + 297, + 1688 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { \\operatorname* { l i m } _ { \\varrho \\to 0 ^ { - } } f ( \\varrho ) = - 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 505, + 834, + 527, + 834, + 527, + 855, + 505, + 855 + ], + "score": 0.64, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 297, + 1656, + 399, + 1656, + 399, + 1689, + 297, + 1689 + ], + "score": 0.44, + "latex": "\\mathrm { l i m } _ { \\varrho \\to 0 ^ { - } }" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1023.0, + 1102.0, + 1023.0, + 1102.0, + 1050.0, + 980.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1043.0, + 618.0, + 1043.0, + 618.0, + 1076.0, + 562.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1042.0, + 1052.0, + 1042.0, + 1052.0, + 1076.0, + 981.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1086.0, + 619.0, + 1086.0, + 619.0, + 1121.0, + 559.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1133.0, + 618.0, + 1133.0, + 618.0, + 1165.0, + 562.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1177.0, + 618.0, + 1177.0, + 618.0, + 1210.0, + 562.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1185.0, + 569.0, + 1185.0, + 569.0, + 1227.0, + 544.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1221.0, + 619.0, + 1221.0, + 619.0, + 1256.0, + 561.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1268.0, + 618.0, + 1268.0, + 618.0, + 1299.0, + 562.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1311.0, + 618.0, + 1311.0, + 618.0, + 1344.0, + 561.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1357.0, + 618.0, + 1357.0, + 618.0, + 1389.0, + 562.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1394.0, + 644.0, + 1394.0, + 644.0, + 1417.0, + 624.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1391.0, + 733.0, + 1391.0, + 733.0, + 1419.0, + 682.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1391.0, + 808.0, + 1391.0, + 808.0, + 1419.0, + 758.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1391.0, + 884.0, + 1391.0, + 884.0, + 1419.0, + 832.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1391.0, + 959.0, + 1391.0, + 959.0, + 1419.0, + 909.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1391.0, + 1039.0, + 1391.0, + 1039.0, + 1419.0, + 979.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1391.0, + 1111.0, + 1391.0, + 1111.0, + 1419.0, + 1055.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1409.0, + 892.0, + 1409.0, + 892.0, + 1441.0, + 823.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1884.0, + 672.0, + 1884.0, + 672.0, + 1920.0, + 296.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1824.0, + 547.0, + 1824.0, + 547.0, + 1857.0, + 297.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 746.0, + 514.0, + 746.0, + 514.0, + 796.0, + 293.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 827.0, + 504.0, + 827.0, + 504.0, + 860.0, + 297.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 827.0, + 696.0, + 827.0, + 696.0, + 860.0, + 528.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1465.0, + 1104.0, + 1465.0, + 1104.0, + 1501.0, + 294.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1465.0, + 1404.0, + 1465.0, + 1404.0, + 1501.0, + 1181.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1497.0, + 1031.0, + 1497.0, + 1031.0, + 1534.0, + 294.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1588.0, + 839.0, + 1588.0, + 839.0, + 1633.0, + 292.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1588.0, + 1151.0, + 1588.0, + 1151.0, + 1633.0, + 1064.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1588.0, + 1405.0, + 1588.0, + 1405.0, + 1633.0, + 1254.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1619.0, + 593.0, + 1619.0, + 593.0, + 1665.0, + 290.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1619.0, + 1042.0, + 1619.0, + 1042.0, + 1665.0, + 1001.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 1619.0, + 1321.0, + 1619.0, + 1321.0, + 1665.0, + 1158.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1619.0, + 1407.0, + 1619.0, + 1407.0, + 1665.0, + 1394.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1652.0, + 296.0, + 1652.0, + 296.0, + 1692.0, + 292.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1652.0, + 581.0, + 1652.0, + 581.0, + 1692.0, + 532.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1652.0, + 1158.0, + 1652.0, + 1158.0, + 1692.0, + 795.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1652.0, + 1407.0, + 1652.0, + 1407.0, + 1692.0, + 1215.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1684.0, + 361.0, + 1684.0, + 361.0, + 1720.0, + 294.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1684.0, + 760.0, + 1684.0, + 760.0, + 1720.0, + 444.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1684.0, + 1012.0, + 1684.0, + 1012.0, + 1720.0, + 838.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1684.0, + 1406.0, + 1684.0, + 1406.0, + 1720.0, + 1088.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1715.0, + 1253.0, + 1715.0, + 1253.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1715.0, + 1406.0, + 1715.0, + 1406.0, + 1751.0, + 1310.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1743.0, + 1307.0, + 1743.0, + 1307.0, + 1783.0, + 294.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1975.0, + 294.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2004.0, + 297.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 473.0, + 2001.0, + 473.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 2001.0, + 888.0, + 2001.0, + 888.0, + 2037.0, + 615.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 456.0, + 1406.0, + 456.0, + 1406.0, + 493.0, + 295.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 485.0, + 1405.0, + 485.0, + 1405.0, + 526.0, + 320.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 519.0, + 676.0, + 519.0, + 676.0, + 552.0, + 324.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 327.0, + 1402.0, + 327.0, + 1402.0, + 364.0, + 294.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 358.0, + 1406.0, + 358.0, + 1406.0, + 394.0, + 322.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 391.0, + 628.0, + 391.0, + 628.0, + 421.0, + 325.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 585.0, + 1405.0, + 585.0, + 1405.0, + 622.0, + 295.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 617.0, + 1405.0, + 617.0, + 1405.0, + 651.0, + 322.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 647.0, + 463.0, + 647.0, + 463.0, + 678.0, + 324.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 260.0, + 1275.0, + 260.0, + 1275.0, + 296.0, + 322.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 885.0, + 468.0, + 885.0, + 468.0, + 925.0, + 295.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 885.0, + 1084.0, + 885.0, + 1084.0, + 925.0, + 541.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1465.0, + 1104.0, + 1465.0, + 1104.0, + 1501.0, + 294.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1465.0, + 1404.0, + 1465.0, + 1404.0, + 1501.0, + 1181.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1497.0, + 1031.0, + 1497.0, + 1031.0, + 1534.0, + 294.0, + 1534.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1276, + 1406, + 1276, + 1406, + 1523, + 296, + 1523 + ], + "score": 0.983 + }, + { + "category_id": 5, + "poly": [ + 298, + 1587, + 1403, + 1587, + 1403, + 1827, + 298, + 1827 + ], + "score": 0.981, + "html": "
Experiment for MLP case on BostonTheoretical Upper Bound (in seconds)Exp. Convergence Time to within 10e-9 (in seconds)Metric rmse
L1L2Lyap.L1L2Lyap.
dataset M=0.1~1.97e111277.66906.61755.590.0950.1450.092
M= 0.2~ 5.67e101297.111097.62914.780.0960.1460.093
M= 0.3~2.73e101290.481197.53998.050.0930.1470.101
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1009, + 1404, + 1009, + 1404, + 1133, + 299, + 1133 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 362, + 434, + 1338, + 434, + 1338, + 807, + 362, + 807 + ], + "score": 0.973 + }, + { + "category_id": 4, + "poly": [ + 296, + 832, + 1405, + 832, + 1405, + 954, + 296, + 954 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 299, + 299, + 1404, + 299, + 1404, + 362, + 299, + 362 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 297, + 1848, + 1403, + 1848, + 1403, + 1972, + 297, + 1972 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.854 + }, + { + "category_id": 0, + "poly": [ + 299, + 1207, + 1163, + 1207, + 1163, + 1239, + 299, + 1239 + ], + "score": 0.846 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.841 + }, + { + "category_id": 0, + "poly": [ + 298, + 230, + 1006, + 230, + 1006, + 261, + 298, + 261 + ], + "score": 0.493 + }, + { + "category_id": 7, + "poly": [ + 297, + 1848, + 1403, + 1848, + 1403, + 1972, + 297, + 1972 + ], + "score": 0.146 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 853, + 76, + 853, + 104, + 300, + 104 + ], + "score": 0.139 + }, + { + "category_id": 0, + "poly": [ + 297, + 230, + 1005, + 230, + 1005, + 261, + 297, + 261 + ], + "score": 0.112 + }, + { + "category_id": 13, + "poly": [ + 1284, + 1369, + 1394, + 1369, + 1394, + 1398, + 1284, + 1398 + ], + "score": 0.9, + "latex": "M = 0 . 3" + }, + { + "category_id": 13, + "poly": [ + 415, + 924, + 509, + 924, + 509, + 953, + 415, + 953 + ], + "score": 0.89, + "latex": "\\alpha = 0 . 8" + }, + { + "category_id": 13, + "poly": [ + 664, + 924, + 775, + 924, + 775, + 953, + 664, + 953 + ], + "score": 0.88, + "latex": "/ \\operatorname { k } = 0 . 0 1" + }, + { + "category_id": 13, + "poly": [ + 862, + 1042, + 897, + 1042, + 897, + 1071, + 862, + 1071 + ], + "score": 0.88, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 954, + 1041, + 989, + 1041, + 989, + 1071, + 954, + 1071 + ], + "score": 0.88, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1492, + 1156, + 1492, + 1156, + 1521, + 1122, + 1521 + ], + "score": 0.88, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1115, + 1369, + 1226, + 1369, + 1226, + 1398, + 1115, + 1398 + ], + "score": 0.87, + "latex": "M = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 490, + 1491, + 524, + 1491, + 524, + 1521, + 490, + 1521 + ], + "score": 0.87, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 994, + 1369, + 1103, + 1369, + 1103, + 1398, + 994, + 1398 + ], + "score": 0.86, + "latex": "M = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 846, + 1911, + 880, + 1911, + 880, + 1941, + 846, + 1941 + ], + "score": 0.81, + "latex": "L _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 891, + 1912, + 926, + 1912, + 926, + 1941, + 891, + 1941 + ], + "score": 0.74, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 365, + 1881, + 398, + 1881, + 398, + 1909, + 365, + 1909 + ], + "score": 0.55, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 916, + 1658, + 952, + 1658, + 952, + 1690, + 916, + 1690 + ], + "score": 0.34, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1135, + 1658, + 1169, + 1658, + 1169, + 1690, + 1135, + 1690 + ], + "score": 0.33, + "latex": "L _ { 1 }" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 432.0, + 408.0, + 432.0, + 408.0, + 457.0, + 375.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 447.0, + 634.0, + 447.0, + 634.0, + 473.0, + 546.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 432.0, + 736.0, + 432.0, + 736.0, + 457.0, + 703.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 440.0, + 963.0, + 440.0, + 963.0, + 475.0, + 872.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 431.0, + 1172.0, + 431.0, + 1172.0, + 472.0, + 1029.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 462.0, + 634.0, + 462.0, + 634.0, + 489.0, + 562.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 461.0, + 961.0, + 461.0, + 961.0, + 489.0, + 887.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 460.0, + 1175.0, + 460.0, + 1175.0, + 490.0, + 1069.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 482.0, + 674.0, + 482.0, + 674.0, + 505.0, + 579.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 481.0, + 749.0, + 481.0, + 749.0, + 512.0, + 737.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 482.0, + 1001.0, + 482.0, + 1001.0, + 505.0, + 906.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 480.0, + 1215.0, + 480.0, + 1215.0, + 506.0, + 1093.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 494.0, + 407.0, + 494.0, + 407.0, + 520.0, + 375.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 494.0, + 737.0, + 494.0, + 737.0, + 520.0, + 702.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 493.0, + 1066.0, + 493.0, + 1066.0, + 523.0, + 1027.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 546.0, + 420.0, + 546.0, + 420.0, + 661.0, + 352.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 561.0, + 736.0, + 561.0, + 736.0, + 648.0, + 689.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 544.0, + 1076.0, + 544.0, + 1076.0, + 662.0, + 1003.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 688.0, + 407.0, + 688.0, + 407.0, + 713.0, + 375.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 688.0, + 736.0, + 688.0, + 736.0, + 713.0, + 702.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 688.0, + 1065.0, + 688.0, + 1065.0, + 713.0, + 1029.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 743.0, + 665.0, + 743.0, + 665.0, + 799.0, + 369.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 743.0, + 992.0, + 743.0, + 992.0, + 799.0, + 696.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 742.0, + 1319.0, + 742.0, + 1319.0, + 799.0, + 1023.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 789.0, + 563.0, + 789.0, + 563.0, + 813.0, + 525.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 785.0, + 890.0, + 785.0, + 890.0, + 814.0, + 850.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 787.0, + 1217.0, + 787.0, + 1217.0, + 813.0, + 1181.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.25, + 646.0, + 1340.25, + 646.0, + 1340.25, + 668.5, + 1160.25, + 668.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 833.0, + 1405.0, + 833.0, + 1405.0, + 865.0, + 295.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 862.0, + 1405.0, + 862.0, + 1405.0, + 895.0, + 294.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 893.0, + 1405.0, + 893.0, + 1405.0, + 928.0, + 294.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 925.0, + 414.0, + 925.0, + 414.0, + 955.0, + 295.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 925.0, + 663.0, + 925.0, + 663.0, + 955.0, + 510.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1207.0, + 1168.0, + 1207.0, + 1168.0, + 1242.0, + 296.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1008.0, + 228.0, + 1008.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1849.0, + 1403.0, + 1849.0, + 1403.0, + 1882.0, + 295.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1878.0, + 364.0, + 1878.0, + 364.0, + 1916.0, + 292.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1878.0, + 1407.0, + 1878.0, + 1407.0, + 1916.0, + 399.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 845.0, + 1909.0, + 845.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1909.0, + 890.0, + 1909.0, + 890.0, + 1947.0, + 881.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1909.0, + 1407.0, + 1909.0, + 1407.0, + 1947.0, + 927.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1940.0, + 993.0, + 1940.0, + 993.0, + 1974.0, + 294.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1008.0, + 228.0, + 1008.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1278.0, + 1406.0, + 1278.0, + 1406.0, + 1312.0, + 296.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1307.0, + 1406.0, + 1307.0, + 1406.0, + 1340.0, + 295.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1339.0, + 1404.0, + 1339.0, + 1404.0, + 1373.0, + 294.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1367.0, + 993.0, + 1367.0, + 993.0, + 1404.0, + 295.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1367.0, + 1114.0, + 1367.0, + 1114.0, + 1404.0, + 1104.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 1367.0, + 1283.0, + 1367.0, + 1283.0, + 1404.0, + 1227.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1367.0, + 1407.0, + 1367.0, + 1407.0, + 1404.0, + 1395.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1397.0, + 1406.0, + 1397.0, + 1406.0, + 1436.0, + 293.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1427.0, + 1406.0, + 1427.0, + 1406.0, + 1466.0, + 293.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1461.0, + 1406.0, + 1461.0, + 1406.0, + 1495.0, + 295.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1491.0, + 489.0, + 1491.0, + 489.0, + 1525.0, + 294.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1491.0, + 1121.0, + 1491.0, + 1121.0, + 1525.0, + 525.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1491.0, + 1315.0, + 1491.0, + 1315.0, + 1525.0, + 1157.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1010.0, + 1403.0, + 1010.0, + 1403.0, + 1043.0, + 295.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1042.0, + 861.0, + 1042.0, + 861.0, + 1075.0, + 295.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1042.0, + 953.0, + 1042.0, + 953.0, + 1075.0, + 898.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1042.0, + 1405.0, + 1042.0, + 1405.0, + 1075.0, + 990.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1107.0, + 295.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1101.0, + 906.0, + 1101.0, + 906.0, + 1136.0, + 294.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 299.0, + 1406.0, + 299.0, + 1406.0, + 336.0, + 294.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 331.0, + 1081.0, + 331.0, + 1081.0, + 364.0, + 293.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1849.0, + 1403.0, + 1849.0, + 1403.0, + 1882.0, + 295.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1878.0, + 364.0, + 1878.0, + 364.0, + 1916.0, + 292.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1878.0, + 1407.0, + 1878.0, + 1407.0, + 1916.0, + 399.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 845.0, + 1909.0, + 845.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1909.0, + 890.0, + 1909.0, + 890.0, + 1947.0, + 881.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1909.0, + 1407.0, + 1909.0, + 1407.0, + 1947.0, + 927.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1940.0, + 993.0, + 1940.0, + 993.0, + 1974.0, + 294.0, + 1974.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 331, + 1142, + 1367, + 1142, + 1367, + 1881, + 331, + 1881 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 353, + 366, + 1340, + 366, + 1340, + 717, + 353, + 717 + ], + "score": 0.96 + }, + { + "category_id": 4, + "poly": [ + 297, + 1901, + 1404, + 1901, + 1404, + 2026, + 297, + 2026 + ], + "score": 0.96 + }, + { + "category_id": 4, + "poly": [ + 296, + 744, + 1406, + 744, + 1406, + 839, + 296, + 839 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 295, + 942, + 1405, + 942, + 1405, + 1067, + 295, + 1067 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 297, + 229, + 1036, + 229, + 1036, + 262, + 297, + 262 + ], + "score": 0.71 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 854, + 75, + 854, + 105, + 297, + 105 + ], + "score": 0.656 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 854, + 75, + 854, + 105, + 297, + 105 + ], + "score": 0.277 + }, + { + "category_id": 4, + "poly": [ + 295, + 942, + 1405, + 942, + 1405, + 1067, + 295, + 1067 + ], + "score": 0.195 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1036, + 229, + 1036, + 262, + 297, + 262 + ], + "score": 0.143 + }, + { + "category_id": 13, + "poly": [ + 460, + 1994, + 574, + 1994, + 574, + 2023, + 460, + 2023 + ], + "score": 0.9, + "latex": "\\Delta x = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 667, + 1994, + 780, + 1994, + 780, + 2023, + 667, + 2023 + ], + "score": 0.89, + "latex": "\\Delta x = 1 . 2" + }, + { + "category_id": 13, + "poly": [ + 298, + 1994, + 412, + 1994, + 412, + 2023, + 298, + 2023 + ], + "score": 0.89, + "latex": "\\Delta x = 0 . 4" + }, + { + "category_id": 13, + "poly": [ + 909, + 1964, + 1025, + 1964, + 1025, + 1993, + 909, + 1993 + ], + "score": 0.88, + "latex": "\\Delta x = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1206, + 1964, + 1354, + 1964, + 1354, + 1994, + 1206, + 1994 + ], + "score": 0.87, + "latex": "( \\mathrm { c } ) \\Delta x = 0 . 3" + }, + { + "category_id": 13, + "poly": [ + 457, + 1934, + 499, + 1934, + 499, + 1961, + 457, + 1961 + ], + "score": 0.86, + "latex": "\\Delta x" + }, + { + "category_id": 13, + "poly": [ + 1338, + 945, + 1405, + 945, + 1405, + 975, + 1338, + 975 + ], + "score": 0.84, + "latex": "M =" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1963, + 1194, + 1963, + 1194, + 1994, + 1077, + 1994 + ], + "score": 0.83, + "latex": "\\Delta x = 0 . 2" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1144.0, + 395.0, + 1144.0, + 395.0, + 1175.0, + 352.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1144.0, + 738.0, + 1144.0, + 738.0, + 1175.0, + 694.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1144.0, + 1079.0, + 1144.0, + 1079.0, + 1174.0, + 1035.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1204.0, + 395.0, + 1204.0, + 395.0, + 1236.0, + 351.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1204.0, + 737.0, + 1204.0, + 737.0, + 1235.0, + 694.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1204.0, + 1079.0, + 1204.0, + 1079.0, + 1235.0, + 1035.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1237.0, + 403.0, + 1237.0, + 403.0, + 1310.0, + 326.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1227.0, + 754.0, + 1227.0, + 754.0, + 1391.0, + 658.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 1236.0, + 1087.0, + 1236.0, + 1087.0, + 1310.0, + 1011.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1273.0, + 409.0, + 1273.0, + 409.0, + 1388.0, + 319.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1326.0, + 737.0, + 1326.0, + 737.0, + 1357.0, + 694.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 1285.0, + 1079.0, + 1285.0, + 1079.0, + 1375.0, + 1016.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1369.0, + 517.0, + 1369.0, + 517.0, + 1405.0, + 440.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1369.0, + 859.0, + 1369.0, + 859.0, + 1405.0, + 781.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1369.0, + 1201.0, + 1369.0, + 1201.0, + 1405.0, + 1123.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1385.0, + 395.0, + 1385.0, + 395.0, + 1417.0, + 351.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1393.0, + 517.0, + 1393.0, + 517.0, + 1430.0, + 442.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1384.0, + 739.0, + 1384.0, + 739.0, + 1419.0, + 692.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1393.0, + 859.0, + 1393.0, + 859.0, + 1430.0, + 783.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1385.0, + 1079.0, + 1385.0, + 1079.0, + 1417.0, + 1035.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1394.0, + 1202.0, + 1394.0, + 1202.0, + 1430.0, + 1124.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1421.0, + 576.0, + 1421.0, + 576.0, + 1452.0, + 441.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1423.0, + 918.0, + 1423.0, + 918.0, + 1451.0, + 785.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1423.0, + 1258.0, + 1423.0, + 1258.0, + 1451.0, + 1126.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1454.0, + 505.0, + 1454.0, + 505.0, + 1486.0, + 466.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1455.0, + 606.0, + 1455.0, + 606.0, + 1485.0, + 557.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1454.0, + 849.0, + 1454.0, + 849.0, + 1486.0, + 810.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1454.0, + 954.0, + 1454.0, + 954.0, + 1487.0, + 906.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1454.0, + 1189.0, + 1454.0, + 1189.0, + 1486.0, + 1150.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1455.0, + 1290.0, + 1455.0, + 1290.0, + 1485.0, + 1243.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1470.0, + 611.0, + 1470.0, + 611.0, + 1507.0, + 445.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1470.0, + 953.0, + 1470.0, + 953.0, + 1507.0, + 787.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1471.0, + 1294.0, + 1471.0, + 1294.0, + 1506.0, + 1129.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1492.0, + 553.0, + 1492.0, + 553.0, + 1525.0, + 515.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1491.0, + 895.0, + 1491.0, + 895.0, + 1525.0, + 857.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1492.0, + 1237.0, + 1492.0, + 1237.0, + 1525.0, + 1199.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1507.0, + 396.0, + 1507.0, + 396.0, + 1540.0, + 353.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1507.0, + 738.0, + 1507.0, + 738.0, + 1539.0, + 694.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1506.0, + 1082.0, + 1506.0, + 1082.0, + 1540.0, + 1034.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1538.0, + 403.0, + 1538.0, + 403.0, + 1554.0, + 390.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1568.0, + 395.0, + 1568.0, + 395.0, + 1600.0, + 351.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1568.0, + 737.0, + 1568.0, + 737.0, + 1598.0, + 693.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1568.0, + 1079.0, + 1568.0, + 1079.0, + 1600.0, + 1035.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1601.0, + 409.0, + 1601.0, + 409.0, + 1751.0, + 319.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1590.0, + 757.0, + 1590.0, + 757.0, + 1755.0, + 659.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1609.0, + 1079.0, + 1609.0, + 1079.0, + 1739.0, + 1018.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1746.0, + 396.0, + 1746.0, + 396.0, + 1782.0, + 350.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1733.0, + 517.0, + 1733.0, + 517.0, + 1793.0, + 440.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1747.0, + 739.0, + 1747.0, + 739.0, + 1783.0, + 692.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1733.0, + 859.0, + 1733.0, + 859.0, + 1793.0, + 781.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1747.0, + 1081.0, + 1747.0, + 1081.0, + 1782.0, + 1034.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1734.0, + 1202.0, + 1734.0, + 1202.0, + 1793.0, + 1123.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1786.0, + 577.0, + 1786.0, + 577.0, + 1817.0, + 443.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1786.0, + 920.0, + 1786.0, + 920.0, + 1817.0, + 784.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1786.0, + 1262.0, + 1786.0, + 1262.0, + 1817.0, + 1126.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1818.0, + 507.0, + 1818.0, + 507.0, + 1850.0, + 467.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1819.0, + 610.0, + 1819.0, + 610.0, + 1849.0, + 561.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1802.0, + 750.0, + 1802.0, + 750.0, + 1852.0, + 691.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1818.0, + 843.0, + 1818.0, + 843.0, + 1850.0, + 804.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1819.0, + 942.0, + 1819.0, + 942.0, + 1849.0, + 893.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1801.0, + 1091.0, + 1801.0, + 1091.0, + 1853.0, + 983.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1818.0, + 1184.0, + 1818.0, + 1184.0, + 1850.0, + 1144.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1818.0, + 1279.0, + 1818.0, + 1279.0, + 1849.0, + 1230.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 1819.0, + 1368.0, + 1819.0, + 1368.0, + 1850.0, + 1321.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1834.0, + 611.0, + 1834.0, + 611.0, + 1869.0, + 445.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1833.0, + 953.0, + 1833.0, + 953.0, + 1870.0, + 787.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1834.0, + 1295.0, + 1834.0, + 1295.0, + 1869.0, + 1129.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1855.0, + 554.0, + 1855.0, + 554.0, + 1887.0, + 514.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1856.0, + 895.0, + 1856.0, + 895.0, + 1888.0, + 856.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1856.0, + 1235.0, + 1856.0, + 1235.0, + 1886.0, + 1202.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1444.0, + 399.0, + 1444.0, + 399.0, + 1484.5, + 355.0, + 1484.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.75, + 1443.5, + 741.75, + 1443.5, + 741.75, + 1483.5, + 695.75, + 1483.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1443.5, + 1084.0, + 1443.5, + 1084.0, + 1484.5, + 1038.0, + 1484.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.25, + 1521.0, + 746.25, + 1521.0, + 746.25, + 1539.0, + 725.25, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.25, + 1806.5, + 400.25, + 1806.5, + 400.25, + 1847.0, + 354.25, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 367.0, + 596.0, + 367.0, + 596.0, + 418.0, + 373.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 365.0, + 1325.0, + 365.0, + 1325.0, + 421.0, + 599.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 540.0, + 481.0, + 540.0, + 481.0, + 592.0, + 378.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 543.0, + 597.0, + 543.0, + 597.0, + 591.0, + 497.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 544.0, + 720.0, + 544.0, + 720.0, + 591.0, + 620.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 545.0, + 845.0, + 545.0, + 845.0, + 588.0, + 745.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 543.0, + 973.0, + 543.0, + 973.0, + 590.0, + 863.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 545.0, + 1104.0, + 545.0, + 1104.0, + 588.0, + 991.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 543.0, + 1328.0, + 543.0, + 1328.0, + 592.0, + 1108.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1901.0, + 1406.0, + 1901.0, + 1406.0, + 1938.0, + 294.0, + 1938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1933.0, + 456.0, + 1933.0, + 456.0, + 1967.0, + 291.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1933.0, + 1405.0, + 1933.0, + 1405.0, + 1967.0, + 500.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1965.0, + 908.0, + 1965.0, + 908.0, + 1997.0, + 294.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1965.0, + 1076.0, + 1965.0, + 1076.0, + 1997.0, + 1026.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1965.0, + 1205.0, + 1965.0, + 1205.0, + 1997.0, + 1195.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1965.0, + 1404.0, + 1965.0, + 1404.0, + 1997.0, + 1355.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1990.0, + 297.0, + 1990.0, + 297.0, + 2028.0, + 294.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1990.0, + 459.0, + 1990.0, + 459.0, + 2028.0, + 413.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1990.0, + 666.0, + 1990.0, + 666.0, + 2028.0, + 575.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1990.0, + 793.0, + 1990.0, + 793.0, + 2028.0, + 781.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 743.0, + 1407.0, + 743.0, + 1407.0, + 781.0, + 294.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 772.0, + 1406.0, + 772.0, + 1406.0, + 812.0, + 291.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 802.0, + 452.0, + 802.0, + 452.0, + 847.0, + 293.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 1038.0, + 229.0, + 1038.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 945.0, + 1337.0, + 945.0, + 1337.0, + 978.0, + 294.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 974.0, + 1405.0, + 974.0, + 1405.0, + 1009.0, + 293.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1006.0, + 1406.0, + 1006.0, + 1406.0, + 1040.0, + 293.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1037.0, + 570.0, + 1037.0, + 570.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 945.0, + 1337.0, + 945.0, + 1337.0, + 978.0, + 294.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 974.0, + 1405.0, + 974.0, + 1405.0, + 1009.0, + 293.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1006.0, + 1406.0, + 1006.0, + 1406.0, + 1040.0, + 293.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1037.0, + 570.0, + 1037.0, + 570.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 1038.0, + 229.0, + 1038.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 333, + 226, + 1366, + 226, + 1366, + 964, + 333, + 964 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 296, + 993, + 1404, + 993, + 1404, + 1116, + 296, + 1116 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.845 + }, + { + "category_id": 2, + "poly": [ + 300, + 77, + 852, + 77, + 852, + 104, + 300, + 104 + ], + "score": 0.829 + }, + { + "category_id": 13, + "poly": [ + 909, + 1055, + 1025, + 1055, + 1025, + 1084, + 909, + 1084 + ], + "score": 0.9, + "latex": "\\Delta x = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 460, + 1085, + 574, + 1085, + 574, + 1114, + 460, + 1114 + ], + "score": 0.89, + "latex": "\\Delta x = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 667, + 1085, + 781, + 1085, + 781, + 1114, + 667, + 1114 + ], + "score": 0.89, + "latex": "\\Delta x = 1 . 2" + }, + { + "category_id": 13, + "poly": [ + 298, + 1085, + 412, + 1085, + 412, + 1114, + 298, + 1114 + ], + "score": 0.89, + "latex": "\\Delta x = 0 . 4" + }, + { + "category_id": 13, + "poly": [ + 1206, + 1055, + 1354, + 1055, + 1354, + 1086, + 1206, + 1086 + ], + "score": 0.87, + "latex": "( \\mathrm { c } ) \\Delta x = 0 . 3" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1054, + 1195, + 1054, + 1195, + 1084, + 1077, + 1084 + ], + "score": 0.86, + "latex": "\\Delta x = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 458, + 1025, + 499, + 1025, + 499, + 1052, + 458, + 1052 + ], + "score": 0.83, + "latex": "\\Delta x" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 230.0, + 394.0, + 230.0, + 394.0, + 257.0, + 355.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 228.0, + 739.0, + 228.0, + 739.0, + 259.0, + 694.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 228.0, + 1081.0, + 228.0, + 1081.0, + 258.0, + 1036.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 288.0, + 396.0, + 288.0, + 396.0, + 320.0, + 353.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 288.0, + 737.0, + 288.0, + 737.0, + 320.0, + 695.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 288.0, + 1080.0, + 288.0, + 1080.0, + 320.0, + 1036.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 330.0, + 395.0, + 330.0, + 395.0, + 455.0, + 335.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 330.0, + 736.0, + 330.0, + 736.0, + 456.0, + 677.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 328.0, + 1080.0, + 328.0, + 1080.0, + 457.0, + 1017.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 409.0, + 396.0, + 409.0, + 396.0, + 441.0, + 354.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 409.0, + 737.0, + 409.0, + 737.0, + 441.0, + 696.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 407.0, + 1080.0, + 407.0, + 1080.0, + 443.0, + 1036.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 468.0, + 396.0, + 468.0, + 396.0, + 503.0, + 351.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 453.0, + 518.0, + 453.0, + 518.0, + 515.0, + 442.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 468.0, + 739.0, + 468.0, + 739.0, + 503.0, + 693.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 456.0, + 861.0, + 456.0, + 861.0, + 515.0, + 783.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 468.0, + 1080.0, + 468.0, + 1080.0, + 503.0, + 1035.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 454.0, + 1202.0, + 454.0, + 1202.0, + 513.0, + 1125.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 507.0, + 576.0, + 507.0, + 576.0, + 534.0, + 445.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 507.0, + 919.0, + 507.0, + 919.0, + 534.0, + 787.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 507.0, + 1259.0, + 507.0, + 1259.0, + 534.0, + 1129.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 525.0, + 407.0, + 525.0, + 407.0, + 573.0, + 351.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 538.0, + 505.0, + 538.0, + 505.0, + 570.0, + 468.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 540.0, + 606.0, + 540.0, + 606.0, + 570.0, + 559.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 540.0, + 848.0, + 540.0, + 848.0, + 569.0, + 813.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 540.0, + 953.0, + 540.0, + 953.0, + 570.0, + 907.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 540.0, + 1187.0, + 540.0, + 1187.0, + 569.0, + 1153.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 539.0, + 1291.0, + 539.0, + 1291.0, + 570.0, + 1244.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 553.0, + 611.0, + 553.0, + 611.0, + 592.0, + 446.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 553.0, + 954.0, + 553.0, + 954.0, + 592.0, + 787.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 553.0, + 1295.0, + 553.0, + 1295.0, + 592.0, + 1131.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 577.0, + 554.0, + 577.0, + 554.0, + 608.0, + 516.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 576.0, + 896.0, + 576.0, + 896.0, + 608.0, + 858.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 578.0, + 1234.0, + 578.0, + 1234.0, + 607.0, + 1202.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 591.0, + 395.0, + 591.0, + 395.0, + 622.0, + 354.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 591.0, + 741.0, + 591.0, + 741.0, + 622.0, + 695.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 590.0, + 1096.0, + 590.0, + 1096.0, + 624.0, + 1035.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 650.0, + 397.0, + 650.0, + 397.0, + 685.0, + 351.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 651.0, + 737.0, + 651.0, + 737.0, + 683.0, + 695.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 651.0, + 1080.0, + 651.0, + 1080.0, + 683.0, + 1036.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 679.0, + 410.0, + 679.0, + 410.0, + 831.0, + 312.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 695.0, + 737.0, + 695.0, + 737.0, + 818.0, + 676.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 692.0, + 1080.0, + 692.0, + 1080.0, + 820.0, + 1017.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 772.0, + 737.0, + 772.0, + 737.0, + 803.0, + 695.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 771.0, + 1078.0, + 771.0, + 1078.0, + 803.0, + 1036.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 831.0, + 396.0, + 831.0, + 396.0, + 866.0, + 351.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 816.0, + 519.0, + 816.0, + 519.0, + 877.0, + 442.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 831.0, + 740.0, + 831.0, + 740.0, + 866.0, + 693.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 817.0, + 861.0, + 817.0, + 861.0, + 878.0, + 783.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 831.0, + 1081.0, + 831.0, + 1081.0, + 866.0, + 1035.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 817.0, + 1202.0, + 817.0, + 1202.0, + 876.0, + 1124.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 870.0, + 578.0, + 870.0, + 578.0, + 901.0, + 444.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 870.0, + 920.0, + 870.0, + 920.0, + 901.0, + 785.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 870.0, + 1263.0, + 870.0, + 1263.0, + 901.0, + 1128.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 890.0, + 406.0, + 890.0, + 406.0, + 935.0, + 350.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 901.0, + 507.0, + 901.0, + 507.0, + 933.0, + 469.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 903.0, + 610.0, + 903.0, + 610.0, + 933.0, + 564.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 903.0, + 842.0, + 903.0, + 842.0, + 932.0, + 807.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 903.0, + 943.0, + 903.0, + 943.0, + 933.0, + 894.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 885.0, + 1092.0, + 885.0, + 1092.0, + 937.0, + 985.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 902.0, + 1183.0, + 902.0, + 1183.0, + 933.0, + 1145.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 902.0, + 1279.0, + 902.0, + 1279.0, + 933.0, + 1231.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 903.0, + 1368.0, + 903.0, + 1368.0, + 933.0, + 1323.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 916.0, + 610.0, + 916.0, + 610.0, + 951.0, + 446.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 916.0, + 954.0, + 916.0, + 954.0, + 953.0, + 789.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 920.0, + 1294.0, + 920.0, + 1294.0, + 951.0, + 1132.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 939.0, + 556.0, + 939.0, + 556.0, + 971.0, + 516.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 939.0, + 896.0, + 939.0, + 896.0, + 971.0, + 858.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 939.0, + 1235.0, + 939.0, + 1235.0, + 972.0, + 1201.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.25, + 528.0, + 743.25, + 528.0, + 743.25, + 567.0, + 697.25, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 527.5, + 1085.0, + 527.5, + 1085.0, + 568.0, + 1036.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.75, + 889.0, + 742.75, + 889.0, + 742.75, + 931.0, + 697.75, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 994.0, + 1405.0, + 994.0, + 1405.0, + 1027.0, + 295.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1024.0, + 457.0, + 1024.0, + 457.0, + 1057.0, + 291.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1024.0, + 1405.0, + 1024.0, + 1405.0, + 1057.0, + 500.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1056.0, + 908.0, + 1056.0, + 908.0, + 1088.0, + 294.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1056.0, + 1076.0, + 1056.0, + 1076.0, + 1088.0, + 1026.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1056.0, + 1205.0, + 1056.0, + 1205.0, + 1088.0, + 1196.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1056.0, + 1404.0, + 1056.0, + 1404.0, + 1088.0, + 1355.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1085.0, + 459.0, + 1085.0, + 459.0, + 1117.0, + 413.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1085.0, + 666.0, + 1085.0, + 666.0, + 1117.0, + 575.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1085.0, + 792.0, + 1085.0, + 792.0, + 1117.0, + 782.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 856.0, + 73.0, + 856.0, + 107.0, + 298.0, + 107.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/WWRBHhH158K/WWRBHhH158K_layout.pdf b/parse/train/WWRBHhH158K/WWRBHhH158K_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f0e6f2d3227ee4568aed546d84d29ec16ea6d806 --- /dev/null +++ b/parse/train/WWRBHhH158K/WWRBHhH158K_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a51cf7824db2328bf2ff63e70a7e435fd8ecdd012fe91f4aa0d3142294c57ca6 +size 2187649 diff --git a/parse/train/WWRBHhH158K/WWRBHhH158K_origin.pdf b/parse/train/WWRBHhH158K/WWRBHhH158K_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9fef26e87a2a78dd03c29dd8deeb30329b3f956a --- /dev/null +++ b/parse/train/WWRBHhH158K/WWRBHhH158K_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39e54a82115ddc2c33713783443a25959bb3c53b9d9dc70790d4cd4994805ee2 +size 2027991 diff --git a/parse/train/WWRBHhH158K/WWRBHhH158K_span.pdf b/parse/train/WWRBHhH158K/WWRBHhH158K_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d6ede203df83279fb966cfd6f2c27a010867b1b1 --- /dev/null +++ b/parse/train/WWRBHhH158K/WWRBHhH158K_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9425e5b63b7a6130b126287f2524acb292bdeaf041a44402c5b190dfc4142d7 +size 2197914 diff --git a/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_layout.pdf b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..19c273c7f0b09a648ed09d472d6691be6d83590d --- /dev/null +++ b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27253f9a07928d5b065620af751ffb58241d3c74ab26645d4377c00aec834b47 +size 4946889 diff --git a/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_origin.pdf b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7edb37b891831f0a6a1f44e520495c2bb8182cd4 --- /dev/null +++ b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:828ccde86b80e3b0a4a7171d1d198eace2a068b372745a52a480af2b9b3061f5 +size 4776241 diff --git a/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_span.pdf b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fc48792c73dabe7f0b08f54b61d71459923d73e6 --- /dev/null +++ b/parse/train/YDGJ5YExiw6/YDGJ5YExiw6_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bfe0d4818e6ff91040d0d1beecbf0f153b5859018a0edf0a0b8135c28bf7536 +size 4954643 diff --git a/parse/train/bYi_2708mKK/bYi_2708mKK_layout.pdf b/parse/train/bYi_2708mKK/bYi_2708mKK_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d95026aaea6593c9d053dbe7b37db72f23d4e469 --- /dev/null +++ b/parse/train/bYi_2708mKK/bYi_2708mKK_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f9ca715be6ac7d5d3bdf02c4ea8b1f2459621b635694f85b4f29926ebcf8985 +size 436613 diff --git a/parse/train/bYi_2708mKK/bYi_2708mKK_origin.pdf b/parse/train/bYi_2708mKK/bYi_2708mKK_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e75ba0d14dd852a820695f3f6e92bc7afdaad124 --- /dev/null +++ b/parse/train/bYi_2708mKK/bYi_2708mKK_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fa285d23bfcb0d70ee33b13c4306381ba6993ca0d7a1d6e57ad41e69258d30b +size 287492 diff --git a/parse/train/bYi_2708mKK/bYi_2708mKK_span.pdf b/parse/train/bYi_2708mKK/bYi_2708mKK_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..527625d3503bf2a35330ca198d7d059187266340 --- /dev/null +++ b/parse/train/bYi_2708mKK/bYi_2708mKK_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78be368aaec11c3e60fc4d40f9426a78a4c5a0be0dbc7ffc8abcade96561bad9 +size 440624 diff --git a/parse/train/dUEpGV2mhf/dUEpGV2mhf_layout.pdf b/parse/train/dUEpGV2mhf/dUEpGV2mhf_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..30c281e837908e5ac7e50bc6c3c83fd4fa385b03 --- /dev/null +++ b/parse/train/dUEpGV2mhf/dUEpGV2mhf_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d1eeb75c1e0e0b78fc0cab50958423e25c1ca15481dd6b0bd60bd821f825718 +size 1296250 diff --git a/parse/train/dUEpGV2mhf/dUEpGV2mhf_origin.pdf b/parse/train/dUEpGV2mhf/dUEpGV2mhf_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..76d05efdd75700479e7b6c2a8d798cd5ae7ae43f --- /dev/null +++ b/parse/train/dUEpGV2mhf/dUEpGV2mhf_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5db4d9ba3ede2a0f9854bece198b6ad9e1d4b559c5d48d09505cfef6d7517760 +size 1092690 diff --git a/parse/train/dUEpGV2mhf/dUEpGV2mhf_span.pdf b/parse/train/dUEpGV2mhf/dUEpGV2mhf_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c5c07ee6354f98da5134791c4eb28c45e67044fa --- /dev/null +++ b/parse/train/dUEpGV2mhf/dUEpGV2mhf_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f7d78f53a3ed006b0f4e6f28c161489c658f6804cecad683f058d70995f77b0 +size 1311763 diff --git a/parse/train/hQGRD1Zael7/hQGRD1Zael7_layout.pdf b/parse/train/hQGRD1Zael7/hQGRD1Zael7_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..266ae46d9335909201a47fefa74dd8ec7ca80a24 --- /dev/null +++ b/parse/train/hQGRD1Zael7/hQGRD1Zael7_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b26af177964880edb45ad0be6249ff64bb0a2692cfbcf6e2543099b71abfeb1a +size 747332 diff --git a/parse/train/hQGRD1Zael7/hQGRD1Zael7_origin.pdf b/parse/train/hQGRD1Zael7/hQGRD1Zael7_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4e29ac554cfde1944b6bbd9ff86394c608d2846a --- /dev/null +++ b/parse/train/hQGRD1Zael7/hQGRD1Zael7_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da0a1fd8940d94662c9559478163891045f572bf36039bcb26d49f74ee34a0f5 +size 579449 diff --git a/parse/train/jnkE5c5f9m/jnkE5c5f9m_layout.pdf b/parse/train/jnkE5c5f9m/jnkE5c5f9m_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..afd587e6fe85f0dd5014c2baa1118dbbf02960cf --- /dev/null +++ b/parse/train/jnkE5c5f9m/jnkE5c5f9m_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a69aef0c9393d13fc75b73522233e642059a79a1b2a15896e710d153ebb5aff +size 1760173 diff --git a/parse/train/jnkE5c5f9m/jnkE5c5f9m_origin.pdf b/parse/train/jnkE5c5f9m/jnkE5c5f9m_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..44677ce4ad55d5bdb7757a1029515273ead6b37a --- /dev/null +++ b/parse/train/jnkE5c5f9m/jnkE5c5f9m_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9549202306a626547b61a53b6ebd755ad4af3a6bb68d235bf9a85a855c7cd9d6 +size 1525902 diff --git a/parse/train/jnkE5c5f9m/jnkE5c5f9m_span.pdf b/parse/train/jnkE5c5f9m/jnkE5c5f9m_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..092afbda41391cd48f31e1a407b8cddd9f320593 --- /dev/null +++ b/parse/train/jnkE5c5f9m/jnkE5c5f9m_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b9428a0b5e4a7c885f2b6448ef5214df999aa6c13624540593ad434511abc69 +size 1772330 diff --git a/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_layout.pdf b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..49e8bec54cf994ea8b489ac29263cb8749c8bdd7 --- /dev/null +++ b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4781ef0f62ae628245dfef3b139d8c114f051a9ba829c5765866935be5d17e25 +size 3475894 diff --git a/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_origin.pdf b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9658ce7142775fd91e026b98f594ff300cb75641 --- /dev/null +++ b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c218129f65f4841e5b0723f3d7aa4045069d8c01038b7039dcac0d2ba8659a7 +size 3391685 diff --git a/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_span.pdf b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7f2c7914c4383bb8d31d00f8522b1e6654714d68 --- /dev/null +++ b/parse/train/kziQtP-nGqzDb/kziQtP-nGqzDb_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85fe132a1ee47c365c4b148ce44fd336be581853b6a91157bb857a4a0f65920e +size 3480558 diff --git a/parse/train/m1CD7tPubNy/m1CD7tPubNy.md b/parse/train/m1CD7tPubNy/m1CD7tPubNy.md new file mode 100644 index 0000000000000000000000000000000000000000..2eeb3735cff1590af1f115165dc76520d25ff0ee --- /dev/null +++ b/parse/train/m1CD7tPubNy/m1CD7tPubNy.md @@ -0,0 +1,692 @@ +# MIND THE PAD – CNNS CAN DEVELOP BLIND SPOTS + +Bilal Alsallakh Facebook AI + +Narine Kokhlikyan Facebook AI + +Vivek Miglani Facebook AI + +Jun Yuan NYU + +Orion Reblitz-Richardson Facebook AI + +# ABSTRACT + +We show how feature maps in convolutional networks are susceptible to spatial bias. Due to a combination of architectural choices, the activation at certain locations is systematically elevated or weakened. The major source of this bias is the padding mechanism. Depending on several aspects of convolution arithmetic, this mechanism can apply the padding unevenly, leading to asymmetries in the learned weights. We demonstrate how such bias can be detrimental to certain tasks such as small object detection: the activation is suppressed if the stimulus lies in the impacted area, leading to blind spots and misdetection. We propose solutions to mitigate spatial bias and demonstrate how they can improve model accuracy. + +# 1 MOTIVATION + +Convolutional neural networks (CNNs) serve as feature extractors for a wide variety of machinelearning tasks. Little attention has been paid to the spatial distribution of activation in the feature maps a CNN computes. Our interest in analyzing this distribution is triggered by mysterious failure cases of a traffic light detector: The detector successfully detects a small but visible traffic light in a road scene. However, it fails completely in detecting the same traffic light in the next frame captured by the ego-vehicle. The major difference between both frames is a limited shift along the vertical dimension as the vehicle moves forward. Therefore, the drastic difference in object detection is surprising given that CNNs are often assumed to have a high degree of translation invariance [8; 17]. + +The spatial distribution of activation in feature maps varies with the input. Nevertheless, by closely examining this distribution for a large number of samples, we found consistent patterns among them, often in the form of artifacts that do not resemble any input features. This work aims to analyze the root cause of such artifacts and their impact on CNNs. We show that these artifacts are responsible for the mysterious failure cases mentioned earlier, as they can induce ‘blind spots’ for the object detection head. Our contributions are: + +• Demonstrating how the padding mechanism can induce spatial bias in CNNs (Section 2). +• Demonstrating how spatial bias can impair downstream tasks (Section 3). +• Identifying uneven application of 0-padding as a resolvable source of bias (Section 5). +Relating the padding mechanism with the foveation behavior of CNNs (Section 6). +• Providing recommendations to mitigate spatial bias and demonstrating how this can prevent blind spots and boost model accuracy. + +# 2 THE EMERGENCE OF SPATIAL BIAS IN CNNS + +Our aim is to determine to which extent activation magnitude in CNN feature maps is influenced by location. We demonstrate our analysis on a publicly-available traffic-light detection model $\pmb { \mathbb { B } } \pmb { \ 6 } \|$ . This model implements the SSD architecture $\left[ \left[ 2 6 \right] \right]$ in TensorFlow $\mathbb { I I }$ , using MobileNet-v1 $\mathbb { \lVert \lambda \rVert }$ as a feature extractor. The model is trained on the BSTLD dataset $\pmb { \Vert 4 \Vert }$ which annotates traffic lights in road scenes. Figure $\bigstar$ shows two example scenes from the dataset. For each scene, we show two feature maps computed by two filters in the $1 1 ^ { \mathrm { t h } }$ convolutional layer. This layer contains 512 filters whose feature maps are used directly by the first box predictor in the SSD to detect small objects. + +![](images/b526bfe207aac3286b6c1d5a6a8d0a326615dced8dd84ad446f05157528bdb45.jpg) +Figure 1: Averaging feature maps per input (column marginal) and per filter (row marginal) in the last convolutional layer of a traffic light detector. Color indicates activation strength (the brighter, the higher), revealing line artifacts in the maps. These artifacts are the manifestation of spatial bias. + +The bottom row in Figure $\bigstar$ shows the average response of each of the two aforementioned filters, computed over the test set in BSTLD. The first filter seems to respond mainly to features in the top half of the input, while the second filter responds mainly to street areas. There are visible lines in the two average maps that do not seem to resemble any scene features and are consistently present in the individual feature maps. We analyzed the prevalence of these line artifacts in the feature maps of all 512 filters. The right column in Figure $\bigstar$ shows the average of these maps per scene, as well as over the entire test set (see supplemental for all 512 maps). The artifacts are largely visible in the average maps, with variations per scene depending on which individual maps are dominant. + +A useful way to make the artifacts stand out is to neutralize scene features by computing the feature maps for a zero-valued input. Figure $2$ depicts the resulting average map for each convolutional layer after applying ReLU units. The first average map is constant as we expect with a 0-valued input. The second map is also constant except for a 1-pixel boundary where the value is lower at the left border and higher at the other three borders. We magnify the corners to make these deviations visible. The border deviations increase in thickness and in variance at subsequent layers, creating multiple line artifacts at each border. These artifacts become quite pronounced at ReLU 8 where they start to propagate inwards, resembling the ones in Figure 1. + +![](images/9ad02d0bb96ca163c5aaed3822f6466e3fe37e0cde18654db995d27f877d7803.jpg) +Figure 2: Activation maps for a 0 input, averaged over each layer’s filters (title format: $\mathrm { H } { \times } \mathrm { W } { \times } \mathrm { C } )$ + +It is evident that the 1-pixel border variations in the second map are caused by the padding mechanism in use. This mechanism pads the output of the previous layer with a 1-pixel 0-valued border in order to maintain the size of the feature map after applying $3 { \tt X } 3$ convolutional. The maps in the first layer are not impacted because the input we feed is zero valued. Subsequent layers, however, are increasingly impacted by the padding, as preceding bias terms do not warrant 0-valued input. + +It is noticeable in Figure 2 that the artifacts caused by the padding differ across the four borders. To investigate this asymmetry, we analyze the convolutional kernels (often called filters) that produce the feature maps. Figure 3 depicts a per-layer mean of these $3 { \tt X } 3$ kernels. These mean kernels exhibit different degrees of asymmetry in the spatial distribution of their weights. For example, the kernels in L1 assign (on average) a negative weight at the left border, and a positive weight at the bottom. This directly impacts the padding-induced variation at each border. Such asymmetries are related to uneven application of padding as we explain in Section 5. + +![](images/fccaabdbce7408640032adb86b3858809bd5967c675d8bc0ad51217199678338.jpg) +Figure 3: Mean kernel per convolutional layer. All kernels are $3 \times 3$ , the titles show their counts. + +# 3 IMPLICATIONS OF SPATIAL BIAS + +We demonstrate how feature-map artifacts can cause blind spots for the SSD model. Similar issues arise in several small-object detectors, e.g., for faces and masks, as well as in pixel-oriented tasks such as semantic segmentation and image inpainting (see supplemental for examples). + +Figure $\sharp$ illustrates how the SSD predicts small objects based on the feature maps of the 11-th convolutional layer. The SSD uses the pixel positions in these maps as anchors of object proposals. Each proposal is scored by the SSD to represent a target category, with ”background“ being an implicit category that is crucial to exclude irrelevant parts of the input. In addition to these scores, the SSD computes a bounding box to localize the predicted object at each anchor. We examine object proposals computed at 1:2 aspect ratio, as they resemble the shape of most traffic lights in the dataset. We visualize the resulting score maps both for the background category and for traffic lights, when feeding a 0-valued input to the SSD. We also visualize the bounding boxes of these proposals in the image space. The SSD predicts the image content to be of background category at all anchor locations, as evident from the value range in both score maps. Such predictions are expected with an input that contains no traffic lights. However, the line artifacts in the feature maps have a strong impact on the score maps. These artifacts elevate the likelihood of anchors closer to the top to be classified as background (see the yellow band in the background score map). Conversely, these anchors have significantly lower scores for the traffic light category, compared with other anchors in the feature map. Such difference in the impact on the target categories is due to the different weights the SSD assigns to the feature maps for each target. As a result, the artifacts lead to potential blind spots in which the scores for certain categories are artificially muted. + +![](images/2f90b37ac2e03a969222bf159083fe214d6629973648c95433a912a506b514b4.jpg) +Figure 4: The formation of blind spots in SSD, illustrated via its box predictor internals with a zero-valued input. The predictor uses spatial anchors to detect and localize the target object at $4 5 \times 8 0$ possible locations based on 512 feature maps. Certain anchors are predisposed to predict background due to feature-map artifacts, as evident in the logit maps. Traffic lights at the corresponding location cannot be detected as demonstrated with a real scene (middle one in the bottom). + +![](images/855129a5fdeb22a83f45d33356680dd0ee5b663a0bc83617313ee7598eec1429.jpg) +Figure 5: (a) A map showing via color the detection score the SSD computes for a traffic light when present at various locations. The detection is muted when the stimulus lies in the area impacted by the artifacts. (b) The same map after changing the padding method to SYMMETRIC. The detection scores are rather constant except for periodic variations due to the SSD’s reliance on anchors. + +To validate whether or not the blind spots hinder object detection, we examine road scenes that contain highly-visible traffic light instances in the impacted area. Figure 4-bottom shows an example of such a scene. The SSD computes a low detection score of $7 \%$ when the traffic light lies in the blind spot (see middle image), far below the detection false-positive cutoff. Shifting the scene image upwards or downwards makes the instance detectable with a high score as long as it lies outside the blind spot. This explains the failure cases mentioned in Section $\mathbf { \overline { { \mathbb { D } } } }$ To further validate this effect, we run the SSD on baseline images that each contains one traffic light instance at a specific location in the input. We store the detection score for each instance. Figure 5a depicts the computed scores in a 2D map. It is evident that the model fails to detect the traffic light instance exactly when it is located within the “blind spot” band. The artifacts further disrupt the localization of the objects as evident in the top-right plot in Figure 4 which shows per-anchor object proposals computed for a 0 input. + +# 4 REMINDER: WHY IS PADDING NEEDED IN CNNS? + +# Padding is applied at most convolutional layers in CNNs to serve two fundamental purposes: + +Maintaining feature map size A padding that satisfies this property is often described as SAME or HALF padding. FULL padding expands the maps by kernel size - 1 along each dimension. VALID padding performs no padding, eroding the maps by the same amount. SAME padding is important to (1) design deep networks that can handle arbitrary input size (a challenge in the presence of gradual erosion), (2) maintain the aspect ratio of non-square input, and (3) concatenate feature maps from different layers as in Inception [39] and ResNet $\mathbf { \bar { \rho } }$ models. + +Reducing information bias against the boundary Consider a $3 \times 3$ kernel applied to a 2D input. An input location at least 2 pixels away from the boundary contributes to nine local convolution operations when computing the feature map. On the other hand, the corner is involved only one time under VALID padding, four times under a 1-pixel SAME 0-padding, and nine times under a 2-pixel FULL 0-padding. With SAME 0-padding, the cumulative contribution differences among the input pixels grow exponentially over the CNN layers. We refer to such uneven treatment of input pixels as the foveation behavior of the padding mechanism and elaborate on this in Section 6. + +We next explore solutions to the issues that cause padding to induce spatial bias. + +![](images/64a4b6b4f56e3df3a1dac55598e3c785be3dbfaa6420ebc60f4670b52e23635c.jpg) +Figure 6: (a) Illustrating the problem of uneven padding when down-sampling at a stride of 2. The padding along x-axis is consumed only at the left side. (b) Mean $3 \times 3$ filters in three ResNet models, trained on ImageNet with two input sizes. Color encodes average weight (green is positive). A size that induces uneven padding (top row) can lead to asymmetries, esp. around down-sampling layers. These asymmetries are mitigated when the input size induces no uneven padding (bottom row). + +# 5 ELIMINATING UNEVEN APPLICATION OF PADDING + +While useful to reduce bias against the boundary, applying padding at down-sampling layers can lead to asymmetry in CNN internals. Figure $6 \mathrm { a }$ illustrates the source of this asymmetry when strided convolution is used for downsampling: At one side of the feature map, the padding is consumed by the kernel while at the other side it is not. To warrant even application of padding throughout the CNN, the following must hold at all $d$ down-sampling layers, where $( h _ { i } , w _ { i } )$ is the output shape at the i-th layer with $\overline { { k } } _ { i } ^ { h } \times k _ { i } ^ { w }$ as kernel size, $( s _ { i } ^ { h } , s _ { i } ^ { w } )$ as strides, and $\mathbf { \Sigma } = ( p _ { i } ^ { h } , p _ { i } ^ { w } )$ as padding amount (refer to appendix $\boxed { \mathrm { A } }$ for a proof): + +$$ +\forall i \in \{ 1 , \ldots , d \} : h _ { i - 1 } = s _ { i } ^ { h } \cdot ( h _ { i } - 1 ) + k _ { i } ^ { h } - 2 \cdot p _ { i } ^ { h } \quad \wedge \quad w _ { i - 1 } = s _ { i } ^ { w } \cdot ( w _ { i } - 1 ) + k _ { i } ^ { w } - 2 \cdot p _ { i } ^ { w } \quad . +$$ + +The values $h _ { 0 }$ and $w _ { 0 }$ represent the CNN input dimensions. The above constraints are not always satisfied during training or inference with arbitrary input dimensions. For example, ImageNet classifiers based on ResNet $\pmb { \mathbb { I } } \pmb { \mathcal { 2 } } \Vert$ and MobileNet $\pmb { \mathbb { I } } \pmb { \overbrace { 3 } } \|$ contain five down-sampling layers $\mathit { a } = 5$ ) that apply 1-pixel 0-padding before performing 2-strided convolution. To avoid uneven application of padding, the input to these CNNs must satisfy the following, as explained in appendix A: + +$$ +h _ { 0 } = a _ { 1 } \times 2 ^ { d } + 1 = 3 2 \cdot a _ { 1 } + 1 \quad \mathrm { a n d } \quad w _ { 0 } = a _ { 2 } \times 2 ^ { d } + 1 = 3 2 \cdot a _ { 2 } + 1 \quad \mathrm { w h e r e } \quad a _ { 1 } , a _ { 2 } \in \mathbb { N } ^ { + } +$$ + +The traditional $\bigstar$ and prevalent input size for training ImageNet models is $2 2 4 \times 2 2 4$ . This size violates Eq. 2, leading to uneven padding at every down-sampling layer in ResNet and MobileNet models where 0-padding is effectively applied only at the left and top sides of layer input. This over-represents zeros at the top and left sides of $3 \times 3$ feature-map patches the filters are convolved with during training. The top row of Figure $6 { \mathsf { b } }$ shows per-layer mean filters in three ResNet models in PyTorch $\mathbb { \lVert 3 3 \rVert }$ , pre-trained on ImageNet with $2 2 4 \times 2 2 4$ images. In all of these models, a few of the mean filters, adjacent to down-sampling layers, exhibit stark asymmetry about their centers. + +We increase the image size to $2 2 5 \times 2 2 5$ without introducing additional image information2. This size satisfies Eq. 2, warranting even application of padding at every downsampling layer in the above models. Retraining the models with this size strongly reduces this asymmetry as evident in the bottom row of Figure $\bar { 6 } 6$ . This, in turn, visibly boosts the accuracy in all models we experimented with as we report in Table 1. The accuracy did not improve further when we retrained two of the models, ResNet-18 and ResNet-34, on $2 2 6 \times 2 2 6$ images. This provides evidence that the boost is due to eliminating uneven padding and not merely due to increasing the input size. + +Replacing 0-padding with a padding method that reuses feature map values can alleviate the asymmetry in the learned filters in the presence of unevenly applied padding. Another possibility is to use a rigid downsampling kernel, such as max-pooling, instead of a learned one. Appendix $\boxed { \dot { \mathbf { C } } }$ demonstrates both possibilities. Finally, antialiasing before downsampling $\mathbb { \lVert \boldsymbol { 4 3 } \rVert }$ can strongly reduce the asymmetry as we elaborate in Section 8 and in Appendix E. + +Table 1: Top-1 (and top-5) accuracy of five ImageNet classifiers trained with different input sizes. + +
Input Size ²MobileNetResNet-18ResNet-34ResNet-50ResNet-101
224×22468.19 (88.44)69.93 (89.22)73.30 (91.42)75.65 (92.47)77.37 (93.56)
225×22568.80 (88.78)70.27 (89.52)73.72 (91.58)76.01 (92.90)77.67 (93.81)
+ +Even when no padding is applied $( p _ { i } ^ { h } = 0$ or $p _ { i } ^ { w } = 0 ,$ ), an input size that does no satisfy Eq. 1 can lead to uneven erosion of feature maps, in turn, reducing the contribution of pixels from the impacted sides $( { \mathrm { F i g ~ } } 7 { \mathrm { \rho } }$ . Satisfying $\mathrm { E q } \ 1$ imposes a restriction on input size, e.g., to values in increments of $2 ^ { d } = 3 2$ with the above models ${ \mathrm { 1 9 3 \times 1 9 3 } }$ , $2 2 5 \times 2 2 5$ , $2 5 7 \times 2 5 7$ , ...). Depending on the application domain, this can be guaranteed either by resizing an input to the closest increment, or by padding it accordingly with suited values. + +# 6 PADDING MECHANISM AND FOVEATION + +By foveation we mean the unequal involvement of input pixels in convolutional operations throughout the CNN. Padding plays a fundamental role in the foveation behavior of CNNs. We visualize this behavior by means of a foveation map that counts for each input pixel the number of convolutional paths through which it can propagate information to the CNN output. We obtain these counts by computing the effective receptive field $\pmb { \pmb { 2 8 } }$ for the sum of the final convolutional layer after assigning all weights in the network to 1 (code in supplemental). Neutralizing the weights is essential to obtain per-pixel counts of input-output paths that reflect the foveation behavior. + +![](images/d463a723cb881cdd5e445297eabd8a8780cc37159812695efe551818975b2e04.jpg) +Figure 7: Foveation behavior of different padding methods applied to VGG-19 [37], and illustrated in a $5 1 2 \times 5 1 2$ input space (unless otherwise stated). Color represents the number of paths to the output for each input pixel. (a) The difference between VALID, FULL, and SAME 0-padding. (b) SAME alternatives to 0-padding. (c) Dilation amplifies foveation of SAME 0-padding. (d) Strides can lead to checkerboard patterns. (e) Foveation effects are more extensive in smaller inputs (relative to input size) and are sensitive to uneven padding. + +Figure $7 \mathrm { a }$ shows the extensive foveation effect when no padding is applied. The diminishing contribution of vast areas of the input explains the drastic drop in accuracy recently observed under VALID padding $\mathbb { \lVert \boldsymbol { 1 6 } \rVert }$ . In contrast, FULL 0-padding does not incur foveation, however, at the cost of increasing the output size after each layer, making it impractical as explained in Section $\mathbb { H }$ SAME 0-padding incurs moderate foveation at the periphery, whose absolute extent depends on the number of convolutional layers and their filter sizes. Its relative extent depends on the input size: the larger the input, the larger the ratio of the constant area in yellow (refer to appendix B for a detailed example). + +Figure 7b shows the foveation behavior of alternatives to SAME 0-padding that have roots in wavelet analysis [19] and image processing $ { \mathbb { E } } { \ b { \mathbb { Z } } } ] $ . Mirror padding mirrors pixels at the boundary to fill the padding area. When the border is included (SYMMETRIC mode in TensorFlow) all input pixels have an equal number of input-output paths 3, resulting in a uniform foveation map. When the border is not included (REFLECT mode both in PyTorch and in TensorFlow), the map exhibits bias against the border and towards a contour in its proximity. This bias is amplified over multiple layers. Replication padding exhibits the opposite bias when the padding area is wider than 1 pixel. This is because it replicates the outer 1-pixel border multiple times to fill this area 3. The method is equivalent to SYMMETRIC if the padding area is 1-pixel wide. Circular padding wraps opposing borders, enabling the kernels to seamlessly operate on the boundary and resulting in a uniform map. Partial Convolution $\lVert 2 2 \rVert$ has been proposed as a padding method that treats pixels outside the original image as missing values and rescales the computed convolutions accordingly $\mathbb { \left[ \left. 2 3 \right] \right. }$ . Its foveation behavior resembles reflective padding 3. Distribution padding $\pmb { \mathbb { B } } 0 \|$ resizes the input to fill the padding area around the original feature map, aiming at preserving the distribution of the map. Its foveation map is largely uniform, except for the corners and edges. + +Impact of input size Besides influencing the relative extent of foveation effects, the input size also determines the presence of uneven padding (or uneven feature-map erosion), as we discussed in Section 5. Figure $\textcircled { 7 } \textcircled { \times }$ shows the foveation map for VGG-19 with a $1 2 7 \times 1 2 7$ input. This input violates Eq. 1 at every downsampling layer (appendix $\mathbf { A } )$ , leading to successive feature map erosion at the bottom and right sides which is reflected in the foveation map (see appendix B for a detailed example). The bottom-right part of the input is hence less involved in the CNN computations. + +Impact of dilation We assign a dilation factor of 2 to all VGG-19 convolutional layers. While this exponentially increases the receptive field of the neurons at deeper layers $\pm 2 \|$ , dilation doubles the extent of the non-uniform peripheral areas that emerge with SAME 0-padding as evident in Figure $\textcircled { 7 } \textcircled { < }$ SYMMETRIC and circular padding maintain uniform foveation maps regardless of dilation 3. In contrast, dilation increases the complexity of these maps for REFLECT and replication padding. + +Impact of strides Whether learned on based on pooling, downsampling layers can amplify the impact of succeeding convolutional layers on foveation behaviour. Furthermore, these layers can cause input pixels to vary in the count of their input-output paths. This can happen when the kernel size is not divisible by the stride, leading to a checkerboard pattern in the foveation maps. This manifests in ResNet models as we illustrate in appendix B. In VGG-19, all max-pooling layers use a stride of 2 and kernel size of 2. Changing the kernel size to 3 leads to a checkerboard pattern as evident in Figure 7d. Such effects were shown to impact pixel-oriented tasks $\mathbb { B } 2 \mathbb { I }$ . + +The padding technique and its foveation behaviour have direct impact on feature-map artifacts (Section $\bar { 7 } )$ , and on the ability of CNNs to encode spatial information (Section $^ { 8 ) }$ . Understanding the foveation behavior is key to determine how suited a padding method is for a given task. For example, small object detection is known to be challenging close to the boundary $\left[ \left[ 2 6 \right] \right]$ , in part due to the foveation behavior of SAME 0-padding. In Figure $\bar { 5 } 6$ , we change the padding method in the SSD to SYMMETRIC. The stimulus is noticeably more detectable at the boundary, compared with 0-padding $^ 4 \cdot$ In contrast, ImageNet classification is less sensitive to foveation effects because the target objects are mostly located away from the periphery. Nevertheless, the padding method was shown to impact classification accuracy $\mathbb { \left[ \left. 2 3 \right] \right. }$ because it still affects feature map artifacts. + +# 7 PADDING METHODS AND FEATURE MAP ARTIFACTS + +It is also noticeable that the score map in Figure $5 \mathsf { b }$ is more uniform than in Figure $\textcircled { 5 } \textcircled { \times }$ . In particular, under SYMMETRIC padding the model is able to detect traffic lights placed in the blind spots of the original 0-padded model. To verify whether the line artifacts in Figure $\bigstar$ are mitigated, we inspect the mean feature maps of the adapted model. With a constant input, SYMMETRIC padding warrants constant maps throughout the CNN because it reuses the border to fill the padding area. Instead, we average these maps over 30 samples generated uniformly at random. Figure 8 depicts the mean maps which are largely uniform, unlike the case with 0-padding. + +![](images/57e698bb07346f6a27d58d62665ad1ecd8d2ee482ba665e630bbd20159d302cc.jpg) +Figure 8: The same feature maps in Figure 2, generated under mirror padding and averaged over 30 randomly-generated input samples. The line artifacts induced by 0-padding are largely mitigated. + +To further analyze the impact of SYMMETRIC padding, we retrain the adapted model following the original training protocol. This significantly improves the average precision (AP) as reported in Table $2$ under different overlap thresholds (matching IoU), confirming that small object detection is particularly sensitive to feature-map artifacts. + +Table 2: Performance of the SSD traffic light detector, trained under two different padding schemes. + +
Average Precision (AP)AP@.20I0UAP@ .50I0UAP@.75I0UAP@.90I0U
Zero Padding80.24%49.58%3.7%0.007%
Mirror Padding83.20%57%8.44%0.02%
+ +Of the padding methods listed in Section $6 ,$ mirror padding in both SYMMETRIC and REFLECT modes, PartialConv, and circular padding are generally effective at reducing feature map artifacts that emerge under zero padding, in particular salient line patterns. In contrast, distribution padding can induce significant artifacts. Refer to appendix D for comparative examples of artifacts under the aforementioned padding schemes. + +Artifact magnitude and propagation While feature-map artifacts are induced by the padding mechanism at the boundary, their magnitude and inward propagation in the maps are impacted by several architectural aspects of CNNs. In particular, certain normalization schemes such as batchnorm [15] tend to limit the range of variation within a feature map and to relatively harmonize this range across different maps. This, in turn, impacts how possible artifacts in these maps accumulate when they are processed by the next convolutional layer. Similarly, artifacts that manifest after applying ReLU units are of a positive sign. These factors were instrumental in the formation of potential blind spots described in Section 3. We hence recommend to involve non-convolutional layers when inspecting the feature maps. Besides having possible impact on artifact magnitude, several aspects of convolution arithmetic, such as filter size and dilation factors, can also impact the spatial propagation of these artifacts. + +# 8 RELATED FINDINGS AND TAKEAWAYS + +Handling the boundary is an inherent challenge when dealing with spatial data [9]. Mean padding is known to cause visual artifacts in traditional image processing, with alternative methods proposed to mitigate them $\mathbb { \lVert 2 4 \rVert }$ . CNNs have been often assumed to deal with such effects implicitly. Innamorati et al $\dot { [ \lVert { 4 } \rVert }$ propose learning separate sets of filters dedicated to the boundaries to avoid impacting the weights learned by regular filters. A grouped padding strategy, proposed to support $2 \times 2$ filters $\mathbf { \bar { \textmu } }$ , offers avenues to mitigate uneven padding and corresponding skewness in foveation maps without restrictions on input size (see our note in appendix $\mathbf { B }$ for explanation). Finally, insights from signal and image processing [10; 11] could inspire further CNN padding schemes. + +Zero padding has been recently linked to CNNs’ ability to encode position information [7; 16; 18; 29]. In contrast, circular padding was shown to limit this ability $\mathbb { \left[ \bigcirc \right] }$ and to boost shift invariance $\begin{array} { r l } { { \bigl \| \overline { { 3 5 } } \bigr \| } } & { { } } \end{array}$ . The input sizes in those studies do induce uneven padding. This can be, in part, the underlying mechanism behind the aforementioned ability. Whether or not this ability is desirable depends on the task, with several methods proposed to explicitly encode spatial information [5; 6; 20; 25; 29; 31]. + +Downsampling using max-pooling or strided convolution has been shown to impact shift invariance in CNNs by incurring aliasing effects [3; 38; 43]. These effects can manifest in the same symptoms we reported in Section $^ { 1 , }$ albeit for a different reason. Zhang [43] demonstrated how blurring the feature maps before subsampling mitigates aliasing effects and improves ImageNet classification accuracy of various popular CNNs. We analyzed the mean filters in antialiased MobileNet and ResNet models pre-trained on ImageNet under 0-padding, with $2 2 4 \times 2 2 4$ as input size (refer to Appendix $\mathrm { E } )$ . We found that antialiasing can also mitigate the asymmetry of mean filters that exhibited high asymmetry in the baseline models, especially at deeper layers. This is remarkable given that these models are trained on $2 2 4 \times 2 2 4$ images, which incurs one-sided zero padding at every downsampling layer. This could, in part, be attributed to the ability of the BlurPool operator used in antialiased CNN to smoothen the acuity of zero-padded borders, in turn, reducing the value imbalance incurred by one-sided padding. Further analysis is needed to examine the interaction between padding and aliasing effects in CNNs and to establish possible synergy between antialiasing and eliminating uneven application of padding. + +Luo et al $\pmb { \Vert 2 8 \Vert }$ drew connections between effective receptive fields and foveated vision. Our analysis links foveation behavior with the padding scheme and suggests that it might occur implicitly in CNNs when using VALID or SAME 0-padding, without the need for explicit mechanisms [2; 21]. Furthermore, it explains the drastic accuracy drop noted by $\boxed { 1 0 }$ under VALID padding, which is amplified by feature map erosion. + +Choosing a padding method SAME 0-padding is by far the most widely-used method. Compared with other methods, it can enable as much as $5 0 \%$ faster training and inference. Problem-specific constraints can dictate different choices [34; 35; 40]. In the lack of a universally superior padding method, we recommend considering multiple ones while paying attention to the nature of the data and the task, as well as to the following aspects: + +• Feature-map statistics: 0-padding can alter the value distribution within the feature maps and can shift their mean value in the presence of ReLU units. The alternatives presented in Section 6 tend to preserve this distribution, thanks to reusing existing values in the maps. Foveation behavior: 0-padding might not be suited for tasks that require high precision at the periphery, unlike circular and SYMMETRIC mirror padding. Interference with image semantics (esp. with a padding amount $> 1$ pixel): For example, circular padding could introduce border discontinuities unless the input is panoramic $| \widehat { \mathsf { B } } \widehat { \mathsf { S } } \|$ . • Potential to induce feature map artifacts: All alternatives to 0-padding induce relatively fewer artifacts, except for Distribution padding $\textcircled { \lVert { 3 0 } \rVert }$ (see appendix ${ \bf D } )$ + +We also recommend eliminating uneven padding at downsampling layers both at training and at inference time, as we illustrated in Section $\boxed { 5 }$ This is especially important when zero padding is applied and the downsampling is learned. The scripts used to generate the visualizations in this paper are available in the supplemental as well as at http://mind-the-pad.github.io. + +Summary We demonstrated how the padding mechanism can induce spatial bias in CNNs, in the form of skewed kernels and feature-map artifacts. These artifacts can be highly pronounced with the widely-used 0-padding when applied unevenly at the four sides of the feature maps. We demonstrated how such uneven padding can inherently take place in state-of-the-art CNNs, and how the artifacts it causes can be detrimental to certain tasks such as small object detection. We provided visualization methods to expose these artifacts and to analyze the implication of various padding schemes on boundary pixels. We further proposed solutions to eliminate uneven padding and to mitigate spatial bias in CNNs. Further work is needed to closely examine the implications of spatial bias and foveation in various applications (see supplementary for examples), as well as padding impact on recurrent models and 1-D CNNs. + +# ACKNOWLEDGEMENT + +We are thankful to Ross Girshick for providing useful recommendations and experiment ideas, and to Shubham Muttepawar for implementing an interactive tool out of our analysis scripts, guided by our front-end specialist Edward Wang and our AI user-experience designer Sara Zhang. + +REFERENCES +[1] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, et al. TensorFlow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016. +[2] E. Akbas and M. P. Eckstein. Object detection through search with a foveated visual system. PLoS computational biology, 13(10):e1005743, 2017. +[3] A. Azulay and Y. Weiss. Why do deep convolutional networks generalize so poorly to small image transformations? Journal of Machine Learning Research (JMLR), 20(184):1–25, 2019. +[4] K. Behrendt, L. Novak, and R. Botros. A deep learning approach to traffic lights: Detection, tracking, and classification. In Robotics and Automation (ICRA), 2017 IEEE International Conference on, pp. 1370–1377. IEEE, 2017. +[5] C.-A. Brust, S. Sickert, M. Simon, E. Rodner, and J. Denzler. Convolutional patch networks with spatial prior for road detection and urban scene understanding. In International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISAPP), 2015. +[6] G. F. Elsayed, P. Ramachandran, J. Shlens, and S. Kornblith. Revisiting spatial invariance with low-rank local connectivity. In International Conference on Machine Learning (ICML), 2020. +[7] J. Geiping, H. Bauermeister, H. Droge, and M. Moeller. Inverting gradients–how easy is it to ¨ break privacy in federated learning? arXiv preprint arXiv:2003.14053, 2020. +[8] R. Gens and P. M. Domingos. Deep symmetry networks. In Advances in neural information processing systems (NeurIPS), pp. 2537–2545, 2014. +[9] D. Griffith and C. Amrhein. An evaluation of correction techniques for boundary effects in spatial statistical analysis: traditional methods. Geographical Analysis, 15(4):352–360, 1983. +[10] V. Gupta and N. Ramani. A note on convolution and padding for two-dimensional data. Geophysical Prospecting, 26(1):214–217, 1978. +[11] L. Hamey. A functional approach to border handling in image processing. In International Conference on Digital Image Computing: Techniques and Applications, pp. 1–8, 2015. +[12] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016. +[13] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017. +[14] C. Innamorati, T. Ritschel, T. Weyrich, and N. J. Mitra. Learning on the edge: Investigating boundary filters in CNNs. International Journal of Computer Vision (IJCV), pp. 1–10, 2019. +[15] S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning (ICML), pp. 448– 456, 2015. +[16] M. A. Islam, S. Jia, and N. D. Bruce. How much position information do convolutional neural networks encode? In International Conference on Learning Representations (ICLR), 2020. +[17] M. Jaderberg, K. Simonyan, A. Zisserman, et al. Spatial transformer networks. In Advances in neural information processing systems (NeurIPS), pp. 2017–2025, 2015. +[18] O. S. Kayhan and J. C. van Gemert. On translation invariance in CNNs: Convolutional layers can exploit absolute spatial location. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), 2020. +[19] T. L. Kijewski-Correa. Full-scale measurements and system identification: A time-frequency perspective. PhD thesis, University of Notre Dame., 2003. +[20] I. Kim, W. Baek, and S. Kim. Spatially attentive output layer for image classification. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), 2020. +[21] H. Larochelle and G. E. Hinton. Learning to combine foveal glimpses with a third-order boltzmann machine. In Advances in neural information processing systems (NeurIPS), pp. 1243–1251, 2010. +[22] G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro. Image inpainting for irregular holes using partial convolutions. In European Conference on Computer Vision, 2018. +[23] G. Liu, K. J. Shih, T.-C. Wang, F. A. Reda, K. Sapra, Z. Yu, A. Tao, and B. Catanzaro. Partial convolution based padding. In arXiv preprint arXiv:1811.11718, 2018. +[24] R. Liu and J. Jia. Reducing boundary artifacts in image deconvolution. In IEEE International Conference on Image Processing (ICIP), pp. 505–508, 2008. +[25] R. Liu, J. Lehman, P. Molino, F. P. Such, E. Frank, A. Sergeev, and J. Yosinski. An intriguing failing of convolutional neural networks and the CoordConv solution. In Advances in Neural Information Processing Systems (NeurIPS), pp. 9605–9616, 2018. +[26] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. SSD: Single shot multibox detector. In European Conference on Computer Vision, pp. 21–37, 2016. +[27] S. Lou, X. Jiang, and P. J. Scott. Fast algorithm for morphological filters. Journal of Physics: Conference Series, 311(1):012001, 2011. +[28] W. Luo, Y. Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in deep convolutional neural networks. In Advances in Neural Information Processing Systems (NeurIPS), pp. 4898–4906, 2016. +[29] R. Murase, M. Suganuma, and T. Okatani. How can cnns use image position for segmentation? arXiv preprint arXiv:2005.03463, 2020. +[30] A.-D. Nguyen, S. Choi, W. Kim, S. Ahn, J. Kim, and S. Lee. Distribution padding in convolutional neural networks. In IEEE International Conference on Image Processing (ICIP), pp. 4275–4279, 2019. +[31] D. Novotny, S. Albanie, D. Larlus, and A. Vedaldi. Semi-convolutional operators for instance segmentation. In European Conference on Computer Vision (ECCV), pp. 86–102, 2018. +[32] A. Odena, V. Dumoulin, and C. Olah. Deconvolution and checkerboard artifacts. Distill, 1 (10):e3, 2016. +[33] A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, et al. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems (NeurIPS), pp. 8024–8035, 2019. +[34] P. O. Pinheiro, T.-Y. Lin, R. Collobert, and P. Dollar. Learning to refine object segments. In ´ European Conference on Computer Vision (ECCV), pp. 75–91, 2016. +[35] S. Schubert, P. Neubert, J. Poschmann, and P. Pretzel. Circular convolutional neural networks ¨ for panoramic images and laser data. In IEEE Intelligent Vehicles Symposium (IV), pp. 653– 660, 2019. +[36] E. Shalnov. BSTLD-demo: A sample project to train and evaluate model on BSTLD. https: //github.com/e-sha/BSTLD_demo, 2019. +[37] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. In International Conference on Learning Representations (ICLR), 2015. +[38] G. Sundaramoorthi and T. E. Wang. Translation insensitive CNNs. arXiv preprint arXiv:1911.11238, 2019. +[39] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. Going deeper with convolutions. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), pp. 1–9, 2015. +[40] S. Vashishth, S. Sanyal, V. Nitin, N. Agrawal, and P. Talukdar. InteractE: Improving convolution-based knowledge graph embeddings by increasing feature interactions. In AAAI conference on Artifical Intelligence, 2020. +[41] S. Wu, G. Wang, P. Tang, F. Chen, and L. Shi. Convolution with even-sized kernels and symmetric padding. In Advances in Neural Information Processing Systems (NeurIPS), pp. 1192–1203, 2019. +[42] F. Yu and V. Koltun. Multi-scale context aggregation by dilated convolutions. In International Conference on Learning Representations (ICLR), 2016. +[43] R. Zhang. Making convolutional networks shift-invariant again. In International Conference on Machine Learning (ICML), 2019. + +# A ELIMINATING UNEVEN APPLICATION OF PADDING + +Consider a CNN with $d$ downsampling layers, $L _ { 1 } , L _ { 2 } , . . . , L _ { d }$ . To simplify the analysis and without loss of generality we assume that the kernels in these layers are of square shape and that all other layers maintain their input size. We denote by $s _ { i }$ and $k _ { i }$ the stride and kernel size of layer $L _ { i }$ . We denote by $h _ { i }$ and $w _ { i }$ the dimensions of the feature maps computed by $L _ { i }$ . We denote by $h _ { 0 }$ and $w _ { 0 }$ the size of the CNN input. We examine the conditions to warrant no uneven application of padding along the height dimension. Parallel conditions apply to the width dimension. + +We denote by $\bar { h } _ { i }$ the height of the padded input to $L _ { i }$ . The effective portion $\hat { h } _ { i } \leq \bar { h } _ { i }$ of this amount processed by the convolutional filters in $L _ { i }$ is equal to: + +$$ +\hat { h } _ { i } = s _ { i } \cdot \left( h _ { i } - 1 \right) + k _ { i } +$$ + +Our goal is to warrant that $\hat { h } _ { i } = \bar { h } _ { i }$ to prevent information loss and to avoid uneven padding along the vertical dimension when the unconsumed part $\bar { h } _ { i } - \hat { h } _ { i } < s _ { i }$ is an odd number. + +Since the non-downsampling layers maintain their input size, we can formulate the height of the padded input as follows: + +$$ +\bar { h } _ { i } = h _ { i - 1 } + 2 \cdot p _ { i } +$$ + +where $p _ { i }$ is the amount of padding applied at the top and at the bottom of the input in $L _ { i }$ . Accordingly, we can warrant no uneven padding if the following holds: + +$$ +\forall i \in [ 1 . . d ] : \quad h _ { i - 1 } = s _ { i } \cdot ( h _ { i } - 1 ) + k _ { i } - 2 \cdot p _ { i } +$$ + +Example 1: ResNet-18 This network contains five downsampling layers ( $\mathrm { : } d = 5$ ) all of which use a stride of 2. Despite performing downsampling, all of these layers apply a padding amount entailed by SAME padding to avoid information bias against the boundary. In four of these layers having $3 \times 3$ kernels $k _ { i } = 3 ,$ ), the amount used is $p _ { i } = 1$ . For the first layer having $7 \times 7$ kernels, this amount is equal to 3. In both cases, the term $k _ { i } - 2 \cdot p _ { i }$ in Eq. $3$ is equal to 1. To warrant no uneven padding along the vertical dimension, the heights of the feature maps at downsampling layers should hence satisfy: + +$$ +\forall i \in [ 1 . . d ] : \quad h _ { i - 1 } = 2 \cdot ( h _ { i } - 1 ) + 1 = 2 \cdot h _ { i } - 1 +$$ + +Accordingly, the input height should satisfy: + +$$ +h _ { 0 } = 2 ^ { d } \cdot h _ { d } - ( 2 ^ { d } - 1 ) = 2 ^ { d } \cdot ( h _ { d } - 1 ) + 1 +$$ + +where $h _ { d }$ is the height of the final feature map, and can be any natural number larger than 1 to avoid a degenerate case of a $1 \times 1$ input. The same holds for the input width: + +$$ +w _ { 0 } = 2 ^ { d } \cdot ( w _ { d } - 1 ) + 1 +$$ + +A $2 2 5 \times 2 2 5$ input satisfies these constraints since $2 2 5 = 2 ^ { 5 } \cdot 7 + 1$ , yielding even padding in all five downsampling layers and output feature maps of size $8 \times 8$ . + +Example 2: VGG-16 This network contains five max-pooling layers $( d = 5$ ) all of which use a stride of 2 and a kernel size of 2 and apply no padding. To warrant no uneven padding along the vertical dimension, the heights of the feature maps at all of these layers should hence satisfy: + +$$ +\forall i \in [ 1 . . d ] : \quad h _ { i - 1 } = 2 \cdot ( h _ { i } - 1 ) + 2 = 2 \cdot h _ { i } +$$ + +Accordingly, the input dimensions should satisfy: + +$$ +h _ { 0 } = 2 ^ { d } \cdot h _ { d } \quad \mathrm { a n d } \quad w _ { 0 } = 2 ^ { d } \cdot w _ { d } +$$ + +A $2 2 4 \times 2 2 4$ input satisfies these constraints since $2 2 4 = 2 ^ { 5 } \cdot 7$ , causing no feature-map erosion at any downsampling layer and resulting in output feature maps of size $7 \times 7$ . + +# B THE EXTENT OF FOVEATION UNDER SAME 0-PADDING + +We illustrate how the absolute extent of foveation under SAME 0-padding depends on the number of convolutional layers, and how its relative extent depends on the input size. + +In the following maps, color represents the number of paths to the CNN output for each input pixel. Note: The checkerboard pattern is caused by downsampling layers in ResNet that use $3 \times 3$ kernels and a stride of 2. + +![](images/230d99b41516d96c2c03aec57cab7c869711ad4d96a7315fd9c45cd64c50a0f0.jpg) +Figure 9: The foveation maps of two ResNet architectures under 0 padding, illustrated with a $2 2 5 \times 2 2 5$ input. Compared with ResNet-50, ResNet-101 has twice the number of convolutional layers with non-unitary filter sizes. Accordingly, the extent of the foveation effect is doubled. + +![](images/a41bda8b2c50acbca4b5cd99ab0cb07019bf3ee67df5aae70ce1fa81a6316f40.jpg) +Figure 10: The foveation maps of ResNet-50 under 0 padding, illustrated with inputs of different size. The smaller the input, the larger the relative extent of foveation. + +In the next figure, we illustrate how uneven application of padding impacts the foveation maps. Note: It is possible to rectify the skewness in the 2nd foveation map by alternating the side where one-sided padding is applied between successive downsampling layers. This, however, does not mitigate the skewness in the learned filters (see next Section). + +![](images/b6915c9095fb0ab1b1b632c64deab332d79ccd94ad4eb92251f1454b4674d7bd.jpg) +Figure 11: The foveation maps of ResNet-50 under 0 padding, illustrated with two input sizes. With a $2 5 7 \times 2 5 7$ input, the padding is evenly applied at all downsampling layers, leading to a symmetric foveation map. With a $2 5 6 \times 2 5 6$ input, the padding is applied only to the left and top sides of feature maps at all downsampling layers, which limits the number of convolutional input-output paths for pixels in the bottom and right sides as evident in the skewed foveation map. + +# C THE IMPACT OF THE PADDING METHOD ON LEARNED WEIGHTS + +In the presence of uneven application of padding, 0-padding causes skewness in the learned weights because the filters are exposed more frequently to feature-map patches with zeros at their top and left sides. Redundancy methods such as circular or mirror padding mitigate such skewness because they fill the padding areas with values taken from the feature maps. PartialConv also mitigates such skewness because it assumes the pixels in the padding area are missing, and rescales the partial convolutional sum to account for them. Below we show the effectiveness of these alternatives in mitigating the skewness in three ResNet architectures. + +![](images/c9b6cd065fd2a6ff0aad0928693bb5a1ef91f3a05a54734b812f2ef68cb1a3cc.jpg) + +![](images/28bf70ac734a23168ae18bc3334f51e2d26eca67e7b3e2d076b3dfd9008e4880.jpg) +(b) Mean filters of ResNet-50 trained on $2 2 4 \times 2 2 4$ images under two padding methods, reaching $7 6 . 1 5 \%$ top-1 accuracy under 0-padding and $7 6 . 6 1 \%$ top-1 accuracy under PartialConv. + +(a) Mean filters of ResNet-18 trained on $2 2 4 \times 2 2 4$ images under two padding methods, reaching $6 9 . 9 3 \%$ top-1 accuracy under 0-padding and $7 0 . 2 8 \%$ top-1 accuracy under circular padding. + +Figure 12: Mean filters of two ResNet models trained on ImageNet with $2 2 4 \times 2 2 4$ images. The input size causes uneven application of padding, leading to frequent asymmetries in the mean filters under 0 padding. We illustrate how two alternatives, circular padding and PartialConv $\pmb { \left. \pmb { \left. \bar { 2 3 } \right. } \right. }$ , enable learning highly-symmetric mean filters despite the uneven application of padding. + +![](images/7bc6aeb0c0d5fa125d64fbf452bdfc569b0089cb7b988e4d3f582824309d0153.jpg) +Figure 13: Mean filters of ResNet-101 trained on ImageNet with $2 2 4 \times 2 2 4$ images under both 0- padding and PartialConv $\mathbb { \lVert 2 3 \rVert }$ . The input size causes uneven application of padding, leading to frequent asymmetries in the mean filters under 0 padding. In contrast, PartialConv produces highly symmetric mean filters, thanks for its treatment of pixels outside the feature map as missing values. + +What if no padding is applied during downsampling? VGG models perform downsampling using $2 \times 2$ pooling layers that do not apply any padding. Accordingly, the mean filters do not exhibit significant skewness, even if the input size does not satisfy Eq 4: + +![](images/51a8af5576a494d4ff89c4d3741816ce78e27e93e3c83213ba9b0e683e37e048.jpg) +Figure 14: Mean filters of VGG-16 trained on ImageNet under different conditions. Most mean filters exhibit high symmetry when trained with $2 2 5 \times 2 2 5$ images where the size violates Eq. 4. + +# D THE IMPACT OF PADDING METHODS ON FEATURE-MAP ARTIFACTS + +We show per-layer mean feature maps in ResNet-18 under different padding methods. The mean maps are averaged over 20 input samples generated at random. + +![](images/4dc8a25d06b953aa2901c2738d70c12d28f3f668e42fe6df8e21854cbd416e19.jpg) +Figure 15: Feature map artifacts under zero padding. Line artifacts accumulate to become significant and asymmetric at deeper layers. + +![](images/6640c3d2bb4a8864228e5cb9e90262456ad14527bc623e13533134115c042001.jpg) +Figure 16: Circular padding largely preserves the randomness and mitigates line artifacts. + +![](images/f3cd7587862b9845b41afe3205965e60be50ddde1f43d11c94bed84f9ba5ee93.jpg) +Figure 17: SYMMETRIC mirror padding also preserves the randomness and mitigates line artifacts. + +![](images/56552f00a0e6677217affe158f48306a5f1d42e3ca61577ae6dcc64f4fc464fc.jpg) +Figure 18: REFLECT mirror padding also preserves the randomness and mitigates line artifacts. + +![](images/72d9bfe380a72b51f7220f1be45381b387d02402c6fb51a1ea2bb58ba8ea21f1.jpg) + +Figure 19: PartialConv $\mathbb { \left[ \left. 2 3 \right] \right. }$ highly preserves the symmetry of the feature maps. The scaling factors it uses can break the randomness at the boundary. + +![](images/3813e0f0f363045277b94f3a816f45c0ab5f295c9700bc4537785b52027e2f6d.jpg) +Figure 20: Feature map artifacts of a VGG-19 model under Distribution Padding (interpolation mode) $\textcircled { \lvert 3 0 \rvert }$ . Due to multiple resize operations used to fill the padding area, the artifacts grow from the boundary inwards. We use a saturated constant input to make the effect visible. + +# E THE IMPACT OF ANTIALIASING ON THE LEARNED WEIGHTS + +We demonstrate how antialiasing [43] significantly reduces the asymmetry of mean filters around downsampling layers, even in the presence of unevenly-applied zero padding. + +![](images/2012189e75de9944003498504b5db09216e0eb5f175070e1ae807135b096f5fd.jpg) +Figure 21: Mean filters of four models trained on ImageNet with $2 2 4 \times 2 2 4$ images under 0-padding both without and with antialiasing. + +![](images/57e3f9aa779c1d5920a44c13b73025feb076f2064e727af4c763cdde601bb09a.jpg) +Figure 22: Mean filters of two models trained on ImageNet with $2 2 4 \times 2 2 4$ images under 0-padding both without and with antialiasing. + +# F FOVEATION ANALYSIS OF PADDING ALGORITHMS + +Refer to http://mind-the-pad.github.io for an interactive and animated visual illustration of padding algorithms and their foveation behavior. This appendix serves as a print version. + +Among the SAME padding algorithms we discussed in the manuscript, two algorithms warrant that each input pixel is involved in an equal number of convolutional operations, leading to uniform foveation maps: circular padding and SYMMETRIC mirror padding. In contrast, this number varies under zero padding, REFLECT mirror padding, replication padding, and partial convolution. + +We illustrate in detail how each padding algorithm treats the input pixels. For this purpose we illustrate step by step how each pixel is processed by the convolutional kernel. We choose a set of pixels that are sufficient to expose the behavior of the respective algorithm. This set spans an area within two or three pixels from the boundary that encompasses all relevant cases for the analysis and is situated at the top-left corner. The behavior at the other corners is analogous. + +All illustrations use a stride of 1. Except for VALID, all configurations warrant SAME padding. + +• VALID Padding: This algorithm is illustrated on a $3 \times 3$ kernel without dilation. A larger kernel size or dilation factor will increase the foveation effect. Zero Padding: This algorithm is illustrated on a $3 \times 3$ kernel without dilation. A larger kernel size or dilation factor will increase the foveation effect. Circular Padding: This algorithm is illustrated on a $3 \times 3$ kernel without dilation. It is straightforward to prove that the algorithm warrants equal treatment of the pixels irrespective of the kernel size or dilation factor. This is because it effectively applies circular convolution: Once the kernel hits one side, it can seamlessly operate on the pixels of the other side. Circular convolution hence renders the feature map as infinite to the kernel, warranting that edge pixels are treated in the same manner as interior pixels. +. Mirror Padding (SYMMETRIC): This algorithm warrants that each pixel is involved in the same number of convolutional operations. It is important to notice that, unlike under circular convolution, these operations do not utilize the kernel pixels uniformly as we demonstrate in detail. We illustrate the algorithm behavior under the following settings: – $3 \times 3$ kernel and dilation factor of 1. +– $5 \times 5$ kernel and dilation factor of 1. +– $3 \times 3$ kernel and dilation factor of 2. +– $2 \times 2$ kernel and dilation factor of 1, along with a grouped padding strategy to compensate for uneven padding $\pm \amalg$ . +– $4 \times 4$ kernel size and dilation factor of 1, along with a grouped padding strategy. + +• Mirror Padding (REFLECT): This algorithm is illustrated on a $3 \times 3$ kernel without dilation. + +Replication Padding: This algorithm is illustrated on a $5 \times 5$ kernel without dilation. We choose this kernel size since a $3 \times 3$ kernel under SAME padding would render the algorithm equivalent to SYMMETRIC mirror padding. +• Partial Convolution: This algorithm is illustrated on a $3 \times 3$ kernel without dilation. Its foveation behavior is analogous to REFLECT mirror padding. + +# VALID Padding Illustrated on a 3x3 kernel + +# Input + +# # of conv ops each pixel is involved in + +
abC
def
gh
+ +
12333
24666
36999
36999
36999
+ +# Which kernel cells these ops utilize? + +![](images/b2f7ea17392566a942f1830fef3e08688e93e3711a4ea27135d9a2b9cf7d34e6.jpg) + +![](images/74f48cb87d1f6bdd6666b649c425ea1dfaa5ae5f8b7d6d1487b47797a64dddd6.jpg) +Detailed Illustration of how the counts are derived + +Convolutions involving (d): rotated version of (b) + +Convolutions involving (e) + +![](images/075393ac4f0fb1926036df9540e55b75341726cc0c0ce597fe3930694431dbc7.jpg) +Convolutions involving (f) + +![](images/a9884b5f8b96c555974a7c1a3e43a6404f579230ff72e5c6ffb7dbcd48eee7a4.jpg) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +# Zero Padding + +# Illustrated on 3x3 kernel and 1-pixel padding + +# Original Input + +Padded Input + +
000• ·
0ab
0Cd
0·
• =
+ +# of conv ops each pixel is involved in + +
46666
69999
6999
69999
69999
+ +
ab··
Cd
·
+ +# Which kernel cells these ops utilize? + +![](images/05776e978b31b6a3e6f1341bc05f18a6de2a45877c2b627e7689f8e71e5ec1b5.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/2db3478b53918d06b972e15db60b69b5650f9e2b06fc4d4bbc8dc175ac75984c.jpg) + +Convolutions involving (b) + +![](images/63ee4d6ede122d55ac9f5d26ee63c1d6ddf758f4267f0d482b84beda20114155.jpg) + +Other border cases are translation or rotation of (a) or (b) + +# Circular Padding + +Illustrated on 3x3 kernel and 1-pixel padding + +![](images/312ea301340019a6276321b4c734de7b4bca19c3c6e7a0942faf86f2dfa2b9d9.jpg) + +# Which kernel cells these ops utilize? + +a 1 1 1 b 1 1 1 +1 1 1 1 1 1 +1 1 1 1 1 1 +sum = 9 sum = 9 +uniform uniform + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/315c801266b735e20fd2dda8208ed2b4da076ddadc7c67b0b3f5105bfd9d73fd.jpg) + +Other border cases are translation or rotation of (a) or (b) + +# Mirror Padding (SYMMETRIC) + +Illustrated on 3x3 kernel and 1-pixel padding + +![](images/6ab53447be0e84af51087dd7f95b2e53e57f5b4c85508482b2d0499f14281ec1.jpg) + +# Which kernel cells these ops utilize? + +![](images/6122910bbc215ed57d822669b52bf0365aa06feda5eeb930d346877584b23468.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/207ecdbeda64fb86e5c2fb8bd0fca5301fca26b3c149e3e0de0f78f19b80b6fc.jpg) + +![](images/ce9ae3f6f155e9103378337a6607d52884729cba759d261dd7aa7d104d009ab0.jpg) +Convolutions involving (b) + +![](images/7ff8f52e1f1fa7323a591bc9045a4daa5fbdbf91e09c8b8c0c73e8c62390c4f3.jpg) + +Convolutions involving (e) + +![](images/635b17420ad7db5958a7c27e978a4071b990979ca04dc4ace8f486033de20cd5.jpg) + +Convolutions involving (f) + +![](images/83b97cfd9b41665e91f8e411ba270b85c9a125588a02eeca4e21f821f73b4bdb.jpg) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +![](images/8a853098ca11735df8a765434a1c5d4f3771546980790927ebd6e4534ec0b669.jpg) + +# Mirror Padding (SYMMETRIC) + +Illustrated on 5x5 kernel and 2-pixel padding + +# Which kernel cells these ops utilize? + +![](images/2ed647a62055c626bfc6be0708fddbc84405c35e0aa5605dfe2983f80d1917e0.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/fb93f2025462e471e18bab29c553707d796f5a5df450e689eb1d800b265197c9.jpg) +Convolutions involving (b) + +![](images/d6b1ab684ccc9d8b4d0633f56a62178f4ad4cceda1c73f8111bc9eb705c2d5a9.jpg) +Convolutions involving (c) + +![](images/923c01362a6ed3e205d71bfc24d478c52b3681aa29cd39b01495efc03ffbd84f.jpg) + +Convolutions involving (e) + +![](images/7e697dd45efbc10ee0c5668063cefbca79bec51cfe0a5bfd6d8827c821509c0a.jpg) + +![](images/9a8f10a070589ef06b9f1a5b0651e3ebe32e891395ebc4fbf96356dc972aba0c.jpg) +Convolutions involving (f) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +![](images/6ab1a20295a0fdba167ef4c254d75ebac90a68354c628ae8597a45a489da49ce.jpg) + +# Mirror Padding (SYMMETRIC) + +Illustrated on $\mathbf { 3 \times 3 }$ kernel and 1-pixel padding with dilation factor of 2 + +# Which kernel cells these ops utilize? + +![](images/d99e99f62153187884e0a145ec5dd4f0201fb8255b44b2c45aa57158fc39430c.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/749f38dde001c33c4de5e80e1053df61ebc232047d92d82b0267074b9fdafae1.jpg) +Convolutions involving (b) + +![](images/3d93f43f0bbad03a7c07d066a148fcedbe3c2528cec4208b76899c07614a04d1.jpg) +Convolutions involving (c) + +![](images/0cda8ee3d259cb24f0c45181ae446820fe9b1146d356cd59c256c8ab595dfe26.jpg) + +Convolutions involving (e) + +![](images/bd8b805d273315488c5e4f9d9c491d79b396d892db1a7990f4d7f5ab6b0f6113.jpg) + +![](images/0c0ebb9c5742e01fd04882c2956ef19826f0f171ae4291c9a9ae0ed4d2866900.jpg) +Convolutions involving (f) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +# Mirror Padding (SYMMETRIC) with Grouping + +Illustrated on $\pmb { 2 \times 2 }$ kernel and 1-pixel padding A grouped padding strategy is applied to balance uneven padding (Wu et al 2019) + +![](images/3cb8ab9203b09b8e6b320ed18ac7056ddcb59f83631f0a18b5a139b841c32530.jpg) + +Number of conv ops each pixel is involved in + +![](images/ac51aae79692b49ae5a912d11b076b5c573aa842106e0cc6e722bb121f69aed2.jpg) + +Which kernel cells these ops utilize? + +![](images/5c6f2c7fbee023779650ad2f065adc250d53bccea9b4cedc28b2d3322622dff4.jpg) + +![](images/2c5c68905e8fca1d112cde596ce4b53cb4b79975e54fc8cea24edcdef086be98.jpg) +Detailed Illustration of how the counts are derived + +![](images/ecf7ba29a9b7754a971b14400d054da81dbb097f5b1facc5aeaa574edbabe472.jpg) +Convolutions involving (b) + +# Mirror Padding (SYMMETRIC) with Grouping Illustrated on 4x4 kernel and 1-pixel padding + +
Original InputPadded at top-leftPadded at bottom-Padded at top-rightPadded at bottom left corner
edde fddefright corner
abCbabaabaC
deedb eC fbC fa da deC faab
hhgd ghie hd gd ge highde
gggh
+ +Number of conv ops each pixel is involved in + +
Padded at top-leftPadded at bottom-leftPadded at top-rightPadded at bottom-rightAverage (grouped padding strategy)
252520202015151212121515202020991212121616161616
252520202015151212121515202020991212121616161616
20201616162020161616121216161612121616161616161616
20201616162020161616121216161612121616161616161616
20201616 1620201616121216161612121616161616161616
+ +![](images/3c18f875aebd47ed9b36050ea6c6f9e4ee9a4c0390f1a3d6358113ae4f001675.jpg) +Which kernel cells these ops utilize? + +# Replication Padding + +![](images/305042bd7a20cd1eafbe805182fa540e7e2ce9a422379289d9ee15436767c91c.jpg) + +Illustrated on 5x5 kernel and 2-pixel padding + +# Which kernel cells these ops utilize? + +![](images/4a3779ad3fcb3e7be7745952303c7cebdc2bebab9c0d386a100a95ee8c53f573.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/bf7eec628f94a9d43e2e8a7d0f391f88d89ee6a81d3ab11c5aa917df496a5a5f.jpg) + +Convolutions involving (b) + +![](images/9787549e926a66cc5f95e14c83661906e3e54f96ed8952a4ab6af039ad13d70a.jpg) +Convolutions involving (c) + +![](images/f3e1205ae4b631ab25b18d5297cbc480dd1ff5d33e0f8ade664d2ce85347cfe8.jpg) + +![](images/80193f24e2b38ecaf141f9d7a2e076285ced3991ba0e63c46aed2978b3d183e8.jpg) +Convolutions involving (e) + +![](images/91bfaefdb8c2ebf071ebe5893fed27de19e40373f9ce1eba526fc7fbf1186d9c.jpg) +Convolutions involving (f) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +# Mirror Padding (REFLECT) + +Illustrated on 3x3 kernel and 1-pixel padding + +# of conv ops each pixel is involved in + +Original Input + +
abC
def
gh
• ·
+ +
edef. • . •
babC= =
edef
hgh
= =
+ +
48666
816121212
612999
612999
612999
+ +# Which kernel cells these ops utilize? + +![](images/d37f8d090f0d4bb551a64bb18b98a508db3428a6185e94f93a10792f876b3330.jpg) + +# Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/a8960c789ef1b038c5adb0e2e1fc943e5a08c712af011e46e3f21af3294ed5ba.jpg) + +Convolutions involving (b) + +![](images/4d298c0343a75ab39732efadb3565e384d2c110e83cefde748e91d4bfb8226d9.jpg) + +Convolutions involving (c) + +![](images/9a44ed4932a742181366a04189d4ecc324022b23ef90542a89c1c7c432db1ba0.jpg) + +Convolutions involving (e) + +![](images/1ab66efe5cac10d4fc9f6ae0a5a8eff9871f5a5a1e76816a55c37644030707da.jpg) + +Convolutions involving (f) + +![](images/9c8456965691c401ca270510270cca64da0ffbfd29e905a6210b07a3a22f0387.jpg) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment + +# Partial Convolution Illustrated on a 3x3 kernel + +# Input + +# Weighted # of conv ops each pixel is involved in + +
abC
def
gh
+ +
6.258.757.57.57.5
8.75 12.2510.5 10.510.5
7.510.5999
7.510.5999
7.510.5999
+ +# Which kernel cells these ops utilize? + +![](images/e44405b931feb473f680ba423538de1d9620da5536ae1e8dac14e2710ecdbfba.jpg) +Detailed Illustration of how the counts are derived + +Convolutions involving (a) + +![](images/b7d84fb1a287aa41e70703547a09d1f42b59c2cdbaeced40b094c93ee2bd4ec3.jpg) +Convolutions involving (b) + +![](images/d9c752239c01303d3667d6eb1e568e38601275535c9e3d462c560285f3ba00ad.jpg) +Convolutions involving (c) + +![](images/075465da09bf54db467ff6536761903b1e0c7278cd14b3304dfab0d3c9fd8eaa.jpg) + +![](images/c5b335e14940e3ec3c3e7a0b95ef1060da39442c5d1eba43d8d09eeed1a09da0.jpg) +Convolutions involving (e) + +Convolutions involving (g): Rotated version of (c) + +Convolutions involving (h): Rotated version of (f) + +Convolutions involving (i): Regular uniform treatment \ No newline at end of file diff --git a/parse/train/m1CD7tPubNy/m1CD7tPubNy_content_list.json b/parse/train/m1CD7tPubNy/m1CD7tPubNy_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..99c3f50b1a33b4019a04043b5f7e6213c55ce441 --- /dev/null +++ b/parse/train/m1CD7tPubNy/m1CD7tPubNy_content_list.json @@ -0,0 +1,3507 @@ +[ + { + "type": "text", + "text": "MIND THE PAD – CNNS CAN DEVELOP BLIND SPOTS ", + "text_level": 1, + "bbox": [ + 173, + 98, + 821, + 121 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Bilal Alsallakh Facebook AI ", + "bbox": [ + 183, + 145, + 289, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Narine Kokhlikyan Facebook AI ", + "bbox": [ + 361, + 145, + 498, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Vivek Miglani Facebook AI ", + "bbox": [ + 570, + 145, + 671, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Jun Yuan NYU ", + "bbox": [ + 743, + 145, + 813, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Orion Reblitz-Richardson Facebook AI ", + "bbox": [ + 183, + 194, + 367, + 222 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 260, + 544, + 275 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We show how feature maps in convolutional networks are susceptible to spatial bias. Due to a combination of architectural choices, the activation at certain locations is systematically elevated or weakened. The major source of this bias is the padding mechanism. Depending on several aspects of convolution arithmetic, this mechanism can apply the padding unevenly, leading to asymmetries in the learned weights. We demonstrate how such bias can be detrimental to certain tasks such as small object detection: the activation is suppressed if the stimulus lies in the impacted area, leading to blind spots and misdetection. We propose solutions to mitigate spatial bias and demonstrate how they can improve model accuracy. ", + "bbox": [ + 232, + 289, + 766, + 415 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 MOTIVATION ", + "text_level": 1, + "bbox": [ + 176, + 439, + 313, + 455 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Convolutional neural networks (CNNs) serve as feature extractors for a wide variety of machinelearning tasks. Little attention has been paid to the spatial distribution of activation in the feature maps a CNN computes. Our interest in analyzing this distribution is triggered by mysterious failure cases of a traffic light detector: The detector successfully detects a small but visible traffic light in a road scene. However, it fails completely in detecting the same traffic light in the next frame captured by the ego-vehicle. The major difference between both frames is a limited shift along the vertical dimension as the vehicle moves forward. Therefore, the drastic difference in object detection is surprising given that CNNs are often assumed to have a high degree of translation invariance [8; 17]. ", + "bbox": [ + 174, + 469, + 825, + 582 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The spatial distribution of activation in feature maps varies with the input. Nevertheless, by closely examining this distribution for a large number of samples, we found consistent patterns among them, often in the form of artifacts that do not resemble any input features. This work aims to analyze the root cause of such artifacts and their impact on CNNs. We show that these artifacts are responsible for the mysterious failure cases mentioned earlier, as they can induce ‘blind spots’ for the object detection head. Our contributions are: ", + "bbox": [ + 174, + 588, + 825, + 672 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "• Demonstrating how the padding mechanism can induce spatial bias in CNNs (Section 2). \n• Demonstrating how spatial bias can impair downstream tasks (Section 3). \n• Identifying uneven application of 0-padding as a resolvable source of bias (Section 5). \nRelating the padding mechanism with the foveation behavior of CNNs (Section 6). \n• Providing recommendations to mitigate spatial bias and demonstrating how this can prevent blind spots and boost model accuracy. ", + "bbox": [ + 215, + 683, + 823, + 776 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "2 THE EMERGENCE OF SPATIAL BIAS IN CNNS ", + "text_level": 1, + "bbox": [ + 174, + 795, + 584, + 811 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Our aim is to determine to which extent activation magnitude in CNN feature maps is influenced by location. We demonstrate our analysis on a publicly-available traffic-light detection model $\\pmb { \\mathbb { B } } \\pmb { \\ 6 } \\|$ . This model implements the SSD architecture $\\left[ \\left[ 2 6 \\right] \\right]$ in TensorFlow $\\mathbb { I I }$ , using MobileNet-v1 $\\mathbb { \\lVert \\lambda \\rVert }$ as a feature extractor. The model is trained on the BSTLD dataset $\\pmb { \\Vert 4 \\Vert }$ which annotates traffic lights in road scenes. Figure $\\bigstar$ shows two example scenes from the dataset. For each scene, we show two feature maps computed by two filters in the $1 1 ^ { \\mathrm { t h } }$ convolutional layer. This layer contains 512 filters whose feature maps are used directly by the first box predictor in the SSD to detect small objects. ", + "bbox": [ + 174, + 825, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/b526bfe207aac3286b6c1d5a6a8d0a326615dced8dd84ad446f05157528bdb45.jpg", + "image_caption": [ + "Figure 1: Averaging feature maps per input (column marginal) and per filter (row marginal) in the last convolutional layer of a traffic light detector. Color indicates activation strength (the brighter, the higher), revealing line artifacts in the maps. These artifacts are the manifestation of spatial bias. " + ], + "image_footnote": [], + "bbox": [ + 194, + 99, + 799, + 342 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The bottom row in Figure $\\bigstar$ shows the average response of each of the two aforementioned filters, computed over the test set in BSTLD. The first filter seems to respond mainly to features in the top half of the input, while the second filter responds mainly to street areas. There are visible lines in the two average maps that do not seem to resemble any scene features and are consistently present in the individual feature maps. We analyzed the prevalence of these line artifacts in the feature maps of all 512 filters. The right column in Figure $\\bigstar$ shows the average of these maps per scene, as well as over the entire test set (see supplemental for all 512 maps). The artifacts are largely visible in the average maps, with variations per scene depending on which individual maps are dominant. ", + "bbox": [ + 173, + 429, + 825, + 541 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "A useful way to make the artifacts stand out is to neutralize scene features by computing the feature maps for a zero-valued input. Figure $2$ depicts the resulting average map for each convolutional layer after applying ReLU units. The first average map is constant as we expect with a 0-valued input. The second map is also constant except for a 1-pixel boundary where the value is lower at the left border and higher at the other three borders. We magnify the corners to make these deviations visible. The border deviations increase in thickness and in variance at subsequent layers, creating multiple line artifacts at each border. These artifacts become quite pronounced at ReLU 8 where they start to propagate inwards, resembling the ones in Figure 1. ", + "bbox": [ + 173, + 547, + 826, + 660 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/9ad02d0bb96ca163c5aaed3822f6466e3fe37e0cde18654db995d27f877d7803.jpg", + "image_caption": [ + "Figure 2: Activation maps for a 0 input, averaged over each layer’s filters (title format: $\\mathrm { H } { \\times } \\mathrm { W } { \\times } \\mathrm { C } )$ " + ], + "image_footnote": [], + "bbox": [ + 192, + 680, + 810, + 888 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "It is evident that the 1-pixel border variations in the second map are caused by the padding mechanism in use. This mechanism pads the output of the previous layer with a 1-pixel 0-valued border in order to maintain the size of the feature map after applying $3 { \\tt X } 3$ convolutional. The maps in the first layer are not impacted because the input we feed is zero valued. Subsequent layers, however, are increasingly impacted by the padding, as preceding bias terms do not warrant 0-valued input. ", + "bbox": [ + 173, + 103, + 825, + 174 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "It is noticeable in Figure 2 that the artifacts caused by the padding differ across the four borders. To investigate this asymmetry, we analyze the convolutional kernels (often called filters) that produce the feature maps. Figure 3 depicts a per-layer mean of these $3 { \\tt X } 3$ kernels. These mean kernels exhibit different degrees of asymmetry in the spatial distribution of their weights. For example, the kernels in L1 assign (on average) a negative weight at the left border, and a positive weight at the bottom. This directly impacts the padding-induced variation at each border. Such asymmetries are related to uneven application of padding as we explain in Section 5. ", + "bbox": [ + 173, + 180, + 825, + 280 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/fccaabdbce7408640032adb86b3858809bd5967c675d8bc0ad51217199678338.jpg", + "image_caption": [ + "Figure 3: Mean kernel per convolutional layer. All kernels are $3 \\times 3$ , the titles show their counts. " + ], + "image_footnote": [], + "bbox": [ + 184, + 295, + 816, + 340 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 IMPLICATIONS OF SPATIAL BIAS ", + "text_level": 1, + "bbox": [ + 176, + 410, + 475, + 426 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We demonstrate how feature-map artifacts can cause blind spots for the SSD model. Similar issues arise in several small-object detectors, e.g., for faces and masks, as well as in pixel-oriented tasks such as semantic segmentation and image inpainting (see supplemental for examples). ", + "bbox": [ + 174, + 443, + 825, + 484 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Figure $\\sharp$ illustrates how the SSD predicts small objects based on the feature maps of the 11-th convolutional layer. The SSD uses the pixel positions in these maps as anchors of object proposals. Each proposal is scored by the SSD to represent a target category, with ”background“ being an implicit category that is crucial to exclude irrelevant parts of the input. In addition to these scores, the SSD computes a bounding box to localize the predicted object at each anchor. We examine object proposals computed at 1:2 aspect ratio, as they resemble the shape of most traffic lights in the dataset. We visualize the resulting score maps both for the background category and for traffic lights, when feeding a 0-valued input to the SSD. We also visualize the bounding boxes of these proposals in the image space. The SSD predicts the image content to be of background category at all anchor locations, as evident from the value range in both score maps. Such predictions are expected with an input that contains no traffic lights. However, the line artifacts in the feature maps have a strong impact on the score maps. These artifacts elevate the likelihood of anchors closer to the top to be classified as background (see the yellow band in the background score map). Conversely, these anchors have significantly lower scores for the traffic light category, compared with other anchors in the feature map. Such difference in the impact on the target categories is due to the different weights the SSD assigns to the feature maps for each target. As a result, the artifacts lead to potential blind spots in which the scores for certain categories are artificially muted. ", + "bbox": [ + 173, + 491, + 825, + 561 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/2f90b37ac2e03a969222bf159083fe214d6629973648c95433a912a506b514b4.jpg", + "image_caption": [ + "Figure 4: The formation of blind spots in SSD, illustrated via its box predictor internals with a zero-valued input. The predictor uses spatial anchors to detect and localize the target object at $4 5 \\times 8 0$ possible locations based on 512 feature maps. Certain anchors are predisposed to predict background due to feature-map artifacts, as evident in the logit maps. Traffic lights at the corresponding location cannot be detected as demonstrated with a real scene (middle one in the bottom). " + ], + "image_footnote": [], + "bbox": [ + 179, + 578, + 820, + 837 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/855129a5fdeb22a83f45d33356680dd0ee5b663a0bc83617313ee7598eec1429.jpg", + "image_caption": [ + "Figure 5: (a) A map showing via color the detection score the SSD computes for a traffic light when present at various locations. The detection is muted when the stimulus lies in the area impacted by the artifacts. (b) The same map after changing the padding method to SYMMETRIC. The detection scores are rather constant except for periodic variations due to the SSD’s reliance on anchors. " + ], + "image_footnote": [], + "bbox": [ + 181, + 102, + 815, + 220 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 315, + 825, + 483 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To validate whether or not the blind spots hinder object detection, we examine road scenes that contain highly-visible traffic light instances in the impacted area. Figure 4-bottom shows an example of such a scene. The SSD computes a low detection score of $7 \\%$ when the traffic light lies in the blind spot (see middle image), far below the detection false-positive cutoff. Shifting the scene image upwards or downwards makes the instance detectable with a high score as long as it lies outside the blind spot. This explains the failure cases mentioned in Section $\\mathbf { \\overline { { \\mathbb { D } } } }$ To further validate this effect, we run the SSD on baseline images that each contains one traffic light instance at a specific location in the input. We store the detection score for each instance. Figure 5a depicts the computed scores in a 2D map. It is evident that the model fails to detect the traffic light instance exactly when it is located within the “blind spot” band. The artifacts further disrupt the localization of the objects as evident in the top-right plot in Figure 4 which shows per-anchor object proposals computed for a 0 input. ", + "bbox": [ + 174, + 489, + 825, + 642 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4 REMINDER: WHY IS PADDING NEEDED IN CNNS? ", + "text_level": 1, + "bbox": [ + 174, + 662, + 625, + 679 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Padding is applied at most convolutional layers in CNNs to serve two fundamental purposes: ", + "text_level": 1, + "bbox": [ + 173, + 694, + 790, + 709 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Maintaining feature map size A padding that satisfies this property is often described as SAME or HALF padding. FULL padding expands the maps by kernel size - 1 along each dimension. VALID padding performs no padding, eroding the maps by the same amount. SAME padding is important to (1) design deep networks that can handle arbitrary input size (a challenge in the presence of gradual erosion), (2) maintain the aspect ratio of non-square input, and (3) concatenate feature maps from different layers as in Inception [39] and ResNet $\\mathbf { \\bar { \\rho } }$ models. ", + "bbox": [ + 174, + 715, + 825, + 799 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Reducing information bias against the boundary Consider a $3 \\times 3$ kernel applied to a 2D input. An input location at least 2 pixels away from the boundary contributes to nine local convolution operations when computing the feature map. On the other hand, the corner is involved only one time under VALID padding, four times under a 1-pixel SAME 0-padding, and nine times under a 2-pixel FULL 0-padding. With SAME 0-padding, the cumulative contribution differences among the input pixels grow exponentially over the CNN layers. We refer to such uneven treatment of input pixels as the foveation behavior of the padding mechanism and elaborate on this in Section 6. ", + "bbox": [ + 174, + 805, + 825, + 902 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We next explore solutions to the issues that cause padding to induce spatial bias. ", + "bbox": [ + 174, + 909, + 699, + 924 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/64a4b6b4f56e3df3a1dac55598e3c785be3dbfaa6420ebc60f4670b52e23635c.jpg", + "image_caption": [ + "Figure 6: (a) Illustrating the problem of uneven padding when down-sampling at a stride of 2. The padding along x-axis is consumed only at the left side. (b) Mean $3 \\times 3$ filters in three ResNet models, trained on ImageNet with two input sizes. Color encodes average weight (green is positive). A size that induces uneven padding (top row) can lead to asymmetries, esp. around down-sampling layers. These asymmetries are mitigated when the input size induces no uneven padding (bottom row). " + ], + "image_footnote": [], + "bbox": [ + 176, + 98, + 825, + 295 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "5 ELIMINATING UNEVEN APPLICATION OF PADDING ", + "text_level": 1, + "bbox": [ + 174, + 400, + 627, + 417 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "While useful to reduce bias against the boundary, applying padding at down-sampling layers can lead to asymmetry in CNN internals. Figure $6 \\mathrm { a }$ illustrates the source of this asymmetry when strided convolution is used for downsampling: At one side of the feature map, the padding is consumed by the kernel while at the other side it is not. To warrant even application of padding throughout the CNN, the following must hold at all $d$ down-sampling layers, where $( h _ { i } , w _ { i } )$ is the output shape at the i-th layer with $\\overline { { k } } _ { i } ^ { h } \\times k _ { i } ^ { w }$ as kernel size, $( s _ { i } ^ { h } , s _ { i } ^ { w } )$ as strides, and $\\mathbf { \\Sigma } = ( p _ { i } ^ { h } , p _ { i } ^ { w } )$ as padding amount (refer to appendix $\\boxed { \\mathrm { A } }$ for a proof): ", + "bbox": [ + 173, + 426, + 825, + 526 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/440b1a1d2d31299d8b44bc2e556555e7b2c2d7ed4e01ad4a9fc301cffdc84cf8.jpg", + "text": "$$\n\\forall i \\in \\{ 1 , \\ldots , d \\} : h _ { i - 1 } = s _ { i } ^ { h } \\cdot ( h _ { i } - 1 ) + k _ { i } ^ { h } - 2 \\cdot p _ { i } ^ { h } \\quad \\wedge \\quad w _ { i - 1 } = s _ { i } ^ { w } \\cdot ( w _ { i } - 1 ) + k _ { i } ^ { w } - 2 \\cdot p _ { i } ^ { w } \\quad .\n$$", + "text_format": "latex", + "bbox": [ + 179, + 532, + 799, + 553 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The values $h _ { 0 }$ and $w _ { 0 }$ represent the CNN input dimensions. The above constraints are not always satisfied during training or inference with arbitrary input dimensions. For example, ImageNet classifiers based on ResNet $\\pmb { \\mathbb { I } } \\pmb { \\mathcal { 2 } } \\Vert$ and MobileNet $\\pmb { \\mathbb { I } } \\pmb { \\overbrace { 3 } } \\|$ contain five down-sampling layers $\\mathit { a } = 5$ ) that apply 1-pixel 0-padding before performing 2-strided convolution. To avoid uneven application of padding, the input to these CNNs must satisfy the following, as explained in appendix A: ", + "bbox": [ + 174, + 559, + 825, + 630 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/88b26002b6ffe50d6fd4677e7bc4524ce5aa8f38e3573e166b6a758a9313ea20.jpg", + "text": "$$\nh _ { 0 } = a _ { 1 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 1 } + 1 \\quad \\mathrm { a n d } \\quad w _ { 0 } = a _ { 2 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 2 } + 1 \\quad \\mathrm { w h e r e } \\quad a _ { 1 } , a _ { 2 } \\in \\mathbb { N } ^ { + }\n$$", + "text_format": "latex", + "bbox": [ + 179, + 636, + 800, + 656 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The traditional $\\bigstar$ and prevalent input size for training ImageNet models is $2 2 4 \\times 2 2 4$ . This size violates Eq. 2, leading to uneven padding at every down-sampling layer in ResNet and MobileNet models where 0-padding is effectively applied only at the left and top sides of layer input. This over-represents zeros at the top and left sides of $3 \\times 3$ feature-map patches the filters are convolved with during training. The top row of Figure $6 { \\mathsf { b } }$ shows per-layer mean filters in three ResNet models in PyTorch $\\mathbb { \\lVert 3 3 \\rVert }$ , pre-trained on ImageNet with $2 2 4 \\times 2 2 4$ images. In all of these models, a few of the mean filters, adjacent to down-sampling layers, exhibit stark asymmetry about their centers. ", + "bbox": [ + 173, + 670, + 825, + 770 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We increase the image size to $2 2 5 \\times 2 2 5$ without introducing additional image information2. This size satisfies Eq. 2, warranting even application of padding at every downsampling layer in the above models. Retraining the models with this size strongly reduces this asymmetry as evident in the bottom row of Figure $\\bar { 6 } 6$ . This, in turn, visibly boosts the accuracy in all models we experimented with as we report in Table 1. The accuracy did not improve further when we retrained two of the models, ResNet-18 and ResNet-34, on $2 2 6 \\times 2 2 6$ images. This provides evidence that the boost is due to eliminating uneven padding and not merely due to increasing the input size. ", + "bbox": [ + 173, + 772, + 825, + 872 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Replacing 0-padding with a padding method that reuses feature map values can alleviate the asymmetry in the learned filters in the presence of unevenly applied padding. Another possibility is to use a rigid downsampling kernel, such as max-pooling, instead of a learned one. Appendix $\\boxed { \\dot { \\mathbf { C } } }$ demonstrates both possibilities. Finally, antialiasing before downsampling $\\mathbb { \\lVert \\boldsymbol { 4 3 } \\rVert }$ can strongly reduce the asymmetry as we elaborate in Section 8 and in Appendix E. ", + "bbox": [ + 174, + 103, + 825, + 175 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/1139c80765c9433cc09b53b2ba194f3b7aee4b315e6a05560053d3f686007e5b.jpg", + "table_caption": [ + "Table 1: Top-1 (and top-5) accuracy of five ImageNet classifiers trained with different input sizes. " + ], + "table_footnote": [], + "table_body": "
Input Size ²MobileNetResNet-18ResNet-34ResNet-50ResNet-101
224×22468.19 (88.44)69.93 (89.22)73.30 (91.42)75.65 (92.47)77.37 (93.56)
225×22568.80 (88.78)70.27 (89.52)73.72 (91.58)76.01 (92.90)77.67 (93.81)
", + "bbox": [ + 184, + 210, + 812, + 270 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Even when no padding is applied $( p _ { i } ^ { h } = 0$ or $p _ { i } ^ { w } = 0 ,$ ), an input size that does no satisfy Eq. 1 can lead to uneven erosion of feature maps, in turn, reducing the contribution of pixels from the impacted sides $( { \\mathrm { F i g ~ } } 7 { \\mathrm { \\rho } }$ . Satisfying $\\mathrm { E q } \\ 1$ imposes a restriction on input size, e.g., to values in increments of $2 ^ { d } = 3 2$ with the above models ${ \\mathrm { 1 9 3 \\times 1 9 3 } }$ , $2 2 5 \\times 2 2 5$ , $2 5 7 \\times 2 5 7$ , ...). Depending on the application domain, this can be guaranteed either by resizing an input to the closest increment, or by padding it accordingly with suited values. ", + "bbox": [ + 173, + 282, + 825, + 367 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "6 PADDING MECHANISM AND FOVEATION ", + "text_level": 1, + "bbox": [ + 174, + 386, + 540, + 402 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "By foveation we mean the unequal involvement of input pixels in convolutional operations throughout the CNN. Padding plays a fundamental role in the foveation behavior of CNNs. We visualize this behavior by means of a foveation map that counts for each input pixel the number of convolutional paths through which it can propagate information to the CNN output. We obtain these counts by computing the effective receptive field $\\pmb { \\pmb { 2 8 } }$ for the sum of the final convolutional layer after assigning all weights in the network to 1 (code in supplemental). Neutralizing the weights is essential to obtain per-pixel counts of input-output paths that reflect the foveation behavior. ", + "bbox": [ + 173, + 417, + 825, + 515 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/d463a723cb881cdd5e445297eabd8a8780cc37159812695efe551818975b2e04.jpg", + "image_caption": [ + "Figure 7: Foveation behavior of different padding methods applied to VGG-19 [37], and illustrated in a $5 1 2 \\times 5 1 2$ input space (unless otherwise stated). Color represents the number of paths to the output for each input pixel. (a) The difference between VALID, FULL, and SAME 0-padding. (b) SAME alternatives to 0-padding. (c) Dilation amplifies foveation of SAME 0-padding. (d) Strides can lead to checkerboard patterns. (e) Foveation effects are more extensive in smaller inputs (relative to input size) and are sensitive to uneven padding. " + ], + "image_footnote": [], + "bbox": [ + 174, + 525, + 821, + 695 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Figure $7 \\mathrm { a }$ shows the extensive foveation effect when no padding is applied. The diminishing contribution of vast areas of the input explains the drastic drop in accuracy recently observed under VALID padding $\\mathbb { \\lVert \\boldsymbol { 1 6 } \\rVert }$ . In contrast, FULL 0-padding does not incur foveation, however, at the cost of increasing the output size after each layer, making it impractical as explained in Section $\\mathbb { H }$ SAME 0-padding incurs moderate foveation at the periphery, whose absolute extent depends on the number of convolutional layers and their filter sizes. Its relative extent depends on the input size: the larger the input, the larger the ratio of the constant area in yellow (refer to appendix B for a detailed example). ", + "bbox": [ + 173, + 810, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Figure 7b shows the foveation behavior of alternatives to SAME 0-padding that have roots in wavelet analysis [19] and image processing $ { \\mathbb { E } } { \\ b { \\mathbb { Z } } } ] $ . Mirror padding mirrors pixels at the boundary to fill the padding area. When the border is included (SYMMETRIC mode in TensorFlow) all input pixels have an equal number of input-output paths 3, resulting in a uniform foveation map. When the border is not included (REFLECT mode both in PyTorch and in TensorFlow), the map exhibits bias against the border and towards a contour in its proximity. This bias is amplified over multiple layers. Replication padding exhibits the opposite bias when the padding area is wider than 1 pixel. This is because it replicates the outer 1-pixel border multiple times to fill this area 3. The method is equivalent to SYMMETRIC if the padding area is 1-pixel wide. Circular padding wraps opposing borders, enabling the kernels to seamlessly operate on the boundary and resulting in a uniform map. Partial Convolution $\\lVert 2 2 \\rVert$ has been proposed as a padding method that treats pixels outside the original image as missing values and rescales the computed convolutions accordingly $\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }$ . Its foveation behavior resembles reflective padding 3. Distribution padding $\\pmb { \\mathbb { B } } 0 \\|$ resizes the input to fill the padding area around the original feature map, aiming at preserving the distribution of the map. Its foveation map is largely uniform, except for the corners and edges. ", + "bbox": [ + 171, + 103, + 825, + 313 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Impact of input size Besides influencing the relative extent of foveation effects, the input size also determines the presence of uneven padding (or uneven feature-map erosion), as we discussed in Section 5. Figure $\\textcircled { 7 } \\textcircled { \\times }$ shows the foveation map for VGG-19 with a $1 2 7 \\times 1 2 7$ input. This input violates Eq. 1 at every downsampling layer (appendix $\\mathbf { A } )$ , leading to successive feature map erosion at the bottom and right sides which is reflected in the foveation map (see appendix B for a detailed example). The bottom-right part of the input is hence less involved in the CNN computations. ", + "bbox": [ + 174, + 321, + 825, + 405 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Impact of dilation We assign a dilation factor of 2 to all VGG-19 convolutional layers. While this exponentially increases the receptive field of the neurons at deeper layers $\\pm 2 \\|$ , dilation doubles the extent of the non-uniform peripheral areas that emerge with SAME 0-padding as evident in Figure $\\textcircled { 7 } \\textcircled { < }$ SYMMETRIC and circular padding maintain uniform foveation maps regardless of dilation 3. In contrast, dilation increases the complexity of these maps for REFLECT and replication padding. ", + "bbox": [ + 174, + 414, + 825, + 484 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Impact of strides Whether learned on based on pooling, downsampling layers can amplify the impact of succeeding convolutional layers on foveation behaviour. Furthermore, these layers can cause input pixels to vary in the count of their input-output paths. This can happen when the kernel size is not divisible by the stride, leading to a checkerboard pattern in the foveation maps. This manifests in ResNet models as we illustrate in appendix B. In VGG-19, all max-pooling layers use a stride of 2 and kernel size of 2. Changing the kernel size to 3 leads to a checkerboard pattern as evident in Figure 7d. Such effects were shown to impact pixel-oriented tasks $\\mathbb { B } 2 \\mathbb { I }$ . ", + "bbox": [ + 174, + 493, + 825, + 592 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "The padding technique and its foveation behaviour have direct impact on feature-map artifacts (Section $\\bar { 7 } )$ , and on the ability of CNNs to encode spatial information (Section $^ { 8 ) }$ . Understanding the foveation behavior is key to determine how suited a padding method is for a given task. For example, small object detection is known to be challenging close to the boundary $\\left[ \\left[ 2 6 \\right] \\right]$ , in part due to the foveation behavior of SAME 0-padding. In Figure $\\bar { 5 } 6$ , we change the padding method in the SSD to SYMMETRIC. The stimulus is noticeably more detectable at the boundary, compared with 0-padding $^ 4 \\cdot$ In contrast, ImageNet classification is less sensitive to foveation effects because the target objects are mostly located away from the periphery. Nevertheless, the padding method was shown to impact classification accuracy $\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }$ because it still affects feature map artifacts. ", + "bbox": [ + 173, + 597, + 825, + 723 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "7 PADDING METHODS AND FEATURE MAP ARTIFACTS ", + "text_level": 1, + "bbox": [ + 176, + 743, + 642, + 760 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "It is also noticeable that the score map in Figure $5 \\mathsf { b }$ is more uniform than in Figure $\\textcircled { 5 } \\textcircled { \\times }$ . In particular, under SYMMETRIC padding the model is able to detect traffic lights placed in the blind spots of the original 0-padded model. To verify whether the line artifacts in Figure $\\bigstar$ are mitigated, we inspect the mean feature maps of the adapted model. With a constant input, SYMMETRIC padding warrants constant maps throughout the CNN because it reuses the border to fill the padding area. Instead, we average these maps over 30 samples generated uniformly at random. Figure 8 depicts the mean maps which are largely uniform, unlike the case with 0-padding. ", + "bbox": [ + 173, + 775, + 825, + 873 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/57e698bb07346f6a27d58d62665ad1ecd8d2ee482ba665e630bbd20159d302cc.jpg", + "image_caption": [ + "Figure 8: The same feature maps in Figure 2, generated under mirror padding and averaged over 30 randomly-generated input samples. The line artifacts induced by 0-padding are largely mitigated. " + ], + "image_footnote": [], + "bbox": [ + 176, + 99, + 823, + 199 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "To further analyze the impact of SYMMETRIC padding, we retrain the adapted model following the original training protocol. This significantly improves the average precision (AP) as reported in Table $2$ under different overlap thresholds (matching IoU), confirming that small object detection is particularly sensitive to feature-map artifacts. ", + "bbox": [ + 174, + 268, + 825, + 325 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/1de57b9720d30158817e2f1a1ba78dd730499b7691b727f819178fba8c13d39e.jpg", + "table_caption": [ + "Table 2: Performance of the SSD traffic light detector, trained under two different padding schemes. " + ], + "table_footnote": [], + "table_body": "
Average Precision (AP)AP@.20I0UAP@ .50I0UAP@.75I0UAP@.90I0U
Zero Padding80.24%49.58%3.7%0.007%
Mirror Padding83.20%57%8.44%0.02%
", + "bbox": [ + 210, + 368, + 789, + 422 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Of the padding methods listed in Section $6 ,$ mirror padding in both SYMMETRIC and REFLECT modes, PartialConv, and circular padding are generally effective at reducing feature map artifacts that emerge under zero padding, in particular salient line patterns. In contrast, distribution padding can induce significant artifacts. Refer to appendix D for comparative examples of artifacts under the aforementioned padding schemes. ", + "bbox": [ + 174, + 439, + 825, + 508 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Artifact magnitude and propagation While feature-map artifacts are induced by the padding mechanism at the boundary, their magnitude and inward propagation in the maps are impacted by several architectural aspects of CNNs. In particular, certain normalization schemes such as batchnorm [15] tend to limit the range of variation within a feature map and to relatively harmonize this range across different maps. This, in turn, impacts how possible artifacts in these maps accumulate when they are processed by the next convolutional layer. Similarly, artifacts that manifest after applying ReLU units are of a positive sign. These factors were instrumental in the formation of potential blind spots described in Section 3. We hence recommend to involve non-convolutional layers when inspecting the feature maps. Besides having possible impact on artifact magnitude, several aspects of convolution arithmetic, such as filter size and dilation factors, can also impact the spatial propagation of these artifacts. ", + "bbox": [ + 173, + 526, + 825, + 679 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "8 RELATED FINDINGS AND TAKEAWAYS ", + "text_level": 1, + "bbox": [ + 174, + 702, + 522, + 718 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Handling the boundary is an inherent challenge when dealing with spatial data [9]. Mean padding is known to cause visual artifacts in traditional image processing, with alternative methods proposed to mitigate them $\\mathbb { \\lVert 2 4 \\rVert }$ . CNNs have been often assumed to deal with such effects implicitly. Innamorati et al $\\dot { [ \\lVert { 4 } \\rVert }$ propose learning separate sets of filters dedicated to the boundaries to avoid impacting the weights learned by regular filters. A grouped padding strategy, proposed to support $2 \\times 2$ filters $\\mathbf { \\bar { \\textmu } }$ , offers avenues to mitigate uneven padding and corresponding skewness in foveation maps without restrictions on input size (see our note in appendix $\\mathbf { B }$ for explanation). Finally, insights from signal and image processing [10; 11] could inspire further CNN padding schemes. ", + "bbox": [ + 173, + 734, + 825, + 848 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Zero padding has been recently linked to CNNs’ ability to encode position information [7; 16; 18; 29]. In contrast, circular padding was shown to limit this ability $\\mathbb { \\left[ \\bigcirc \\right] }$ and to boost shift invariance $\\begin{array} { r l } { { \\bigl \\| \\overline { { 3 5 } } \\bigr \\| } } & { { } } \\end{array}$ . The input sizes in those studies do induce uneven padding. This can be, in part, the underlying mechanism behind the aforementioned ability. Whether or not this ability is desirable depends on the task, with several methods proposed to explicitly encode spatial information [5; 6; 20; 25; 29; 31]. ", + "bbox": [ + 174, + 853, + 823, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Downsampling using max-pooling or strided convolution has been shown to impact shift invariance in CNNs by incurring aliasing effects [3; 38; 43]. These effects can manifest in the same symptoms we reported in Section $^ { 1 , }$ albeit for a different reason. Zhang [43] demonstrated how blurring the feature maps before subsampling mitigates aliasing effects and improves ImageNet classification accuracy of various popular CNNs. We analyzed the mean filters in antialiased MobileNet and ResNet models pre-trained on ImageNet under 0-padding, with $2 2 4 \\times 2 2 4$ as input size (refer to Appendix $\\mathrm { E } )$ . We found that antialiasing can also mitigate the asymmetry of mean filters that exhibited high asymmetry in the baseline models, especially at deeper layers. This is remarkable given that these models are trained on $2 2 4 \\times 2 2 4$ images, which incurs one-sided zero padding at every downsampling layer. This could, in part, be attributed to the ability of the BlurPool operator used in antialiased CNN to smoothen the acuity of zero-padded borders, in turn, reducing the value imbalance incurred by one-sided padding. Further analysis is needed to examine the interaction between padding and aliasing effects in CNNs and to establish possible synergy between antialiasing and eliminating uneven application of padding. ", + "bbox": [ + 174, + 103, + 825, + 297 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Luo et al $\\pmb { \\Vert 2 8 \\Vert }$ drew connections between effective receptive fields and foveated vision. Our analysis links foveation behavior with the padding scheme and suggests that it might occur implicitly in CNNs when using VALID or SAME 0-padding, without the need for explicit mechanisms [2; 21]. Furthermore, it explains the drastic accuracy drop noted by $\\boxed { 1 0 }$ under VALID padding, which is amplified by feature map erosion. ", + "bbox": [ + 174, + 305, + 825, + 375 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Choosing a padding method SAME 0-padding is by far the most widely-used method. Compared with other methods, it can enable as much as $5 0 \\%$ faster training and inference. Problem-specific constraints can dictate different choices [34; 35; 40]. In the lack of a universally superior padding method, we recommend considering multiple ones while paying attention to the nature of the data and the task, as well as to the following aspects: ", + "bbox": [ + 176, + 390, + 825, + 459 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "• Feature-map statistics: 0-padding can alter the value distribution within the feature maps and can shift their mean value in the presence of ReLU units. The alternatives presented in Section 6 tend to preserve this distribution, thanks to reusing existing values in the maps. Foveation behavior: 0-padding might not be suited for tasks that require high precision at the periphery, unlike circular and SYMMETRIC mirror padding. Interference with image semantics (esp. with a padding amount $> 1$ pixel): For example, circular padding could introduce border discontinuities unless the input is panoramic $| \\widehat { \\mathsf { B } } \\widehat { \\mathsf { S } } \\|$ . • Potential to induce feature map artifacts: All alternatives to 0-padding induce relatively fewer artifacts, except for Distribution padding $\\textcircled { \\lVert { 3 0 } \\rVert }$ (see appendix ${ \\bf D } )$ ", + "bbox": [ + 215, + 470, + 825, + 611 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We also recommend eliminating uneven padding at downsampling layers both at training and at inference time, as we illustrated in Section $\\boxed { 5 }$ This is especially important when zero padding is applied and the downsampling is learned. The scripts used to generate the visualizations in this paper are available in the supplemental as well as at http://mind-the-pad.github.io. ", + "bbox": [ + 174, + 621, + 825, + 678 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Summary We demonstrated how the padding mechanism can induce spatial bias in CNNs, in the form of skewed kernels and feature-map artifacts. These artifacts can be highly pronounced with the widely-used 0-padding when applied unevenly at the four sides of the feature maps. We demonstrated how such uneven padding can inherently take place in state-of-the-art CNNs, and how the artifacts it causes can be detrimental to certain tasks such as small object detection. We provided visualization methods to expose these artifacts and to analyze the implication of various padding schemes on boundary pixels. We further proposed solutions to eliminate uneven padding and to mitigate spatial bias in CNNs. Further work is needed to closely examine the implications of spatial bias and foveation in various applications (see supplementary for examples), as well as padding impact on recurrent models and 1-D CNNs. ", + "bbox": [ + 174, + 691, + 825, + 829 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "ACKNOWLEDGEMENT ", + "text_level": 1, + "bbox": [ + 176, + 852, + 356, + 866 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We are thankful to Ross Girshick for providing useful recommendations and experiment ideas, and to Shubham Muttepawar for implementing an interactive tool out of our analysis scripts, guided by our front-end specialist Edward Wang and our AI user-experience designer Sara Zhang. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "REFERENCES \n[1] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, et al. TensorFlow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016. \n[2] E. Akbas and M. P. Eckstein. Object detection through search with a foveated visual system. PLoS computational biology, 13(10):e1005743, 2017. \n[3] A. Azulay and Y. Weiss. Why do deep convolutional networks generalize so poorly to small image transformations? Journal of Machine Learning Research (JMLR), 20(184):1–25, 2019. \n[4] K. Behrendt, L. Novak, and R. Botros. A deep learning approach to traffic lights: Detection, tracking, and classification. In Robotics and Automation (ICRA), 2017 IEEE International Conference on, pp. 1370–1377. IEEE, 2017. \n[5] C.-A. Brust, S. Sickert, M. Simon, E. Rodner, and J. Denzler. Convolutional patch networks with spatial prior for road detection and urban scene understanding. In International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISAPP), 2015. \n[6] G. F. Elsayed, P. Ramachandran, J. Shlens, and S. Kornblith. Revisiting spatial invariance with low-rank local connectivity. In International Conference on Machine Learning (ICML), 2020. \n[7] J. Geiping, H. Bauermeister, H. Droge, and M. Moeller. Inverting gradients–how easy is it to ¨ break privacy in federated learning? arXiv preprint arXiv:2003.14053, 2020. \n[8] R. Gens and P. M. Domingos. Deep symmetry networks. In Advances in neural information processing systems (NeurIPS), pp. 2537–2545, 2014. \n[9] D. Griffith and C. Amrhein. An evaluation of correction techniques for boundary effects in spatial statistical analysis: traditional methods. Geographical Analysis, 15(4):352–360, 1983. \n[10] V. Gupta and N. Ramani. A note on convolution and padding for two-dimensional data. Geophysical Prospecting, 26(1):214–217, 1978. \n[11] L. Hamey. A functional approach to border handling in image processing. In International Conference on Digital Image Computing: Techniques and Applications, pp. 1–8, 2015. \n[12] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016. \n[13] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017. \n[14] C. Innamorati, T. Ritschel, T. Weyrich, and N. J. Mitra. Learning on the edge: Investigating boundary filters in CNNs. International Journal of Computer Vision (IJCV), pp. 1–10, 2019. \n[15] S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning (ICML), pp. 448– 456, 2015. \n[16] M. A. Islam, S. Jia, and N. D. Bruce. How much position information do convolutional neural networks encode? In International Conference on Learning Representations (ICLR), 2020. \n[17] M. Jaderberg, K. Simonyan, A. Zisserman, et al. Spatial transformer networks. In Advances in neural information processing systems (NeurIPS), pp. 2017–2025, 2015. \n[18] O. S. Kayhan and J. C. van Gemert. On translation invariance in CNNs: Convolutional layers can exploit absolute spatial location. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), 2020. \n[19] T. L. Kijewski-Correa. Full-scale measurements and system identification: A time-frequency perspective. PhD thesis, University of Notre Dame., 2003. \n[20] I. Kim, W. Baek, and S. Kim. Spatially attentive output layer for image classification. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), 2020. \n[21] H. Larochelle and G. E. Hinton. Learning to combine foveal glimpses with a third-order boltzmann machine. In Advances in neural information processing systems (NeurIPS), pp. 1243–1251, 2010. \n[22] G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro. Image inpainting for irregular holes using partial convolutions. In European Conference on Computer Vision, 2018. \n[23] G. Liu, K. J. Shih, T.-C. Wang, F. A. Reda, K. Sapra, Z. Yu, A. Tao, and B. Catanzaro. Partial convolution based padding. In arXiv preprint arXiv:1811.11718, 2018. \n[24] R. Liu and J. Jia. Reducing boundary artifacts in image deconvolution. In IEEE International Conference on Image Processing (ICIP), pp. 505–508, 2008. \n[25] R. Liu, J. Lehman, P. Molino, F. P. Such, E. Frank, A. Sergeev, and J. Yosinski. An intriguing failing of convolutional neural networks and the CoordConv solution. In Advances in Neural Information Processing Systems (NeurIPS), pp. 9605–9616, 2018. \n[26] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. SSD: Single shot multibox detector. In European Conference on Computer Vision, pp. 21–37, 2016. \n[27] S. Lou, X. Jiang, and P. J. Scott. Fast algorithm for morphological filters. Journal of Physics: Conference Series, 311(1):012001, 2011. \n[28] W. Luo, Y. Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in deep convolutional neural networks. In Advances in Neural Information Processing Systems (NeurIPS), pp. 4898–4906, 2016. \n[29] R. Murase, M. Suganuma, and T. Okatani. How can cnns use image position for segmentation? arXiv preprint arXiv:2005.03463, 2020. \n[30] A.-D. Nguyen, S. Choi, W. Kim, S. Ahn, J. Kim, and S. Lee. Distribution padding in convolutional neural networks. In IEEE International Conference on Image Processing (ICIP), pp. 4275–4279, 2019. \n[31] D. Novotny, S. Albanie, D. Larlus, and A. Vedaldi. Semi-convolutional operators for instance segmentation. In European Conference on Computer Vision (ECCV), pp. 86–102, 2018. \n[32] A. Odena, V. Dumoulin, and C. Olah. Deconvolution and checkerboard artifacts. Distill, 1 (10):e3, 2016. \n[33] A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, et al. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems (NeurIPS), pp. 8024–8035, 2019. \n[34] P. O. Pinheiro, T.-Y. Lin, R. Collobert, and P. Dollar. Learning to refine object segments. In ´ European Conference on Computer Vision (ECCV), pp. 75–91, 2016. \n[35] S. Schubert, P. Neubert, J. Poschmann, and P. Pretzel. Circular convolutional neural networks ¨ for panoramic images and laser data. In IEEE Intelligent Vehicles Symposium (IV), pp. 653– 660, 2019. \n[36] E. Shalnov. BSTLD-demo: A sample project to train and evaluate model on BSTLD. https: //github.com/e-sha/BSTLD_demo, 2019. \n[37] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. In International Conference on Learning Representations (ICLR), 2015. \n[38] G. Sundaramoorthi and T. E. Wang. Translation insensitive CNNs. arXiv preprint arXiv:1911.11238, 2019. \n[39] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. Going deeper with convolutions. In IEEE conference on Computer Vision and Pattern Recognition (CVPR), pp. 1–9, 2015. \n[40] S. Vashishth, S. Sanyal, V. Nitin, N. Agrawal, and P. Talukdar. InteractE: Improving convolution-based knowledge graph embeddings by increasing feature interactions. In AAAI conference on Artifical Intelligence, 2020. \n[41] S. Wu, G. Wang, P. Tang, F. Chen, and L. Shi. Convolution with even-sized kernels and symmetric padding. In Advances in Neural Information Processing Systems (NeurIPS), pp. 1192–1203, 2019. \n[42] F. Yu and V. Koltun. Multi-scale context aggregation by dilated convolutions. In International Conference on Learning Representations (ICLR), 2016. \n[43] R. Zhang. Making convolutional networks shift-invariant again. In International Conference on Machine Learning (ICML), 2019. ", + "bbox": [ + 173, + 74, + 828, + 928 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 60, + 828, + 931 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 103, + 826, + 325 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A ELIMINATING UNEVEN APPLICATION OF PADDING ", + "text_level": 1, + "bbox": [ + 174, + 102, + 632, + 118 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Consider a CNN with $d$ downsampling layers, $L _ { 1 } , L _ { 2 } , . . . , L _ { d }$ . To simplify the analysis and without loss of generality we assume that the kernels in these layers are of square shape and that all other layers maintain their input size. We denote by $s _ { i }$ and $k _ { i }$ the stride and kernel size of layer $L _ { i }$ . We denote by $h _ { i }$ and $w _ { i }$ the dimensions of the feature maps computed by $L _ { i }$ . We denote by $h _ { 0 }$ and $w _ { 0 }$ the size of the CNN input. We examine the conditions to warrant no uneven application of padding along the height dimension. Parallel conditions apply to the width dimension. ", + "bbox": [ + 174, + 132, + 825, + 218 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "We denote by $\\bar { h } _ { i }$ the height of the padded input to $L _ { i }$ . The effective portion $\\hat { h } _ { i } \\leq \\bar { h } _ { i }$ of this amount processed by the convolutional filters in $L _ { i }$ is equal to: ", + "bbox": [ + 173, + 226, + 823, + 255 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/4899e2975dfc90d0a8c069ee157694d33649b53ae558b1470547ef4b9d531efb.jpg", + "text": "$$\n\\hat { h } _ { i } = s _ { i } \\cdot \\left( h _ { i } - 1 \\right) + k _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 421, + 261, + 575, + 281 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Our goal is to warrant that $\\hat { h } _ { i } = \\bar { h } _ { i }$ to prevent information loss and to avoid uneven padding along the vertical dimension when the unconsumed part $\\bar { h } _ { i } - \\hat { h } _ { i } < s _ { i }$ is an odd number. ", + "bbox": [ + 168, + 290, + 825, + 321 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Since the non-downsampling layers maintain their input size, we can formulate the height of the padded input as follows: ", + "bbox": [ + 171, + 327, + 823, + 354 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/797456e257ded21d46fedaac314e9e3c7963e74ea1f2282d7215bc5e8eece948.jpg", + "text": "$$\n\\bar { h } _ { i } = h _ { i - 1 } + 2 \\cdot p _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 436, + 352, + 562, + 371 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "where $p _ { i }$ is the amount of padding applied at the top and at the bottom of the input in $L _ { i }$ . Accordingly, we can warrant no uneven padding if the following holds: ", + "bbox": [ + 173, + 375, + 823, + 402 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/a6ffbc359fab9bb87c045b8de51352b91c855be9bb9ef831966ddb179afca6e9.jpg", + "text": "$$\n\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = s _ { i } \\cdot ( h _ { i } - 1 ) + k _ { i } - 2 \\cdot p _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 333, + 410, + 665, + 428 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Example 1: ResNet-18 This network contains five downsampling layers ( $\\mathrm { : } d = 5$ ) all of which use a stride of 2. Despite performing downsampling, all of these layers apply a padding amount entailed by SAME padding to avoid information bias against the boundary. In four of these layers having $3 \\times 3$ kernels $k _ { i } = 3 ,$ ), the amount used is $p _ { i } = 1$ . For the first layer having $7 \\times 7$ kernels, this amount is equal to 3. In both cases, the term $k _ { i } - 2 \\cdot p _ { i }$ in Eq. $3$ is equal to 1. To warrant no uneven padding along the vertical dimension, the heights of the feature maps at downsampling layers should hence satisfy: ", + "bbox": [ + 173, + 440, + 825, + 537 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/f4c20619d23c945d28ff9fb47cf60037cd7d565436bbf46e60e9e1fb979e72bd.jpg", + "text": "$$\n\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 1 = 2 \\cdot h _ { i } - 1\n$$", + "text_format": "latex", + "bbox": [ + 321, + 536, + 673, + 554 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Accordingly, the input height should satisfy: ", + "bbox": [ + 174, + 556, + 465, + 571 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/6fe3f2125e90aca84b006cdd0eaf043e7d529e887d0d0da2d4895e8e626c76ee.jpg", + "text": "$$\nh _ { 0 } = 2 ^ { d } \\cdot h _ { d } - ( 2 ^ { d } - 1 ) = 2 ^ { d } \\cdot ( h _ { d } - 1 ) + 1\n$$", + "text_format": "latex", + "bbox": [ + 349, + 577, + 647, + 597 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "where $h _ { d }$ is the height of the final feature map, and can be any natural number larger than 1 to avoid a degenerate case of a $1 \\times 1$ input. The same holds for the input width: ", + "bbox": [ + 169, + 603, + 823, + 632 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/9cb47d9168d465e0eb73d6b4ce4a5b226c2dd4a1e46fd194b85d4a9a707400ee.jpg", + "text": "$$\nw _ { 0 } = 2 ^ { d } \\cdot ( w _ { d } - 1 ) + 1\n$$", + "text_format": "latex", + "bbox": [ + 418, + 637, + 580, + 657 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "A $2 2 5 \\times 2 2 5$ input satisfies these constraints since $2 2 5 = 2 ^ { 5 } \\cdot 7 + 1$ , yielding even padding in all five downsampling layers and output feature maps of size $8 \\times 8$ . ", + "bbox": [ + 173, + 670, + 823, + 700 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Example 2: VGG-16 This network contains five max-pooling layers $( d = 5$ ) all of which use a stride of 2 and a kernel size of 2 and apply no padding. To warrant no uneven padding along the vertical dimension, the heights of the feature maps at all of these layers should hence satisfy: ", + "bbox": [ + 173, + 714, + 825, + 757 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/6e698528c3d9d0dab1ced918089a3dd957d677e240f61ace78c08761eec3fef4.jpg", + "text": "$$\n\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 2 = 2 \\cdot h _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 336, + 763, + 660, + 781 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Accordingly, the input dimensions should satisfy: ", + "bbox": [ + 174, + 787, + 500, + 803 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/3cd98ef77ef0095452f1a67be1696c066ece51daf7d0ea9598b0c464ca7d450f.jpg", + "text": "$$\nh _ { 0 } = 2 ^ { d } \\cdot h _ { d } \\quad \\mathrm { a n d } \\quad w _ { 0 } = 2 ^ { d } \\cdot w _ { d }\n$$", + "text_format": "latex", + "bbox": [ + 382, + 809, + 616, + 827 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "A $2 2 4 \\times 2 2 4$ input satisfies these constraints since $2 2 4 = 2 ^ { 5 } \\cdot 7$ , causing no feature-map erosion at any downsampling layer and resulting in output feature maps of size $7 \\times 7$ . ", + "bbox": [ + 173, + 834, + 831, + 864 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "B THE EXTENT OF FOVEATION UNDER SAME 0-PADDING ", + "text_level": 1, + "bbox": [ + 174, + 102, + 663, + 118 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We illustrate how the absolute extent of foveation under SAME 0-padding depends on the number of convolutional layers, and how its relative extent depends on the input size. ", + "bbox": [ + 176, + 133, + 821, + 162 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "In the following maps, color represents the number of paths to the CNN output for each input pixel. Note: The checkerboard pattern is caused by downsampling layers in ResNet that use $3 \\times 3$ kernels and a stride of 2. ", + "bbox": [ + 174, + 169, + 825, + 210 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/230d99b41516d96c2c03aec57cab7c869711ad4d96a7315fd9c45cd64c50a0f0.jpg", + "image_caption": [ + "Figure 9: The foveation maps of two ResNet architectures under 0 padding, illustrated with a $2 2 5 \\times 2 2 5$ input. Compared with ResNet-50, ResNet-101 has twice the number of convolutional layers with non-unitary filter sizes. Accordingly, the extent of the foveation effect is doubled. " + ], + "image_footnote": [], + "bbox": [ + 287, + 223, + 714, + 372 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/a41bda8b2c50acbca4b5cd99ab0cb07019bf3ee67df5aae70ce1fa81a6316f40.jpg", + "image_caption": [ + "Figure 10: The foveation maps of ResNet-50 under 0 padding, illustrated with inputs of different size. The smaller the input, the larger the relative extent of foveation. " + ], + "image_footnote": [], + "bbox": [ + 176, + 449, + 823, + 553 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "In the next figure, we illustrate how uneven application of padding impacts the foveation maps. Note: It is possible to rectify the skewness in the 2nd foveation map by alternating the side where one-sided padding is applied between successive downsampling layers. This, however, does not mitigate the skewness in the learned filters (see next Section). ", + "bbox": [ + 171, + 612, + 826, + 669 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/b6915c9095fb0ab1b1b632c64deab332d79ccd94ad4eb92251f1454b4674d7bd.jpg", + "image_caption": [ + "Figure 11: The foveation maps of ResNet-50 under 0 padding, illustrated with two input sizes. With a $2 5 7 \\times 2 5 7$ input, the padding is evenly applied at all downsampling layers, leading to a symmetric foveation map. With a $2 5 6 \\times 2 5 6$ input, the padding is applied only to the left and top sides of feature maps at all downsampling layers, which limits the number of convolutional input-output paths for pixels in the bottom and right sides as evident in the skewed foveation map. " + ], + "image_footnote": [], + "bbox": [ + 281, + 681, + 714, + 829 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "C THE IMPACT OF THE PADDING METHOD ON LEARNED WEIGHTS", + "text_level": 1, + "bbox": [ + 174, + 102, + 743, + 118 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "In the presence of uneven application of padding, 0-padding causes skewness in the learned weights because the filters are exposed more frequently to feature-map patches with zeros at their top and left sides. Redundancy methods such as circular or mirror padding mitigate such skewness because they fill the padding areas with values taken from the feature maps. PartialConv also mitigates such skewness because it assumes the pixels in the padding area are missing, and rescales the partial convolutional sum to account for them. Below we show the effectiveness of these alternatives in mitigating the skewness in three ResNet architectures. ", + "bbox": [ + 173, + 133, + 825, + 231 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/c9b6cd065fd2a6ff0aad0928693bb5a1ef91f3a05a54734b812f2ef68cb1a3cc.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 179, + 246, + 460, + 358 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/28bf70ac734a23168ae18bc3334f51e2d26eca67e7b3e2d076b3dfd9008e4880.jpg", + "image_caption": [ + "(b) Mean filters of ResNet-50 trained on $2 2 4 \\times 2 2 4$ images under two padding methods, reaching $7 6 . 1 5 \\%$ top-1 accuracy under 0-padding and $7 6 . 6 1 \\%$ top-1 accuracy under PartialConv. " + ], + "image_footnote": [], + "bbox": [ + 540, + 244, + 813, + 358 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "(a) Mean filters of ResNet-18 trained on $2 2 4 \\times 2 2 4$ images under two padding methods, reaching $6 9 . 9 3 \\%$ top-1 accuracy under 0-padding and $7 0 . 2 8 \\%$ top-1 accuracy under circular padding. ", + "bbox": [ + 173, + 364, + 468, + 415 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Figure 12: Mean filters of two ResNet models trained on ImageNet with $2 2 4 \\times 2 2 4$ images. The input size causes uneven application of padding, leading to frequent asymmetries in the mean filters under 0 padding. We illustrate how two alternatives, circular padding and PartialConv $\\pmb { \\left. \\pmb { \\left. \\bar { 2 3 } \\right. } \\right. }$ , enable learning highly-symmetric mean filters despite the uneven application of padding. ", + "bbox": [ + 173, + 426, + 825, + 483 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/7bc6aeb0c0d5fa125d64fbf452bdfc569b0089cb7b988e4d3f582824309d0153.jpg", + "image_caption": [ + "Figure 13: Mean filters of ResNet-101 trained on ImageNet with $2 2 4 \\times 2 2 4$ images under both 0- padding and PartialConv $\\mathbb { \\lVert 2 3 \\rVert }$ . The input size causes uneven application of padding, leading to frequent asymmetries in the mean filters under 0 padding. In contrast, PartialConv produces highly symmetric mean filters, thanks for its treatment of pixels outside the feature map as missing values. " + ], + "image_footnote": [], + "bbox": [ + 191, + 505, + 810, + 626 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "What if no padding is applied during downsampling? VGG models perform downsampling using $2 \\times 2$ pooling layers that do not apply any padding. Accordingly, the mean filters do not exhibit significant skewness, even if the input size does not satisfy Eq 4: ", + "bbox": [ + 174, + 717, + 825, + 760 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/51a8af5576a494d4ff89c4d3741816ce78e27e93e3c83213ba9b0e683e37e048.jpg", + "image_caption": [ + "Figure 14: Mean filters of VGG-16 trained on ImageNet under different conditions. Most mean filters exhibit high symmetry when trained with $2 2 5 \\times 2 2 5$ images where the size violates Eq. 4. " + ], + "image_footnote": [], + "bbox": [ + 228, + 786, + 769, + 880 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "D THE IMPACT OF PADDING METHODS ON FEATURE-MAP ARTIFACTS ", + "text_level": 1, + "bbox": [ + 171, + 102, + 774, + 118 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We show per-layer mean feature maps in ResNet-18 under different padding methods. The mean maps are averaged over 20 input samples generated at random. ", + "bbox": [ + 174, + 133, + 825, + 162 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/4dc8a25d06b953aa2901c2738d70c12d28f3f668e42fe6df8e21854cbd416e19.jpg", + "image_caption": [ + "Figure 15: Feature map artifacts under zero padding. Line artifacts accumulate to become significant and asymmetric at deeper layers. " + ], + "image_footnote": [], + "bbox": [ + 178, + 179, + 823, + 275 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/6640c3d2bb4a8864228e5cb9e90262456ad14527bc623e13533134115c042001.jpg", + "image_caption": [ + "Figure 16: Circular padding largely preserves the randomness and mitigates line artifacts. " + ], + "image_footnote": [], + "bbox": [ + 174, + 337, + 820, + 434 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/f3cd7587862b9845b41afe3205965e60be50ddde1f43d11c94bed84f9ba5ee93.jpg", + "image_caption": [ + "Figure 17: SYMMETRIC mirror padding also preserves the randomness and mitigates line artifacts. " + ], + "image_footnote": [], + "bbox": [ + 178, + 483, + 823, + 579 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/56552f00a0e6677217affe158f48306a5f1d42e3ca61577ae6dcc64f4fc464fc.jpg", + "image_caption": [ + "Figure 18: REFLECT mirror padding also preserves the randomness and mitigates line artifacts. " + ], + "image_footnote": [], + "bbox": [ + 178, + 630, + 823, + 727 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/72d9bfe380a72b51f7220f1be45381b387d02402c6fb51a1ea2bb58ba8ea21f1.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 178, + 775, + 823, + 871 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Figure 19: PartialConv $\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }$ highly preserves the symmetry of the feature maps. The scaling factors it uses can break the randomness at the boundary. ", + "bbox": [ + 173, + 882, + 823, + 911 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/3813e0f0f363045277b94f3a816f45c0ab5f295c9700bc4537785b52027e2f6d.jpg", + "image_caption": [ + "Figure 20: Feature map artifacts of a VGG-19 model under Distribution Padding (interpolation mode) $\\textcircled { \\lvert 3 0 \\rvert }$ . Due to multiple resize operations used to fill the padding area, the artifacts grow from the boundary inwards. We use a saturated constant input to make the effect visible. " + ], + "image_footnote": [], + "bbox": [ + 174, + 99, + 825, + 231 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "E THE IMPACT OF ANTIALIASING ON THE LEARNED WEIGHTS ", + "text_level": 1, + "bbox": [ + 174, + 308, + 710, + 324 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We demonstrate how antialiasing [43] significantly reduces the asymmetry of mean filters around downsampling layers, even in the presence of unevenly-applied zero padding. ", + "bbox": [ + 173, + 338, + 826, + 367 + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/2012189e75de9944003498504b5db09216e0eb5f175070e1ae807135b096f5fd.jpg", + "image_caption": [ + "Figure 21: Mean filters of four models trained on ImageNet with $2 2 4 \\times 2 2 4$ images under 0-padding both without and with antialiasing. " + ], + "image_footnote": [], + "bbox": [ + 183, + 380, + 810, + 587 + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/57e3f9aa779c1d5920a44c13b73025feb076f2064e727af4c763cdde601bb09a.jpg", + "image_caption": [ + "Figure 22: Mean filters of two models trained on ImageNet with $2 2 4 \\times 2 2 4$ images under 0-padding both without and with antialiasing. " + ], + "image_footnote": [], + "bbox": [ + 212, + 646, + 779, + 871 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "F FOVEATION ANALYSIS OF PADDING ALGORITHMS ", + "text_level": 1, + "bbox": [ + 174, + 102, + 625, + 118 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Refer to http://mind-the-pad.github.io for an interactive and animated visual illustration of padding algorithms and their foveation behavior. This appendix serves as a print version. ", + "bbox": [ + 174, + 133, + 820, + 161 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Among the SAME padding algorithms we discussed in the manuscript, two algorithms warrant that each input pixel is involved in an equal number of convolutional operations, leading to uniform foveation maps: circular padding and SYMMETRIC mirror padding. In contrast, this number varies under zero padding, REFLECT mirror padding, replication padding, and partial convolution. ", + "bbox": [ + 174, + 169, + 825, + 226 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We illustrate in detail how each padding algorithm treats the input pixels. For this purpose we illustrate step by step how each pixel is processed by the convolutional kernel. We choose a set of pixels that are sufficient to expose the behavior of the respective algorithm. This set spans an area within two or three pixels from the boundary that encompasses all relevant cases for the analysis and is situated at the top-left corner. The behavior at the other corners is analogous. ", + "bbox": [ + 174, + 232, + 825, + 301 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "All illustrations use a stride of 1. Except for VALID, all configurations warrant SAME padding. ", + "bbox": [ + 178, + 308, + 794, + 323 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "• VALID Padding: This algorithm is illustrated on a $3 \\times 3$ kernel without dilation. A larger kernel size or dilation factor will increase the foveation effect. Zero Padding: This algorithm is illustrated on a $3 \\times 3$ kernel without dilation. A larger kernel size or dilation factor will increase the foveation effect. Circular Padding: This algorithm is illustrated on a $3 \\times 3$ kernel without dilation. It is straightforward to prove that the algorithm warrants equal treatment of the pixels irrespective of the kernel size or dilation factor. This is because it effectively applies circular convolution: Once the kernel hits one side, it can seamlessly operate on the pixels of the other side. Circular convolution hence renders the feature map as infinite to the kernel, warranting that edge pixels are treated in the same manner as interior pixels. \n. Mirror Padding (SYMMETRIC): This algorithm warrants that each pixel is involved in the same number of convolutional operations. It is important to notice that, unlike under circular convolution, these operations do not utilize the kernel pixels uniformly as we demonstrate in detail. We illustrate the algorithm behavior under the following settings: – $3 \\times 3$ kernel and dilation factor of 1. \n– $5 \\times 5$ kernel and dilation factor of 1. \n– $3 \\times 3$ kernel and dilation factor of 2. \n– $2 \\times 2$ kernel and dilation factor of 1, along with a grouped padding strategy to compensate for uneven padding $\\pm \\amalg$ . \n– $4 \\times 4$ kernel size and dilation factor of 1, along with a grouped padding strategy. ", + "bbox": [ + 214, + 335, + 825, + 545 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "", + "bbox": [ + 245, + 549, + 823, + 642 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "• Mirror Padding (REFLECT): This algorithm is illustrated on a $3 \\times 3$ kernel without dilation. ", + "bbox": [ + 217, + 647, + 825, + 675 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Replication Padding: This algorithm is illustrated on a $5 \\times 5$ kernel without dilation. We choose this kernel size since a $3 \\times 3$ kernel under SAME padding would render the algorithm equivalent to SYMMETRIC mirror padding. \n• Partial Convolution: This algorithm is illustrated on a $3 \\times 3$ kernel without dilation. Its foveation behavior is analogous to REFLECT mirror padding. ", + "bbox": [ + 217, + 679, + 825, + 755 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "VALID Padding Illustrated on a 3x3 kernel ", + "text_level": 1, + "bbox": [ + 73, + 95, + 558, + 121 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Input ", + "text_level": 1, + "bbox": [ + 133, + 143, + 179, + 160 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "# of conv ops each pixel is involved in ", + "text_level": 1, + "bbox": [ + 264, + 138, + 416, + 169 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/1bd1e81c91dba38729b05ee03816c771ae6ca8a523f03363c9811b15a7413f26.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
abC
def
gh
", + "bbox": [ + 102, + 180, + 209, + 266 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/a59efdde31063700c0c4e43e5424da3e21980c2ec1190cc5622cbcddd3171f09.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
12333
24666
36999
36999
36999
", + "bbox": [ + 272, + 178, + 405, + 265 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 81, + 275, + 366, + 291 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/b2f7ea17392566a942f1830fef3e08688e93e3711a4ea27135d9a2b9cf7d34e6.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 98, + 297, + 831, + 378 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/74f48cb87d1f6bdd6666b649c425ea1dfaa5ae5f8b7d6d1487b47797a64dddd6.jpg", + "image_caption": [ + "Detailed Illustration of how the counts are derived " + ], + "image_footnote": [], + "bbox": [ + 93, + 434, + 893, + 532 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions involving (d): rotated version of (b) ", + "bbox": [ + 93, + 561, + 375, + 575 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions involving (e) ", + "bbox": [ + 93, + 590, + 246, + 603 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/075393ac4f0fb1926036df9540e55b75341726cc0c0ce597fe3930694431dbc7.jpg", + "image_caption": [ + "Convolutions involving (f) " + ], + "image_footnote": [], + "bbox": [ + 107, + 616, + 849, + 690 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/a9884b5f8b96c555974a7c1a3e43a6404f579230ff72e5c6ffb7dbcd48eee7a4.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 114, + 744, + 856, + 819 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 96, + 845, + 382, + 859 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 93, + 875, + 377, + 888 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 93, + 904, + 401, + 919 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Zero Padding ", + "text_level": 1, + "bbox": [ + 75, + 94, + 264, + 119 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Illustrated on 3x3 kernel and 1-pixel padding ", + "text_level": 1, + "bbox": [ + 75, + 136, + 493, + 155 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Original Input ", + "text_level": 1, + "bbox": [ + 73, + 231, + 184, + 247 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/556bf414f5fa773bb6e62cc6bf9a5be9d31d1ce35ef4ed652f3e5eb2c42a6fc1.jpg", + "table_caption": [ + "Padded Input " + ], + "table_footnote": [], + "table_body": "
000• ·
0ab
0Cd
0·
• =
", + "bbox": [ + 223, + 253, + 356, + 344 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/f558934c361d8576a111002b479ccad6f1b9812730e47cc3b73814ad1b83231e.jpg", + "table_caption": [ + "# of conv ops each pixel is involved in " + ], + "table_footnote": [], + "table_body": "
46666
69999
6999
69999
69999
", + "bbox": [ + 411, + 263, + 526, + 344 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/359a2e24fb2010e55e4bae06e157a5c8ccb87b3102cb9f9cc8606187601612e7.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
ab··
Cd
·
", + "bbox": [ + 71, + 263, + 184, + 343 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 73, + 373, + 357, + 388 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/05776e978b31b6a3e6f1341bc05f18a6de2a45877c2b627e7689f8e71e5ec1b5.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 161, + 414, + 661, + 520 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 75, + 570, + 465, + 587 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 73, + 614, + 227, + 627 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/2db3478b53918d06b972e15db60b69b5650f9e2b06fc4d4bbc8dc175ac75984c.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 114, + 635, + 632, + 715 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Convolutions involving (b) ", + "bbox": [ + 70, + 751, + 223, + 763 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/63ee4d6ede122d55ac9f5d26ee63c1d6ddf758f4267f0d482b84beda20114155.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 112, + 771, + 900, + 852 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Other border cases are translation or rotation of (a) or (b) ", + "bbox": [ + 70, + 887, + 406, + 900 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Circular Padding ", + "text_level": 1, + "bbox": [ + 75, + 94, + 313, + 121 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Illustrated on 3x3 kernel and 1-pixel padding ", + "bbox": [ + 73, + 136, + 493, + 155 + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/312ea301340019a6276321b4c734de7b4bca19c3c6e7a0942faf86f2dfa2b9d9.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 158, + 169, + 812, + 383 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 165, + 395, + 450, + 411 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "a 1 1 1 b 1 1 1 \n1 1 1 1 1 1 \n1 1 1 1 1 1 \nsum = 9 sum = 9 \nuniform uniform ", + "bbox": [ + 163, + 421, + 398, + 527 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 156, + 544, + 547, + 560 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 161, + 569, + 315, + 582 + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/315c801266b735e20fd2dda8208ed2b4da076ddadc7c67b0b3f5105bfd9d73fd.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 161, + 577, + 759, + 925 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Other border cases are translation or rotation of (a) or (b) ", + "bbox": [ + 166, + 935, + 501, + 949 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Mirror Padding (SYMMETRIC) ", + "text_level": 1, + "bbox": [ + 75, + 94, + 491, + 121 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Illustrated on 3x3 kernel and 1-pixel padding ", + "bbox": [ + 75, + 136, + 493, + 155 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/6ab53447be0e84af51087dd7f95b2e53e57f5b4c85508482b2d0499f14281ec1.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 81, + 200, + 573, + 335 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 88, + 357, + 372, + 372 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/6122910bbc215ed57d822669b52bf0365aa06feda5eeb930d346877584b23468.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 84, + 378, + 844, + 472 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 88, + 502, + 478, + 520 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 127, + 558, + 281, + 570 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/207ecdbeda64fb86e5c2fb8bd0fca5301fca26b3c149e3e0de0f78f19b80b6fc.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 127, + 579, + 627, + 661 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/ce9ae3f6f155e9103378337a6607d52884729cba759d261dd7aa7d104d009ab0.jpg", + "image_caption": [ + "Convolutions involving (b) " + ], + "image_footnote": [], + "bbox": [ + 127, + 717, + 885, + 800 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/7ff8f52e1f1fa7323a591bc9045a4daa5fbdbf91e09c8b8c0c73e8c62390c4f3.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 132, + 857, + 901, + 939 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Convolutions involving (e) ", + "bbox": [ + 81, + 162, + 232, + 174 + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/635b17420ad7db5958a7c27e978a4071b990979ca04dc4ace8f486033de20cd5.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 125, + 199, + 872, + 381 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Convolutions involving (f) ", + "bbox": [ + 78, + 407, + 228, + 420 + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/83b97cfd9b41665e91f8e411ba270b85c9a125588a02eeca4e21f821f73b4bdb.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 130, + 436, + 856, + 618 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 76, + 714, + 362, + 727 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 75, + 743, + 357, + 757 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 75, + 773, + 383, + 786 + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/8a853098ca11735df8a765434a1c5d4f3771546980790927ebd6e4534ec0b669.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 120, + 101, + 393, + 202 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Mirror Padding (SYMMETRIC) ", + "text_level": 1, + "bbox": [ + 472, + 99, + 890, + 126 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Illustrated on 5x5 kernel and 2-pixel padding ", + "bbox": [ + 472, + 141, + 890, + 160 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 116, + 209, + 400, + 224 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/2ed647a62055c626bfc6be0708fddbc84405c35e0aa5605dfe2983f80d1917e0.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 148, + 229, + 831, + 304 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 114, + 316, + 504, + 332 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 116, + 339, + 267, + 352 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/fb93f2025462e471e18bab29c553707d796f5a5df450e689eb1d800b265197c9.jpg", + "image_caption": [ + "Convolutions involving (b) " + ], + "image_footnote": [], + "bbox": [ + 173, + 357, + 782, + 498 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/d6b1ab684ccc9d8b4d0633f56a62178f4ad4cceda1c73f8111bc9eb705c2d5a9.jpg", + "image_caption": [ + "Convolutions involving (c) " + ], + "image_footnote": [], + "bbox": [ + 171, + 535, + 818, + 678 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/923c01362a6ed3e205d71bfc24d478c52b3681aa29cd39b01495efc03ffbd84f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 171, + 715, + 818, + 933 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Convolutions involving (e) ", + "bbox": [ + 129, + 106, + 281, + 118 + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/7e697dd45efbc10ee0c5668063cefbca79bec51cfe0a5bfd6d8827c821509c0a.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 171, + 140, + 777, + 388 + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/9a8f10a070589ef06b9f1a5b0651e3ebe32e891395ebc4fbf96356dc972aba0c.jpg", + "image_caption": [ + "Convolutions involving (f) " + ], + "image_footnote": [], + "bbox": [ + 155, + 446, + 777, + 770 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 117, + 820, + 401, + 834 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 116, + 849, + 398, + 863 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 112, + 880, + 423, + 893 + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/6ab1a20295a0fdba167ef4c254d75ebac90a68354c628ae8597a45a489da49ce.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 124, + 101, + 413, + 195 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Mirror Padding (SYMMETRIC) ", + "text_level": 1, + "bbox": [ + 472, + 99, + 890, + 126 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Illustrated on $\\mathbf { 3 \\times 3 }$ kernel and 1-pixel padding with dilation factor of 2 ", + "bbox": [ + 472, + 141, + 890, + 176 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 116, + 204, + 398, + 219 + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/d99e99f62153187884e0a145ec5dd4f0201fb8255b44b2c45aa57158fc39430c.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 150, + 223, + 620, + 287 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 109, + 299, + 501, + 314 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 127, + 324, + 281, + 338 + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/749f38dde001c33c4de5e80e1053df61ebc232047d92d82b0267074b9fdafae1.jpg", + "image_caption": [ + "Convolutions involving (b) " + ], + "image_footnote": [], + "bbox": [ + 174, + 344, + 803, + 497 + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/3d93f43f0bbad03a7c07d066a148fcedbe3c2528cec4208b76899c07614a04d1.jpg", + "image_caption": [ + "Convolutions involving (c) " + ], + "image_footnote": [], + "bbox": [ + 174, + 535, + 846, + 688 + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/0cda8ee3d259cb24f0c45181ae446820fe9b1146d356cd59c256c8ab595dfe26.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 176, + 719, + 846, + 949 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Convolutions involving (e) ", + "bbox": [ + 129, + 106, + 281, + 118 + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/bd8b805d273315488c5e4f9d9c491d79b396d892db1a7990f4d7f5ab6b0f6113.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 174, + 128, + 803, + 387 + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/0c0ebb9c5742e01fd04882c2956ef19826f0f171ae4291c9a9ae0ed4d2866900.jpg", + "image_caption": [ + "Convolutions involving (f) " + ], + "image_footnote": [], + "bbox": [ + 173, + 448, + 805, + 789 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 117, + 820, + 401, + 834 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 116, + 849, + 398, + 863 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 112, + 880, + 423, + 893 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "Mirror Padding (SYMMETRIC) with Grouping ", + "text_level": 1, + "bbox": [ + 73, + 94, + 699, + 121 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "Illustrated on $\\pmb { 2 \\times 2 }$ kernel and 1-pixel padding A grouped padding strategy is applied to balance uneven padding (Wu et al 2019) ", + "bbox": [ + 75, + 137, + 836, + 172 + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/3cb8ab9203b09b8e6b320ed18ac7056ddcb59f83631f0a18b5a139b841c32530.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 107, + 215, + 787, + 335 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "Number of conv ops each pixel is involved in ", + "bbox": [ + 93, + 348, + 446, + 364 + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/ac51aae79692b49ae5a912d11b076b5c573aa842106e0cc6e722bb121f69aed2.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 246, + 381, + 933, + 486 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "bbox": [ + 96, + 503, + 380, + 518 + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/5c6f2c7fbee023779650ad2f065adc250d53bccea9b4cedc28b2d3322622dff4.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 78, + 554, + 918, + 939 + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/2c5c68905e8fca1d112cde596ce4b53cb4b79975e54fc8cea24edcdef086be98.jpg", + "image_caption": [ + "Detailed Illustration of how the counts are derived " + ], + "image_footnote": [], + "bbox": [ + 119, + 160, + 911, + 388 + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/ecf7ba29a9b7754a971b14400d054da81dbb097f5b1facc5aeaa574edbabe472.jpg", + "image_caption": [ + "Convolutions involving (b) " + ], + "image_footnote": [], + "bbox": [ + 120, + 462, + 887, + 751 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Mirror Padding (SYMMETRIC) with Grouping Illustrated on 4x4 kernel and 1-pixel padding ", + "text_level": 1, + "bbox": [ + 76, + 85, + 699, + 128 + ], + "page_idx": 29 + }, + { + "type": "table", + "img_path": "images/fb8c094e8c999c997ba4d08ba2ab639d22f53c597775025b2c2401ae4a44b15a.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Original InputPadded at top-leftPadded at bottom-Padded at top-rightPadded at bottom left corner
edde fddefright corner
abCbabaabaC
deedb eC fbC fa da deC faab
hhgd ghie hd gd ge highde
gggh
", + "bbox": [ + 186, + 143, + 820, + 252 + ], + "page_idx": 29 + }, + { + "type": "table", + "img_path": "images/aaa852930f4bed4db64a9515a48e23a8fe7d82f7409fd5f982abdff6de16aa11.jpg", + "table_caption": [ + "Number of conv ops each pixel is involved in " + ], + "table_footnote": [], + "table_body": "
Padded at top-leftPadded at bottom-leftPadded at top-rightPadded at bottom-rightAverage (grouped padding strategy)
252520202015151212121515202020991212121616161616
252520202015151212121515202020991212121616161616
20201616162020161616121216161612121616161616161616
20201616162020161616121216161612121616161616161616
20201616 1620201616121216161612121616161616161616
", + "bbox": [ + 186, + 285, + 808, + 391 + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/3c18f875aebd47ed9b36050ea6c6f9e4ee9a4c0390f1a3d6358113ae4f001675.jpg", + "image_caption": [ + "Which kernel cells these ops utilize? " + ], + "image_footnote": [], + "bbox": [ + 217, + 430, + 761, + 963 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Replication Padding ", + "text_level": 1, + "bbox": [ + 537, + 99, + 825, + 125 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/305042bd7a20cd1eafbe805182fa540e7e2ce9a422379289d9ee15436767c91c.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 122, + 101, + 415, + 200 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Illustrated on 5x5 kernel and 2-pixel padding ", + "bbox": [ + 472, + 141, + 890, + 160 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 127, + 212, + 413, + 227 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/4a3779ad3fcb3e7be7745952303c7cebdc2bebab9c0d386a100a95ee8c53f573.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 155, + 228, + 826, + 304 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 119, + 321, + 511, + 337 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 132, + 353, + 287, + 367 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/bf7eec628f94a9d43e2e8a7d0f391f88d89ee6a81d3ab11c5aa917df496a5a5f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 179, + 366, + 781, + 511 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "Convolutions involving (b) ", + "bbox": [ + 132, + 523, + 285, + 536 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/9787549e926a66cc5f95e14c83661906e3e54f96ed8952a4ab6af039ad13d70a.jpg", + "image_caption": [ + "Convolutions involving (c) " + ], + "image_footnote": [], + "bbox": [ + 179, + 545, + 813, + 685 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/f3e1205ae4b631ab25b18d5297cbc480dd1ff5d33e0f8ade664d2ce85347cfe8.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 179, + 724, + 816, + 940 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/80193f24e2b38ecaf141f9d7a2e076285ced3991ba0e63c46aed2978b3d183e8.jpg", + "image_caption": [ + "Convolutions involving (e) " + ], + "image_footnote": [], + "bbox": [ + 179, + 154, + 776, + 396 + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/91bfaefdb8c2ebf071ebe5893fed27de19e40373f9ce1eba526fc7fbf1186d9c.jpg", + "image_caption": [ + "Convolutions involving (f) " + ], + "image_footnote": [], + "bbox": [ + 179, + 467, + 774, + 787 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 116, + 820, + 400, + 833 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 116, + 849, + 398, + 863 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 112, + 880, + 423, + 893 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "Mirror Padding (REFLECT) ", + "text_level": 1, + "bbox": [ + 75, + 94, + 450, + 121 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Illustrated on 3x3 kernel and 1-pixel padding ", + "bbox": [ + 75, + 136, + 493, + 155 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "# of conv ops each pixel is involved in ", + "bbox": [ + 397, + 224, + 549, + 253 + ], + "page_idx": 32 + }, + { + "type": "table", + "img_path": "images/66c037e5000c5f15328d5a862d9327138356030ac9ad265f688caf6674843321.jpg", + "table_caption": [ + "Original Input " + ], + "table_footnote": [], + "table_body": "
abC
def
gh
• ·
", + "bbox": [ + 71, + 265, + 184, + 343 + ], + "page_idx": 32 + }, + { + "type": "table", + "img_path": "images/52cce5556c6a9fed6b3cfd88ed27294598ef69edd8569c27a9700b92554ad13c.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
edef. • . •
babC= =
edef
hgh
= =
", + "bbox": [ + 225, + 253, + 357, + 343 + ], + "page_idx": 32 + }, + { + "type": "table", + "img_path": "images/fb4c72f78124a1cbb79a80404101d7e9d2f6771809c49d980bf98f947cb55cc8.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
48666
816121212
612999
612999
612999
", + "bbox": [ + 411, + 263, + 526, + 343 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 76, + 353, + 361, + 367 + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/d37f8d090f0d4bb551a64bb18b98a508db3428a6185e94f93a10792f876b3330.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 71, + 375, + 857, + 453 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Detailed Illustration of how the counts are derived ", + "text_level": 1, + "bbox": [ + 75, + 496, + 465, + 511 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 73, + 539, + 227, + 553 + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/a8960c789ef1b038c5adb0e2e1fc943e5a08c712af011e46e3f21af3294ed5ba.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 116, + 560, + 633, + 640 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Convolutions involving (b) ", + "bbox": [ + 70, + 675, + 223, + 689 + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/4d298c0343a75ab39732efadb3565e384d2c110e83cefde748e91d4bfb8226d9.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 114, + 695, + 901, + 776 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Convolutions involving (c) ", + "bbox": [ + 71, + 813, + 223, + 824 + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/9a44ed4932a742181366a04189d4ecc324022b23ef90542a89c1c7c432db1ba0.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 120, + 832, + 921, + 912 + ], + "page_idx": 32 + }, + { + "type": "text", + "text": "Convolutions involving (e) ", + "bbox": [ + 83, + 237, + 235, + 250 + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/1ab66efe5cac10d4fc9f6ae0a5a8eff9871f5a5a1e76816a55c37644030707da.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 124, + 271, + 887, + 446 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Convolutions involving (f) ", + "bbox": [ + 86, + 479, + 236, + 492 + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/9c8456965691c401ca270510270cca64da0ffbfd29e905a6210b07a3a22f0387.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 129, + 505, + 877, + 680 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 78, + 714, + 362, + 727 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 75, + 743, + 357, + 757 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 75, + 773, + 383, + 786 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Partial Convolution Illustrated on a 3x3 kernel ", + "text_level": 1, + "bbox": [ + 75, + 95, + 614, + 121 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Input ", + "text_level": 1, + "bbox": [ + 133, + 145, + 178, + 160 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Weighted # of conv ops each pixel is involved in ", + "text_level": 1, + "bbox": [ + 240, + 138, + 429, + 169 + ], + "page_idx": 34 + }, + { + "type": "table", + "img_path": "images/29f6d4252d6ee1cd5aab6da7481bcee47d1724469c368dd7e3e702b2ce66f4a8.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
abC
def
gh
", + "bbox": [ + 101, + 179, + 209, + 266 + ], + "page_idx": 34 + }, + { + "type": "table", + "img_path": "images/234f6997e14ab6a62e25a1c2ec989ecd397bad2952f8a45ffd4fac96eb40a809.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
6.258.757.57.57.5
8.75 12.2510.5 10.510.5
7.510.5999
7.510.5999
7.510.5999
", + "bbox": [ + 274, + 179, + 405, + 265 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Which kernel cells these ops utilize? ", + "text_level": 1, + "bbox": [ + 81, + 276, + 366, + 291 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/e44405b931feb473f680ba423538de1d9620da5536ae1e8dac14e2710ecdbfba.jpg", + "image_caption": [ + "Detailed Illustration of how the counts are derived " + ], + "image_footnote": [], + "bbox": [ + 96, + 297, + 831, + 387 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Convolutions involving (a) ", + "bbox": [ + 99, + 434, + 253, + 448 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/b7d84fb1a287aa41e70703547a09d1f42b59c2cdbaeced40b094c93ee2bd4ec3.jpg", + "image_caption": [ + "Convolutions involving (b) " + ], + "image_footnote": [], + "bbox": [ + 99, + 439, + 625, + 575 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/d9c752239c01303d3667d6eb1e568e38601275535c9e3d462c560285f3ba00ad.jpg", + "image_caption": [ + "Convolutions involving (c) " + ], + "image_footnote": [], + "bbox": [ + 98, + 619, + 875, + 738 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/075465da09bf54db467ff6536761903b1e0c7278cd14b3304dfab0d3c9fd8eaa.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 109, + 785, + 897, + 888 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/c5b335e14940e3ec3c3e7a0b95ef1060da39442c5d1eba43d8d09eeed1a09da0.jpg", + "image_caption": [ + "Convolutions involving (e) " + ], + "image_footnote": [], + "bbox": [ + 93, + 178, + 880, + 703 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Convolutions involving (g): Rotated version of (c) ", + "bbox": [ + 116, + 770, + 400, + 784 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Convolutions involving (h): Rotated version of (f) ", + "bbox": [ + 114, + 800, + 398, + 814 + ], + "page_idx": 35 + }, + { + "type": "text", + "text": "Convolutions involving (i): Regular uniform treatment ", + "bbox": [ + 112, + 829, + 423, + 843 + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/parse/train/m1CD7tPubNy/m1CD7tPubNy_middle.json b/parse/train/m1CD7tPubNy/m1CD7tPubNy_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..e181b2c49b0e8c352d95d494c8ac951097e9837a --- /dev/null +++ b/parse/train/m1CD7tPubNy/m1CD7tPubNy_middle.json @@ -0,0 +1,62546 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 78, + 503, + 96 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 505, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 505, + 98 + ], + "score": 1.0, + "content": "MIND THE PAD – CNNS CAN DEVELOP BLIND SPOTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 112, + 115, + 177, + 137 + ], + "lines": [ + { + "bbox": [ + 112, + 115, + 178, + 126 + ], + "spans": [ + { + "bbox": [ + 112, + 115, + 178, + 126 + ], + "score": 1.0, + "content": "Bilal Alsallakh", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 112, + 127, + 167, + 137 + ], + "spans": [ + { + "bbox": [ + 112, + 127, + 167, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 221, + 115, + 305, + 137 + ], + "lines": [ + { + "bbox": [ + 221, + 114, + 306, + 127 + ], + "spans": [ + { + "bbox": [ + 221, + 114, + 306, + 127 + ], + "score": 1.0, + "content": "Narine Kokhlikyan", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 221, + 127, + 276, + 137 + ], + "spans": [ + { + "bbox": [ + 221, + 127, + 276, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 349, + 115, + 411, + 137 + ], + "lines": [ + { + "bbox": [ + 349, + 114, + 413, + 128 + ], + "spans": [ + { + "bbox": [ + 349, + 114, + 413, + 128 + ], + "score": 1.0, + "content": "Vivek Miglani", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 349, + 126, + 405, + 137 + ], + "spans": [ + { + "bbox": [ + 349, + 126, + 405, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 455, + 115, + 498, + 137 + ], + "lines": [ + { + "bbox": [ + 454, + 114, + 500, + 127 + ], + "spans": [ + { + "bbox": [ + 454, + 114, + 500, + 127 + ], + "score": 1.0, + "content": "Jun Yuan", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 454, + 125, + 481, + 138 + ], + "spans": [ + { + "bbox": [ + 454, + 125, + 481, + 138 + ], + "score": 1.0, + "content": "NYU", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 112, + 154, + 225, + 176 + ], + "lines": [ + { + "bbox": [ + 112, + 153, + 226, + 166 + ], + "spans": [ + { + "bbox": [ + 112, + 153, + 226, + 166 + ], + "score": 1.0, + "content": "Orion Reblitz-Richardson", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 112, + 165, + 167, + 176 + ], + "spans": [ + { + "bbox": [ + 112, + 165, + 167, + 176 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 276, + 204, + 336, + 220 + ], + "spans": [ + { + "bbox": [ + 276, + 204, + 336, + 220 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 142, + 229, + 469, + 329 + ], + "lines": [ + { + "bbox": [ + 142, + 229, + 469, + 242 + ], + "spans": [ + { + "bbox": [ + 142, + 229, + 469, + 242 + ], + "score": 1.0, + "content": "We show how feature maps in convolutional networks are susceptible to spatial", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 241, + 469, + 252 + ], + "spans": [ + { + "bbox": [ + 142, + 241, + 469, + 252 + ], + "score": 1.0, + "content": "bias. Due to a combination of architectural choices, the activation at certain loca-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 252, + 469, + 263 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 469, + 263 + ], + "score": 1.0, + "content": "tions is systematically elevated or weakened. The major source of this bias is the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 263, + 470, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 470, + 275 + ], + "score": 1.0, + "content": "padding mechanism. Depending on several aspects of convolution arithmetic, this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "score": 1.0, + "content": "mechanism can apply the padding unevenly, leading to asymmetries in the learned", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 285, + 469, + 296 + ], + "spans": [ + { + "bbox": [ + 142, + 285, + 469, + 296 + ], + "score": 1.0, + "content": "weights. We demonstrate how such bias can be detrimental to certain tasks such", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 296, + 469, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 469, + 307 + ], + "score": 1.0, + "content": "as small object detection: the activation is suppressed if the stimulus lies in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 307, + 469, + 319 + ], + "spans": [ + { + "bbox": [ + 142, + 307, + 469, + 319 + ], + "score": 1.0, + "content": "impacted area, leading to blind spots and misdetection. We propose solutions to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 317, + 450, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 317, + 450, + 331 + ], + "score": 1.0, + "content": "mitigate spatial bias and demonstrate how they can improve model accuracy.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 348, + 192, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 194, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 194, + 363 + ], + "score": 1.0, + "content": "1 MOTIVATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 372, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "Convolutional neural networks (CNNs) serve as feature extractors for a wide variety of machine-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "learning tasks. Little attention has been paid to the spatial distribution of activation in the feature", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "maps a CNN computes. Our interest in analyzing this distribution is triggered by mysterious failure", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 406, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 418 + ], + "score": 1.0, + "content": "cases of a traffic light detector: The detector successfully detects a small but visible traffic light in a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "road scene. However, it fails completely in detecting the same traffic light in the next frame captured", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 428, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 440 + ], + "score": 1.0, + "content": "by the ego-vehicle. The major difference between both frames is a limited shift along the vertical", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "dimension as the vehicle moves forward. Therefore, the drastic difference in object detection is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 450, + 503, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 503, + 461 + ], + "score": 1.0, + "content": "surprising given that CNNs are often assumed to have a high degree of translation invariance [8; 17].", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "The spatial distribution of activation in feature maps varies with the input. Nevertheless, by closely", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 477, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 490 + ], + "score": 1.0, + "content": "examining this distribution for a large number of samples, we found consistent patterns among them,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 489, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 500 + ], + "score": 1.0, + "content": "often in the form of artifacts that do not resemble any input features. This work aims to analyze the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 499, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 511 + ], + "score": 1.0, + "content": "root cause of such artifacts and their impact on CNNs. We show that these artifacts are responsible", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "for the mysterious failure cases mentioned earlier, as they can induce ‘blind spots’ for the object", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 520, + 260, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 260, + 534 + ], + "score": 1.0, + "content": "detection head. Our contributions are:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 132, + 541, + 504, + 615 + ], + "lines": [ + { + "bbox": [ + 131, + 540, + 498, + 555 + ], + "spans": [ + { + "bbox": [ + 131, + 540, + 498, + 555 + ], + "score": 1.0, + "content": "• Demonstrating how the padding mechanism can induce spatial bias in CNNs (Section 2).", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 132, + 554, + 435, + 567 + ], + "spans": [ + { + "bbox": [ + 132, + 554, + 435, + 567 + ], + "score": 1.0, + "content": "• Demonstrating how spatial bias can impair downstream tasks (Section 3).", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 131, + 566, + 487, + 580 + ], + "spans": [ + { + "bbox": [ + 131, + 566, + 487, + 580 + ], + "score": 1.0, + "content": "• Identifying uneven application of 0-padding as a resolvable source of bias (Section 5).", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 137, + 579, + 473, + 591 + ], + "spans": [ + { + "bbox": [ + 137, + 579, + 473, + 591 + ], + "score": 1.0, + "content": "Relating the padding mechanism with the foveation behavior of CNNs (Section 6).", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 133, + 591, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 133, + 591, + 505, + 606 + ], + "score": 1.0, + "content": "• Providing recommendations to mitigate spatial bias and demonstrating how this can prevent", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 141, + 603, + 296, + 616 + ], + "spans": [ + { + "bbox": [ + 141, + 603, + 296, + 616 + ], + "score": 1.0, + "content": "blind spots and boost model accuracy.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.5 + }, + { + "type": "title", + "bbox": [ + 107, + 630, + 358, + 643 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 359, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 359, + 644 + ], + "score": 1.0, + "content": "2 THE EMERGENCE OF SPATIAL BIAS IN CNNS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Our aim is to determine to which extent activation magnitude in CNN feature maps is influenced", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 483, + 678 + ], + "score": 1.0, + "content": "by location. We demonstrate our analysis on a publicly-available traffic-light detection model", + "type": "text" + }, + { + "bbox": [ + 484, + 665, + 501, + 677 + ], + "score": 0.38, + "content": "\\pmb { \\mathbb { B } } \\pmb { \\ 6 } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 665, + 504, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 289, + 689 + ], + "score": 1.0, + "content": "This model implements the SSD architecture", + "type": "text" + }, + { + "bbox": [ + 289, + 677, + 307, + 688 + ], + "score": 0.88, + "content": "\\left[ \\left[ 2 6 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 676, + 368, + 689 + ], + "score": 1.0, + "content": "in TensorFlow", + "type": "text" + }, + { + "bbox": [ + 369, + 677, + 381, + 688 + ], + "score": 0.73, + "content": "\\mathbb { I I }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 676, + 468, + 689 + ], + "score": 1.0, + "content": ", using MobileNet-v1", + "type": "text" + }, + { + "bbox": [ + 468, + 676, + 486, + 688 + ], + "score": 0.77, + "content": "\\mathbb { \\lVert \\lambda \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "as a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 360, + 699 + ], + "score": 1.0, + "content": "feature extractor. The model is trained on the BSTLD dataset", + "type": "text" + }, + { + "bbox": [ + 360, + 687, + 373, + 699 + ], + "score": 0.64, + "content": "\\pmb { \\Vert 4 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "which annotates traffic lights in", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 188, + 711 + ], + "score": 1.0, + "content": "road scenes. Figure", + "type": "text" + }, + { + "bbox": [ + 189, + 698, + 198, + 711 + ], + "score": 0.5, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "shows two example scenes from the dataset. For each scene, we show two", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 283, + 722 + ], + "score": 1.0, + "content": "feature maps computed by two filters in the", + "type": "text" + }, + { + "bbox": [ + 283, + 709, + 300, + 720 + ], + "score": 0.86, + "content": "1 1 ^ { \\mathrm { t h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "convolutional layer. This layer contains 512 filters", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 721, + 497, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 497, + 733 + ], + "score": 1.0, + "content": "whose feature maps are used directly by the first box predictor in the SSD to detect small objects.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 78, + 503, + 96 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 505, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 505, + 98 + ], + "score": 1.0, + "content": "MIND THE PAD – CNNS CAN DEVELOP BLIND SPOTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 112, + 115, + 177, + 137 + ], + "lines": [ + { + "bbox": [ + 112, + 115, + 178, + 126 + ], + "spans": [ + { + "bbox": [ + 112, + 115, + 178, + 126 + ], + "score": 1.0, + "content": "Bilal Alsallakh", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 112, + 127, + 167, + 137 + ], + "spans": [ + { + "bbox": [ + 112, + 127, + 167, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 112, + 115, + 178, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 221, + 115, + 305, + 137 + ], + "lines": [ + { + "bbox": [ + 221, + 114, + 306, + 127 + ], + "spans": [ + { + "bbox": [ + 221, + 114, + 306, + 127 + ], + "score": 1.0, + "content": "Narine Kokhlikyan", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 221, + 127, + 276, + 137 + ], + "spans": [ + { + "bbox": [ + 221, + 127, + 276, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 221, + 114, + 306, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 349, + 115, + 411, + 137 + ], + "lines": [ + { + "bbox": [ + 349, + 114, + 413, + 128 + ], + "spans": [ + { + "bbox": [ + 349, + 114, + 413, + 128 + ], + "score": 1.0, + "content": "Vivek Miglani", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 349, + 126, + 405, + 137 + ], + "spans": [ + { + "bbox": [ + 349, + 126, + 405, + 137 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 349, + 114, + 413, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 455, + 115, + 498, + 137 + ], + "lines": [ + { + "bbox": [ + 454, + 114, + 500, + 127 + ], + "spans": [ + { + "bbox": [ + 454, + 114, + 500, + 127 + ], + "score": 1.0, + "content": "Jun Yuan", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 454, + 125, + 481, + 138 + ], + "spans": [ + { + "bbox": [ + 454, + 125, + 481, + 138 + ], + "score": 1.0, + "content": "NYU", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 454, + 114, + 500, + 138 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 154, + 225, + 176 + ], + "lines": [ + { + "bbox": [ + 112, + 153, + 226, + 166 + ], + "spans": [ + { + "bbox": [ + 112, + 153, + 226, + 166 + ], + "score": 1.0, + "content": "Orion Reblitz-Richardson", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 112, + 165, + 167, + 176 + ], + "spans": [ + { + "bbox": [ + 112, + 165, + 167, + 176 + ], + "score": 1.0, + "content": "Facebook AI", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 112, + 153, + 226, + 176 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 276, + 204, + 336, + 220 + ], + "spans": [ + { + "bbox": [ + 276, + 204, + 336, + 220 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 142, + 229, + 469, + 329 + ], + "lines": [ + { + "bbox": [ + 142, + 229, + 469, + 242 + ], + "spans": [ + { + "bbox": [ + 142, + 229, + 469, + 242 + ], + "score": 1.0, + "content": "We show how feature maps in convolutional networks are susceptible to spatial", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 241, + 469, + 252 + ], + "spans": [ + { + "bbox": [ + 142, + 241, + 469, + 252 + ], + "score": 1.0, + "content": "bias. Due to a combination of architectural choices, the activation at certain loca-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 252, + 469, + 263 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 469, + 263 + ], + "score": 1.0, + "content": "tions is systematically elevated or weakened. The major source of this bias is the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 263, + 470, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 470, + 275 + ], + "score": 1.0, + "content": "padding mechanism. Depending on several aspects of convolution arithmetic, this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "score": 1.0, + "content": "mechanism can apply the padding unevenly, leading to asymmetries in the learned", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 285, + 469, + 296 + ], + "spans": [ + { + "bbox": [ + 142, + 285, + 469, + 296 + ], + "score": 1.0, + "content": "weights. We demonstrate how such bias can be detrimental to certain tasks such", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 296, + 469, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 469, + 307 + ], + "score": 1.0, + "content": "as small object detection: the activation is suppressed if the stimulus lies in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 307, + 469, + 319 + ], + "spans": [ + { + "bbox": [ + 142, + 307, + 469, + 319 + ], + "score": 1.0, + "content": "impacted area, leading to blind spots and misdetection. We propose solutions to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 317, + 450, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 317, + 450, + 331 + ], + "score": 1.0, + "content": "mitigate spatial bias and demonstrate how they can improve model accuracy.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16, + "bbox_fs": [ + 141, + 229, + 470, + 331 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 348, + 192, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 194, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 194, + 363 + ], + "score": 1.0, + "content": "1 MOTIVATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 372, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "Convolutional neural networks (CNNs) serve as feature extractors for a wide variety of machine-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "learning tasks. Little attention has been paid to the spatial distribution of activation in the feature", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "maps a CNN computes. Our interest in analyzing this distribution is triggered by mysterious failure", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 406, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 418 + ], + "score": 1.0, + "content": "cases of a traffic light detector: The detector successfully detects a small but visible traffic light in a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "road scene. However, it fails completely in detecting the same traffic light in the next frame captured", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 428, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 440 + ], + "score": 1.0, + "content": "by the ego-vehicle. The major difference between both frames is a limited shift along the vertical", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "dimension as the vehicle moves forward. Therefore, the drastic difference in object detection is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 450, + 503, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 503, + 461 + ], + "score": 1.0, + "content": "surprising given that CNNs are often assumed to have a high degree of translation invariance [8; 17].", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 372, + 506, + 461 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "The spatial distribution of activation in feature maps varies with the input. Nevertheless, by closely", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 477, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 490 + ], + "score": 1.0, + "content": "examining this distribution for a large number of samples, we found consistent patterns among them,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 489, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 500 + ], + "score": 1.0, + "content": "often in the form of artifacts that do not resemble any input features. This work aims to analyze the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 499, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 511 + ], + "score": 1.0, + "content": "root cause of such artifacts and their impact on CNNs. We show that these artifacts are responsible", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "for the mysterious failure cases mentioned earlier, as they can induce ‘blind spots’ for the object", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 520, + 260, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 260, + 534 + ], + "score": 1.0, + "content": "detection head. Our contributions are:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 466, + 505, + 534 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 541, + 504, + 615 + ], + "lines": [ + { + "bbox": [ + 131, + 540, + 498, + 555 + ], + "spans": [ + { + "bbox": [ + 131, + 540, + 498, + 555 + ], + "score": 1.0, + "content": "• Demonstrating how the padding mechanism can induce spatial bias in CNNs (Section 2).", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 554, + 435, + 567 + ], + "spans": [ + { + "bbox": [ + 132, + 554, + 435, + 567 + ], + "score": 1.0, + "content": "• Demonstrating how spatial bias can impair downstream tasks (Section 3).", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 566, + 487, + 580 + ], + "spans": [ + { + "bbox": [ + 131, + 566, + 487, + 580 + ], + "score": 1.0, + "content": "• Identifying uneven application of 0-padding as a resolvable source of bias (Section 5).", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 137, + 579, + 473, + 591 + ], + "spans": [ + { + "bbox": [ + 137, + 579, + 473, + 591 + ], + "score": 1.0, + "content": "Relating the padding mechanism with the foveation behavior of CNNs (Section 6).", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 133, + 591, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 133, + 591, + 505, + 606 + ], + "score": 1.0, + "content": "• Providing recommendations to mitigate spatial bias and demonstrating how this can prevent", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 603, + 296, + 616 + ], + "spans": [ + { + "bbox": [ + 141, + 603, + 296, + 616 + ], + "score": 1.0, + "content": "blind spots and boost model accuracy.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + } + ], + "index": 38.5, + "bbox_fs": [ + 131, + 540, + 505, + 616 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 630, + 358, + 643 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 359, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 359, + 644 + ], + "score": 1.0, + "content": "2 THE EMERGENCE OF SPATIAL BIAS IN CNNS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Our aim is to determine to which extent activation magnitude in CNN feature maps is influenced", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 483, + 678 + ], + "score": 1.0, + "content": "by location. We demonstrate our analysis on a publicly-available traffic-light detection model", + "type": "text" + }, + { + "bbox": [ + 484, + 665, + 501, + 677 + ], + "score": 0.38, + "content": "\\pmb { \\mathbb { B } } \\pmb { \\ 6 } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 665, + 504, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 289, + 689 + ], + "score": 1.0, + "content": "This model implements the SSD architecture", + "type": "text" + }, + { + "bbox": [ + 289, + 677, + 307, + 688 + ], + "score": 0.88, + "content": "\\left[ \\left[ 2 6 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 676, + 368, + 689 + ], + "score": 1.0, + "content": "in TensorFlow", + "type": "text" + }, + { + "bbox": [ + 369, + 677, + 381, + 688 + ], + "score": 0.73, + "content": "\\mathbb { I I }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 676, + 468, + 689 + ], + "score": 1.0, + "content": ", using MobileNet-v1", + "type": "text" + }, + { + "bbox": [ + 468, + 676, + 486, + 688 + ], + "score": 0.77, + "content": "\\mathbb { \\lVert \\lambda \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "as a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 360, + 699 + ], + "score": 1.0, + "content": "feature extractor. The model is trained on the BSTLD dataset", + "type": "text" + }, + { + "bbox": [ + 360, + 687, + 373, + 699 + ], + "score": 0.64, + "content": "\\pmb { \\Vert 4 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "which annotates traffic lights in", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 188, + 711 + ], + "score": 1.0, + "content": "road scenes. Figure", + "type": "text" + }, + { + "bbox": [ + 189, + 698, + 198, + 711 + ], + "score": 0.5, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "shows two example scenes from the dataset. For each scene, we show two", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 283, + 722 + ], + "score": 1.0, + "content": "feature maps computed by two filters in the", + "type": "text" + }, + { + "bbox": [ + 283, + 709, + 300, + 720 + ], + "score": 0.86, + "content": "1 1 ^ { \\mathrm { t h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "convolutional layer. This layer contains 512 filters", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 721, + 497, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 497, + 733 + ], + "score": 1.0, + "content": "whose feature maps are used directly by the first box predictor in the SSD to detect small objects.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 654, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 79, + 489, + 271 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 79, + 489, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 79, + 489, + 271 + ], + "spans": [ + { + "bbox": [ + 119, + 79, + 489, + 271 + ], + "score": 0.977, + "type": "image", + "image_path": "b526bfe207aac3286b6c1d5a6a8d0a326615dced8dd84ad446f05157528bdb45.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 79, + 489, + 143.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 143.0, + 489, + 207.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 207.0, + 489, + 271.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 280, + 504, + 315 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Figure 1: Averaging feature maps per input (column marginal) and per filter (row marginal) in the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "last convolutional layer of a traffic light detector. Color indicates activation strength (the brighter,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "the higher), revealing line artifacts in the maps. These artifacts are the manifestation of spatial bias.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 212, + 353 + ], + "score": 1.0, + "content": "The bottom row in Figure", + "type": "text" + }, + { + "bbox": [ + 212, + 340, + 222, + 353 + ], + "score": 0.3, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "shows the average response of each of the two aforementioned filters,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 351, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 506, + 365 + ], + "score": 1.0, + "content": "computed over the test set in BSTLD. The first filter seems to respond mainly to features in the top", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "half of the input, while the second filter responds mainly to street areas. There are visible lines in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "the two average maps that do not seem to resemble any scene features and are consistently present", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "in the individual feature maps. We analyzed the prevalence of these line artifacts in the feature maps", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 394, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 394, + 288, + 409 + ], + "score": 1.0, + "content": "of all 512 filters. The right column in Figure", + "type": "text" + }, + { + "bbox": [ + 288, + 395, + 298, + 408 + ], + "score": 0.45, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 394, + 506, + 409 + ], + "score": 1.0, + "content": "shows the average of these maps per scene, as well", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 407, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 419 + ], + "score": 1.0, + "content": "as over the entire test set (see supplemental for all 512 maps). The artifacts are largely visible in the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 418, + 474, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 474, + 430 + ], + "score": 1.0, + "content": "average maps, with variations per scene depending on which individual maps are dominant.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 506, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 446 + ], + "score": 1.0, + "content": "A useful way to make the artifacts stand out is to neutralize scene features by computing the feature", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 444, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 261, + 458 + ], + "score": 1.0, + "content": "maps for a zero-valued input. Figure", + "type": "text" + }, + { + "bbox": [ + 261, + 444, + 271, + 458 + ], + "score": 0.59, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 445, + 506, + 458 + ], + "score": 1.0, + "content": "depicts the resulting average map for each convolutional", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "layer after applying ReLU units. The first average map is constant as we expect with a 0-valued", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "input. The second map is also constant except for a 1-pixel boundary where the value is lower at the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "score": 1.0, + "content": "left border and higher at the other three borders. We magnify the corners to make these deviations", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "score": 1.0, + "content": "visible. The border deviations increase in thickness and in variance at subsequent layers, creating", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "multiple line artifacts at each border. These artifacts become quite pronounced at ReLU 8 where", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 511, + 362, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 362, + 524 + ], + "score": 1.0, + "content": "they start to propagate inwards, resembling the ones in Figure 1.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5 + }, + { + "type": "image", + "bbox": [ + 118, + 539, + 496, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 118, + 539, + 496, + 704 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 118, + 539, + 496, + 704 + ], + "spans": [ + { + "bbox": [ + 118, + 539, + 496, + 704 + ], + "score": 0.967, + "type": "image", + "image_path": "9ad02d0bb96ca163c5aaed3822f6466e3fe37e0cde18654db995d27f877d7803.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 118, + 539, + 496, + 594.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 118, + 594.0, + 496, + 649.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 118, + 649.0, + 496, + 704.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 717, + 502, + 729 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 716, + 498, + 731 + ], + "spans": [ + { + "bbox": [ + 107, + 716, + 455, + 731 + ], + "score": 1.0, + "content": "Figure 2: Activation maps for a 0 input, averaged over each layer’s filters (title format:", + "type": "text" + }, + { + "bbox": [ + 456, + 717, + 498, + 728 + ], + "score": 0.86, + "content": "\\mathrm { H } { \\times } \\mathrm { W } { \\times } \\mathrm { C } )", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25 + } + ], + "index": 24.0 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 79, + 489, + 271 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 79, + 489, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 79, + 489, + 271 + ], + "spans": [ + { + "bbox": [ + 119, + 79, + 489, + 271 + ], + "score": 0.977, + "type": "image", + "image_path": "b526bfe207aac3286b6c1d5a6a8d0a326615dced8dd84ad446f05157528bdb45.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 79, + 489, + 143.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 143.0, + 489, + 207.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 207.0, + 489, + 271.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 280, + 504, + 315 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Figure 1: Averaging feature maps per input (column marginal) and per filter (row marginal) in the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "last convolutional layer of a traffic light detector. Color indicates activation strength (the brighter,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "the higher), revealing line artifacts in the maps. These artifacts are the manifestation of spatial bias.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 212, + 353 + ], + "score": 1.0, + "content": "The bottom row in Figure", + "type": "text" + }, + { + "bbox": [ + 212, + 340, + 222, + 353 + ], + "score": 0.3, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "shows the average response of each of the two aforementioned filters,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 351, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 506, + 365 + ], + "score": 1.0, + "content": "computed over the test set in BSTLD. The first filter seems to respond mainly to features in the top", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "half of the input, while the second filter responds mainly to street areas. There are visible lines in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "the two average maps that do not seem to resemble any scene features and are consistently present", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "in the individual feature maps. We analyzed the prevalence of these line artifacts in the feature maps", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 394, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 394, + 288, + 409 + ], + "score": 1.0, + "content": "of all 512 filters. The right column in Figure", + "type": "text" + }, + { + "bbox": [ + 288, + 395, + 298, + 408 + ], + "score": 0.45, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 394, + 506, + 409 + ], + "score": 1.0, + "content": "shows the average of these maps per scene, as well", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 407, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 419 + ], + "score": 1.0, + "content": "as over the entire test set (see supplemental for all 512 maps). The artifacts are largely visible in the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 418, + 474, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 474, + 430 + ], + "score": 1.0, + "content": "average maps, with variations per scene depending on which individual maps are dominant.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 340, + 506, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 506, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 446 + ], + "score": 1.0, + "content": "A useful way to make the artifacts stand out is to neutralize scene features by computing the feature", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 444, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 261, + 458 + ], + "score": 1.0, + "content": "maps for a zero-valued input. Figure", + "type": "text" + }, + { + "bbox": [ + 261, + 444, + 271, + 458 + ], + "score": 0.59, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 445, + 506, + 458 + ], + "score": 1.0, + "content": "depicts the resulting average map for each convolutional", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "layer after applying ReLU units. The first average map is constant as we expect with a 0-valued", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "input. The second map is also constant except for a 1-pixel boundary where the value is lower at the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 490 + ], + "score": 1.0, + "content": "left border and higher at the other three borders. We magnify the corners to make these deviations", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "score": 1.0, + "content": "visible. The border deviations increase in thickness and in variance at subsequent layers, creating", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "multiple line artifacts at each border. These artifacts become quite pronounced at ReLU 8 where", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 511, + 362, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 362, + 524 + ], + "score": 1.0, + "content": "they start to propagate inwards, resembling the ones in Figure 1.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 434, + 506, + 524 + ] + }, + { + "type": "image", + "bbox": [ + 118, + 539, + 496, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 118, + 539, + 496, + 704 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 118, + 539, + 496, + 704 + ], + "spans": [ + { + "bbox": [ + 118, + 539, + 496, + 704 + ], + "score": 0.967, + "type": "image", + "image_path": "9ad02d0bb96ca163c5aaed3822f6466e3fe37e0cde18654db995d27f877d7803.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 118, + 539, + 496, + 594.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 118, + 594.0, + 496, + 649.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 118, + 649.0, + 496, + 704.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 717, + 502, + 729 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 716, + 498, + 731 + ], + "spans": [ + { + "bbox": [ + 107, + 716, + 455, + 731 + ], + "score": 1.0, + "content": "Figure 2: Activation maps for a 0 input, averaged over each layer’s filters (title format:", + "type": "text" + }, + { + "bbox": [ + 456, + 717, + 498, + 728 + ], + "score": 0.86, + "content": "\\mathrm { H } { \\times } \\mathrm { W } { \\times } \\mathrm { C } )", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25 + } + ], + "index": 24.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 504, + 95 + ], + "score": 1.0, + "content": "It is evident that the 1-pixel border variations in the second map are caused by the padding mecha-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "nism in use. This mechanism pads the output of the previous layer with a 1-pixel 0-valued border", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 358, + 117 + ], + "score": 1.0, + "content": "in order to maintain the size of the feature map after applying", + "type": "text" + }, + { + "bbox": [ + 358, + 105, + 375, + 115 + ], + "score": 0.29, + "content": "3 { \\tt X } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "convolutional. The maps in the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "score": 1.0, + "content": "first layer are not impacted because the input we feed is zero valued. Subsequent layers, however,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 495, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 495, + 140 + ], + "score": 1.0, + "content": "are increasingly impacted by the padding, as preceding bias terms do not warrant 0-valued input.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "It is noticeable in Figure 2 that the artifacts caused by the padding differ across the four borders. To", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "investigate this asymmetry, we analyze the convolutional kernels (often called filters) that produce", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 343, + 179 + ], + "score": 1.0, + "content": "the feature maps. Figure 3 depicts a per-layer mean of these", + "type": "text" + }, + { + "bbox": [ + 343, + 165, + 360, + 176 + ], + "score": 0.35, + "content": "3 { \\tt X } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 165, + 506, + 179 + ], + "score": 1.0, + "content": "kernels. These mean kernels exhibit", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "different degrees of asymmetry in the spatial distribution of their weights. For example, the kernels", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 186, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 201 + ], + "score": 1.0, + "content": "in L1 assign (on average) a negative weight at the left border, and a positive weight at the bottom.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "This directly impacts the padding-induced variation at each border. Such asymmetries are related to", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 210, + 337, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 337, + 222 + ], + "score": 1.0, + "content": "uneven application of padding as we explain in Section 5.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 113, + 234, + 500, + 270 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 234, + 500, + 270 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 234, + 500, + 270 + ], + "spans": [ + { + "bbox": [ + 113, + 234, + 500, + 270 + ], + "score": 0.945, + "type": "image", + "image_path": "fccaabdbce7408640032adb86b3858809bd5967c675d8bc0ad51217199678338.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 113, + 234, + 500, + 246.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 113, + 246.0, + 500, + 258.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 113, + 258.0, + 500, + 270.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 108, + 283, + 498, + 295 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 282, + 500, + 297 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 361, + 297 + ], + "score": 1.0, + "content": "Figure 3: Mean kernel per convolutional layer. All kernels are", + "type": "text" + }, + { + "bbox": [ + 361, + 284, + 385, + 294 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 282, + 500, + 297 + ], + "score": 1.0, + "content": ", the titles show their counts.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "title", + "bbox": [ + 108, + 325, + 291, + 338 + ], + "lines": [ + { + "bbox": [ + 104, + 324, + 293, + 340 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 293, + 340 + ], + "score": 1.0, + "content": "3 IMPLICATIONS OF SPATIAL BIAS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 351, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "We demonstrate how feature-map artifacts can cause blind spots for the SSD model. Similar issues", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "arise in several small-object detectors, e.g., for faces and masks, as well as in pixel-oriented tasks", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 373, + 451, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 451, + 385 + ], + "score": 1.0, + "content": "such as semantic segmentation and image inpainting (see supplemental for examples).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 134, + 402 + ], + "score": 1.0, + "content": "Figure", + "type": "text" + }, + { + "bbox": [ + 135, + 389, + 145, + 403 + ], + "score": 0.57, + "content": "\\sharp", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "illustrates how the SSD predicts small objects based on the feature maps of the 11-th", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 401, + 504, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 504, + 412 + ], + "score": 1.0, + "content": "convolutional layer. The SSD uses the pixel positions in these maps as anchors of object proposals.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "Each proposal is scored by the SSD to represent a target category, with ”background“ being an", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 422, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 436 + ], + "score": 1.0, + "content": "implicit category that is crucial to exclude irrelevant parts of the input. In addition to these scores,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "the SSD computes a bounding box to localize the predicted object at each anchor. We examine", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "image", + "bbox": [ + 110, + 458, + 502, + 663 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 458, + 502, + 663 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 458, + 502, + 663 + ], + "spans": [ + { + "bbox": [ + 110, + 458, + 502, + 663 + ], + "score": 0.965, + "type": "image", + "image_path": "2f90b37ac2e03a969222bf159083fe214d6629973648c95433a912a506b514b4.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 110, + 458, + 502, + 526.3333333333334 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 110, + 526.3333333333334, + 502, + 594.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 110, + 594.6666666666667, + 502, + 663.0000000000001 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 673, + 505, + 729 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "Figure 4: The formation of blind spots in SSD, illustrated via its box predictor internals with a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 685, + 505, + 697 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 505, + 697 + ], + "score": 1.0, + "content": "zero-valued input. The predictor uses spatial anchors to detect and localize the target object at", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 695, + 505, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 695, + 141, + 706 + ], + "score": 0.89, + "content": "4 5 \\times 8 0", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 695, + 505, + 708 + ], + "score": 1.0, + "content": "possible locations based on 512 feature maps. Certain anchors are predisposed to predict", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 706, + 505, + 719 + ], + "spans": [ + { + "bbox": [ + 105, + 706, + 505, + 719 + ], + "score": 1.0, + "content": "background due to feature-map artifacts, as evident in the logit maps. Traffic lights at the corre-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 717, + 505, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 717, + 505, + 730 + ], + "score": 1.0, + "content": "sponding location cannot be detected as demonstrated with a real scene (middle one in the bottom).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "index": 28.0 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 504, + 95 + ], + "score": 1.0, + "content": "It is evident that the 1-pixel border variations in the second map are caused by the padding mecha-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "nism in use. This mechanism pads the output of the previous layer with a 1-pixel 0-valued border", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 358, + 117 + ], + "score": 1.0, + "content": "in order to maintain the size of the feature map after applying", + "type": "text" + }, + { + "bbox": [ + 358, + 105, + 375, + 115 + ], + "score": 0.29, + "content": "3 { \\tt X } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "convolutional. The maps in the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "score": 1.0, + "content": "first layer are not impacted because the input we feed is zero valued. Subsequent layers, however,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 495, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 495, + 140 + ], + "score": 1.0, + "content": "are increasingly impacted by the padding, as preceding bias terms do not warrant 0-valued input.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 83, + 505, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "It is noticeable in Figure 2 that the artifacts caused by the padding differ across the four borders. To", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "investigate this asymmetry, we analyze the convolutional kernels (often called filters) that produce", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 343, + 179 + ], + "score": 1.0, + "content": "the feature maps. Figure 3 depicts a per-layer mean of these", + "type": "text" + }, + { + "bbox": [ + 343, + 165, + 360, + 176 + ], + "score": 0.35, + "content": "3 { \\tt X } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 165, + 506, + 179 + ], + "score": 1.0, + "content": "kernels. These mean kernels exhibit", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "different degrees of asymmetry in the spatial distribution of their weights. For example, the kernels", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 186, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 201 + ], + "score": 1.0, + "content": "in L1 assign (on average) a negative weight at the left border, and a positive weight at the bottom.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "This directly impacts the padding-induced variation at each border. Such asymmetries are related to", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 210, + 337, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 337, + 222 + ], + "score": 1.0, + "content": "uneven application of padding as we explain in Section 5.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 143, + 506, + 222 + ] + }, + { + "type": "image", + "bbox": [ + 113, + 234, + 500, + 270 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 234, + 500, + 270 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 234, + 500, + 270 + ], + "spans": [ + { + "bbox": [ + 113, + 234, + 500, + 270 + ], + "score": 0.945, + "type": "image", + "image_path": "fccaabdbce7408640032adb86b3858809bd5967c675d8bc0ad51217199678338.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 113, + 234, + 500, + 246.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 113, + 246.0, + 500, + 258.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 113, + 258.0, + 500, + 270.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 108, + 283, + 498, + 295 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 282, + 500, + 297 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 361, + 297 + ], + "score": 1.0, + "content": "Figure 3: Mean kernel per convolutional layer. All kernels are", + "type": "text" + }, + { + "bbox": [ + 361, + 284, + 385, + 294 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 282, + 500, + 297 + ], + "score": 1.0, + "content": ", the titles show their counts.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "title", + "bbox": [ + 108, + 325, + 291, + 338 + ], + "lines": [ + { + "bbox": [ + 104, + 324, + 293, + 340 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 293, + 340 + ], + "score": 1.0, + "content": "3 IMPLICATIONS OF SPATIAL BIAS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 351, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "We demonstrate how feature-map artifacts can cause blind spots for the SSD model. Similar issues", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "arise in several small-object detectors, e.g., for faces and masks, as well as in pixel-oriented tasks", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 373, + 451, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 451, + 385 + ], + "score": 1.0, + "content": "such as semantic segmentation and image inpainting (see supplemental for examples).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 351, + 505, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 134, + 402 + ], + "score": 1.0, + "content": "Figure", + "type": "text" + }, + { + "bbox": [ + 135, + 389, + 145, + 403 + ], + "score": 0.57, + "content": "\\sharp", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "illustrates how the SSD predicts small objects based on the feature maps of the 11-th", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 401, + 504, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 504, + 412 + ], + "score": 1.0, + "content": "convolutional layer. The SSD uses the pixel positions in these maps as anchors of object proposals.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "Each proposal is scored by the SSD to represent a target category, with ”background“ being an", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 422, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 436 + ], + "score": 1.0, + "content": "implicit category that is crucial to exclude irrelevant parts of the input. In addition to these scores,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "the SSD computes a bounding box to localize the predicted object at each anchor. We examine", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "object proposals computed at 1:2 aspect ratio, as they resemble the shape of most traffic lights in the", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 262, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 505, + 274 + ], + "score": 1.0, + "content": "dataset. We visualize the resulting score maps both for the background category and for traffic lights,", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "when feeding a 0-valued input to the SSD. We also visualize the bounding boxes of these proposals", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "in the image space. The SSD predicts the image content to be of background category at all anchor", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "locations, as evident from the value range in both score maps. Such predictions are expected with", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 304, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 320 + ], + "score": 1.0, + "content": "an input that contains no traffic lights. However, the line artifacts in the feature maps have a strong", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "impact on the score maps. These artifacts elevate the likelihood of anchors closer to the top to be", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "classified as background (see the yellow band in the background score map). Conversely, these", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "anchors have significantly lower scores for the traffic light category, compared with other anchors in", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "the feature map. Such difference in the impact on the target categories is due to the different weights", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 360, + 506, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 506, + 373 + ], + "score": 1.0, + "content": "the SSD assigns to the feature maps for each target. As a result, the artifacts lead to potential blind", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 383, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 383, + 383 + ], + "score": 1.0, + "content": "spots in which the scores for certain categories are artificially muted.", + "type": "text", + "cross_page": true + } + ], + "index": 18 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 389, + 505, + 447 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 458, + 502, + 663 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 458, + 502, + 663 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 458, + 502, + 663 + ], + "spans": [ + { + "bbox": [ + 110, + 458, + 502, + 663 + ], + "score": 0.965, + "type": "image", + "image_path": "2f90b37ac2e03a969222bf159083fe214d6629973648c95433a912a506b514b4.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 110, + 458, + 502, + 526.3333333333334 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 110, + 526.3333333333334, + 502, + 594.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 110, + 594.6666666666667, + 502, + 663.0000000000001 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 673, + 505, + 729 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "Figure 4: The formation of blind spots in SSD, illustrated via its box predictor internals with a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 685, + 505, + 697 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 505, + 697 + ], + "score": 1.0, + "content": "zero-valued input. The predictor uses spatial anchors to detect and localize the target object at", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 695, + 505, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 695, + 141, + 706 + ], + "score": 0.89, + "content": "4 5 \\times 8 0", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 695, + 505, + 708 + ], + "score": 1.0, + "content": "possible locations based on 512 feature maps. Certain anchors are predisposed to predict", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 706, + 505, + 719 + ], + "spans": [ + { + "bbox": [ + 105, + 706, + 505, + 719 + ], + "score": 1.0, + "content": "background due to feature-map artifacts, as evident in the logit maps. Traffic lights at the corre-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 717, + 505, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 717, + 505, + 730 + ], + "score": 1.0, + "content": "sponding location cannot be detected as demonstrated with a real scene (middle one in the bottom).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "index": 28.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 81, + 499, + 175 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 81, + 499, + 175 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 81, + 499, + 175 + ], + "spans": [ + { + "bbox": [ + 111, + 81, + 499, + 175 + ], + "score": 0.963, + "type": "image", + "image_path": "855129a5fdeb22a83f45d33356680dd0ee5b663a0bc83617313ee7598eec1429.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 81, + 499, + 112.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 112.33333333333333, + 499, + 143.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 143.66666666666666, + 499, + 175.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 505, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Figure 5: (a) A map showing via color the detection score the SSD computes for a traffic light when", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "present at various locations. The detection is muted when the stimulus lies in the area impacted by", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "the artifacts. (b) The same map after changing the padding method to SYMMETRIC. The detection", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 218, + 480, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 480, + 229 + ], + "score": 1.0, + "content": "scores are rather constant except for periodic variations due to the SSD’s reliance on anchors.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "object proposals computed at 1:2 aspect ratio, as they resemble the shape of most traffic lights in the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 262, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 505, + 274 + ], + "score": 1.0, + "content": "dataset. We visualize the resulting score maps both for the background category and for traffic lights,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "when feeding a 0-valued input to the SSD. We also visualize the bounding boxes of these proposals", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "in the image space. The SSD predicts the image content to be of background category at all anchor", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "locations, as evident from the value range in both score maps. Such predictions are expected with", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 304, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 320 + ], + "score": 1.0, + "content": "an input that contains no traffic lights. However, the line artifacts in the feature maps have a strong", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "impact on the score maps. These artifacts elevate the likelihood of anchors closer to the top to be", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "classified as background (see the yellow band in the background score map). Conversely, these", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "anchors have significantly lower scores for the traffic light category, compared with other anchors in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "the feature map. Such difference in the impact on the target categories is due to the different weights", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 360, + 506, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 506, + 373 + ], + "score": 1.0, + "content": "the SSD assigns to the feature maps for each target. As a result, the artifacts lead to potential blind", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 383, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 383, + 383 + ], + "score": 1.0, + "content": "spots in which the scores for certain categories are artificially muted.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "To validate whether or not the blind spots hinder object detection, we examine road scenes that", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 399, + 504, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 504, + 411 + ], + "score": 1.0, + "content": "contain highly-visible traffic light instances in the impacted area. Figure 4-bottom shows an example", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 358, + 422 + ], + "score": 1.0, + "content": "of such a scene. The SSD computes a low detection score of", + "type": "text" + }, + { + "bbox": [ + 358, + 410, + 373, + 421 + ], + "score": 0.87, + "content": "7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "when the traffic light lies in the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "score": 1.0, + "content": "blind spot (see middle image), far below the detection false-positive cutoff. Shifting the scene image", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "upwards or downwards makes the instance detectable with a high score as long as it lies outside the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 359, + 455 + ], + "score": 1.0, + "content": "blind spot. This explains the failure cases mentioned in Section", + "type": "text" + }, + { + "bbox": [ + 359, + 443, + 369, + 456 + ], + "score": 0.38, + "content": "\\mathbf { \\overline { { \\mathbb { D } } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "To further validate this effect, we", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "run the SSD on baseline images that each contains one traffic light instance at a specific location in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 464, + 507, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 507, + 478 + ], + "score": 1.0, + "content": "the input. We store the detection score for each instance. Figure 5a depicts the computed scores in a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "2D map. It is evident that the model fails to detect the traffic light instance exactly when it is located", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "score": 1.0, + "content": "within the “blind spot” band. The artifacts further disrupt the localization of the objects as evident", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 499, + 495, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 495, + 510 + ], + "score": 1.0, + "content": "in the top-right plot in Figure 4 which shows per-anchor object proposals computed for a 0 input.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 107, + 525, + 383, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 384, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 384, + 540 + ], + "score": 1.0, + "content": "4 REMINDER: WHY IS PADDING NEEDED IN CNNS?", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "title", + "bbox": [ + 106, + 550, + 484, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 484, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 484, + 564 + ], + "score": 1.0, + "content": "Padding is applied at most convolutional layers in CNNs to serve two fundamental purposes:", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 567, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "Maintaining feature map size A padding that satisfies this property is often described as SAME or", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "score": 1.0, + "content": "HALF padding. FULL padding expands the maps by kernel size - 1 along each dimension.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 590, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 602 + ], + "score": 1.0, + "content": "VALID padding performs no padding, eroding the maps by the same amount. SAME padding is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 601, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 505, + 612 + ], + "score": 1.0, + "content": "important to (1) design deep networks that can handle arbitrary input size (a challenge in the presence", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "of gradual erosion), (2) maintain the aspect ratio of non-square input, and (3) concatenate feature", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 622, + 397, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 344, + 634 + ], + "score": 1.0, + "content": "maps from different layers as in Inception [39] and ResNet", + "type": "text" + }, + { + "bbox": [ + 344, + 622, + 361, + 633 + ], + "score": 0.55, + "content": "\\mathbf { \\bar { \\rho } }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 622, + 397, + 634 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 504, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 367, + 650 + ], + "score": 1.0, + "content": "Reducing information bias against the boundary Consider a", + "type": "text" + }, + { + "bbox": [ + 368, + 638, + 388, + 649 + ], + "score": 0.86, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 638, + 504, + 650 + ], + "score": 1.0, + "content": "kernel applied to a 2D input.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "An input location at least 2 pixels away from the boundary contributes to nine local convolution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "operations when computing the feature map. On the other hand, the corner is involved only one time", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "under VALID padding, four times under a 1-pixel SAME 0-padding, and nine times under a 2-pixel", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "FULL 0-padding. With SAME 0-padding, the cumulative contribution differences among the input", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 693, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 505, + 705 + ], + "score": 1.0, + "content": "pixels grow exponentially over the CNN layers. We refer to such uneven treatment of input pixels", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 704, + 453, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 704, + 453, + 716 + ], + "score": 1.0, + "content": "as the foveation behavior of the padding mechanism and elaborate on this in Section 6.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 720, + 428, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 718, + 429, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 429, + 734 + ], + "score": 1.0, + "content": "We next explore solutions to the issues that cause padding to induce spatial bias.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 81, + 499, + 175 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 81, + 499, + 175 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 81, + 499, + 175 + ], + "spans": [ + { + "bbox": [ + 111, + 81, + 499, + 175 + ], + "score": 0.963, + "type": "image", + "image_path": "855129a5fdeb22a83f45d33356680dd0ee5b663a0bc83617313ee7598eec1429.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 81, + 499, + 112.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 112.33333333333333, + 499, + 143.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 143.66666666666666, + 499, + 175.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 505, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Figure 5: (a) A map showing via color the detection score the SSD computes for a traffic light when", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "present at various locations. The detection is muted when the stimulus lies in the area impacted by", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "the artifacts. (b) The same map after changing the padding method to SYMMETRIC. The detection", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 218, + 480, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 480, + 229 + ], + "score": 1.0, + "content": "scores are rather constant except for periodic variations due to the SSD’s reliance on anchors.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 383 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 105, + 250, + 506, + 383 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "To validate whether or not the blind spots hinder object detection, we examine road scenes that", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 399, + 504, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 504, + 411 + ], + "score": 1.0, + "content": "contain highly-visible traffic light instances in the impacted area. Figure 4-bottom shows an example", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 358, + 422 + ], + "score": 1.0, + "content": "of such a scene. The SSD computes a low detection score of", + "type": "text" + }, + { + "bbox": [ + 358, + 410, + 373, + 421 + ], + "score": 0.87, + "content": "7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "when the traffic light lies in the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "score": 1.0, + "content": "blind spot (see middle image), far below the detection false-positive cutoff. Shifting the scene image", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "upwards or downwards makes the instance detectable with a high score as long as it lies outside the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 359, + 455 + ], + "score": 1.0, + "content": "blind spot. This explains the failure cases mentioned in Section", + "type": "text" + }, + { + "bbox": [ + 359, + 443, + 369, + 456 + ], + "score": 0.38, + "content": "\\mathbf { \\overline { { \\mathbb { D } } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "To further validate this effect, we", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "run the SSD on baseline images that each contains one traffic light instance at a specific location in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 464, + 507, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 507, + 478 + ], + "score": 1.0, + "content": "the input. We store the detection score for each instance. Figure 5a depicts the computed scores in a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "2D map. It is evident that the model fails to detect the traffic light instance exactly when it is located", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "score": 1.0, + "content": "within the “blind spot” band. The artifacts further disrupt the localization of the objects as evident", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 499, + 495, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 495, + 510 + ], + "score": 1.0, + "content": "in the top-right plot in Figure 4 which shows per-anchor object proposals computed for a 0 input.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 388, + 507, + 510 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 525, + 383, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 384, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 384, + 540 + ], + "score": 1.0, + "content": "4 REMINDER: WHY IS PADDING NEEDED IN CNNS?", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "title", + "bbox": [ + 106, + 550, + 484, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 484, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 484, + 564 + ], + "score": 1.0, + "content": "Padding is applied at most convolutional layers in CNNs to serve two fundamental purposes:", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 567, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "Maintaining feature map size A padding that satisfies this property is often described as SAME or", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "score": 1.0, + "content": "HALF padding. FULL padding expands the maps by kernel size - 1 along each dimension.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 590, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 602 + ], + "score": 1.0, + "content": "VALID padding performs no padding, eroding the maps by the same amount. SAME padding is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 601, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 505, + 612 + ], + "score": 1.0, + "content": "important to (1) design deep networks that can handle arbitrary input size (a challenge in the presence", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "of gradual erosion), (2) maintain the aspect ratio of non-square input, and (3) concatenate feature", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 622, + 397, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 344, + 634 + ], + "score": 1.0, + "content": "maps from different layers as in Inception [39] and ResNet", + "type": "text" + }, + { + "bbox": [ + 344, + 622, + 361, + 633 + ], + "score": 0.55, + "content": "\\mathbf { \\bar { \\rho } }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 622, + 397, + 634 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 567, + 506, + 634 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 504, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 367, + 650 + ], + "score": 1.0, + "content": "Reducing information bias against the boundary Consider a", + "type": "text" + }, + { + "bbox": [ + 368, + 638, + 388, + 649 + ], + "score": 0.86, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 638, + 504, + 650 + ], + "score": 1.0, + "content": "kernel applied to a 2D input.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "An input location at least 2 pixels away from the boundary contributes to nine local convolution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "operations when computing the feature map. On the other hand, the corner is involved only one time", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "under VALID padding, four times under a 1-pixel SAME 0-padding, and nine times under a 2-pixel", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "FULL 0-padding. With SAME 0-padding, the cumulative contribution differences among the input", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 693, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 505, + 705 + ], + "score": 1.0, + "content": "pixels grow exponentially over the CNN layers. We refer to such uneven treatment of input pixels", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 704, + 453, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 704, + 453, + 716 + ], + "score": 1.0, + "content": "as the foveation behavior of the padding mechanism and elaborate on this in Section 6.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 638, + 506, + 716 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 720, + 428, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 718, + 429, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 429, + 734 + ], + "score": 1.0, + "content": "We next explore solutions to the issues that cause padding to induce spatial bias.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 718, + 429, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 78, + 505, + 234 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 78, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 78, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 108, + 78, + 505, + 234 + ], + "score": 0.974, + "type": "image", + "image_path": "64a4b6b4f56e3df3a1dac55598e3c785be3dbfaa6420ebc60f4670b52e23635c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 78, + 505, + 130.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 130.0, + 505, + 182.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 182.0, + 505, + 234.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 240, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "Figure 6: (a) Illustrating the problem of uneven padding when down-sampling at a stride of 2. The", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 361, + 264 + ], + "score": 1.0, + "content": "padding along x-axis is consumed only at the left side. (b) Mean", + "type": "text" + }, + { + "bbox": [ + 362, + 252, + 383, + 263 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "filters in three ResNet models,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "trained on ImageNet with two input sizes. Color encodes average weight (green is positive). A size", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 272, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 272, + 505, + 288 + ], + "score": 1.0, + "content": "that induces uneven padding (top row) can lead to asymmetries, esp. around down-sampling layers.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 285, + 488, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 488, + 298 + ], + "score": 1.0, + "content": "These asymmetries are mitigated when the input size induces no uneven padding (bottom row).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 317, + 384, + 331 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 385, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 385, + 332 + ], + "score": 1.0, + "content": "5 ELIMINATING UNEVEN APPLICATION OF PADDING", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 337, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 353 + ], + "score": 1.0, + "content": "While useful to reduce bias against the boundary, applying padding at down-sampling layers can", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 281, + 363 + ], + "score": 1.0, + "content": "lead to asymmetry in CNN internals. Figure", + "type": "text" + }, + { + "bbox": [ + 282, + 350, + 294, + 363 + ], + "score": 0.53, + "content": "6 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "illustrates the source of this asymmetry when strided", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "score": 1.0, + "content": "convolution is used for downsampling: At one side of the feature map, the padding is consumed by", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "the kernel while at the other side it is not. To warrant even application of padding throughout the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 254, + 396 + ], + "score": 1.0, + "content": "CNN, the following must hold at all", + "type": "text" + }, + { + "bbox": [ + 254, + 384, + 261, + 393 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 382, + 383, + 396 + ], + "score": 1.0, + "content": "down-sampling layers, where", + "type": "text" + }, + { + "bbox": [ + 383, + 383, + 416, + 394 + ], + "score": 0.88, + "content": "( h _ { i } , w _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 382, + 506, + 396 + ], + "score": 1.0, + "content": "is the output shape at", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 391, + 507, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 391, + 182, + 409 + ], + "score": 1.0, + "content": "the i-th layer with", + "type": "text" + }, + { + "bbox": [ + 182, + 393, + 219, + 406 + ], + "score": 0.92, + "content": "\\overline { { k } } _ { i } ^ { h } \\times k _ { i } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 391, + 281, + 409 + ], + "score": 1.0, + "content": "as kernel size,", + "type": "text" + }, + { + "bbox": [ + 282, + 393, + 315, + 406 + ], + "score": 0.92, + "content": "( s _ { i } ^ { h } , s _ { i } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 391, + 377, + 409 + ], + "score": 1.0, + "content": "as strides, and", + "type": "text" + }, + { + "bbox": [ + 377, + 394, + 424, + 406 + ], + "score": 0.91, + "content": "\\mathbf { \\Sigma } = ( p _ { i } ^ { h } , p _ { i } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 391, + 507, + 409 + ], + "score": 1.0, + "content": "as padding amount", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 404, + 242, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 179, + 418 + ], + "score": 1.0, + "content": "(refer to appendix", + "type": "text" + }, + { + "bbox": [ + 179, + 406, + 191, + 418 + ], + "score": 0.58, + "content": "\\boxed { \\mathrm { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 404, + 242, + 418 + ], + "score": 1.0, + "content": "for a proof):", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 422, + 489, + 438 + ], + "lines": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "spans": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "score": 0.92, + "content": "\\forall i \\in \\{ 1 , \\ldots , d \\} : h _ { i - 1 } = s _ { i } ^ { h } \\cdot ( h _ { i } - 1 ) + k _ { i } ^ { h } - 2 \\cdot p _ { i } ^ { h } \\quad \\wedge \\quad w _ { i - 1 } = s _ { i } ^ { w } \\cdot ( w _ { i } - 1 ) + k _ { i } ^ { w } - 2 \\cdot p _ { i } ^ { w } \\quad .", + "type": "interline_equation", + "image_path": "440b1a1d2d31299d8b44bc2e556555e7b2c2d7ed4e01ad4a9fc301cffdc84cf8.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 153, + 456 + ], + "score": 1.0, + "content": "The values", + "type": "text" + }, + { + "bbox": [ + 153, + 444, + 165, + 455 + ], + "score": 0.88, + "content": "h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 444, + 183, + 456 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 184, + 446, + 196, + 455 + ], + "score": 0.86, + "content": "w _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 444, + 505, + 456 + ], + "score": 1.0, + "content": "represent the CNN input dimensions. The above constraints are not always", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 453, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 468 + ], + "score": 1.0, + "content": "satisfied during training or inference with arbitrary input dimensions. For example, ImageNet clas-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 207, + 478 + ], + "score": 1.0, + "content": "sifiers based on ResNet", + "type": "text" + }, + { + "bbox": [ + 208, + 465, + 226, + 477 + ], + "score": 0.79, + "content": "\\pmb { \\mathbb { I } } \\pmb { \\mathcal { 2 } } \\Vert", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 466, + 291, + 478 + ], + "score": 1.0, + "content": "and MobileNet", + "type": "text" + }, + { + "bbox": [ + 291, + 465, + 309, + 477 + ], + "score": 0.78, + "content": "\\pmb { \\mathbb { I } } \\pmb { \\overbrace { 3 } } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 466, + 455, + 478 + ], + "score": 1.0, + "content": "contain five down-sampling layers", + "type": "text" + }, + { + "bbox": [ + 455, + 466, + 483, + 476 + ], + "score": 0.88, + "content": "\\mathit { a } = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 466, + 505, + 478 + ], + "score": 1.0, + "content": ") that", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 475, + 507, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 507, + 489 + ], + "score": 1.0, + "content": "apply 1-pixel 0-padding before performing 2-strided convolution. To avoid uneven application of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 487, + 463, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 463, + 501 + ], + "score": 1.0, + "content": "padding, the input to these CNNs must satisfy the following, as explained in appendix A:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 504, + 490, + 520 + ], + "lines": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "spans": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "score": 0.89, + "content": "h _ { 0 } = a _ { 1 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 1 } + 1 \\quad \\mathrm { a n d } \\quad w _ { 0 } = a _ { 2 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 2 } + 1 \\quad \\mathrm { w h e r e } \\quad a _ { 1 } , a _ { 2 } \\in \\mathbb { N } ^ { + }", + "type": "interline_equation", + "image_path": "88b26002b6ffe50d6fd4677e7bc4524ce5aa8f38e3573e166b6a758a9313ea20.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 169, + 545 + ], + "score": 1.0, + "content": "The traditional", + "type": "text" + }, + { + "bbox": [ + 169, + 530, + 178, + 545 + ], + "score": 0.8, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 531, + 417, + 545 + ], + "score": 1.0, + "content": "and prevalent input size for training ImageNet models is", + "type": "text" + }, + { + "bbox": [ + 418, + 532, + 457, + 543 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 531, + 506, + 545 + ], + "score": 1.0, + "content": ". This size", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 543, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 556 + ], + "score": 1.0, + "content": "violates Eq. 2, leading to uneven padding at every down-sampling layer in ResNet and MobileNet", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "models where 0-padding is effectively applied only at the left and top sides of layer input. This", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 299, + 578 + ], + "score": 1.0, + "content": "over-represents zeros at the top and left sides of", + "type": "text" + }, + { + "bbox": [ + 299, + 566, + 323, + 576 + ], + "score": 0.91, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "feature-map patches the filters are convolved", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 280, + 589 + ], + "score": 1.0, + "content": "with during training. The top row of Figure", + "type": "text" + }, + { + "bbox": [ + 281, + 576, + 294, + 588 + ], + "score": 0.79, + "content": "6 { \\mathsf { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "shows per-layer mean filters in three ResNet models", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 154, + 600 + ], + "score": 1.0, + "content": "in PyTorch", + "type": "text" + }, + { + "bbox": [ + 154, + 587, + 172, + 599 + ], + "score": 0.69, + "content": "\\mathbb { \\lVert 3 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 587, + 298, + 600 + ], + "score": 1.0, + "content": ", pre-trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 299, + 587, + 338, + 598 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "images. In all of these models, a few of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 598, + 489, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 489, + 611 + ], + "score": 1.0, + "content": "the mean filters, adjacent to down-sampling layers, exhibit stark asymmetry about their centers.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 231, + 625 + ], + "score": 1.0, + "content": "We increase the image size to", + "type": "text" + }, + { + "bbox": [ + 231, + 613, + 271, + 624 + ], + "score": 0.89, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 612, + 505, + 625 + ], + "score": 1.0, + "content": "without introducing additional image information2. This", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "size satisfies Eq. 2, warranting even application of padding at every downsampling layer in the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "above models. Retraining the models with this size strongly reduces this asymmetry as evident in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 205, + 658 + ], + "score": 1.0, + "content": "the bottom row of Figure", + "type": "text" + }, + { + "bbox": [ + 205, + 646, + 218, + 658 + ], + "score": 0.39, + "content": "\\bar { 6 } 6", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 646, + 505, + 658 + ], + "score": 1.0, + "content": ". This, in turn, visibly boosts the accuracy in all models we experimented", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 657, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 506, + 669 + ], + "score": 1.0, + "content": "with as we report in Table 1. The accuracy did not improve further when we retrained two of the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 668, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 263, + 680 + ], + "score": 1.0, + "content": "models, ResNet-18 and ResNet-34, on", + "type": "text" + }, + { + "bbox": [ + 264, + 668, + 308, + 678 + ], + "score": 0.89, + "content": "2 2 6 \\times 2 2 6", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 668, + 506, + 680 + ], + "score": 1.0, + "content": "images. This provides evidence that the boost is", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 679, + 437, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 679, + 437, + 692 + ], + "score": 1.0, + "content": "due to eliminating uneven padding and not merely due to increasing the input size.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 700, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 117, + 698, + 488, + 713 + ], + "spans": [ + { + "bbox": [ + 117, + 698, + 488, + 713 + ], + "score": 1.0, + "content": "1 This size has been used to facilitate model comparison on ImageNet, since the inception of AlexNet.", + "type": "text" + } + ] + }, + { + "bbox": [ + 117, + 708, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 117, + 708, + 506, + 725 + ], + "score": 1.0, + "content": "2 This is done via constant padding. The side to pad with one pixel is chosen at random to balance out the", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 721, + 493, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 493, + 734 + ], + "score": 1.0, + "content": "application of padding at both sides over the training set. No additional padding is applied at further layers.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 78, + 505, + 234 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 78, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 78, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 108, + 78, + 505, + 234 + ], + "score": 0.974, + "type": "image", + "image_path": "64a4b6b4f56e3df3a1dac55598e3c785be3dbfaa6420ebc60f4670b52e23635c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 78, + 505, + 130.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 130.0, + 505, + 182.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 182.0, + 505, + 234.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 240, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "Figure 6: (a) Illustrating the problem of uneven padding when down-sampling at a stride of 2. The", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 361, + 264 + ], + "score": 1.0, + "content": "padding along x-axis is consumed only at the left side. (b) Mean", + "type": "text" + }, + { + "bbox": [ + 362, + 252, + 383, + 263 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "filters in three ResNet models,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "trained on ImageNet with two input sizes. Color encodes average weight (green is positive). A size", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 272, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 272, + 505, + 288 + ], + "score": 1.0, + "content": "that induces uneven padding (top row) can lead to asymmetries, esp. around down-sampling layers.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 285, + 488, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 488, + 298 + ], + "score": 1.0, + "content": "These asymmetries are mitigated when the input size induces no uneven padding (bottom row).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "title", + "bbox": [ + 107, + 317, + 384, + 331 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 385, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 385, + 332 + ], + "score": 1.0, + "content": "5 ELIMINATING UNEVEN APPLICATION OF PADDING", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 337, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 353 + ], + "score": 1.0, + "content": "While useful to reduce bias against the boundary, applying padding at down-sampling layers can", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 281, + 363 + ], + "score": 1.0, + "content": "lead to asymmetry in CNN internals. Figure", + "type": "text" + }, + { + "bbox": [ + 282, + 350, + 294, + 363 + ], + "score": 0.53, + "content": "6 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "illustrates the source of this asymmetry when strided", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "score": 1.0, + "content": "convolution is used for downsampling: At one side of the feature map, the padding is consumed by", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "the kernel while at the other side it is not. To warrant even application of padding throughout the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 254, + 396 + ], + "score": 1.0, + "content": "CNN, the following must hold at all", + "type": "text" + }, + { + "bbox": [ + 254, + 384, + 261, + 393 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 382, + 383, + 396 + ], + "score": 1.0, + "content": "down-sampling layers, where", + "type": "text" + }, + { + "bbox": [ + 383, + 383, + 416, + 394 + ], + "score": 0.88, + "content": "( h _ { i } , w _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 382, + 506, + 396 + ], + "score": 1.0, + "content": "is the output shape at", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 391, + 507, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 391, + 182, + 409 + ], + "score": 1.0, + "content": "the i-th layer with", + "type": "text" + }, + { + "bbox": [ + 182, + 393, + 219, + 406 + ], + "score": 0.92, + "content": "\\overline { { k } } _ { i } ^ { h } \\times k _ { i } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 391, + 281, + 409 + ], + "score": 1.0, + "content": "as kernel size,", + "type": "text" + }, + { + "bbox": [ + 282, + 393, + 315, + 406 + ], + "score": 0.92, + "content": "( s _ { i } ^ { h } , s _ { i } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 391, + 377, + 409 + ], + "score": 1.0, + "content": "as strides, and", + "type": "text" + }, + { + "bbox": [ + 377, + 394, + 424, + 406 + ], + "score": 0.91, + "content": "\\mathbf { \\Sigma } = ( p _ { i } ^ { h } , p _ { i } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 391, + 507, + 409 + ], + "score": 1.0, + "content": "as padding amount", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 404, + 242, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 179, + 418 + ], + "score": 1.0, + "content": "(refer to appendix", + "type": "text" + }, + { + "bbox": [ + 179, + 406, + 191, + 418 + ], + "score": 0.58, + "content": "\\boxed { \\mathrm { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 404, + 242, + 418 + ], + "score": 1.0, + "content": "for a proof):", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 337, + 507, + 418 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 422, + 489, + 438 + ], + "lines": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "spans": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "score": 0.92, + "content": "\\forall i \\in \\{ 1 , \\ldots , d \\} : h _ { i - 1 } = s _ { i } ^ { h } \\cdot ( h _ { i } - 1 ) + k _ { i } ^ { h } - 2 \\cdot p _ { i } ^ { h } \\quad \\wedge \\quad w _ { i - 1 } = s _ { i } ^ { w } \\cdot ( w _ { i } - 1 ) + k _ { i } ^ { w } - 2 \\cdot p _ { i } ^ { w } \\quad .", + "type": "interline_equation", + "image_path": "440b1a1d2d31299d8b44bc2e556555e7b2c2d7ed4e01ad4a9fc301cffdc84cf8.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 110, + 422, + 489, + 438 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 153, + 456 + ], + "score": 1.0, + "content": "The values", + "type": "text" + }, + { + "bbox": [ + 153, + 444, + 165, + 455 + ], + "score": 0.88, + "content": "h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 444, + 183, + 456 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 184, + 446, + 196, + 455 + ], + "score": 0.86, + "content": "w _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 444, + 505, + 456 + ], + "score": 1.0, + "content": "represent the CNN input dimensions. The above constraints are not always", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 453, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 468 + ], + "score": 1.0, + "content": "satisfied during training or inference with arbitrary input dimensions. For example, ImageNet clas-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 207, + 478 + ], + "score": 1.0, + "content": "sifiers based on ResNet", + "type": "text" + }, + { + "bbox": [ + 208, + 465, + 226, + 477 + ], + "score": 0.79, + "content": "\\pmb { \\mathbb { I } } \\pmb { \\mathcal { 2 } } \\Vert", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 466, + 291, + 478 + ], + "score": 1.0, + "content": "and MobileNet", + "type": "text" + }, + { + "bbox": [ + 291, + 465, + 309, + 477 + ], + "score": 0.78, + "content": "\\pmb { \\mathbb { I } } \\pmb { \\overbrace { 3 } } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 466, + 455, + 478 + ], + "score": 1.0, + "content": "contain five down-sampling layers", + "type": "text" + }, + { + "bbox": [ + 455, + 466, + 483, + 476 + ], + "score": 0.88, + "content": "\\mathit { a } = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 466, + 505, + 478 + ], + "score": 1.0, + "content": ") that", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 475, + 507, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 507, + 489 + ], + "score": 1.0, + "content": "apply 1-pixel 0-padding before performing 2-strided convolution. To avoid uneven application of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 487, + 463, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 463, + 501 + ], + "score": 1.0, + "content": "padding, the input to these CNNs must satisfy the following, as explained in appendix A:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 444, + 507, + 501 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 504, + 490, + 520 + ], + "lines": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "spans": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "score": 0.89, + "content": "h _ { 0 } = a _ { 1 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 1 } + 1 \\quad \\mathrm { a n d } \\quad w _ { 0 } = a _ { 2 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 2 } + 1 \\quad \\mathrm { w h e r e } \\quad a _ { 1 } , a _ { 2 } \\in \\mathbb { N } ^ { + }", + "type": "interline_equation", + "image_path": "88b26002b6ffe50d6fd4677e7bc4524ce5aa8f38e3573e166b6a758a9313ea20.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 110, + 504, + 490, + 520 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 169, + 545 + ], + "score": 1.0, + "content": "The traditional", + "type": "text" + }, + { + "bbox": [ + 169, + 530, + 178, + 545 + ], + "score": 0.8, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 531, + 417, + 545 + ], + "score": 1.0, + "content": "and prevalent input size for training ImageNet models is", + "type": "text" + }, + { + "bbox": [ + 418, + 532, + 457, + 543 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 531, + 506, + 545 + ], + "score": 1.0, + "content": ". This size", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 543, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 556 + ], + "score": 1.0, + "content": "violates Eq. 2, leading to uneven padding at every down-sampling layer in ResNet and MobileNet", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "models where 0-padding is effectively applied only at the left and top sides of layer input. This", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 299, + 578 + ], + "score": 1.0, + "content": "over-represents zeros at the top and left sides of", + "type": "text" + }, + { + "bbox": [ + 299, + 566, + 323, + 576 + ], + "score": 0.91, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "feature-map patches the filters are convolved", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 280, + 589 + ], + "score": 1.0, + "content": "with during training. The top row of Figure", + "type": "text" + }, + { + "bbox": [ + 281, + 576, + 294, + 588 + ], + "score": 0.79, + "content": "6 { \\mathsf { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "shows per-layer mean filters in three ResNet models", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 154, + 600 + ], + "score": 1.0, + "content": "in PyTorch", + "type": "text" + }, + { + "bbox": [ + 154, + 587, + 172, + 599 + ], + "score": 0.69, + "content": "\\mathbb { \\lVert 3 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 587, + 298, + 600 + ], + "score": 1.0, + "content": ", pre-trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 299, + 587, + 338, + 598 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "images. In all of these models, a few of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 598, + 489, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 489, + 611 + ], + "score": 1.0, + "content": "the mean filters, adjacent to down-sampling layers, exhibit stark asymmetry about their centers.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 530, + 506, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 231, + 625 + ], + "score": 1.0, + "content": "We increase the image size to", + "type": "text" + }, + { + "bbox": [ + 231, + 613, + 271, + 624 + ], + "score": 0.89, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 612, + 505, + 625 + ], + "score": 1.0, + "content": "without introducing additional image information2. This", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "size satisfies Eq. 2, warranting even application of padding at every downsampling layer in the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "above models. Retraining the models with this size strongly reduces this asymmetry as evident in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 205, + 658 + ], + "score": 1.0, + "content": "the bottom row of Figure", + "type": "text" + }, + { + "bbox": [ + 205, + 646, + 218, + 658 + ], + "score": 0.39, + "content": "\\bar { 6 } 6", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 646, + 505, + 658 + ], + "score": 1.0, + "content": ". This, in turn, visibly boosts the accuracy in all models we experimented", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 657, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 506, + 669 + ], + "score": 1.0, + "content": "with as we report in Table 1. The accuracy did not improve further when we retrained two of the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 668, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 263, + 680 + ], + "score": 1.0, + "content": "models, ResNet-18 and ResNet-34, on", + "type": "text" + }, + { + "bbox": [ + 264, + 668, + 308, + 678 + ], + "score": 0.89, + "content": "2 2 6 \\times 2 2 6", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 668, + 506, + 680 + ], + "score": 1.0, + "content": "images. This provides evidence that the boost is", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 679, + 437, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 679, + 437, + 692 + ], + "score": 1.0, + "content": "due to eliminating uneven padding and not merely due to increasing the input size.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 612, + 506, + 692 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Replacing 0-padding with a padding method that reuses feature map values can alleviate the asym-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "metry in the learned filters in the presence of unevenly applied padding. Another possibility is to use", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 461, + 117 + ], + "score": 1.0, + "content": "a rigid downsampling kernel, such as max-pooling, instead of a learned one. Appendix", + "type": "text" + }, + { + "bbox": [ + 461, + 104, + 472, + 117 + ], + "score": 0.66, + "content": "\\boxed { \\dot { \\mathbf { C } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "demon-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 387, + 128 + ], + "score": 1.0, + "content": "strates both possibilities. Finally, antialiasing before downsampling", + "type": "text" + }, + { + "bbox": [ + 387, + 115, + 405, + 127 + ], + "score": 0.42, + "content": "\\mathbb { \\lVert \\boldsymbol { 4 3 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "can strongly reduce the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 346, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 346, + 140 + ], + "score": 1.0, + "content": "asymmetry as we elaborate in Section 8 and in Appendix E.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table", + "bbox": [ + 113, + 167, + 497, + 214 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 146, + 501, + 159 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 145, + 501, + 161 + ], + "spans": [ + { + "bbox": [ + 108, + 145, + 501, + 161 + ], + "score": 1.0, + "content": "Table 1: Top-1 (and top-5) accuracy of five ImageNet classifiers trained with different input sizes.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 167, + 497, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 167, + 497, + 214 + ], + "spans": [ + { + "bbox": [ + 113, + 167, + 497, + 214 + ], + "score": 0.97, + "html": "
Input Size ²MobileNetResNet-18ResNet-34ResNet-50ResNet-101
224×22468.19 (88.44)69.93 (89.22)73.30 (91.42)75.65 (92.47)77.37 (93.56)
225×22568.80 (88.78)70.27 (89.52)73.72 (91.58)76.01 (92.90)77.67 (93.81)
", + "type": "table", + "image_path": "1139c80765c9433cc09b53b2ba194f3b7aee4b315e6a05560053d3f686007e5b.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 113, + 167, + 497, + 182.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 182.66666666666666, + 497, + 198.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 198.33333333333331, + 497, + 213.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 245, + 237 + ], + "score": 1.0, + "content": "Even when no padding is applied", + "type": "text" + }, + { + "bbox": [ + 245, + 224, + 278, + 237 + ], + "score": 0.91, + "content": "( p _ { i } ^ { h } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 223, + 290, + 237 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 290, + 225, + 324, + 237 + ], + "score": 0.91, + "content": "p _ { i } ^ { w } = 0 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "), an input size that does no satisfy Eq. 1 can", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "score": 1.0, + "content": "lead to uneven erosion of feature maps, in turn, reducing the contribution of pixels from the impacted", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 245, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 130, + 260 + ], + "score": 1.0, + "content": "sides", + "type": "text" + }, + { + "bbox": [ + 131, + 246, + 160, + 259 + ], + "score": 0.39, + "content": "( { \\mathrm { F i g ~ } } 7 { \\mathrm { \\rho } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 245, + 209, + 260 + ], + "score": 1.0, + "content": ". Satisfying", + "type": "text" + }, + { + "bbox": [ + 210, + 246, + 232, + 259 + ], + "score": 0.61, + "content": "\\mathrm { E q } \\ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 245, + 506, + 260 + ], + "score": 1.0, + "content": "imposes a restriction on input size, e.g., to values in increments of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 255, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 141, + 268 + ], + "score": 0.9, + "content": "2 ^ { d } = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 255, + 235, + 272 + ], + "score": 1.0, + "content": "with the above models", + "type": "text" + }, + { + "bbox": [ + 235, + 258, + 277, + 269 + ], + "score": 0.79, + "content": "{ \\mathrm { 1 9 3 \\times 1 9 3 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 255, + 280, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 281, + 258, + 321, + 269 + ], + "score": 0.82, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 255, + 325, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 325, + 258, + 366, + 269 + ], + "score": 0.7, + "content": "2 5 7 \\times 2 5 7", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 255, + 506, + 272 + ], + "score": 1.0, + "content": ", ...). Depending on the application", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "score": 1.0, + "content": "domain, this can be guaranteed either by resizing an input to the closest increment, or by padding it", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 280, + 232, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 232, + 291 + ], + "score": 1.0, + "content": "accordingly with suited values.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 107, + 306, + 331, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 305, + 332, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 332, + 321 + ], + "score": 1.0, + "content": "6 PADDING MECHANISM AND FOVEATION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "By foveation we mean the unequal involvement of input pixels in convolutional operations through-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "out the CNN. Padding plays a fundamental role in the foveation behavior of CNNs. We visualize this", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "score": 1.0, + "content": "behavior by means of a foveation map that counts for each input pixel the number of convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 364, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 505, + 377 + ], + "score": 1.0, + "content": "paths through which it can propagate information to the CNN output. We obtain these counts by", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 261, + 388 + ], + "score": 1.0, + "content": "computing the effective receptive field", + "type": "text" + }, + { + "bbox": [ + 261, + 375, + 279, + 386 + ], + "score": 0.46, + "content": "\\pmb { \\pmb { 2 8 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "for the sum of the final convolutional layer after assign-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "ing all weights in the network to 1 (code in supplemental). Neutralizing the weights is essential to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 397, + 425, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 425, + 409 + ], + "score": 1.0, + "content": "obtain per-pixel counts of input-output paths that reflect the foveation behavior.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 107, + 416, + 503, + 551 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 416, + 503, + 551 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 416, + 503, + 551 + ], + "spans": [ + { + "bbox": [ + 107, + 416, + 503, + 551 + ], + "score": 0.973, + "type": "image", + "image_path": "d463a723cb881cdd5e445297eabd8a8780cc37159812695efe551818975b2e04.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 107, + 416, + 503, + 461.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 107, + 461.0, + 503, + 506.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 107, + 506.0, + 503, + 551.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 560, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "Figure 7: Foveation behavior of different padding methods applied to VGG-19 [37], and illustrated", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 124, + 584 + ], + "score": 1.0, + "content": "in a", + "type": "text" + }, + { + "bbox": [ + 125, + 572, + 169, + 582 + ], + "score": 0.89, + "content": "5 1 2 \\times 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 571, + 506, + 584 + ], + "score": 1.0, + "content": "input space (unless otherwise stated). Color represents the number of paths to the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "score": 1.0, + "content": "output for each input pixel. (a) The difference between VALID, FULL, and SAME 0-padding. (b)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 505, + 606 + ], + "score": 1.0, + "content": "SAME alternatives to 0-padding. (c) Dilation amplifies foveation of SAME 0-padding. (d) Strides", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "can lead to checkerboard patterns. (e) Foveation effects are more extensive in smaller inputs (relative", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 307, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 307, + 628 + ], + "score": 1.0, + "content": "to input size) and are sensitive to uneven padding.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + } + ], + "index": 26.25 + }, + { + "type": "text", + "bbox": [ + 106, + 642, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 134, + 656 + ], + "score": 1.0, + "content": "Figure", + "type": "text" + }, + { + "bbox": [ + 134, + 643, + 147, + 656 + ], + "score": 0.29, + "content": "7 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "shows the extensive foveation effect when no padding is applied. The diminishing con-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "tribution of vast areas of the input explains the drastic drop in accuracy recently observed under", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 173, + 678 + ], + "score": 1.0, + "content": "VALID padding", + "type": "text" + }, + { + "bbox": [ + 174, + 665, + 191, + 677 + ], + "score": 0.29, + "content": "\\mathbb { \\lVert \\boldsymbol { 1 6 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 666, + 505, + 678 + ], + "score": 1.0, + "content": ". In contrast, FULL 0-padding does not incur foveation, however, at the cost of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 465, + 689 + ], + "score": 1.0, + "content": "increasing the output size after each layer, making it impractical as explained in Section", + "type": "text" + }, + { + "bbox": [ + 465, + 676, + 476, + 690 + ], + "score": 0.45, + "content": "\\mathbb { H }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 678, + 505, + 688 + ], + "score": 1.0, + "content": "SAME", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 504, + 699 + ], + "score": 1.0, + "content": "0-padding incurs moderate foveation at the periphery, whose absolute extent depends on the num-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "ber of convolutional layers and their filter sizes. Its relative extent depends on the input size: the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "larger the input, the larger the ratio of the constant area in yellow (refer to appendix B for a detailed", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 721, + 149, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 149, + 733 + ], + "score": 1.0, + "content": "example).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Replacing 0-padding with a padding method that reuses feature map values can alleviate the asym-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "metry in the learned filters in the presence of unevenly applied padding. Another possibility is to use", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 461, + 117 + ], + "score": 1.0, + "content": "a rigid downsampling kernel, such as max-pooling, instead of a learned one. Appendix", + "type": "text" + }, + { + "bbox": [ + 461, + 104, + 472, + 117 + ], + "score": 0.66, + "content": "\\boxed { \\dot { \\mathbf { C } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "demon-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 387, + 128 + ], + "score": 1.0, + "content": "strates both possibilities. Finally, antialiasing before downsampling", + "type": "text" + }, + { + "bbox": [ + 387, + 115, + 405, + 127 + ], + "score": 0.42, + "content": "\\mathbb { \\lVert \\boldsymbol { 4 3 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "can strongly reduce the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 346, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 346, + 140 + ], + "score": 1.0, + "content": "asymmetry as we elaborate in Section 8 and in Appendix E.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 82, + 505, + 140 + ] + }, + { + "type": "table", + "bbox": [ + 113, + 167, + 497, + 214 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 146, + 501, + 159 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 145, + 501, + 161 + ], + "spans": [ + { + "bbox": [ + 108, + 145, + 501, + 161 + ], + "score": 1.0, + "content": "Table 1: Top-1 (and top-5) accuracy of five ImageNet classifiers trained with different input sizes.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 167, + 497, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 167, + 497, + 214 + ], + "spans": [ + { + "bbox": [ + 113, + 167, + 497, + 214 + ], + "score": 0.97, + "html": "
Input Size ²MobileNetResNet-18ResNet-34ResNet-50ResNet-101
224×22468.19 (88.44)69.93 (89.22)73.30 (91.42)75.65 (92.47)77.37 (93.56)
225×22568.80 (88.78)70.27 (89.52)73.72 (91.58)76.01 (92.90)77.67 (93.81)
", + "type": "table", + "image_path": "1139c80765c9433cc09b53b2ba194f3b7aee4b315e6a05560053d3f686007e5b.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 113, + 167, + 497, + 182.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 182.66666666666666, + 497, + 198.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 198.33333333333331, + 497, + 213.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 245, + 237 + ], + "score": 1.0, + "content": "Even when no padding is applied", + "type": "text" + }, + { + "bbox": [ + 245, + 224, + 278, + 237 + ], + "score": 0.91, + "content": "( p _ { i } ^ { h } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 223, + 290, + 237 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 290, + 225, + 324, + 237 + ], + "score": 0.91, + "content": "p _ { i } ^ { w } = 0 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "), an input size that does no satisfy Eq. 1 can", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 249 + ], + "score": 1.0, + "content": "lead to uneven erosion of feature maps, in turn, reducing the contribution of pixels from the impacted", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 245, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 130, + 260 + ], + "score": 1.0, + "content": "sides", + "type": "text" + }, + { + "bbox": [ + 131, + 246, + 160, + 259 + ], + "score": 0.39, + "content": "( { \\mathrm { F i g ~ } } 7 { \\mathrm { \\rho } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 245, + 209, + 260 + ], + "score": 1.0, + "content": ". Satisfying", + "type": "text" + }, + { + "bbox": [ + 210, + 246, + 232, + 259 + ], + "score": 0.61, + "content": "\\mathrm { E q } \\ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 245, + 506, + 260 + ], + "score": 1.0, + "content": "imposes a restriction on input size, e.g., to values in increments of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 255, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 141, + 268 + ], + "score": 0.9, + "content": "2 ^ { d } = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 255, + 235, + 272 + ], + "score": 1.0, + "content": "with the above models", + "type": "text" + }, + { + "bbox": [ + 235, + 258, + 277, + 269 + ], + "score": 0.79, + "content": "{ \\mathrm { 1 9 3 \\times 1 9 3 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 255, + 280, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 281, + 258, + 321, + 269 + ], + "score": 0.82, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 255, + 325, + 272 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 325, + 258, + 366, + 269 + ], + "score": 0.7, + "content": "2 5 7 \\times 2 5 7", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 255, + 506, + 272 + ], + "score": 1.0, + "content": ", ...). Depending on the application", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 281 + ], + "score": 1.0, + "content": "domain, this can be guaranteed either by resizing an input to the closest increment, or by padding it", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 280, + 232, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 232, + 291 + ], + "score": 1.0, + "content": "accordingly with suited values.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5, + "bbox_fs": [ + 104, + 223, + 506, + 291 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 306, + 331, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 305, + 332, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 332, + 321 + ], + "score": 1.0, + "content": "6 PADDING MECHANISM AND FOVEATION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "By foveation we mean the unequal involvement of input pixels in convolutional operations through-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "out the CNN. Padding plays a fundamental role in the foveation behavior of CNNs. We visualize this", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "score": 1.0, + "content": "behavior by means of a foveation map that counts for each input pixel the number of convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 364, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 505, + 377 + ], + "score": 1.0, + "content": "paths through which it can propagate information to the CNN output. We obtain these counts by", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 261, + 388 + ], + "score": 1.0, + "content": "computing the effective receptive field", + "type": "text" + }, + { + "bbox": [ + 261, + 375, + 279, + 386 + ], + "score": 0.46, + "content": "\\pmb { \\pmb { 2 8 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "for the sum of the final convolutional layer after assign-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "ing all weights in the network to 1 (code in supplemental). Neutralizing the weights is essential to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 397, + 425, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 397, + 425, + 409 + ], + "score": 1.0, + "content": "obtain per-pixel counts of input-output paths that reflect the foveation behavior.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 331, + 506, + 409 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 416, + 503, + 551 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 416, + 503, + 551 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 416, + 503, + 551 + ], + "spans": [ + { + "bbox": [ + 107, + 416, + 503, + 551 + ], + "score": 0.973, + "type": "image", + "image_path": "d463a723cb881cdd5e445297eabd8a8780cc37159812695efe551818975b2e04.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 107, + 416, + 503, + 461.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 107, + 461.0, + 503, + 506.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 107, + 506.0, + 503, + 551.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 560, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "Figure 7: Foveation behavior of different padding methods applied to VGG-19 [37], and illustrated", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 124, + 584 + ], + "score": 1.0, + "content": "in a", + "type": "text" + }, + { + "bbox": [ + 125, + 572, + 169, + 582 + ], + "score": 0.89, + "content": "5 1 2 \\times 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 571, + 506, + 584 + ], + "score": 1.0, + "content": "input space (unless otherwise stated). Color represents the number of paths to the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 596 + ], + "score": 1.0, + "content": "output for each input pixel. (a) The difference between VALID, FULL, and SAME 0-padding. (b)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 505, + 606 + ], + "score": 1.0, + "content": "SAME alternatives to 0-padding. (c) Dilation amplifies foveation of SAME 0-padding. (d) Strides", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "can lead to checkerboard patterns. (e) Foveation effects are more extensive in smaller inputs (relative", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 307, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 307, + 628 + ], + "score": 1.0, + "content": "to input size) and are sensitive to uneven padding.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + } + ], + "index": 26.25 + }, + { + "type": "text", + "bbox": [ + 106, + 642, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 134, + 656 + ], + "score": 1.0, + "content": "Figure", + "type": "text" + }, + { + "bbox": [ + 134, + 643, + 147, + 656 + ], + "score": 0.29, + "content": "7 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "shows the extensive foveation effect when no padding is applied. The diminishing con-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "tribution of vast areas of the input explains the drastic drop in accuracy recently observed under", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 173, + 678 + ], + "score": 1.0, + "content": "VALID padding", + "type": "text" + }, + { + "bbox": [ + 174, + 665, + 191, + 677 + ], + "score": 0.29, + "content": "\\mathbb { \\lVert \\boldsymbol { 1 6 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 666, + 505, + 678 + ], + "score": 1.0, + "content": ". In contrast, FULL 0-padding does not incur foveation, however, at the cost of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 465, + 689 + ], + "score": 1.0, + "content": "increasing the output size after each layer, making it impractical as explained in Section", + "type": "text" + }, + { + "bbox": [ + 465, + 676, + 476, + 690 + ], + "score": 0.45, + "content": "\\mathbb { H }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 678, + 505, + 688 + ], + "score": 1.0, + "content": "SAME", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 504, + 699 + ], + "score": 1.0, + "content": "0-padding incurs moderate foveation at the periphery, whose absolute extent depends on the num-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "ber of convolutional layers and their filter sizes. Its relative extent depends on the input size: the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "larger the input, the larger the ratio of the constant area in yellow (refer to appendix B for a detailed", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 721, + 149, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 149, + 733 + ], + "score": 1.0, + "content": "example).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 643, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Figure 7b shows the foveation behavior of alternatives to SAME 0-padding that have roots in wavelet", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 254, + 106 + ], + "score": 1.0, + "content": "analysis [19] and image processing", + "type": "text" + }, + { + "bbox": [ + 254, + 93, + 272, + 105 + ], + "score": 0.33, + "content": " { \\mathbb { E } } { \\ b { \\mathbb { Z } } } ] ", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 92, + 505, + 106 + ], + "score": 1.0, + "content": ". Mirror padding mirrors pixels at the boundary to fill", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "the padding area. When the border is included (SYMMETRIC mode in TensorFlow) all input pixels", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 297, + 128 + ], + "score": 1.0, + "content": "have an equal number of input-output paths 3,", + "type": "text" + }, + { + "bbox": [ + 299, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "resulting in a uniform foveation map. When the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "border is not included (REFLECT mode both in PyTorch and in TensorFlow), the map exhibits bias", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "against the border and towards a contour in its proximity. This bias is amplified over multiple layers.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "Replication padding exhibits the opposite bias when the padding area is wider than 1 pixel. This", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "is because it replicates the outer 1-pixel border multiple times to fill this area 3. The method is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 169, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 506, + 184 + ], + "score": 1.0, + "content": "equivalent to SYMMETRIC if the padding area is 1-pixel wide. Circular padding wraps opposing", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 506, + 194 + ], + "score": 1.0, + "content": "borders, enabling the kernels to seamlessly operate on the boundary and resulting in a uniform", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 219, + 204 + ], + "score": 1.0, + "content": "map. Partial Convolution", + "type": "text" + }, + { + "bbox": [ + 219, + 191, + 237, + 203 + ], + "score": 0.46, + "content": "\\lVert 2 2 \\rVert", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 192, + 506, + 204 + ], + "score": 1.0, + "content": "has been proposed as a padding method that treats pixels outside", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 469, + 215 + ], + "score": 1.0, + "content": "the original image as missing values and rescales the computed convolutions accordingly", + "type": "text" + }, + { + "bbox": [ + 469, + 203, + 487, + 214 + ], + "score": 0.78, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 204, + 505, + 215 + ], + "score": 1.0, + "content": ". Its", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 406, + 226 + ], + "score": 1.0, + "content": "foveation behavior resembles reflective padding 3. Distribution padding", + "type": "text" + }, + { + "bbox": [ + 406, + 214, + 424, + 225 + ], + "score": 0.76, + "content": "\\pmb { \\mathbb { B } } 0 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "resizes the input to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "fill the padding area around the original feature map, aiming at preserving the distribution of the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 236, + 410, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 410, + 249 + ], + "score": 1.0, + "content": "map. Its foveation map is largely uniform, except for the corners and edges.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "score": 1.0, + "content": "Impact of input size Besides influencing the relative extent of foveation effects, the input size", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 266, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 278 + ], + "score": 1.0, + "content": "also determines the presence of uneven padding (or uneven feature-map erosion), as we discussed", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 276, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 191, + 289 + ], + "score": 1.0, + "content": "in Section 5. Figure", + "type": "text" + }, + { + "bbox": [ + 192, + 276, + 204, + 289 + ], + "score": 0.48, + "content": "\\textcircled { 7 } \\textcircled { \\times }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 276, + 392, + 289 + ], + "score": 1.0, + "content": "shows the foveation map for VGG-19 with a", + "type": "text" + }, + { + "bbox": [ + 392, + 277, + 431, + 288 + ], + "score": 0.89, + "content": "1 2 7 \\times 1 2 7", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 276, + 506, + 289 + ], + "score": 1.0, + "content": "input. This input", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 321, + 301 + ], + "score": 1.0, + "content": "violates Eq. 1 at every downsampling layer (appendix", + "type": "text" + }, + { + "bbox": [ + 321, + 288, + 334, + 300 + ], + "score": 0.8, + "content": "\\mathbf { A } )", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 287, + 505, + 301 + ], + "score": 1.0, + "content": ", leading to successive feature map erosion", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "at the bottom and right sides which is reflected in the foveation map (see appendix B for a detailed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 309, + 482, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 482, + 322 + ], + "score": 1.0, + "content": "example). The bottom-right part of the input is hence less involved in the CNN computations.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 328, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "Impact of dilation We assign a dilation factor of 2 to all VGG-19 convolutional layers. While this", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 402, + 352 + ], + "score": 1.0, + "content": "exponentially increases the receptive field of the neurons at deeper layers", + "type": "text" + }, + { + "bbox": [ + 402, + 339, + 419, + 351 + ], + "score": 0.55, + "content": "\\pm 2 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 339, + 505, + 352 + ], + "score": 1.0, + "content": ", dilation doubles the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 350, + 501, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 489, + 363 + ], + "score": 1.0, + "content": "extent of the non-uniform peripheral areas that emerge with SAME 0-padding as evident in Figure", + "type": "text" + }, + { + "bbox": [ + 489, + 350, + 501, + 363 + ], + "score": 0.25, + "content": "\\textcircled { 7 } \\textcircled { < }", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 491, + 375 + ], + "score": 1.0, + "content": "SYMMETRIC and circular padding maintain uniform foveation maps regardless of dilation 3.", + "type": "text" + }, + { + "bbox": [ + 494, + 363, + 505, + 372 + ], + "score": 1.0, + "content": "In", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 372, + 491, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 491, + 385 + ], + "score": 1.0, + "content": "contrast, dilation increases the complexity of these maps for REFLECT and replication padding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 505, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Impact of strides Whether learned on based on pooling, downsampling layers can amplify the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "impact of succeeding convolutional layers on foveation behaviour. Furthermore, these layers can", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "cause input pixels to vary in the count of their input-output paths. This can happen when the kernel", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "size is not divisible by the stride, leading to a checkerboard pattern in the foveation maps. This", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "manifests in ResNet models as we illustrate in appendix B. In VGG-19, all max-pooling layers use", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "a stride of 2 and kernel size of 2. Changing the kernel size to 3 leads to a checkerboard pattern as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 456, + 434, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 414, + 470 + ], + "score": 1.0, + "content": "evident in Figure 7d. Such effects were shown to impact pixel-oriented tasks", + "type": "text" + }, + { + "bbox": [ + 414, + 456, + 432, + 468 + ], + "score": 0.61, + "content": "\\mathbb { B } 2 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 456, + 434, + 470 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 474, + 504, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 504, + 486 + ], + "score": 1.0, + "content": "The padding technique and its foveation behaviour have direct impact on feature-map artifacts (Sec-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 484, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 123, + 498 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 124, + 484, + 135, + 497 + ], + "score": 0.67, + "content": "\\bar { 7 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 484, + 411, + 498 + ], + "score": 1.0, + "content": ", and on the ability of CNNs to encode spatial information (Section", + "type": "text" + }, + { + "bbox": [ + 411, + 484, + 421, + 497 + ], + "score": 0.37, + "content": "^ { 8 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 484, + 506, + 498 + ], + "score": 1.0, + "content": ". Understanding the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "foveation behavior is key to determine how suited a padding method is for a given task. For ex-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 435, + 519 + ], + "score": 1.0, + "content": "ample, small object detection is known to be challenging close to the boundary", + "type": "text" + }, + { + "bbox": [ + 435, + 506, + 453, + 518 + ], + "score": 0.78, + "content": "\\left[ \\left[ 2 6 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 507, + 505, + 519 + ], + "score": 1.0, + "content": ", in part due", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 517, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 334, + 532 + ], + "score": 1.0, + "content": "to the foveation behavior of SAME 0-padding. In Figure", + "type": "text" + }, + { + "bbox": [ + 334, + 518, + 347, + 530 + ], + "score": 0.8, + "content": "\\bar { 5 } 6", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 517, + 506, + 532 + ], + "score": 1.0, + "content": ", we change the padding method in the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 529, + 504, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 504, + 541 + ], + "score": 1.0, + "content": "SSD to SYMMETRIC. The stimulus is noticeably more detectable at the boundary, compared with", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 148, + 553 + ], + "score": 1.0, + "content": "0-padding", + "type": "text" + }, + { + "bbox": [ + 149, + 538, + 158, + 552 + ], + "score": 0.38, + "content": "^ 4 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 539, + 506, + 553 + ], + "score": 1.0, + "content": "In contrast, ImageNet classification is less sensitive to foveation effects because the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "target objects are mostly located away from the periphery. Nevertheless, the padding method was", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 560, + 461, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 267, + 574 + ], + "score": 1.0, + "content": "shown to impact classification accuracy", + "type": "text" + }, + { + "bbox": [ + 267, + 561, + 285, + 573 + ], + "score": 0.7, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 560, + 461, + 574 + ], + "score": 1.0, + "content": "because it still affects feature map artifacts.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 108, + 589, + 393, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 394, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 394, + 604 + ], + "score": 1.0, + "content": "7 PADDING METHODS AND FEATURE MAP ARTIFACTS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 298, + 627 + ], + "score": 1.0, + "content": "It is also noticeable that the score map in Figure", + "type": "text" + }, + { + "bbox": [ + 298, + 614, + 311, + 627 + ], + "score": 0.73, + "content": "5 \\mathsf { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 614, + 435, + 627 + ], + "score": 1.0, + "content": "is more uniform than in Figure", + "type": "text" + }, + { + "bbox": [ + 435, + 614, + 447, + 627 + ], + "score": 0.61, + "content": "\\textcircled { 5 } \\textcircled { \\times }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 614, + 505, + 627 + ], + "score": 1.0, + "content": ". In particular,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "under SYMMETRIC padding the model is able to detect traffic lights placed in the blind spots of the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 392, + 650 + ], + "score": 1.0, + "content": "original 0-padded model. To verify whether the line artifacts in Figure", + "type": "text" + }, + { + "bbox": [ + 393, + 636, + 402, + 649 + ], + "score": 0.54, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "are mitigated, we inspect", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 647, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 504, + 660 + ], + "score": 1.0, + "content": "the mean feature maps of the adapted model. With a constant input, SYMMETRIC padding warrants", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 659, + 504, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 504, + 670 + ], + "score": 1.0, + "content": "constant maps throughout the CNN because it reuses the border to fill the padding area. Instead,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "we average these maps over 30 samples generated uniformly at random. Figure 8 depicts the mean", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 680, + 365, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 365, + 694 + ], + "score": 1.0, + "content": "maps which are largely uniform, unlike the case with 0-padding.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 116, + 697, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 116, + 697, + 506, + 713 + ], + "score": 1.0, + "content": "3 Refer to appendix F or to http://mind-the-pad.github.io for visual illustration and further", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 709, + 272, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 272, + 722 + ], + "score": 1.0, + "content": "theoretical analysis of the foveation behavior.", + "type": "text" + } + ] + }, + { + "bbox": [ + 117, + 718, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 117, + 718, + 506, + 734 + ], + "score": 1.0, + "content": "4 Since the input size causes uneven application of padding, the right and bottom borders are still challenging.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Figure 7b shows the foveation behavior of alternatives to SAME 0-padding that have roots in wavelet", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 254, + 106 + ], + "score": 1.0, + "content": "analysis [19] and image processing", + "type": "text" + }, + { + "bbox": [ + 254, + 93, + 272, + 105 + ], + "score": 0.33, + "content": " { \\mathbb { E } } { \\ b { \\mathbb { Z } } } ] ", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 92, + 505, + 106 + ], + "score": 1.0, + "content": ". Mirror padding mirrors pixels at the boundary to fill", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "the padding area. When the border is included (SYMMETRIC mode in TensorFlow) all input pixels", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 297, + 128 + ], + "score": 1.0, + "content": "have an equal number of input-output paths 3,", + "type": "text" + }, + { + "bbox": [ + 299, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "resulting in a uniform foveation map. When the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "border is not included (REFLECT mode both in PyTorch and in TensorFlow), the map exhibits bias", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "against the border and towards a contour in its proximity. This bias is amplified over multiple layers.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "Replication padding exhibits the opposite bias when the padding area is wider than 1 pixel. This", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "is because it replicates the outer 1-pixel border multiple times to fill this area 3. The method is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 169, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 506, + 184 + ], + "score": 1.0, + "content": "equivalent to SYMMETRIC if the padding area is 1-pixel wide. Circular padding wraps opposing", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 506, + 194 + ], + "score": 1.0, + "content": "borders, enabling the kernels to seamlessly operate on the boundary and resulting in a uniform", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 219, + 204 + ], + "score": 1.0, + "content": "map. Partial Convolution", + "type": "text" + }, + { + "bbox": [ + 219, + 191, + 237, + 203 + ], + "score": 0.46, + "content": "\\lVert 2 2 \\rVert", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 192, + 506, + 204 + ], + "score": 1.0, + "content": "has been proposed as a padding method that treats pixels outside", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 469, + 215 + ], + "score": 1.0, + "content": "the original image as missing values and rescales the computed convolutions accordingly", + "type": "text" + }, + { + "bbox": [ + 469, + 203, + 487, + 214 + ], + "score": 0.78, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 204, + 505, + 215 + ], + "score": 1.0, + "content": ". Its", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 406, + 226 + ], + "score": 1.0, + "content": "foveation behavior resembles reflective padding 3. Distribution padding", + "type": "text" + }, + { + "bbox": [ + 406, + 214, + 424, + 225 + ], + "score": 0.76, + "content": "\\pmb { \\mathbb { B } } 0 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "resizes the input to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "fill the padding area around the original feature map, aiming at preserving the distribution of the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 236, + 410, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 410, + 249 + ], + "score": 1.0, + "content": "map. Its foveation map is largely uniform, except for the corners and edges.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 82, + 506, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "score": 1.0, + "content": "Impact of input size Besides influencing the relative extent of foveation effects, the input size", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 266, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 506, + 278 + ], + "score": 1.0, + "content": "also determines the presence of uneven padding (or uneven feature-map erosion), as we discussed", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 276, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 191, + 289 + ], + "score": 1.0, + "content": "in Section 5. Figure", + "type": "text" + }, + { + "bbox": [ + 192, + 276, + 204, + 289 + ], + "score": 0.48, + "content": "\\textcircled { 7 } \\textcircled { \\times }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 276, + 392, + 289 + ], + "score": 1.0, + "content": "shows the foveation map for VGG-19 with a", + "type": "text" + }, + { + "bbox": [ + 392, + 277, + 431, + 288 + ], + "score": 0.89, + "content": "1 2 7 \\times 1 2 7", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 276, + 506, + 289 + ], + "score": 1.0, + "content": "input. This input", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 287, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 321, + 301 + ], + "score": 1.0, + "content": "violates Eq. 1 at every downsampling layer (appendix", + "type": "text" + }, + { + "bbox": [ + 321, + 288, + 334, + 300 + ], + "score": 0.8, + "content": "\\mathbf { A } )", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 287, + 505, + 301 + ], + "score": 1.0, + "content": ", leading to successive feature map erosion", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "at the bottom and right sides which is reflected in the foveation map (see appendix B for a detailed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 309, + 482, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 482, + 322 + ], + "score": 1.0, + "content": "example). The bottom-right part of the input is hence less involved in the CNN computations.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 255, + 506, + 322 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 328, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "Impact of dilation We assign a dilation factor of 2 to all VGG-19 convolutional layers. While this", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 402, + 352 + ], + "score": 1.0, + "content": "exponentially increases the receptive field of the neurons at deeper layers", + "type": "text" + }, + { + "bbox": [ + 402, + 339, + 419, + 351 + ], + "score": 0.55, + "content": "\\pm 2 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 339, + 505, + 352 + ], + "score": 1.0, + "content": ", dilation doubles the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 350, + 501, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 489, + 363 + ], + "score": 1.0, + "content": "extent of the non-uniform peripheral areas that emerge with SAME 0-padding as evident in Figure", + "type": "text" + }, + { + "bbox": [ + 489, + 350, + 501, + 363 + ], + "score": 0.25, + "content": "\\textcircled { 7 } \\textcircled { < }", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 491, + 375 + ], + "score": 1.0, + "content": "SYMMETRIC and circular padding maintain uniform foveation maps regardless of dilation 3.", + "type": "text" + }, + { + "bbox": [ + 494, + 363, + 505, + 372 + ], + "score": 1.0, + "content": "In", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 372, + 491, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 491, + 385 + ], + "score": 1.0, + "content": "contrast, dilation increases the complexity of these maps for REFLECT and replication padding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 328, + 505, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 505, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Impact of strides Whether learned on based on pooling, downsampling layers can amplify the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "impact of succeeding convolutional layers on foveation behaviour. Furthermore, these layers can", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "cause input pixels to vary in the count of their input-output paths. This can happen when the kernel", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "size is not divisible by the stride, leading to a checkerboard pattern in the foveation maps. This", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "manifests in ResNet models as we illustrate in appendix B. In VGG-19, all max-pooling layers use", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "a stride of 2 and kernel size of 2. Changing the kernel size to 3 leads to a checkerboard pattern as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 456, + 434, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 414, + 470 + ], + "score": 1.0, + "content": "evident in Figure 7d. Such effects were shown to impact pixel-oriented tasks", + "type": "text" + }, + { + "bbox": [ + 414, + 456, + 432, + 468 + ], + "score": 0.61, + "content": "\\mathbb { B } 2 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 456, + 434, + 470 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 390, + 505, + 470 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 474, + 504, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 504, + 486 + ], + "score": 1.0, + "content": "The padding technique and its foveation behaviour have direct impact on feature-map artifacts (Sec-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 484, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 123, + 498 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 124, + 484, + 135, + 497 + ], + "score": 0.67, + "content": "\\bar { 7 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 484, + 411, + 498 + ], + "score": 1.0, + "content": ", and on the ability of CNNs to encode spatial information (Section", + "type": "text" + }, + { + "bbox": [ + 411, + 484, + 421, + 497 + ], + "score": 0.37, + "content": "^ { 8 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 484, + 506, + 498 + ], + "score": 1.0, + "content": ". Understanding the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "foveation behavior is key to determine how suited a padding method is for a given task. For ex-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 435, + 519 + ], + "score": 1.0, + "content": "ample, small object detection is known to be challenging close to the boundary", + "type": "text" + }, + { + "bbox": [ + 435, + 506, + 453, + 518 + ], + "score": 0.78, + "content": "\\left[ \\left[ 2 6 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 507, + 505, + 519 + ], + "score": 1.0, + "content": ", in part due", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 517, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 334, + 532 + ], + "score": 1.0, + "content": "to the foveation behavior of SAME 0-padding. In Figure", + "type": "text" + }, + { + "bbox": [ + 334, + 518, + 347, + 530 + ], + "score": 0.8, + "content": "\\bar { 5 } 6", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 517, + 506, + 532 + ], + "score": 1.0, + "content": ", we change the padding method in the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 529, + 504, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 504, + 541 + ], + "score": 1.0, + "content": "SSD to SYMMETRIC. The stimulus is noticeably more detectable at the boundary, compared with", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 148, + 553 + ], + "score": 1.0, + "content": "0-padding", + "type": "text" + }, + { + "bbox": [ + 149, + 538, + 158, + 552 + ], + "score": 0.38, + "content": "^ 4 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 539, + 506, + 553 + ], + "score": 1.0, + "content": "In contrast, ImageNet classification is less sensitive to foveation effects because the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "target objects are mostly located away from the periphery. Nevertheless, the padding method was", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 560, + 461, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 267, + 574 + ], + "score": 1.0, + "content": "shown to impact classification accuracy", + "type": "text" + }, + { + "bbox": [ + 267, + 561, + 285, + 573 + ], + "score": 0.7, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 560, + 461, + 574 + ], + "score": 1.0, + "content": "because it still affects feature map artifacts.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 474, + 506, + 574 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 589, + 393, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 394, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 394, + 604 + ], + "score": 1.0, + "content": "7 PADDING METHODS AND FEATURE MAP ARTIFACTS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 298, + 627 + ], + "score": 1.0, + "content": "It is also noticeable that the score map in Figure", + "type": "text" + }, + { + "bbox": [ + 298, + 614, + 311, + 627 + ], + "score": 0.73, + "content": "5 \\mathsf { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 614, + 435, + 627 + ], + "score": 1.0, + "content": "is more uniform than in Figure", + "type": "text" + }, + { + "bbox": [ + 435, + 614, + 447, + 627 + ], + "score": 0.61, + "content": "\\textcircled { 5 } \\textcircled { \\times }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 614, + 505, + 627 + ], + "score": 1.0, + "content": ". In particular,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "under SYMMETRIC padding the model is able to detect traffic lights placed in the blind spots of the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 392, + 650 + ], + "score": 1.0, + "content": "original 0-padded model. To verify whether the line artifacts in Figure", + "type": "text" + }, + { + "bbox": [ + 393, + 636, + 402, + 649 + ], + "score": 0.54, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "are mitigated, we inspect", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 647, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 504, + 660 + ], + "score": 1.0, + "content": "the mean feature maps of the adapted model. With a constant input, SYMMETRIC padding warrants", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 659, + 504, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 504, + 670 + ], + "score": 1.0, + "content": "constant maps throughout the CNN because it reuses the border to fill the padding area. Instead,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "we average these maps over 30 samples generated uniformly at random. Figure 8 depicts the mean", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 680, + 365, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 365, + 694 + ], + "score": 1.0, + "content": "maps which are largely uniform, unlike the case with 0-padding.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 614, + 506, + 694 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 158 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 158 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 158 + ], + "score": 0.964, + "type": "image", + "image_path": "57e698bb07346f6a27d58d62665ad1ecd8d2ee482ba665e630bbd20159d302cc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 105.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 105.33333333333333, + 504, + 131.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 131.66666666666666, + 504, + 158.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 166, + 505, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 181 + ], + "score": 1.0, + "content": "Figure 8: The same feature maps in Figure 2, generated under mirror padding and averaged over 30", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 178, + 496, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 496, + 191 + ], + "score": 1.0, + "content": "randomly-generated input samples. The line artifacts induced by 0-padding are largely mitigated.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 213, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "To further analyze the impact of SYMMETRIC padding, we retrain the adapted model following the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "original training protocol. This significantly improves the average precision (AP) as reported in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 129, + 249 + ], + "score": 1.0, + "content": "Table", + "type": "text" + }, + { + "bbox": [ + 129, + 235, + 139, + 248 + ], + "score": 0.32, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "under different overlap thresholds (matching IoU), confirming that small object detection is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 289, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 289, + 259 + ], + "score": 1.0, + "content": "particularly sensitive to feature-map artifacts.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "table", + "bbox": [ + 129, + 292, + 483, + 335 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 270, + 504, + 282 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "score": 1.0, + "content": "Table 2: Performance of the SSD traffic light detector, trained under two different padding schemes.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "table_body", + "bbox": [ + 129, + 292, + 483, + 335 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 292, + 483, + 335 + ], + "spans": [ + { + "bbox": [ + 129, + 292, + 483, + 335 + ], + "score": 0.968, + "html": "
Average Precision (AP)AP@.20I0UAP@ .50I0UAP@.75I0UAP@.90I0U
Zero Padding80.24%49.58%3.7%0.007%
Mirror Padding83.20%57%8.44%0.02%
", + "type": "table", + "image_path": "1de57b9720d30158817e2f1a1ba78dd730499b7691b727f819178fba8c13d39e.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 129, + 292, + 483, + 306.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 129, + 306.3333333333333, + 483, + 320.66666666666663 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 129, + 320.66666666666663, + 483, + 334.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 107, + 348, + 505, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 347, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 276, + 361 + ], + "score": 1.0, + "content": "Of the padding methods listed in Section", + "type": "text" + }, + { + "bbox": [ + 277, + 348, + 287, + 361 + ], + "score": 0.61, + "content": "6 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 347, + 505, + 361 + ], + "score": 1.0, + "content": "mirror padding in both SYMMETRIC and REFLECT", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "modes, PartialConv, and circular padding are generally effective at reducing feature map artifacts", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "score": 1.0, + "content": "that emerge under zero padding, in particular salient line patterns. In contrast, distribution padding", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 381, + 504, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 504, + 394 + ], + "score": 1.0, + "content": "can induce significant artifacts. Refer to appendix D for comparative examples of artifacts under the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 392, + 244, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 244, + 405 + ], + "score": 1.0, + "content": "aforementioned padding schemes.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "Artifact magnitude and propagation While feature-map artifacts are induced by the padding", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "mechanism at the boundary, their magnitude and inward propagation in the maps are impacted", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "by several architectural aspects of CNNs. In particular, certain normalization schemes such as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "batchnorm [15] tend to limit the range of variation within a feature map and to relatively harmo-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "nize this range across different maps. This, in turn, impacts how possible artifacts in these maps ac-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "cumulate when they are processed by the next convolutional layer. Similarly, artifacts that manifest", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "after applying ReLU units are of a positive sign. These factors were instrumental in the formation", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "of potential blind spots described in Section 3. We hence recommend to involve non-convolutional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "layers when inspecting the feature maps. Besides having possible impact on artifact magnitude,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "several aspects of convolution arithmetic, such as filter size and dilation factors, can also impact the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 528, + 256, + 539 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 256, + 539 + ], + "score": 1.0, + "content": "spatial propagation of these artifacts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 556, + 320, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 321, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 321, + 571 + ], + "score": 1.0, + "content": "8 RELATED FINDINGS AND TAKEAWAYS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 582, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "score": 1.0, + "content": "Handling the boundary is an inherent challenge when dealing with spatial data [9]. Mean padding is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "known to cause visual artifacts in traditional image processing, with alternative methods proposed to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 164, + 617 + ], + "score": 1.0, + "content": "mitigate them", + "type": "text" + }, + { + "bbox": [ + 164, + 604, + 182, + 616 + ], + "score": 0.29, + "content": "\\mathbb { \\lVert 2 4 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 604, + 505, + 617 + ], + "score": 1.0, + "content": ". CNNs have been often assumed to deal with such effects implicitly. Innamorati", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 126, + 629 + ], + "score": 1.0, + "content": "et al", + "type": "text" + }, + { + "bbox": [ + 126, + 615, + 144, + 627 + ], + "score": 0.54, + "content": "\\dot { [ \\lVert { 4 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "propose learning separate sets of filters dedicated to the boundaries to avoid impacting the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 626, + 504, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 436, + 640 + ], + "score": 1.0, + "content": "weights learned by regular filters. A grouped padding strategy, proposed to support", + "type": "text" + }, + { + "bbox": [ + 437, + 627, + 458, + 637 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 626, + 483, + 640 + ], + "score": 1.0, + "content": "filters", + "type": "text" + }, + { + "bbox": [ + 484, + 626, + 501, + 638 + ], + "score": 0.64, + "content": "\\mathbf { \\bar { \\textmu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 626, + 504, + 640 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "offers avenues to mitigate uneven padding and corresponding skewness in foveation maps without", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 309, + 661 + ], + "score": 1.0, + "content": "restrictions on input size (see our note in appendix", + "type": "text" + }, + { + "bbox": [ + 310, + 649, + 321, + 662 + ], + "score": 0.6, + "content": "\\mathbf { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "for explanation). Finally, insights from signal", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 659, + 410, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 410, + 673 + ], + "score": 1.0, + "content": "and image processing [10; 11] could inspire further CNN padding schemes.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 504, + 689 + ], + "score": 1.0, + "content": "Zero padding has been recently linked to CNNs’ ability to encode position information [7; 16; 18;", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 687, + 503, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 358, + 699 + ], + "score": 1.0, + "content": "29]. In contrast, circular padding was shown to limit this ability", + "type": "text" + }, + { + "bbox": [ + 358, + 687, + 370, + 699 + ], + "score": 0.7, + "content": "\\mathbb { \\left[ \\bigcirc \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 687, + 483, + 699 + ], + "score": 1.0, + "content": "and to boost shift invariance", + "type": "text" + }, + { + "bbox": [ + 484, + 687, + 501, + 699 + ], + "score": 0.44, + "content": "\\begin{array} { r l } { { \\bigl \\| \\overline { { 3 5 } } \\bigr \\| } } & { { } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 687, + 503, + 699 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "score": 1.0, + "content": "The input sizes in those studies do induce uneven padding. This can be, in part, the underlying", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "mechanism behind the aforementioned ability. Whether or not this ability is desirable depends on the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 501, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 501, + 732 + ], + "score": 1.0, + "content": "task, with several methods proposed to explicitly encode spatial information [5; 6; 20; 25; 29; 31].", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 79, + 504, + 158 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 79, + 504, + 158 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 79, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 108, + 79, + 504, + 158 + ], + "score": 0.964, + "type": "image", + "image_path": "57e698bb07346f6a27d58d62665ad1ecd8d2ee482ba665e630bbd20159d302cc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 79, + 504, + 105.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 105.33333333333333, + 504, + 131.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 131.66666666666666, + 504, + 158.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 166, + 505, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 181 + ], + "score": 1.0, + "content": "Figure 8: The same feature maps in Figure 2, generated under mirror padding and averaged over 30", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 178, + 496, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 496, + 191 + ], + "score": 1.0, + "content": "randomly-generated input samples. The line artifacts induced by 0-padding are largely mitigated.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 213, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "To further analyze the impact of SYMMETRIC padding, we retrain the adapted model following the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "original training protocol. This significantly improves the average precision (AP) as reported in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 129, + 249 + ], + "score": 1.0, + "content": "Table", + "type": "text" + }, + { + "bbox": [ + 129, + 235, + 139, + 248 + ], + "score": 0.32, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "under different overlap thresholds (matching IoU), confirming that small object detection is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 289, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 289, + 259 + ], + "score": 1.0, + "content": "particularly sensitive to feature-map artifacts.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 213, + 505, + 259 + ] + }, + { + "type": "table", + "bbox": [ + 129, + 292, + 483, + 335 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 270, + 504, + 282 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "score": 1.0, + "content": "Table 2: Performance of the SSD traffic light detector, trained under two different padding schemes.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "table_body", + "bbox": [ + 129, + 292, + 483, + 335 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 292, + 483, + 335 + ], + "spans": [ + { + "bbox": [ + 129, + 292, + 483, + 335 + ], + "score": 0.968, + "html": "
Average Precision (AP)AP@.20I0UAP@ .50I0UAP@.75I0UAP@.90I0U
Zero Padding80.24%49.58%3.7%0.007%
Mirror Padding83.20%57%8.44%0.02%
", + "type": "table", + "image_path": "1de57b9720d30158817e2f1a1ba78dd730499b7691b727f819178fba8c13d39e.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 129, + 292, + 483, + 306.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 129, + 306.3333333333333, + 483, + 320.66666666666663 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 129, + 320.66666666666663, + 483, + 334.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 107, + 348, + 505, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 347, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 276, + 361 + ], + "score": 1.0, + "content": "Of the padding methods listed in Section", + "type": "text" + }, + { + "bbox": [ + 277, + 348, + 287, + 361 + ], + "score": 0.61, + "content": "6 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 347, + 505, + 361 + ], + "score": 1.0, + "content": "mirror padding in both SYMMETRIC and REFLECT", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "modes, PartialConv, and circular padding are generally effective at reducing feature map artifacts", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "score": 1.0, + "content": "that emerge under zero padding, in particular salient line patterns. In contrast, distribution padding", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 381, + 504, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 504, + 394 + ], + "score": 1.0, + "content": "can induce significant artifacts. Refer to appendix D for comparative examples of artifacts under the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 392, + 244, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 244, + 405 + ], + "score": 1.0, + "content": "aforementioned padding schemes.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 347, + 506, + 405 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "Artifact magnitude and propagation While feature-map artifacts are induced by the padding", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "mechanism at the boundary, their magnitude and inward propagation in the maps are impacted", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "by several architectural aspects of CNNs. In particular, certain normalization schemes such as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "batchnorm [15] tend to limit the range of variation within a feature map and to relatively harmo-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "nize this range across different maps. This, in turn, impacts how possible artifacts in these maps ac-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "cumulate when they are processed by the next convolutional layer. Similarly, artifacts that manifest", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "after applying ReLU units are of a positive sign. These factors were instrumental in the formation", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "of potential blind spots described in Section 3. We hence recommend to involve non-convolutional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "layers when inspecting the feature maps. Besides having possible impact on artifact magnitude,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "several aspects of convolution arithmetic, such as filter size and dilation factors, can also impact the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 528, + 256, + 539 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 256, + 539 + ], + "score": 1.0, + "content": "spatial propagation of these artifacts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 416, + 505, + 539 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 556, + 320, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 321, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 321, + 571 + ], + "score": 1.0, + "content": "8 RELATED FINDINGS AND TAKEAWAYS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 582, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "score": 1.0, + "content": "Handling the boundary is an inherent challenge when dealing with spatial data [9]. Mean padding is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "known to cause visual artifacts in traditional image processing, with alternative methods proposed to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 164, + 617 + ], + "score": 1.0, + "content": "mitigate them", + "type": "text" + }, + { + "bbox": [ + 164, + 604, + 182, + 616 + ], + "score": 0.29, + "content": "\\mathbb { \\lVert 2 4 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 604, + 505, + 617 + ], + "score": 1.0, + "content": ". CNNs have been often assumed to deal with such effects implicitly. Innamorati", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 126, + 629 + ], + "score": 1.0, + "content": "et al", + "type": "text" + }, + { + "bbox": [ + 126, + 615, + 144, + 627 + ], + "score": 0.54, + "content": "\\dot { [ \\lVert { 4 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "propose learning separate sets of filters dedicated to the boundaries to avoid impacting the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 626, + 504, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 436, + 640 + ], + "score": 1.0, + "content": "weights learned by regular filters. A grouped padding strategy, proposed to support", + "type": "text" + }, + { + "bbox": [ + 437, + 627, + 458, + 637 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 626, + 483, + 640 + ], + "score": 1.0, + "content": "filters", + "type": "text" + }, + { + "bbox": [ + 484, + 626, + 501, + 638 + ], + "score": 0.64, + "content": "\\mathbf { \\bar { \\textmu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 626, + 504, + 640 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "offers avenues to mitigate uneven padding and corresponding skewness in foveation maps without", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 309, + 661 + ], + "score": 1.0, + "content": "restrictions on input size (see our note in appendix", + "type": "text" + }, + { + "bbox": [ + 310, + 649, + 321, + 662 + ], + "score": 0.6, + "content": "\\mathbf { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "for explanation). Finally, insights from signal", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 659, + 410, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 410, + 673 + ], + "score": 1.0, + "content": "and image processing [10; 11] could inspire further CNN padding schemes.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 582, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 504, + 689 + ], + "score": 1.0, + "content": "Zero padding has been recently linked to CNNs’ ability to encode position information [7; 16; 18;", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 687, + 503, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 358, + 699 + ], + "score": 1.0, + "content": "29]. In contrast, circular padding was shown to limit this ability", + "type": "text" + }, + { + "bbox": [ + 358, + 687, + 370, + 699 + ], + "score": 0.7, + "content": "\\mathbb { \\left[ \\bigcirc \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 687, + 483, + 699 + ], + "score": 1.0, + "content": "and to boost shift invariance", + "type": "text" + }, + { + "bbox": [ + 484, + 687, + 501, + 699 + ], + "score": 0.44, + "content": "\\begin{array} { r l } { { \\bigl \\| \\overline { { 3 5 } } \\bigr \\| } } & { { } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 687, + 503, + 699 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 506, + 713 + ], + "score": 1.0, + "content": "The input sizes in those studies do induce uneven padding. This can be, in part, the underlying", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "mechanism behind the aforementioned ability. Whether or not this ability is desirable depends on the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 501, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 501, + 732 + ], + "score": 1.0, + "content": "task, with several methods proposed to explicitly encode spatial information [5; 6; 20; 25; 29; 31].", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 676, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Downsampling using max-pooling or strided convolution has been shown to impact shift invariance", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "in CNNs by incurring aliasing effects [3; 38; 43]. These effects can manifest in the same symptoms", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 199, + 117 + ], + "score": 1.0, + "content": "we reported in Section", + "type": "text" + }, + { + "bbox": [ + 200, + 104, + 210, + 117 + ], + "score": 0.62, + "content": "^ { 1 , }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "albeit for a different reason. Zhang [43] demonstrated how blurring the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "feature maps before subsampling mitigates aliasing effects and improves ImageNet classification ac-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "curacy of various popular CNNs. We analyzed the mean filters in antialiased MobileNet and ResNet", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 325, + 151 + ], + "score": 1.0, + "content": "models pre-trained on ImageNet under 0-padding, with", + "type": "text" + }, + { + "bbox": [ + 326, + 137, + 365, + 148 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 137, + 489, + 151 + ], + "score": 1.0, + "content": "as input size (refer to Appendix", + "type": "text" + }, + { + "bbox": [ + 489, + 137, + 501, + 150 + ], + "score": 0.48, + "content": "\\mathrm { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 137, + 505, + 151 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 147, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 505, + 161 + ], + "score": 1.0, + "content": "We found that antialiasing can also mitigate the asymmetry of mean filters that exhibited high asym-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "metry in the baseline models, especially at deeper layers. This is remarkable given that these models", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 163, + 183 + ], + "score": 1.0, + "content": "are trained on", + "type": "text" + }, + { + "bbox": [ + 163, + 170, + 203, + 181 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 170, + 505, + 183 + ], + "score": 1.0, + "content": "images, which incurs one-sided zero padding at every downsampling layer.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "This could, in part, be attributed to the ability of the BlurPool operator used in antialiased CNN", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 191, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 504, + 204 + ], + "score": 1.0, + "content": "to smoothen the acuity of zero-padded borders, in turn, reducing the value imbalance incurred by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "score": 1.0, + "content": "one-sided padding. Further analysis is needed to examine the interaction between padding and alias-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "score": 1.0, + "content": "ing effects in CNNs and to establish possible synergy between antialiasing and eliminating uneven", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 200, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 200, + 238 + ], + "score": 1.0, + "content": "application of padding.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 144, + 254 + ], + "score": 1.0, + "content": "Luo et al", + "type": "text" + }, + { + "bbox": [ + 144, + 241, + 162, + 253 + ], + "score": 0.66, + "content": "\\pmb { \\Vert 2 8 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "drew connections between effective receptive fields and foveated vision. Our analysis", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "links foveation behavior with the padding scheme and suggests that it might occur implicitly in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 504, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 504, + 276 + ], + "score": 1.0, + "content": "CNNs when using VALID or SAME 0-padding, without the need for explicit mechanisms [2; 21].", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 351, + 288 + ], + "score": 1.0, + "content": "Furthermore, it explains the drastic accuracy drop noted by", + "type": "text" + }, + { + "bbox": [ + 352, + 274, + 369, + 286 + ], + "score": 0.68, + "content": "\\boxed { 1 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "under VALID padding, which is", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 285, + 243, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 243, + 298 + ], + "score": 1.0, + "content": "amplified by feature map erosion.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 108, + 309, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 107, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 107, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "Choosing a padding method SAME 0-padding is by far the most widely-used method. Compared", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 107, + 320, + 292, + 332 + ], + "score": 1.0, + "content": "with other methods, it can enable as much as", + "type": "text" + }, + { + "bbox": [ + 293, + 320, + 312, + 330 + ], + "score": 0.86, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "faster training and inference. Problem-specific", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 329, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 505, + 345 + ], + "score": 1.0, + "content": "constraints can dictate different choices [34; 35; 40]. In the lack of a universally superior padding", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "method, we recommend considering multiple ones while paying attention to the nature of the data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 352, + 300, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 300, + 365 + ], + "score": 1.0, + "content": "and the task, as well as to the following aspects:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 132, + 373, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 132, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 132, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "• Feature-map statistics: 0-padding can alter the value distribution within the feature maps", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "and can shift their mean value in the presence of ReLU units. The alternatives presented in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 395, + 499, + 407 + ], + "spans": [ + { + "bbox": [ + 142, + 395, + 499, + 407 + ], + "score": 1.0, + "content": "Section 6 tend to preserve this distribution, thanks to reusing existing values in the maps.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 139, + 409, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 139, + 409, + 506, + 422 + ], + "score": 1.0, + "content": "Foveation behavior: 0-padding might not be suited for tasks that require high precision at", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 420, + 398, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 398, + 434 + ], + "score": 1.0, + "content": "the periphery, unlike circular and SYMMETRIC mirror padding.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 132, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 132, + 435, + 401, + 447 + ], + "score": 1.0, + "content": "Interference with image semantics (esp. with a padding amount", + "type": "text" + }, + { + "bbox": [ + 401, + 435, + 419, + 446 + ], + "score": 0.84, + "content": "> 1", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "pixel): For example,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 446, + 502, + 458 + ], + "spans": [ + { + "bbox": [ + 142, + 446, + 482, + 458 + ], + "score": 1.0, + "content": "circular padding could introduce border discontinuities unless the input is panoramic", + "type": "text" + }, + { + "bbox": [ + 483, + 446, + 500, + 457 + ], + "score": 0.64, + "content": "| \\widehat { \\mathsf { B } } \\widehat { \\mathsf { S } } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 446, + 502, + 458 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 132, + 461, + 504, + 473 + ], + "spans": [ + { + "bbox": [ + 132, + 461, + 504, + 473 + ], + "score": 1.0, + "content": "• Potential to induce feature map artifacts: All alternatives to 0-padding induce relatively", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 471, + 420, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 472, + 332, + 484 + ], + "score": 1.0, + "content": "fewer artifacts, except for Distribution padding", + "type": "text" + }, + { + "bbox": [ + 332, + 471, + 350, + 483 + ], + "score": 0.54, + "content": "\\textcircled { \\lVert { 3 0 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 472, + 407, + 484 + ], + "score": 1.0, + "content": "(see appendix", + "type": "text" + }, + { + "bbox": [ + 407, + 471, + 420, + 484 + ], + "score": 0.39, + "content": "{ \\bf D } )", + "type": "inline_equation" + } + ], + "index": 32 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "We also recommend eliminating uneven padding at downsampling layers both at training and at", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 283, + 515 + ], + "score": 1.0, + "content": "inference time, as we illustrated in Section", + "type": "text" + }, + { + "bbox": [ + 284, + 502, + 294, + 515 + ], + "score": 0.65, + "content": "\\boxed { 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "This is especially important when zero padding is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "score": 1.0, + "content": "applied and the downsampling is learned. The scripts used to generate the visualizations in this", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 493, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 493, + 538 + ], + "score": 1.0, + "content": "paper are available in the supplemental as well as at http://mind-the-pad.github.io.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 548, + 505, + 657 + ], + "lines": [ + { + "bbox": [ + 106, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 560 + ], + "score": 1.0, + "content": "Summary We demonstrated how the padding mechanism can induce spatial bias in CNNs, in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "score": 1.0, + "content": "the form of skewed kernels and feature-map artifacts. These artifacts can be highly pronounced", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "with the widely-used 0-padding when applied unevenly at the four sides of the feature maps. We", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "score": 1.0, + "content": "demonstrated how such uneven padding can inherently take place in state-of-the-art CNNs, and how", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 592, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 506, + 604 + ], + "score": 1.0, + "content": "the artifacts it causes can be detrimental to certain tasks such as small object detection. We provided", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "visualization methods to expose these artifacts and to analyze the implication of various padding", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "schemes on boundary pixels. We further proposed solutions to eliminate uneven padding and to", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "mitigate spatial bias in CNNs. Further work is needed to closely examine the implications of spatial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 634, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 650 + ], + "score": 1.0, + "content": "bias and foveation in various applications (see supplementary for examples), as well as padding", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 647, + 283, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 283, + 658 + ], + "score": 1.0, + "content": "impact on recurrent models and 1-D CNNs.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41.5 + }, + { + "type": "title", + "bbox": [ + 108, + 675, + 218, + 686 + ], + "lines": [ + { + "bbox": [ + 107, + 675, + 220, + 689 + ], + "spans": [ + { + "bbox": [ + 107, + 675, + 220, + 689 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "We are thankful to Ross Girshick for providing useful recommendations and experiment ideas, and", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "to Shubham Muttepawar for implementing an interactive tool out of our analysis scripts, guided by", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 459, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 459, + 734 + ], + "score": 1.0, + "content": "our front-end specialist Edward Wang and our AI user-experience designer Sara Zhang.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Downsampling using max-pooling or strided convolution has been shown to impact shift invariance", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "in CNNs by incurring aliasing effects [3; 38; 43]. These effects can manifest in the same symptoms", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 199, + 117 + ], + "score": 1.0, + "content": "we reported in Section", + "type": "text" + }, + { + "bbox": [ + 200, + 104, + 210, + 117 + ], + "score": 0.62, + "content": "^ { 1 , }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "albeit for a different reason. Zhang [43] demonstrated how blurring the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "feature maps before subsampling mitigates aliasing effects and improves ImageNet classification ac-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "curacy of various popular CNNs. We analyzed the mean filters in antialiased MobileNet and ResNet", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 325, + 151 + ], + "score": 1.0, + "content": "models pre-trained on ImageNet under 0-padding, with", + "type": "text" + }, + { + "bbox": [ + 326, + 137, + 365, + 148 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 137, + 489, + 151 + ], + "score": 1.0, + "content": "as input size (refer to Appendix", + "type": "text" + }, + { + "bbox": [ + 489, + 137, + 501, + 150 + ], + "score": 0.48, + "content": "\\mathrm { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 137, + 505, + 151 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 147, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 505, + 161 + ], + "score": 1.0, + "content": "We found that antialiasing can also mitigate the asymmetry of mean filters that exhibited high asym-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "metry in the baseline models, especially at deeper layers. This is remarkable given that these models", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 163, + 183 + ], + "score": 1.0, + "content": "are trained on", + "type": "text" + }, + { + "bbox": [ + 163, + 170, + 203, + 181 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 170, + 505, + 183 + ], + "score": 1.0, + "content": "images, which incurs one-sided zero padding at every downsampling layer.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "This could, in part, be attributed to the ability of the BlurPool operator used in antialiased CNN", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 191, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 504, + 204 + ], + "score": 1.0, + "content": "to smoothen the acuity of zero-padded borders, in turn, reducing the value imbalance incurred by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "score": 1.0, + "content": "one-sided padding. Further analysis is needed to examine the interaction between padding and alias-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "score": 1.0, + "content": "ing effects in CNNs and to establish possible synergy between antialiasing and eliminating uneven", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 200, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 200, + 238 + ], + "score": 1.0, + "content": "application of padding.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 82, + 506, + 238 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 144, + 254 + ], + "score": 1.0, + "content": "Luo et al", + "type": "text" + }, + { + "bbox": [ + 144, + 241, + 162, + 253 + ], + "score": 0.66, + "content": "\\pmb { \\Vert 2 8 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "drew connections between effective receptive fields and foveated vision. Our analysis", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "links foveation behavior with the padding scheme and suggests that it might occur implicitly in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 504, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 504, + 276 + ], + "score": 1.0, + "content": "CNNs when using VALID or SAME 0-padding, without the need for explicit mechanisms [2; 21].", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 351, + 288 + ], + "score": 1.0, + "content": "Furthermore, it explains the drastic accuracy drop noted by", + "type": "text" + }, + { + "bbox": [ + 352, + 274, + 369, + 286 + ], + "score": 0.68, + "content": "\\boxed { 1 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "under VALID padding, which is", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 285, + 243, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 243, + 298 + ], + "score": 1.0, + "content": "amplified by feature map erosion.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 241, + 506, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 309, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 107, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 107, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "Choosing a padding method SAME 0-padding is by far the most widely-used method. Compared", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 107, + 320, + 292, + 332 + ], + "score": 1.0, + "content": "with other methods, it can enable as much as", + "type": "text" + }, + { + "bbox": [ + 293, + 320, + 312, + 330 + ], + "score": 0.86, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "faster training and inference. Problem-specific", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 329, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 505, + 345 + ], + "score": 1.0, + "content": "constraints can dictate different choices [34; 35; 40]. In the lack of a universally superior padding", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "method, we recommend considering multiple ones while paying attention to the nature of the data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 352, + 300, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 300, + 365 + ], + "score": 1.0, + "content": "and the task, as well as to the following aspects:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 309, + 506, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 373, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 132, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 132, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "• Feature-map statistics: 0-padding can alter the value distribution within the feature maps", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "and can shift their mean value in the presence of ReLU units. The alternatives presented in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 395, + 499, + 407 + ], + "spans": [ + { + "bbox": [ + 142, + 395, + 499, + 407 + ], + "score": 1.0, + "content": "Section 6 tend to preserve this distribution, thanks to reusing existing values in the maps.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 139, + 409, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 139, + 409, + 506, + 422 + ], + "score": 1.0, + "content": "Foveation behavior: 0-padding might not be suited for tasks that require high precision at", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 420, + 398, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 398, + 434 + ], + "score": 1.0, + "content": "the periphery, unlike circular and SYMMETRIC mirror padding.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 132, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 132, + 435, + 401, + 447 + ], + "score": 1.0, + "content": "Interference with image semantics (esp. with a padding amount", + "type": "text" + }, + { + "bbox": [ + 401, + 435, + 419, + 446 + ], + "score": 0.84, + "content": "> 1", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "pixel): For example,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 446, + 502, + 458 + ], + "spans": [ + { + "bbox": [ + 142, + 446, + 482, + 458 + ], + "score": 1.0, + "content": "circular padding could introduce border discontinuities unless the input is panoramic", + "type": "text" + }, + { + "bbox": [ + 483, + 446, + 500, + 457 + ], + "score": 0.64, + "content": "| \\widehat { \\mathsf { B } } \\widehat { \\mathsf { S } } \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 446, + 502, + 458 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 132, + 461, + 504, + 473 + ], + "spans": [ + { + "bbox": [ + 132, + 461, + 504, + 473 + ], + "score": 1.0, + "content": "• Potential to induce feature map artifacts: All alternatives to 0-padding induce relatively", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 471, + 420, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 472, + 332, + 484 + ], + "score": 1.0, + "content": "fewer artifacts, except for Distribution padding", + "type": "text" + }, + { + "bbox": [ + 332, + 471, + 350, + 483 + ], + "score": 0.54, + "content": "\\textcircled { \\lVert { 3 0 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 472, + 407, + 484 + ], + "score": 1.0, + "content": "(see appendix", + "type": "text" + }, + { + "bbox": [ + 407, + 471, + 420, + 484 + ], + "score": 0.39, + "content": "{ \\bf D } )", + "type": "inline_equation" + } + ], + "index": 32 + } + ], + "index": 28, + "bbox_fs": [ + 132, + 372, + 506, + 484 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "We also recommend eliminating uneven padding at downsampling layers both at training and at", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 283, + 515 + ], + "score": 1.0, + "content": "inference time, as we illustrated in Section", + "type": "text" + }, + { + "bbox": [ + 284, + 502, + 294, + 515 + ], + "score": 0.65, + "content": "\\boxed { 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "This is especially important when zero padding is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "score": 1.0, + "content": "applied and the downsampling is learned. The scripts used to generate the visualizations in this", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 493, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 493, + 538 + ], + "score": 1.0, + "content": "paper are available in the supplemental as well as at http://mind-the-pad.github.io.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 491, + 506, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 548, + 505, + 657 + ], + "lines": [ + { + "bbox": [ + 106, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 560 + ], + "score": 1.0, + "content": "Summary We demonstrated how the padding mechanism can induce spatial bias in CNNs, in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "score": 1.0, + "content": "the form of skewed kernels and feature-map artifacts. These artifacts can be highly pronounced", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "with the widely-used 0-padding when applied unevenly at the four sides of the feature maps. We", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "score": 1.0, + "content": "demonstrated how such uneven padding can inherently take place in state-of-the-art CNNs, and how", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 592, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 506, + 604 + ], + "score": 1.0, + "content": "the artifacts it causes can be detrimental to certain tasks such as small object detection. We provided", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "visualization methods to expose these artifacts and to analyze the implication of various padding", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "schemes on boundary pixels. We further proposed solutions to eliminate uneven padding and to", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "mitigate spatial bias in CNNs. Further work is needed to closely examine the implications of spatial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 634, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 650 + ], + "score": 1.0, + "content": "bias and foveation in various applications (see supplementary for examples), as well as padding", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 647, + 283, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 283, + 658 + ], + "score": 1.0, + "content": "impact on recurrent models and 1-D CNNs.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 547, + 506, + 658 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 675, + 218, + 686 + ], + "lines": [ + { + "bbox": [ + 107, + 675, + 220, + 689 + ], + "spans": [ + { + "bbox": [ + 107, + 675, + 220, + 689 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "We are thankful to Ross Girshick for providing useful recommendations and experiment ideas, and", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "to Shubham Muttepawar for implementing an interactive tool out of our analysis scripts, guided by", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 459, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 459, + 734 + ], + "score": 1.0, + "content": "our front-end specialist Edward Wang and our AI user-experience designer Sara Zhang.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 698, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 59, + 507, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 109, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 109, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "[1] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 126, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 126, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "J. Dean, et al. TensorFlow: Large-scale machine learning on heterogeneous distributed sys-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 123, + 315, + 135 + ], + "spans": [ + { + "bbox": [ + 127, + 123, + 315, + 135 + ], + "score": 1.0, + "content": "tems. arXiv preprint arXiv:1603.04467, 2016.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 108, + 139, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 108, + 139, + 506, + 155 + ], + "score": 1.0, + "content": "[2] E. Akbas and M. P. Eckstein. Object detection through search with a foveated visual system.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 153, + 347, + 164 + ], + "spans": [ + { + "bbox": [ + 127, + 153, + 347, + 164 + ], + "score": 1.0, + "content": "PLoS computational biology, 13(10):e1005743, 2017.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 110, + 170, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 506, + 184 + ], + "score": 1.0, + "content": "[3] A. Azulay and Y. Weiss. Why do deep convolutional networks generalize so poorly to small", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 181, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 127, + 181, + 505, + 195 + ], + "score": 1.0, + "content": "image transformations? Journal of Machine Learning Research (JMLR), 20(184):1–25, 2019.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 199, + 507, + 215 + ], + "spans": [ + { + "bbox": [ + 110, + 199, + 507, + 215 + ], + "score": 1.0, + "content": "[4] K. Behrendt, L. Novak, and R. Botros. A deep learning approach to traffic lights: Detection,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 127, + 210, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 127, + 210, + 506, + 225 + ], + "score": 1.0, + "content": "tracking, and classification. In Robotics and Automation (ICRA), 2017 IEEE International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 222, + 308, + 236 + ], + "spans": [ + { + "bbox": [ + 127, + 222, + 308, + 236 + ], + "score": 1.0, + "content": "Conference on, pp. 1370–1377. IEEE, 2017.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "[5] C.-A. Brust, S. Sickert, M. Simon, E. Rodner, and J. Denzler. Convolutional patch networks", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 127, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "with spatial prior for road detection and urban scene understanding. In International Joint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 125, + 273, + 198, + 286 + ], + "spans": [ + { + "bbox": [ + 125, + 273, + 198, + 286 + ], + "score": 1.0, + "content": "(VISAPP), 2015.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 110, + 292, + 506, + 306 + ], + "score": 1.0, + "content": "[6] G. F. Elsayed, P. Ramachandran, J. Shlens, and S. Kornblith. Revisiting spatial invariance with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 127, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 127, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "low-rank local connectivity. In International Conference on Machine Learning (ICML), 2020.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 110, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 110, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "[7] J. Geiping, H. Bauermeister, H. Droge, and M. Moeller. Inverting gradients–how easy is it to ¨", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 127, + 332, + 436, + 346 + ], + "spans": [ + { + "bbox": [ + 127, + 332, + 436, + 346 + ], + "score": 1.0, + "content": "break privacy in federated learning? arXiv preprint arXiv:2003.14053, 2020.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 110, + 351, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 110, + 351, + 506, + 365 + ], + "score": 1.0, + "content": "[8] R. Gens and P. M. Domingos. Deep symmetry networks. In Advances in neural information", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 362, + 343, + 376 + ], + "spans": [ + { + "bbox": [ + 125, + 362, + 343, + 376 + ], + "score": 1.0, + "content": "processing systems (NeurIPS), pp. 2537–2545, 2014.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 108, + 379, + 507, + 394 + ], + "spans": [ + { + "bbox": [ + 108, + 379, + 507, + 394 + ], + "score": 1.0, + "content": "[9] D. Griffith and C. Amrhein. An evaluation of correction techniques for boundary effects in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 125, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "spatial statistical analysis: traditional methods. Geographical Analysis, 15(4):352–360, 1983.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 410, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 424 + ], + "score": 1.0, + "content": "[10] V. Gupta and N. Ramani. A note on convolution and padding for two-dimensional data. Geo-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 421, + 307, + 435 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 307, + 435 + ], + "score": 1.0, + "content": "physical Prospecting, 26(1):214–217, 1978.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 440, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 454 + ], + "score": 1.0, + "content": "[11] L. Hamey. A functional approach to border handling in image processing. In International", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 451, + 477, + 465 + ], + "spans": [ + { + "bbox": [ + 127, + 451, + 477, + 465 + ], + "score": 1.0, + "content": "Conference on Digital Image Computing: Techniques and Applications, pp. 1–8, 2015.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "[12] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In IEEE", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 480, + 472, + 495 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 472, + 495 + ], + "score": 1.0, + "content": "conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "[13] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 126, + 509, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 126, + 509, + 505, + 524 + ], + "score": 1.0, + "content": "H. Adam. MobileNets: Efficient convolutional neural networks for mobile vision applications.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 521, + 290, + 535 + ], + "spans": [ + { + "bbox": [ + 127, + 521, + 290, + 535 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1704.04861, 2017.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 539, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 539, + 505, + 554 + ], + "score": 1.0, + "content": "[14] C. Innamorati, T. Ritschel, T. Weyrich, and N. J. Mitra. Learning on the edge: Investigating", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 127, + 551, + 501, + 565 + ], + "spans": [ + { + "bbox": [ + 127, + 551, + 501, + 565 + ], + "score": 1.0, + "content": "boundary filters in CNNs. International Journal of Computer Vision (IJCV), pp. 1–10, 2019.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "[15] S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 127, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 127, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "internal covariate shift. In International Conference on Machine Learning (ICML), pp. 448–", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 592, + 174, + 604 + ], + "spans": [ + { + "bbox": [ + 127, + 592, + 174, + 604 + ], + "score": 1.0, + "content": "456, 2015.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "score": 1.0, + "content": "[16] M. A. Islam, S. Jia, and N. D. Bruce. How much position information do convolutional neural", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 126, + 620, + 495, + 635 + ], + "spans": [ + { + "bbox": [ + 126, + 620, + 495, + 635 + ], + "score": 1.0, + "content": "networks encode? In International Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "[17] M. Jaderberg, K. Simonyan, A. Zisserman, et al. Spatial transformer networks. In Advances", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 126, + 650, + 431, + 664 + ], + "spans": [ + { + "bbox": [ + 126, + 650, + 431, + 664 + ], + "score": 1.0, + "content": "in neural information processing systems (NeurIPS), pp. 2017–2025, 2015.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "[18] O. S. Kayhan and J. C. van Gemert. On translation invariance in CNNs: Convolutional layers", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 127, + 680, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 127, + 680, + 506, + 694 + ], + "score": 1.0, + "content": "can exploit absolute spatial location. In IEEE conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 692, + 240, + 704 + ], + "spans": [ + { + "bbox": [ + 127, + 692, + 240, + 704 + ], + "score": 1.0, + "content": "Recognition (CVPR), 2020.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 709, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 724 + ], + "score": 1.0, + "content": "[19] T. L. Kijewski-Correa. Full-scale measurements and system identification: A time-frequency", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 721, + 364, + 735 + ], + "spans": [ + { + "bbox": [ + 126, + 721, + 364, + 735 + ], + "score": 1.0, + "content": "perspective. PhD thesis, University of Notre Dame., 2003.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 22.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 106, + 59, + 507, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 109, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "[1] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis,", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 126, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "J. Dean, et al. TensorFlow: Large-scale machine learning on heterogeneous distributed sys-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 123, + 315, + 135 + ], + "spans": [ + { + "bbox": [ + 127, + 123, + 315, + 135 + ], + "score": 1.0, + "content": "tems. arXiv preprint arXiv:1603.04467, 2016.", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 139, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 108, + 139, + 506, + 155 + ], + "score": 1.0, + "content": "[2] E. Akbas and M. P. Eckstein. Object detection through search with a foveated visual system.", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 153, + 347, + 164 + ], + "spans": [ + { + "bbox": [ + 127, + 153, + 347, + 164 + ], + "score": 1.0, + "content": "PLoS computational biology, 13(10):e1005743, 2017.", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 170, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 506, + 184 + ], + "score": 1.0, + "content": "[3] A. Azulay and Y. Weiss. Why do deep convolutional networks generalize so poorly to small", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 181, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 127, + 181, + 505, + 195 + ], + "score": 1.0, + "content": "image transformations? Journal of Machine Learning Research (JMLR), 20(184):1–25, 2019.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 199, + 507, + 215 + ], + "spans": [ + { + "bbox": [ + 110, + 199, + 507, + 215 + ], + "score": 1.0, + "content": "[4] K. Behrendt, L. Novak, and R. Botros. A deep learning approach to traffic lights: Detection,", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 210, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 127, + 210, + 506, + 225 + ], + "score": 1.0, + "content": "tracking, and classification. In Robotics and Automation (ICRA), 2017 IEEE International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 222, + 308, + 236 + ], + "spans": [ + { + "bbox": [ + 127, + 222, + 308, + 236 + ], + "score": 1.0, + "content": "Conference on, pp. 1370–1377. IEEE, 2017.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "[5] C.-A. Brust, S. Sickert, M. Simon, E. Rodner, and J. Denzler. Convolutional patch networks", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 127, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "with spatial prior for road detection and urban scene understanding. In International Joint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 125, + 273, + 198, + 286 + ], + "spans": [ + { + "bbox": [ + 125, + 273, + 198, + 286 + ], + "score": 1.0, + "content": "(VISAPP), 2015.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 110, + 292, + 506, + 306 + ], + "score": 1.0, + "content": "[6] G. F. Elsayed, P. Ramachandran, J. Shlens, and S. Kornblith. Revisiting spatial invariance with", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 127, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "low-rank local connectivity. In International Conference on Machine Learning (ICML), 2020.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 110, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 110, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "[7] J. Geiping, H. Bauermeister, H. Droge, and M. Moeller. Inverting gradients–how easy is it to ¨", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 332, + 436, + 346 + ], + "spans": [ + { + "bbox": [ + 127, + 332, + 436, + 346 + ], + "score": 1.0, + "content": "break privacy in federated learning? arXiv preprint arXiv:2003.14053, 2020.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 351, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 110, + 351, + 506, + 365 + ], + "score": 1.0, + "content": "[8] R. Gens and P. M. Domingos. Deep symmetry networks. In Advances in neural information", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 362, + 343, + 376 + ], + "spans": [ + { + "bbox": [ + 125, + 362, + 343, + 376 + ], + "score": 1.0, + "content": "processing systems (NeurIPS), pp. 2537–2545, 2014.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 379, + 507, + 394 + ], + "spans": [ + { + "bbox": [ + 108, + 379, + 507, + 394 + ], + "score": 1.0, + "content": "[9] D. Griffith and C. Amrhein. An evaluation of correction techniques for boundary effects in", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 125, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "spatial statistical analysis: traditional methods. Geographical Analysis, 15(4):352–360, 1983.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 410, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 424 + ], + "score": 1.0, + "content": "[10] V. Gupta and N. Ramani. A note on convolution and padding for two-dimensional data. Geo-", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 421, + 307, + 435 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 307, + 435 + ], + "score": 1.0, + "content": "physical Prospecting, 26(1):214–217, 1978.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 440, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 454 + ], + "score": 1.0, + "content": "[11] L. Hamey. A functional approach to border handling in image processing. In International", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 451, + 477, + 465 + ], + "spans": [ + { + "bbox": [ + 127, + 451, + 477, + 465 + ], + "score": 1.0, + "content": "Conference on Digital Image Computing: Techniques and Applications, pp. 1–8, 2015.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "[12] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In IEEE", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 480, + 472, + 495 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 472, + 495 + ], + "score": 1.0, + "content": "conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "[13] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 509, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 126, + 509, + 505, + 524 + ], + "score": 1.0, + "content": "H. Adam. MobileNets: Efficient convolutional neural networks for mobile vision applications.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 521, + 290, + 535 + ], + "spans": [ + { + "bbox": [ + 127, + 521, + 290, + 535 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1704.04861, 2017.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 539, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 539, + 505, + 554 + ], + "score": 1.0, + "content": "[14] C. Innamorati, T. Ritschel, T. Weyrich, and N. J. Mitra. Learning on the edge: Investigating", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 551, + 501, + 565 + ], + "spans": [ + { + "bbox": [ + 127, + 551, + 501, + 565 + ], + "score": 1.0, + "content": "boundary filters in CNNs. International Journal of Computer Vision (IJCV), pp. 1–10, 2019.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "[15] S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 127, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "internal covariate shift. In International Conference on Machine Learning (ICML), pp. 448–", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 592, + 174, + 604 + ], + "spans": [ + { + "bbox": [ + 127, + 592, + 174, + 604 + ], + "score": 1.0, + "content": "456, 2015.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "score": 1.0, + "content": "[16] M. A. Islam, S. Jia, and N. D. Bruce. How much position information do convolutional neural", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 620, + 495, + 635 + ], + "spans": [ + { + "bbox": [ + 126, + 620, + 495, + 635 + ], + "score": 1.0, + "content": "networks encode? In International Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "[17] M. Jaderberg, K. Simonyan, A. Zisserman, et al. Spatial transformer networks. In Advances", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 650, + 431, + 664 + ], + "spans": [ + { + "bbox": [ + 126, + 650, + 431, + 664 + ], + "score": 1.0, + "content": "in neural information processing systems (NeurIPS), pp. 2017–2025, 2015.", + "type": "text" + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "[18] O. S. Kayhan and J. C. van Gemert. On translation invariance in CNNs: Convolutional layers", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 680, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 127, + 680, + 506, + 694 + ], + "score": 1.0, + "content": "can exploit absolute spatial location. In IEEE conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 692, + 240, + 704 + ], + "spans": [ + { + "bbox": [ + 127, + 692, + 240, + 704 + ], + "score": 1.0, + "content": "Recognition (CVPR), 2020.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 709, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 724 + ], + "score": 1.0, + "content": "[19] T. L. Kijewski-Correa. Full-scale measurements and system identification: A time-frequency", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 721, + 364, + 735 + ], + "spans": [ + { + "bbox": [ + 126, + 721, + 364, + 735 + ], + "score": 1.0, + "content": "perspective. PhD thesis, University of Notre Dame., 2003.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "[20] I. Kim, W. Baek, and S. Kim. Spatially attentive output layer for image classification. In IEEE", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 94, + 417, + 107 + ], + "spans": [ + { + "bbox": [ + 125, + 94, + 417, + 107 + ], + "score": 1.0, + "content": "conference on Computer Vision and Pattern Recognition (CVPR), 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 114, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "[21] H. Larochelle and G. E. Hinton. Learning to combine foveal glimpses with a third-order", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 123, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 126, + 123, + 505, + 139 + ], + "score": 1.0, + "content": "boltzmann machine. In Advances in neural information processing systems (NeurIPS), pp.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 128, + 136, + 203, + 148 + ], + "spans": [ + { + "bbox": [ + 128, + 136, + 203, + 148 + ], + "score": 1.0, + "content": "1243–1251, 2010.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 154, + 507, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 507, + 170 + ], + "score": 1.0, + "content": "[22] G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro. Image inpainting for", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 167, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 127, + 167, + 505, + 181 + ], + "score": 1.0, + "content": "irregular holes using partial convolutions. In European Conference on Computer Vision, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "[23] G. Liu, K. J. Shih, T.-C. Wang, F. A. Reda, K. Sapra, Z. Yu, A. Tao, and B. Catanzaro. Partial", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 198, + 415, + 211 + ], + "spans": [ + { + "bbox": [ + 126, + 198, + 415, + 211 + ], + "score": 1.0, + "content": "convolution based padding. In arXiv preprint arXiv:1811.11718, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 218, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 231 + ], + "score": 1.0, + "content": "[24] R. Liu and J. Jia. Reducing boundary artifacts in image deconvolution. In IEEE International", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 230, + 373, + 242 + ], + "spans": [ + { + "bbox": [ + 128, + 230, + 373, + 242 + ], + "score": 1.0, + "content": "Conference on Image Processing (ICIP), pp. 505–508, 2008.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 247, + 507, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 507, + 265 + ], + "score": 1.0, + "content": "[25] R. Liu, J. Lehman, P. Molino, F. P. Such, E. Frank, A. Sergeev, and J. Yosinski. An intriguing", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 127, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "failing of convolutional neural networks and the CoordConv solution. In Advances in Neural", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 272, + 393, + 284 + ], + "spans": [ + { + "bbox": [ + 127, + 272, + 393, + 284 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), pp. 9605–9616, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 290, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 306 + ], + "score": 1.0, + "content": "[26] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. SSD: Single", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 303, + 477, + 317 + ], + "spans": [ + { + "bbox": [ + 127, + 303, + 477, + 317 + ], + "score": 1.0, + "content": "shot multibox detector. In European Conference on Computer Vision, pp. 21–37, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "[27] S. Lou, X. Jiang, and P. J. Scott. Fast algorithm for morphological filters. Journal of Physics:", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 334, + 294, + 346 + ], + "spans": [ + { + "bbox": [ + 127, + 334, + 294, + 346 + ], + "score": 1.0, + "content": "Conference Series, 311(1):012001, 2011.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "[28] W. Luo, Y. Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 127, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "deep convolutional neural networks. In Advances in Neural Information Processing Systems", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 376, + 264, + 390 + ], + "spans": [ + { + "bbox": [ + 127, + 376, + 264, + 390 + ], + "score": 1.0, + "content": "(NeurIPS), pp. 4898–4906, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 396, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 506, + 410 + ], + "score": 1.0, + "content": "[29] R. Murase, M. Suganuma, and T. Okatani. How can cnns use image position for segmentation?", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 407, + 292, + 420 + ], + "spans": [ + { + "bbox": [ + 125, + 407, + 292, + 420 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2005.03463, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 428, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 505, + 440 + ], + "score": 1.0, + "content": "[30] A.-D. Nguyen, S. Choi, W. Kim, S. Ahn, J. Kim, and S. Lee. Distribution padding in convo-", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 437, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 126, + 437, + 505, + 453 + ], + "score": 1.0, + "content": "lutional neural networks. In IEEE International Conference on Image Processing (ICIP), pp.", + "type": "text", + "cross_page": true + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 450, + 203, + 462 + ], + "spans": [ + { + "bbox": [ + 127, + 450, + 203, + 462 + ], + "score": 1.0, + "content": "4275–4279, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 467, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 104, + 467, + 507, + 485 + ], + "score": 1.0, + "content": "[31] D. Novotny, S. Albanie, D. Larlus, and A. Vedaldi. Semi-convolutional operators for instance", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 480, + 482, + 495 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 482, + 495 + ], + "score": 1.0, + "content": "segmentation. In European Conference on Computer Vision (ECCV), pp. 86–102, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 514 + ], + "score": 1.0, + "content": "[32] A. Odena, V. Dumoulin, and C. Olah. Deconvolution and checkerboard artifacts. Distill, 1", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 511, + 188, + 525 + ], + "spans": [ + { + "bbox": [ + 127, + 511, + 188, + 525 + ], + "score": 1.0, + "content": "(10):e3, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "score": 1.0, + "content": "[33] A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin,", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 539, + 507, + 559 + ], + "spans": [ + { + "bbox": [ + 124, + 539, + 507, + 559 + ], + "score": 1.0, + "content": "N. Gimelshein, et al. PyTorch: An imperative style, high-performance deep learning library.", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 553, + 486, + 568 + ], + "spans": [ + { + "bbox": [ + 127, + 553, + 486, + 568 + ], + "score": 1.0, + "content": "In Advances in Neural Information Processing Systems (NeurIPS), pp. 8024–8035, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 574, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 107, + 574, + 505, + 587 + ], + "score": 1.0, + "content": "[34] P. O. Pinheiro, T.-Y. Lin, R. Collobert, and P. Dollar. Learning to refine object segments. In ´", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 585, + 406, + 599 + ], + "spans": [ + { + "bbox": [ + 127, + 585, + 406, + 599 + ], + "score": 1.0, + "content": "European Conference on Computer Vision (ECCV), pp. 75–91, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "[35] S. Schubert, P. Neubert, J. Poschmann, and P. Pretzel. Circular convolutional neural networks ¨", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 616, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 127, + 616, + 505, + 631 + ], + "score": 1.0, + "content": "for panoramic images and laser data. In IEEE Intelligent Vehicles Symposium (IV), pp. 653–", + "type": "text", + "cross_page": true + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 627, + 175, + 641 + ], + "spans": [ + { + "bbox": [ + 127, + 627, + 175, + 641 + ], + "score": 1.0, + "content": "660, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 646, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 661 + ], + "score": 1.0, + "content": "[36] E. Shalnov. BSTLD-demo: A sample project to train and evaluate model on BSTLD. https:", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 658, + 333, + 671 + ], + "spans": [ + { + "bbox": [ + 127, + 658, + 333, + 671 + ], + "score": 1.0, + "content": "//github.com/e-sha/BSTLD_demo, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 676, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 505, + 694 + ], + "score": 1.0, + "content": "[37] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recog-", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 689, + 447, + 704 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 447, + 704 + ], + "score": 1.0, + "content": "nition. In International Conference on Learning Representations (ICLR), 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "[38] G. Sundaramoorthi and T. E. Wang. Translation insensitive CNNs. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 127, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "arXiv:1911.11238, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "[39] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "A. Rabinovich. Going deeper with convolutions. In IEEE conference on Computer Vision and", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 127, + 105, + 307, + 118 + ], + "spans": [ + { + "bbox": [ + 127, + 105, + 307, + 118 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), pp. 1–9, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 123, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 506, + 137 + ], + "score": 1.0, + "content": "[40] S. Vashishth, S. Sanyal, V. Nitin, N. Agrawal, and P. Talukdar. InteractE: Improving", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 135, + 506, + 147 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 506, + 147 + ], + "score": 1.0, + "content": "convolution-based knowledge graph embeddings by increasing feature interactions. In AAAI", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 146, + 298, + 158 + ], + "spans": [ + { + "bbox": [ + 127, + 146, + 298, + 158 + ], + "score": 1.0, + "content": "conference on Artifical Intelligence, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "[41] S. Wu, G. Wang, P. Tang, F. Chen, and L. Shi. Convolution with even-sized kernels and", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 174, + 507, + 191 + ], + "spans": [ + { + "bbox": [ + 126, + 174, + 507, + 191 + ], + "score": 1.0, + "content": "symmetric padding. In Advances in Neural Information Processing Systems (NeurIPS), pp.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 128, + 187, + 203, + 198 + ], + "spans": [ + { + "bbox": [ + 128, + 187, + 203, + 198 + ], + "score": 1.0, + "content": "1192–1203, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 204, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 506, + 219 + ], + "score": 1.0, + "content": "[42] F. Yu and V. Koltun. Multi-scale context aggregation by dilated convolutions. In International", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 216, + 352, + 230 + ], + "spans": [ + { + "bbox": [ + 127, + 216, + 352, + 230 + ], + "score": 1.0, + "content": "Conference on Learning Representations (ICLR), 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 234, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 506, + 249 + ], + "score": 1.0, + "content": "[43] R. Zhang. Making convolutional networks shift-invariant again. In International Conference", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 247, + 277, + 259 + ], + "spans": [ + { + "bbox": [ + 127, + 247, + 277, + 259 + ], + "score": 1.0, + "content": "on Machine Learning (ICML), 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 81, + 507, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 48, + 507, + 738 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "[20] I. Kim, W. Baek, and S. Kim. Spatially attentive output layer for image classification. In IEEE", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 125, + 94, + 417, + 107 + ], + "spans": [ + { + "bbox": [ + 125, + 94, + 417, + 107 + ], + "score": 1.0, + "content": "conference on Computer Vision and Pattern Recognition (CVPR), 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 114, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "[21] H. Larochelle and G. E. Hinton. Learning to combine foveal glimpses with a third-order", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 123, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 126, + 123, + 505, + 139 + ], + "score": 1.0, + "content": "boltzmann machine. In Advances in neural information processing systems (NeurIPS), pp.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 128, + 136, + 203, + 148 + ], + "spans": [ + { + "bbox": [ + 128, + 136, + 203, + 148 + ], + "score": 1.0, + "content": "1243–1251, 2010.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 154, + 507, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 507, + 170 + ], + "score": 1.0, + "content": "[22] G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro. Image inpainting for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 167, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 127, + 167, + 505, + 181 + ], + "score": 1.0, + "content": "irregular holes using partial convolutions. In European Conference on Computer Vision, 2018.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "[23] G. Liu, K. J. Shih, T.-C. Wang, F. A. Reda, K. Sapra, Z. Yu, A. Tao, and B. Catanzaro. Partial", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 198, + 415, + 211 + ], + "spans": [ + { + "bbox": [ + 126, + 198, + 415, + 211 + ], + "score": 1.0, + "content": "convolution based padding. In arXiv preprint arXiv:1811.11718, 2018.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 218, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 231 + ], + "score": 1.0, + "content": "[24] R. Liu and J. Jia. Reducing boundary artifacts in image deconvolution. In IEEE International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 128, + 230, + 373, + 242 + ], + "spans": [ + { + "bbox": [ + 128, + 230, + 373, + 242 + ], + "score": 1.0, + "content": "Conference on Image Processing (ICIP), pp. 505–508, 2008.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 247, + 507, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 507, + 265 + ], + "score": 1.0, + "content": "[25] R. Liu, J. Lehman, P. Molino, F. P. Such, E. Frank, A. Sergeev, and J. Yosinski. An intriguing", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 127, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "failing of convolutional neural networks and the CoordConv solution. In Advances in Neural", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 272, + 393, + 284 + ], + "spans": [ + { + "bbox": [ + 127, + 272, + 393, + 284 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), pp. 9605–9616, 2018.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 290, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 306 + ], + "score": 1.0, + "content": "[26] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. SSD: Single", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 303, + 477, + 317 + ], + "spans": [ + { + "bbox": [ + 127, + 303, + 477, + 317 + ], + "score": 1.0, + "content": "shot multibox detector. In European Conference on Computer Vision, pp. 21–37, 2016.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "[27] S. Lou, X. Jiang, and P. J. Scott. Fast algorithm for morphological filters. Journal of Physics:", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 334, + 294, + 346 + ], + "spans": [ + { + "bbox": [ + 127, + 334, + 294, + 346 + ], + "score": 1.0, + "content": "Conference Series, 311(1):012001, 2011.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "[28] W. Luo, Y. Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 127, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 127, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "deep convolutional neural networks. In Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 376, + 264, + 390 + ], + "spans": [ + { + "bbox": [ + 127, + 376, + 264, + 390 + ], + "score": 1.0, + "content": "(NeurIPS), pp. 4898–4906, 2016.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 396, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 506, + 410 + ], + "score": 1.0, + "content": "[29] R. Murase, M. Suganuma, and T. Okatani. How can cnns use image position for segmentation?", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 407, + 292, + 420 + ], + "spans": [ + { + "bbox": [ + 125, + 407, + 292, + 420 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2005.03463, 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 428, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 505, + 440 + ], + "score": 1.0, + "content": "[30] A.-D. Nguyen, S. Choi, W. Kim, S. Ahn, J. Kim, and S. Lee. Distribution padding in convo-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 437, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 126, + 437, + 505, + 453 + ], + "score": 1.0, + "content": "lutional neural networks. In IEEE International Conference on Image Processing (ICIP), pp.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 450, + 203, + 462 + ], + "spans": [ + { + "bbox": [ + 127, + 450, + 203, + 462 + ], + "score": 1.0, + "content": "4275–4279, 2019.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 467, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 104, + 467, + 507, + 485 + ], + "score": 1.0, + "content": "[31] D. Novotny, S. Albanie, D. Larlus, and A. Vedaldi. Semi-convolutional operators for instance", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 127, + 480, + 482, + 495 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 482, + 495 + ], + "score": 1.0, + "content": "segmentation. In European Conference on Computer Vision (ECCV), pp. 86–102, 2018.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 514 + ], + "score": 1.0, + "content": "[32] A. Odena, V. Dumoulin, and C. Olah. Deconvolution and checkerboard artifacts. Distill, 1", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 127, + 511, + 188, + 525 + ], + "spans": [ + { + "bbox": [ + 127, + 511, + 188, + 525 + ], + "score": 1.0, + "content": "(10):e3, 2016.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "score": 1.0, + "content": "[33] A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 124, + 539, + 507, + 559 + ], + "spans": [ + { + "bbox": [ + 124, + 539, + 507, + 559 + ], + "score": 1.0, + "content": "N. Gimelshein, et al. PyTorch: An imperative style, high-performance deep learning library.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 553, + 486, + 568 + ], + "spans": [ + { + "bbox": [ + 127, + 553, + 486, + 568 + ], + "score": 1.0, + "content": "In Advances in Neural Information Processing Systems (NeurIPS), pp. 8024–8035, 2019.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 574, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 107, + 574, + 505, + 587 + ], + "score": 1.0, + "content": "[34] P. O. Pinheiro, T.-Y. Lin, R. Collobert, and P. Dollar. Learning to refine object segments. In ´", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 127, + 585, + 406, + 599 + ], + "spans": [ + { + "bbox": [ + 127, + 585, + 406, + 599 + ], + "score": 1.0, + "content": "European Conference on Computer Vision (ECCV), pp. 75–91, 2016.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "[35] S. Schubert, P. Neubert, J. Poschmann, and P. Pretzel. Circular convolutional neural networks ¨", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 616, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 127, + 616, + 505, + 631 + ], + "score": 1.0, + "content": "for panoramic images and laser data. In IEEE Intelligent Vehicles Symposium (IV), pp. 653–", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 627, + 175, + 641 + ], + "spans": [ + { + "bbox": [ + 127, + 627, + 175, + 641 + ], + "score": 1.0, + "content": "660, 2019.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 646, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 661 + ], + "score": 1.0, + "content": "[36] E. Shalnov. BSTLD-demo: A sample project to train and evaluate model on BSTLD. https:", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 127, + 658, + 333, + 671 + ], + "spans": [ + { + "bbox": [ + 127, + 658, + 333, + 671 + ], + "score": 1.0, + "content": "//github.com/e-sha/BSTLD_demo, 2019.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 676, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 505, + 694 + ], + "score": 1.0, + "content": "[37] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recog-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 127, + 689, + 447, + 704 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 447, + 704 + ], + "score": 1.0, + "content": "nition. In International Conference on Learning Representations (ICLR), 2015.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "[38] G. Sundaramoorthi and T. E. Wang. Translation insensitive CNNs. arXiv preprint", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 127, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "arXiv:1911.11238, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 21.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 761 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "11", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 48, + 507, + 738 + ], + "lines": [], + "index": 21.5, + "bbox_fs": [ + 104, + 82, + 507, + 733 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 506, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "[39] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "A. Rabinovich. Going deeper with convolutions. In IEEE conference on Computer Vision and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 127, + 105, + 307, + 118 + ], + "spans": [ + { + "bbox": [ + 127, + 105, + 307, + 118 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), pp. 1–9, 2015.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 123, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 506, + 137 + ], + "score": 1.0, + "content": "[40] S. Vashishth, S. Sanyal, V. Nitin, N. Agrawal, and P. Talukdar. InteractE: Improving", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 127, + 135, + 506, + 147 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 506, + 147 + ], + "score": 1.0, + "content": "convolution-based knowledge graph embeddings by increasing feature interactions. In AAAI", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 146, + 298, + 158 + ], + "spans": [ + { + "bbox": [ + 127, + 146, + 298, + 158 + ], + "score": 1.0, + "content": "conference on Artifical Intelligence, 2020.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "[41] S. Wu, G. Wang, P. Tang, F. Chen, and L. Shi. Convolution with even-sized kernels and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 174, + 507, + 191 + ], + "spans": [ + { + "bbox": [ + 126, + 174, + 507, + 191 + ], + "score": 1.0, + "content": "symmetric padding. In Advances in Neural Information Processing Systems (NeurIPS), pp.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 128, + 187, + 203, + 198 + ], + "spans": [ + { + "bbox": [ + 128, + 187, + 203, + 198 + ], + "score": 1.0, + "content": "1192–1203, 2019.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 204, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 506, + 219 + ], + "score": 1.0, + "content": "[42] F. Yu and V. Koltun. Multi-scale context aggregation by dilated convolutions. In International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 216, + 352, + 230 + ], + "spans": [ + { + "bbox": [ + 127, + 216, + 352, + 230 + ], + "score": 1.0, + "content": "Conference on Learning Representations (ICLR), 2016.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 234, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 506, + 249 + ], + "score": 1.0, + "content": "[43] R. Zhang. Making convolutional networks shift-invariant again. In International Conference", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 247, + 277, + 259 + ], + "spans": [ + { + "bbox": [ + 127, + 247, + 277, + 259 + ], + "score": 1.0, + "content": "on Machine Learning (ICML), 2019.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 82, + 506, + 258 + ], + "lines": [], + "index": 6, + "bbox_fs": [ + 105, + 82, + 507, + 259 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 387, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 388, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 388, + 96 + ], + "score": 1.0, + "content": "A ELIMINATING UNEVEN APPLICATION OF PADDING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 105, + 505, + 173 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 196, + 119 + ], + "score": 1.0, + "content": "Consider a CNN with", + "type": "text" + }, + { + "bbox": [ + 196, + 107, + 203, + 116 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 106, + 294, + 119 + ], + "score": 1.0, + "content": "downsampling layers,", + "type": "text" + }, + { + "bbox": [ + 295, + 106, + 351, + 118 + ], + "score": 0.93, + "content": "L _ { 1 } , L _ { 2 } , . . . , L _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 106, + 506, + 119 + ], + "score": 1.0, + "content": ". To simplify the analysis and without", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "loss of generality we assume that the kernels in these layers are of square shape and that all other", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 295, + 140 + ], + "score": 1.0, + "content": "layers maintain their input size. We denote by", + "type": "text" + }, + { + "bbox": [ + 296, + 130, + 305, + 139 + ], + "score": 0.85, + "content": "s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 128, + 323, + 140 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 324, + 129, + 334, + 139 + ], + "score": 0.88, + "content": "k _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 128, + 472, + 140 + ], + "score": 1.0, + "content": "the stride and kernel size of layer", + "type": "text" + }, + { + "bbox": [ + 473, + 129, + 484, + 139 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 128, + 505, + 140 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 504, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 148, + 152 + ], + "score": 1.0, + "content": "denote by", + "type": "text" + }, + { + "bbox": [ + 148, + 140, + 159, + 150 + ], + "score": 0.88, + "content": "h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 139, + 177, + 152 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 140, + 189, + 150 + ], + "score": 0.86, + "content": "w _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 139, + 387, + 152 + ], + "score": 1.0, + "content": "the dimensions of the feature maps computed by", + "type": "text" + }, + { + "bbox": [ + 387, + 140, + 398, + 150 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 139, + 461, + 152 + ], + "score": 1.0, + "content": ". We denote by", + "type": "text" + }, + { + "bbox": [ + 461, + 140, + 473, + 150 + ], + "score": 0.88, + "content": "h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 139, + 491, + 152 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 491, + 141, + 504, + 150 + ], + "score": 0.83, + "content": "w _ { 0 }", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 148, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 164 + ], + "score": 1.0, + "content": "the size of the CNN input. We examine the conditions to warrant no uneven application of padding", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 162, + 419, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 419, + 173 + ], + "score": 1.0, + "content": "along the height dimension. Parallel conditions apply to the width dimension.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 179, + 504, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 163, + 192 + ], + "score": 1.0, + "content": "We denote by", + "type": "text" + }, + { + "bbox": [ + 164, + 179, + 174, + 191 + ], + "score": 0.88, + "content": "\\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 179, + 308, + 192 + ], + "score": 1.0, + "content": "the height of the padded input to", + "type": "text" + }, + { + "bbox": [ + 308, + 180, + 319, + 191 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 179, + 410, + 192 + ], + "score": 1.0, + "content": ". The effective portion", + "type": "text" + }, + { + "bbox": [ + 410, + 178, + 443, + 191 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } \\leq \\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "of this amount", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 327, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 267, + 203 + ], + "score": 1.0, + "content": "processed by the convolutional filters in", + "type": "text" + }, + { + "bbox": [ + 268, + 191, + 279, + 201 + ], + "score": 0.88, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 190, + 327, + 203 + ], + "score": 1.0, + "content": "is equal to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 207, + 352, + 223 + ], + "lines": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "spans": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } = s _ { i } \\cdot \\left( h _ { i } - 1 \\right) + k _ { i }", + "type": "interline_equation", + "image_path": "4899e2975dfc90d0a8c069ee157694d33649b53ae558b1470547ef4b9d531efb.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 230, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 227, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 215, + 245 + ], + "score": 1.0, + "content": "Our goal is to warrant that", + "type": "text" + }, + { + "bbox": [ + 216, + 228, + 249, + 242 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } = \\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 227, + 506, + 245 + ], + "score": 1.0, + "content": "to prevent information loss and to avoid uneven padding along", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 241, + 434, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 307, + 255 + ], + "score": 1.0, + "content": "the vertical dimension when the unconsumed part", + "type": "text" + }, + { + "bbox": [ + 307, + 241, + 360, + 254 + ], + "score": 0.93, + "content": "\\bar { h } _ { i } - \\hat { h } _ { i } < s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 242, + 434, + 255 + ], + "score": 1.0, + "content": "is an odd number.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 105, + 259, + 504, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "score": 1.0, + "content": "Since the non-downsampling layers maintain their input size, we can formulate the height of the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 271, + 207, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 207, + 283 + ], + "score": 1.0, + "content": "padded input as follows:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 279, + 344, + 294 + ], + "lines": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "spans": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "score": 0.92, + "content": "\\bar { h } _ { i } = h _ { i - 1 } + 2 \\cdot p _ { i }", + "type": "interline_equation", + "image_path": "797456e257ded21d46fedaac314e9e3c7963e74ea1f2282d7215bc5e8eece948.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 504, + 319 + ], + "lines": [ + { + "bbox": [ + 106, + 297, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 133, + 309 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 298, + 143, + 308 + ], + "score": 0.85, + "content": "p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 297, + 453, + 309 + ], + "score": 1.0, + "content": "is the amount of padding applied at the top and at the bottom of the input in", + "type": "text" + }, + { + "bbox": [ + 453, + 297, + 464, + 308 + ], + "score": 0.89, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 297, + 505, + 309 + ], + "score": 1.0, + "content": ". Accord-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 307, + 364, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 364, + 320 + ], + "score": 1.0, + "content": "ingly, we can warrant no uneven padding if the following holds:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 325, + 407, + 339 + ], + "lines": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "spans": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "score": 0.89, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = s _ { i } \\cdot ( h _ { i } - 1 ) + k _ { i } - 2 \\cdot p _ { i }", + "type": "interline_equation", + "image_path": "a6ffbc359fab9bb87c045b8de51352b91c855be9bb9ef831966ddb179afca6e9.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 410, + 362 + ], + "score": 1.0, + "content": "Example 1: ResNet-18 This network contains five downsampling layers (", + "type": "text" + }, + { + "bbox": [ + 410, + 350, + 435, + 360 + ], + "score": 0.86, + "content": "\\mathrm { : } d = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 349, + 505, + 362 + ], + "score": 1.0, + "content": ") all of which use", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 360, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 506, + 374 + ], + "score": 1.0, + "content": "a stride of 2. Despite performing downsampling, all of these layers apply a padding amount entailed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 371, + 504, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 483, + 384 + ], + "score": 1.0, + "content": "by SAME padding to avoid information bias against the boundary. In four of these layers having", + "type": "text" + }, + { + "bbox": [ + 483, + 372, + 504, + 383 + ], + "score": 0.87, + "content": "3 \\times 3", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 383, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 140, + 395 + ], + "score": 1.0, + "content": "kernels", + "type": "text" + }, + { + "bbox": [ + 141, + 383, + 170, + 393 + ], + "score": 0.87, + "content": "k _ { i } = 3 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 383, + 253, + 395 + ], + "score": 1.0, + "content": "), the amount used is", + "type": "text" + }, + { + "bbox": [ + 254, + 383, + 282, + 394 + ], + "score": 0.9, + "content": "p _ { i } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 383, + 388, + 395 + ], + "score": 1.0, + "content": ". For the first layer having", + "type": "text" + }, + { + "bbox": [ + 388, + 383, + 412, + 393 + ], + "score": 0.89, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 383, + 505, + 395 + ], + "score": 1.0, + "content": "kernels, this amount is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 246, + 407 + ], + "score": 1.0, + "content": "equal to 3. In both cases, the term", + "type": "text" + }, + { + "bbox": [ + 246, + 394, + 289, + 405 + ], + "score": 0.91, + "content": "k _ { i } - 2 \\cdot p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 393, + 316, + 407 + ], + "score": 1.0, + "content": "in Eq.", + "type": "text" + }, + { + "bbox": [ + 316, + 393, + 326, + 406 + ], + "score": 0.83, + "content": "3", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 393, + 505, + 407 + ], + "score": 1.0, + "content": "is equal to 1. To warrant no uneven padding", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "along the vertical dimension, the heights of the feature maps at downsampling layers should hence", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 416, + 138, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 138, + 428 + ], + "score": 1.0, + "content": "satisfy:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 425, + 412, + 439 + ], + "lines": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "spans": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "score": 0.89, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 1 = 2 \\cdot h _ { i } - 1", + "type": "interline_equation", + "image_path": "f4c20619d23c945d28ff9fb47cf60037cd7d565436bbf46e60e9e1fb979e72bd.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 441, + 285, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 441, + 286, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 286, + 455 + ], + "score": 1.0, + "content": "Accordingly, the input height should satisfy:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 457, + 396, + 473 + ], + "lines": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "spans": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "score": 0.93, + "content": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } - ( 2 ^ { d } - 1 ) = 2 ^ { d } \\cdot ( h _ { d } - 1 ) + 1", + "type": "interline_equation", + "image_path": "6fe3f2125e90aca84b006cdd0eaf043e7d529e887d0d0da2d4895e8e626c76ee.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 478, + 504, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 133, + 491 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 479, + 145, + 489 + ], + "score": 0.88, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "is the height of the final feature map, and can be any natural number larger than 1 to avoid", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 489, + 392, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 196, + 501 + ], + "score": 1.0, + "content": "a degenerate case of a", + "type": "text" + }, + { + "bbox": [ + 196, + 490, + 219, + 500 + ], + "score": 0.9, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 489, + 392, + 501 + ], + "score": 1.0, + "content": "input. The same holds for the input width:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "interline_equation", + "bbox": [ + 256, + 505, + 355, + 521 + ], + "lines": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "spans": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "score": 0.92, + "content": "w _ { 0 } = 2 ^ { d } \\cdot ( w _ { d } - 1 ) + 1", + "type": "interline_equation", + "image_path": "9cb47d9168d465e0eb73d6b4ce4a5b226c2dd4a1e46fd194b85d4a9a707400ee.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 504, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 116, + 545 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 116, + 532, + 155, + 543 + ], + "score": 0.87, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 531, + 304, + 545 + ], + "score": 1.0, + "content": "input satisfies these constraints since", + "type": "text" + }, + { + "bbox": [ + 304, + 532, + 371, + 543 + ], + "score": 0.91, + "content": "2 2 5 = 2 ^ { 5 } \\cdot 7 + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 531, + 505, + 545 + ], + "score": 1.0, + "content": ", yielding even padding in all five", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 349, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 321, + 556 + ], + "score": 1.0, + "content": "downsampling layers and output feature maps of size", + "type": "text" + }, + { + "bbox": [ + 321, + 544, + 344, + 554 + ], + "score": 0.88, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 543, + 349, + 556 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 398, + 580 + ], + "score": 1.0, + "content": "Example 2: VGG-16 This network contains five max-pooling layers", + "type": "text" + }, + { + "bbox": [ + 398, + 567, + 426, + 577 + ], + "score": 0.85, + "content": "( d = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 565, + 506, + 580 + ], + "score": 1.0, + "content": ") all of which use a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "stride of 2 and a kernel size of 2 and apply no padding. To warrant no uneven padding along the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 589, + 478, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 478, + 601 + ], + "score": 1.0, + "content": "vertical dimension, the heights of the feature maps at all of these layers should hence satisfy:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 605, + 404, + 619 + ], + "lines": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "spans": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "score": 0.87, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 2 = 2 \\cdot h _ { i }", + "type": "interline_equation", + "image_path": "6e698528c3d9d0dab1ced918089a3dd957d677e240f61ace78c08761eec3fef4.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 624, + 306, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 306, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 306, + 638 + ], + "score": 1.0, + "content": "Accordingly, the input dimensions should satisfy:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 641, + 377, + 655 + ], + "lines": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "spans": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "score": 0.91, + "content": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } \\quad \\mathrm { a n d } \\quad w _ { 0 } = 2 ^ { d } \\cdot w _ { d }", + "type": "interline_equation", + "image_path": "3cd98ef77ef0095452f1a67be1696c066ece51daf7d0ea9598b0c464ca7d450f.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 509, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 116, + 675 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 662, + 156, + 672 + ], + "score": 0.87, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 659, + 308, + 675 + ], + "score": 1.0, + "content": "input satisfies these constraints since", + "type": "text" + }, + { + "bbox": [ + 308, + 661, + 362, + 672 + ], + "score": 0.9, + "content": "2 2 4 = 2 ^ { 5 } \\cdot 7", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 659, + 506, + 675 + ], + "score": 1.0, + "content": ", causing no feature-map erosion at", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 672, + 410, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 381, + 685 + ], + "score": 1.0, + "content": "any downsampling layer and resulting in output feature maps of size", + "type": "text" + }, + { + "bbox": [ + 381, + 673, + 405, + 683 + ], + "score": 0.9, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 672, + 410, + 685 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 387, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 388, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 388, + 96 + ], + "score": 1.0, + "content": "A ELIMINATING UNEVEN APPLICATION OF PADDING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 105, + 505, + 173 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 196, + 119 + ], + "score": 1.0, + "content": "Consider a CNN with", + "type": "text" + }, + { + "bbox": [ + 196, + 107, + 203, + 116 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 106, + 294, + 119 + ], + "score": 1.0, + "content": "downsampling layers,", + "type": "text" + }, + { + "bbox": [ + 295, + 106, + 351, + 118 + ], + "score": 0.93, + "content": "L _ { 1 } , L _ { 2 } , . . . , L _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 106, + 506, + 119 + ], + "score": 1.0, + "content": ". To simplify the analysis and without", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "loss of generality we assume that the kernels in these layers are of square shape and that all other", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 295, + 140 + ], + "score": 1.0, + "content": "layers maintain their input size. We denote by", + "type": "text" + }, + { + "bbox": [ + 296, + 130, + 305, + 139 + ], + "score": 0.85, + "content": "s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 128, + 323, + 140 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 324, + 129, + 334, + 139 + ], + "score": 0.88, + "content": "k _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 128, + 472, + 140 + ], + "score": 1.0, + "content": "the stride and kernel size of layer", + "type": "text" + }, + { + "bbox": [ + 473, + 129, + 484, + 139 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 128, + 505, + 140 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 504, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 148, + 152 + ], + "score": 1.0, + "content": "denote by", + "type": "text" + }, + { + "bbox": [ + 148, + 140, + 159, + 150 + ], + "score": 0.88, + "content": "h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 139, + 177, + 152 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 140, + 189, + 150 + ], + "score": 0.86, + "content": "w _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 139, + 387, + 152 + ], + "score": 1.0, + "content": "the dimensions of the feature maps computed by", + "type": "text" + }, + { + "bbox": [ + 387, + 140, + 398, + 150 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 139, + 461, + 152 + ], + "score": 1.0, + "content": ". We denote by", + "type": "text" + }, + { + "bbox": [ + 461, + 140, + 473, + 150 + ], + "score": 0.88, + "content": "h _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 139, + 491, + 152 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 491, + 141, + 504, + 150 + ], + "score": 0.83, + "content": "w _ { 0 }", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 148, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 164 + ], + "score": 1.0, + "content": "the size of the CNN input. We examine the conditions to warrant no uneven application of padding", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 162, + 419, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 419, + 173 + ], + "score": 1.0, + "content": "along the height dimension. Parallel conditions apply to the width dimension.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 106, + 506, + 173 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 179, + 504, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 163, + 192 + ], + "score": 1.0, + "content": "We denote by", + "type": "text" + }, + { + "bbox": [ + 164, + 179, + 174, + 191 + ], + "score": 0.88, + "content": "\\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 179, + 308, + 192 + ], + "score": 1.0, + "content": "the height of the padded input to", + "type": "text" + }, + { + "bbox": [ + 308, + 180, + 319, + 191 + ], + "score": 0.87, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 179, + 410, + 192 + ], + "score": 1.0, + "content": ". The effective portion", + "type": "text" + }, + { + "bbox": [ + 410, + 178, + 443, + 191 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } \\leq \\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "of this amount", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 327, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 267, + 203 + ], + "score": 1.0, + "content": "processed by the convolutional filters in", + "type": "text" + }, + { + "bbox": [ + 268, + 191, + 279, + 201 + ], + "score": 0.88, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 190, + 327, + 203 + ], + "score": 1.0, + "content": "is equal to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 178, + 505, + 203 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 207, + 352, + 223 + ], + "lines": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "spans": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } = s _ { i } \\cdot \\left( h _ { i } - 1 \\right) + k _ { i }", + "type": "interline_equation", + "image_path": "4899e2975dfc90d0a8c069ee157694d33649b53ae558b1470547ef4b9d531efb.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 258, + 207, + 352, + 223 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 230, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 227, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 215, + 245 + ], + "score": 1.0, + "content": "Our goal is to warrant that", + "type": "text" + }, + { + "bbox": [ + 216, + 228, + 249, + 242 + ], + "score": 0.93, + "content": "\\hat { h } _ { i } = \\bar { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 227, + 506, + 245 + ], + "score": 1.0, + "content": "to prevent information loss and to avoid uneven padding along", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 241, + 434, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 307, + 255 + ], + "score": 1.0, + "content": "the vertical dimension when the unconsumed part", + "type": "text" + }, + { + "bbox": [ + 307, + 241, + 360, + 254 + ], + "score": 0.93, + "content": "\\bar { h } _ { i } - \\hat { h } _ { i } < s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 242, + 434, + 255 + ], + "score": 1.0, + "content": "is an odd number.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 227, + 506, + 255 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 259, + 504, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "score": 1.0, + "content": "Since the non-downsampling layers maintain their input size, we can formulate the height of the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 271, + 207, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 207, + 283 + ], + "score": 1.0, + "content": "padded input as follows:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 259, + 505, + 283 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 279, + 344, + 294 + ], + "lines": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "spans": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "score": 0.92, + "content": "\\bar { h } _ { i } = h _ { i - 1 } + 2 \\cdot p _ { i }", + "type": "interline_equation", + "image_path": "797456e257ded21d46fedaac314e9e3c7963e74ea1f2282d7215bc5e8eece948.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 267, + 279, + 344, + 294 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 504, + 319 + ], + "lines": [ + { + "bbox": [ + 106, + 297, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 133, + 309 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 298, + 143, + 308 + ], + "score": 0.85, + "content": "p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 297, + 453, + 309 + ], + "score": 1.0, + "content": "is the amount of padding applied at the top and at the bottom of the input in", + "type": "text" + }, + { + "bbox": [ + 453, + 297, + 464, + 308 + ], + "score": 0.89, + "content": "L _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 297, + 505, + 309 + ], + "score": 1.0, + "content": ". Accord-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 307, + 364, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 364, + 320 + ], + "score": 1.0, + "content": "ingly, we can warrant no uneven padding if the following holds:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 297, + 505, + 320 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 325, + 407, + 339 + ], + "lines": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "spans": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "score": 0.89, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = s _ { i } \\cdot ( h _ { i } - 1 ) + k _ { i } - 2 \\cdot p _ { i }", + "type": "interline_equation", + "image_path": "a6ffbc359fab9bb87c045b8de51352b91c855be9bb9ef831966ddb179afca6e9.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 204, + 325, + 407, + 339 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 410, + 362 + ], + "score": 1.0, + "content": "Example 1: ResNet-18 This network contains five downsampling layers (", + "type": "text" + }, + { + "bbox": [ + 410, + 350, + 435, + 360 + ], + "score": 0.86, + "content": "\\mathrm { : } d = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 349, + 505, + 362 + ], + "score": 1.0, + "content": ") all of which use", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 360, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 506, + 374 + ], + "score": 1.0, + "content": "a stride of 2. Despite performing downsampling, all of these layers apply a padding amount entailed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 371, + 504, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 483, + 384 + ], + "score": 1.0, + "content": "by SAME padding to avoid information bias against the boundary. In four of these layers having", + "type": "text" + }, + { + "bbox": [ + 483, + 372, + 504, + 383 + ], + "score": 0.87, + "content": "3 \\times 3", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 383, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 140, + 395 + ], + "score": 1.0, + "content": "kernels", + "type": "text" + }, + { + "bbox": [ + 141, + 383, + 170, + 393 + ], + "score": 0.87, + "content": "k _ { i } = 3 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 383, + 253, + 395 + ], + "score": 1.0, + "content": "), the amount used is", + "type": "text" + }, + { + "bbox": [ + 254, + 383, + 282, + 394 + ], + "score": 0.9, + "content": "p _ { i } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 383, + 388, + 395 + ], + "score": 1.0, + "content": ". For the first layer having", + "type": "text" + }, + { + "bbox": [ + 388, + 383, + 412, + 393 + ], + "score": 0.89, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 383, + 505, + 395 + ], + "score": 1.0, + "content": "kernels, this amount is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 246, + 407 + ], + "score": 1.0, + "content": "equal to 3. In both cases, the term", + "type": "text" + }, + { + "bbox": [ + 246, + 394, + 289, + 405 + ], + "score": 0.91, + "content": "k _ { i } - 2 \\cdot p _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 393, + 316, + 407 + ], + "score": 1.0, + "content": "in Eq.", + "type": "text" + }, + { + "bbox": [ + 316, + 393, + 326, + 406 + ], + "score": 0.83, + "content": "3", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 393, + 505, + 407 + ], + "score": 1.0, + "content": "is equal to 1. To warrant no uneven padding", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "along the vertical dimension, the heights of the feature maps at downsampling layers should hence", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 416, + 138, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 138, + 428 + ], + "score": 1.0, + "content": "satisfy:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 349, + 506, + 428 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 425, + 412, + 439 + ], + "lines": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "spans": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "score": 0.89, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 1 = 2 \\cdot h _ { i } - 1", + "type": "interline_equation", + "image_path": "f4c20619d23c945d28ff9fb47cf60037cd7d565436bbf46e60e9e1fb979e72bd.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 197, + 425, + 412, + 439 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 441, + 285, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 441, + 286, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 286, + 455 + ], + "score": 1.0, + "content": "Accordingly, the input height should satisfy:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 441, + 286, + 455 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 457, + 396, + 473 + ], + "lines": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "spans": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "score": 0.93, + "content": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } - ( 2 ^ { d } - 1 ) = 2 ^ { d } \\cdot ( h _ { d } - 1 ) + 1", + "type": "interline_equation", + "image_path": "6fe3f2125e90aca84b006cdd0eaf043e7d529e887d0d0da2d4895e8e626c76ee.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 214, + 457, + 396, + 473 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 478, + 504, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 133, + 491 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 479, + 145, + 489 + ], + "score": 0.88, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "is the height of the final feature map, and can be any natural number larger than 1 to avoid", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 489, + 392, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 196, + 501 + ], + "score": 1.0, + "content": "a degenerate case of a", + "type": "text" + }, + { + "bbox": [ + 196, + 490, + 219, + 500 + ], + "score": 0.9, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 489, + 392, + 501 + ], + "score": 1.0, + "content": "input. The same holds for the input width:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 478, + 505, + 501 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 256, + 505, + 355, + 521 + ], + "lines": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "spans": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "score": 0.92, + "content": "w _ { 0 } = 2 ^ { d } \\cdot ( w _ { d } - 1 ) + 1", + "type": "interline_equation", + "image_path": "9cb47d9168d465e0eb73d6b4ce4a5b226c2dd4a1e46fd194b85d4a9a707400ee.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 256, + 505, + 355, + 521 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 504, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 116, + 545 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 116, + 532, + 155, + 543 + ], + "score": 0.87, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 531, + 304, + 545 + ], + "score": 1.0, + "content": "input satisfies these constraints since", + "type": "text" + }, + { + "bbox": [ + 304, + 532, + 371, + 543 + ], + "score": 0.91, + "content": "2 2 5 = 2 ^ { 5 } \\cdot 7 + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 531, + 505, + 545 + ], + "score": 1.0, + "content": ", yielding even padding in all five", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 349, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 321, + 556 + ], + "score": 1.0, + "content": "downsampling layers and output feature maps of size", + "type": "text" + }, + { + "bbox": [ + 321, + 544, + 344, + 554 + ], + "score": 0.88, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 543, + 349, + 556 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 531, + 505, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 398, + 580 + ], + "score": 1.0, + "content": "Example 2: VGG-16 This network contains five max-pooling layers", + "type": "text" + }, + { + "bbox": [ + 398, + 567, + 426, + 577 + ], + "score": 0.85, + "content": "( d = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 565, + 506, + 580 + ], + "score": 1.0, + "content": ") all of which use a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "stride of 2 and a kernel size of 2 and apply no padding. To warrant no uneven padding along the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 589, + 478, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 478, + 601 + ], + "score": 1.0, + "content": "vertical dimension, the heights of the feature maps at all of these layers should hence satisfy:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 565, + 506, + 601 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 605, + 404, + 619 + ], + "lines": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "spans": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "score": 0.87, + "content": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 2 = 2 \\cdot h _ { i }", + "type": "interline_equation", + "image_path": "6e698528c3d9d0dab1ced918089a3dd957d677e240f61ace78c08761eec3fef4.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 206, + 605, + 404, + 619 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 624, + 306, + 636 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 306, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 306, + 638 + ], + "score": 1.0, + "content": "Accordingly, the input dimensions should satisfy:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 623, + 306, + 638 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 641, + 377, + 655 + ], + "lines": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "spans": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "score": 0.91, + "content": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } \\quad \\mathrm { a n d } \\quad w _ { 0 } = 2 ^ { d } \\cdot w _ { d }", + "type": "interline_equation", + "image_path": "3cd98ef77ef0095452f1a67be1696c066ece51daf7d0ea9598b0c464ca7d450f.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 234, + 641, + 377, + 655 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 509, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 116, + 675 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 662, + 156, + 672 + ], + "score": 0.87, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 659, + 308, + 675 + ], + "score": 1.0, + "content": "input satisfies these constraints since", + "type": "text" + }, + { + "bbox": [ + 308, + 661, + 362, + 672 + ], + "score": 0.9, + "content": "2 2 4 = 2 ^ { 5 } \\cdot 7", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 659, + 506, + 675 + ], + "score": 1.0, + "content": ", causing no feature-map erosion at", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 672, + 410, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 381, + 685 + ], + "score": 1.0, + "content": "any downsampling layer and resulting in output feature maps of size", + "type": "text" + }, + { + "bbox": [ + 381, + 673, + 405, + 683 + ], + "score": 0.9, + "content": "7 \\times 7", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 672, + 410, + 685 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 659, + 506, + 685 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 406, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 406, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 406, + 95 + ], + "score": 1.0, + "content": "B THE EXTENT OF FOVEATION UNDER SAME 0-PADDING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 503, + 129 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 119 + ], + "score": 1.0, + "content": "We illustrate how the absolute extent of foveation under SAME 0-padding depends on the number of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 404, + 130 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 404, + 130 + ], + "score": 1.0, + "content": "convolutional layers, and how its relative extent depends on the input size.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 105, + 134, + 504, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 504, + 146 + ], + "score": 1.0, + "content": "In the following maps, color represents the number of paths to the CNN output for each input pixel.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 449, + 158 + ], + "score": 1.0, + "content": "Note: The checkerboard pattern is caused by downsampling layers in ResNet that use", + "type": "text" + }, + { + "bbox": [ + 450, + 146, + 473, + 156 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "kernels", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 156, + 176, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 176, + 168 + ], + "score": 1.0, + "content": "and a stride of 2.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 176, + 177, + 437, + 295 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 176, + 177, + 437, + 295 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 177, + 437, + 295 + ], + "spans": [ + { + "bbox": [ + 176, + 177, + 437, + 295 + ], + "score": 0.969, + "type": "image", + "image_path": "230d99b41516d96c2c03aec57cab7c869711ad4d96a7315fd9c45cd64c50a0f0.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 176, + 177, + 437, + 216.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 176, + 216.33333333333334, + 437, + 255.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 176, + 255.66666666666669, + 437, + 295.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 305, + 505, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 305, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 506, + 317 + ], + "score": 1.0, + "content": "Figure 9: The foveation maps of two ResNet architectures under 0 padding, illustrated with a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 315, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 146, + 327 + ], + "score": 0.87, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 315, + 505, + 329 + ], + "score": 1.0, + "content": "input. Compared with ResNet-50, ResNet-101 has twice the number of convolutional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 327, + 480, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 480, + 340 + ], + "score": 1.0, + "content": "layers with non-unitary filter sizes. Accordingly, the extent of the foveation effect is doubled.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 108, + 356, + 504, + 438 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 356, + 504, + 438 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 356, + 504, + 438 + ], + "spans": [ + { + "bbox": [ + 108, + 356, + 504, + 438 + ], + "score": 0.963, + "type": "image", + "image_path": "a41bda8b2c50acbca4b5cd99ab0cb07019bf3ee67df5aae70ce1fa81a6316f40.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 108, + 356, + 504, + 383.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 108, + 383.3333333333333, + 504, + 410.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 108, + 410.66666666666663, + 504, + 437.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 447, + 504, + 471 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Figure 10: The foveation maps of ResNet-50 under 0 padding, illustrated with inputs of different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 460, + 384, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 384, + 471 + ], + "score": 1.0, + "content": "size. The smaller the input, the larger the relative extent of foveation.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + } + ], + "index": 14.25 + }, + { + "type": "text", + "bbox": [ + 105, + 485, + 506, + 530 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 487, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 487, + 498 + ], + "score": 1.0, + "content": "In the next figure, we illustrate how uneven application of padding impacts the foveation maps.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 497, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 508 + ], + "score": 1.0, + "content": "Note: It is possible to rectify the skewness in the 2nd foveation map by alternating the side where", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 507, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 506, + 520 + ], + "score": 1.0, + "content": "one-sided padding is applied between successive downsampling layers. This, however, does not", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 519, + 354, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 354, + 531 + ], + "score": 1.0, + "content": "mitigate the skewness in the learned filters (see next Section).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "image", + "bbox": [ + 172, + 540, + 437, + 657 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 172, + 540, + 437, + 657 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 172, + 540, + 437, + 657 + ], + "spans": [ + { + "bbox": [ + 172, + 540, + 437, + 657 + ], + "score": 0.971, + "type": "image", + "image_path": "b6915c9095fb0ab1b1b632c64deab332d79ccd94ad4eb92251f1454b4674d7bd.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 172, + 540, + 437, + 579.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 172, + 579.0, + 437, + 618.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 172, + 618.0, + 437, + 657.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 667, + 506, + 723 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 667, + 504, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 504, + 679 + ], + "score": 1.0, + "content": "Figure 11: The foveation maps of ResNet-50 under 0 padding, illustrated with two input sizes. With", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 113, + 691 + ], + "score": 1.0, + "content": "a", + "type": "text" + }, + { + "bbox": [ + 113, + 678, + 156, + 689 + ], + "score": 0.9, + "content": "2 5 7 \\times 2 5 7", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "input, the padding is evenly applied at all downsampling layers, leading to a symmetric", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 196, + 702 + ], + "score": 1.0, + "content": "foveation map. With a", + "type": "text" + }, + { + "bbox": [ + 197, + 689, + 237, + 700 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "input, the padding is applied only to the left and top sides of feature", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "maps at all downsampling layers, which limits the number of convolutional input-output paths for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 711, + 410, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 410, + 725 + ], + "score": 1.0, + "content": "pixels in the bottom and right sides as evident in the skewed foveation map.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + } + ], + "index": 24.0 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 406, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 406, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 406, + 95 + ], + "score": 1.0, + "content": "B THE EXTENT OF FOVEATION UNDER SAME 0-PADDING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 503, + 129 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 119 + ], + "score": 1.0, + "content": "We illustrate how the absolute extent of foveation under SAME 0-padding depends on the number of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 404, + 130 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 404, + 130 + ], + "score": 1.0, + "content": "convolutional layers, and how its relative extent depends on the input size.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 106, + 105, + 505, + 130 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 105, + 134, + 504, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 504, + 146 + ], + "score": 1.0, + "content": "In the following maps, color represents the number of paths to the CNN output for each input pixel.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 449, + 158 + ], + "score": 1.0, + "content": "Note: The checkerboard pattern is caused by downsampling layers in ResNet that use", + "type": "text" + }, + { + "bbox": [ + 450, + 146, + 473, + 156 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "kernels", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 156, + 176, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 176, + 168 + ], + "score": 1.0, + "content": "and a stride of 2.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 134, + 505, + 168 + ] + }, + { + "type": "image", + "bbox": [ + 176, + 177, + 437, + 295 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 176, + 177, + 437, + 295 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 177, + 437, + 295 + ], + "spans": [ + { + "bbox": [ + 176, + 177, + 437, + 295 + ], + "score": 0.969, + "type": "image", + "image_path": "230d99b41516d96c2c03aec57cab7c869711ad4d96a7315fd9c45cd64c50a0f0.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 176, + 177, + 437, + 216.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 176, + 216.33333333333334, + 437, + 255.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 176, + 255.66666666666669, + 437, + 295.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 305, + 505, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 305, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 506, + 317 + ], + "score": 1.0, + "content": "Figure 9: The foveation maps of two ResNet architectures under 0 padding, illustrated with a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 315, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 146, + 327 + ], + "score": 0.87, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 315, + 505, + 329 + ], + "score": 1.0, + "content": "input. Compared with ResNet-50, ResNet-101 has twice the number of convolutional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 327, + 480, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 480, + 340 + ], + "score": 1.0, + "content": "layers with non-unitary filter sizes. Accordingly, the extent of the foveation effect is doubled.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 108, + 356, + 504, + 438 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 356, + 504, + 438 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 356, + 504, + 438 + ], + "spans": [ + { + "bbox": [ + 108, + 356, + 504, + 438 + ], + "score": 0.963, + "type": "image", + "image_path": "a41bda8b2c50acbca4b5cd99ab0cb07019bf3ee67df5aae70ce1fa81a6316f40.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 108, + 356, + 504, + 383.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 108, + 383.3333333333333, + 504, + 410.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 108, + 410.66666666666663, + 504, + 437.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 447, + 504, + 471 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Figure 10: The foveation maps of ResNet-50 under 0 padding, illustrated with inputs of different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 460, + 384, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 384, + 471 + ], + "score": 1.0, + "content": "size. The smaller the input, the larger the relative extent of foveation.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + } + ], + "index": 14.25 + }, + { + "type": "text", + "bbox": [ + 105, + 485, + 506, + 530 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 487, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 487, + 498 + ], + "score": 1.0, + "content": "In the next figure, we illustrate how uneven application of padding impacts the foveation maps.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 497, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 508 + ], + "score": 1.0, + "content": "Note: It is possible to rectify the skewness in the 2nd foveation map by alternating the side where", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 507, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 506, + 520 + ], + "score": 1.0, + "content": "one-sided padding is applied between successive downsampling layers. This, however, does not", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 519, + 354, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 354, + 531 + ], + "score": 1.0, + "content": "mitigate the skewness in the learned filters (see next Section).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 485, + 506, + 531 + ] + }, + { + "type": "image", + "bbox": [ + 172, + 540, + 437, + 657 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 172, + 540, + 437, + 657 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 172, + 540, + 437, + 657 + ], + "spans": [ + { + "bbox": [ + 172, + 540, + 437, + 657 + ], + "score": 0.971, + "type": "image", + "image_path": "b6915c9095fb0ab1b1b632c64deab332d79ccd94ad4eb92251f1454b4674d7bd.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 172, + 540, + 437, + 579.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 172, + 579.0, + 437, + 618.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 172, + 618.0, + 437, + 657.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 667, + 506, + 723 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 667, + 504, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 504, + 679 + ], + "score": 1.0, + "content": "Figure 11: The foveation maps of ResNet-50 under 0 padding, illustrated with two input sizes. With", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 113, + 691 + ], + "score": 1.0, + "content": "a", + "type": "text" + }, + { + "bbox": [ + 113, + 678, + 156, + 689 + ], + "score": 0.9, + "content": "2 5 7 \\times 2 5 7", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "input, the padding is evenly applied at all downsampling layers, leading to a symmetric", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 196, + 702 + ], + "score": 1.0, + "content": "foveation map. With a", + "type": "text" + }, + { + "bbox": [ + 197, + 689, + 237, + 700 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "input, the padding is applied only to the left and top sides of feature", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "maps at all downsampling layers, which limits the number of convolutional input-output paths for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 711, + 410, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 410, + 725 + ], + "score": 1.0, + "content": "pixels in the bottom and right sides as evident in the skewed foveation map.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + } + ], + "index": 24.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 455, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 457, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 457, + 95 + ], + "score": 1.0, + "content": "C THE IMPACT OF THE PADDING METHOD ON LEARNED WEIGHTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 106, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 104, + 105, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 505, + 120 + ], + "score": 1.0, + "content": "In the presence of uneven application of padding, 0-padding causes skewness in the learned weights", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "score": 1.0, + "content": "because the filters are exposed more frequently to feature-map patches with zeros at their top and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "left sides. Redundancy methods such as circular or mirror padding mitigate such skewness because", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 139, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 505, + 151 + ], + "score": 1.0, + "content": "they fill the padding areas with values taken from the feature maps. PartialConv also mitigates such", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 163 + ], + "score": 1.0, + "content": "skewness because it assumes the pixels in the padding area are missing, and rescales the partial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 161, + 504, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 161, + 504, + 172 + ], + "score": 1.0, + "content": "convolutional sum to account for them. Below we show the effectiveness of these alternatives in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 172, + 325, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 325, + 184 + ], + "score": 1.0, + "content": "mitigating the skewness in three ResNet architectures.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 110, + 195, + 282, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 195, + 282, + 284 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 110, + 195, + 282, + 284 + ], + "spans": [ + { + "bbox": [ + 110, + 195, + 282, + 284 + ], + "score": 0.755, + "type": "image", + "image_path": "c9b6cd065fd2a6ff0aad0928693bb5a1ef91f3a05a54734b812f2ef68cb1a3cc.jpg" + } + ] + } + ], + "index": 13.0, + "virtual_lines": [ + { + "bbox": [ + 110, + 195, + 282, + 209.83333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 209.83333333333334, + 282, + 224.66666666666669 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 110, + 224.66666666666669, + 282, + 239.50000000000003 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 110, + 239.50000000000003, + 282, + 254.33333333333337 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 254.33333333333337, + 282, + 269.1666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 110, + 269.1666666666667, + 282, + 284.0 + ], + "spans": [], + "index": 18 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "image", + "bbox": [ + 331, + 194, + 498, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 331, + 194, + 498, + 284 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 331, + 194, + 498, + 284 + ], + "spans": [ + { + "bbox": [ + 331, + 194, + 498, + 284 + ], + "score": 0.249, + "type": "image", + "image_path": "28bf70ac734a23168ae18bc3334f51e2d26eca67e7b3e2d076b3dfd9008e4880.jpg" + } + ] + } + ], + "index": 14.0, + "virtual_lines": [ + { + "bbox": [ + 331, + 194, + 498, + 209.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 331, + 209.0, + 498, + 224.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 331, + 224.0, + 498, + 239.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 331, + 239.0, + 498, + 254.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 331, + 254.0, + 498, + 269.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 331, + 269.0, + 498, + 284.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 324, + 289, + 505, + 329 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 288, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 323, + 288, + 468, + 299 + ], + "score": 1.0, + "content": "(b) Mean filters of ResNet-50 trained on", + "type": "text" + }, + { + "bbox": [ + 469, + 288, + 504, + 299 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 323, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 323, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "images under two padding methods, reaching", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 325, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 325, + 308, + 354, + 318 + ], + "score": 0.85, + "content": "7 6 . 1 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "top-1 accuracy under 0-padding and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 324, + 318, + 480, + 330 + ], + "spans": [ + { + "bbox": [ + 324, + 319, + 354, + 329 + ], + "score": 0.85, + "content": "7 6 . 6 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 318, + 480, + 330 + ], + "score": 1.0, + "content": "top-1 accuracy under PartialConv.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24.5 + } + ], + "index": 19.25 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 287, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 286, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 250, + 299 + ], + "score": 1.0, + "content": "(a) Mean filters of ResNet-18 trained on", + "type": "text" + }, + { + "bbox": [ + 250, + 288, + 286, + 299 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 298, + 288, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 288, + 310 + ], + "score": 1.0, + "content": "images under two padding methods, reaching", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 307, + 287, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 137, + 318 + ], + "score": 0.85, + "content": "6 9 . 9 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 307, + 287, + 320 + ], + "score": 1.0, + "content": "top-1 accuracy under 0-padding and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 317, + 279, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 137, + 329 + ], + "score": 0.85, + "content": "7 0 . 2 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 317, + 279, + 331 + ], + "score": 1.0, + "content": "top-1 accuracy under circular padding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 409, + 351 + ], + "score": 1.0, + "content": "Figure 12: Mean filters of two ResNet models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 409, + 339, + 449, + 349 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "images. The", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 505, + 361 + ], + "score": 1.0, + "content": "input size causes uneven application of padding, leading to frequent asymmetries in the mean filters", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 455, + 373 + ], + "score": 1.0, + "content": "under 0 padding. We illustrate how two alternatives, circular padding and PartialConv", + "type": "text" + }, + { + "bbox": [ + 455, + 360, + 473, + 372 + ], + "score": 0.74, + "content": "\\pmb { \\left. \\pmb { \\left. \\bar { 2 3 } \\right. } \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 361, + 505, + 372 + ], + "score": 1.0, + "content": ", enable", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 371, + 435, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 435, + 385 + ], + "score": 1.0, + "content": "learning highly-symmetric mean filters despite the uneven application of padding.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "image", + "bbox": [ + 117, + 400, + 496, + 496 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 400, + 496, + 496 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 400, + 496, + 496 + ], + "spans": [ + { + "bbox": [ + 117, + 400, + 496, + 496 + ], + "score": 0.97, + "type": "image", + "image_path": "7bc6aeb0c0d5fa125d64fbf452bdfc569b0089cb7b988e4d3f582824309d0153.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 117, + 400, + 496, + 432.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 117, + 432.0, + 496, + 464.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 117, + 464.0, + 496, + 496.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 505, + 505, + 549 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 374, + 517 + ], + "score": 1.0, + "content": "Figure 13: Mean filters of ResNet-101 trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 375, + 505, + 414, + 515 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "images under both 0-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 211, + 528 + ], + "score": 1.0, + "content": "padding and PartialConv", + "type": "text" + }, + { + "bbox": [ + 212, + 515, + 229, + 527 + ], + "score": 0.32, + "content": "\\mathbb { \\lVert 2 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 515, + 506, + 528 + ], + "score": 1.0, + "content": ". The input size causes uneven application of padding, leading to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 525, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 541 + ], + "score": 1.0, + "content": "frequent asymmetries in the mean filters under 0 padding. In contrast, PartialConv produces highly", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "score": 1.0, + "content": "symmetric mean filters, thanks for its treatment of pixels outside the feature map as missing values.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + } + ], + "index": 34.75 + }, + { + "type": "text", + "bbox": [ + 107, + 568, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "What if no padding is applied during downsampling? VGG models perform downsampling", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 131, + 592 + ], + "score": 1.0, + "content": "using", + "type": "text" + }, + { + "bbox": [ + 132, + 579, + 157, + 590 + ], + "score": 0.88, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "pooling layers that do not apply any padding. Accordingly, the mean filters do not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 589, + 396, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 396, + 604 + ], + "score": 1.0, + "content": "exhibit significant skewness, even if the input size does not satisfy Eq 4:", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "image", + "bbox": [ + 140, + 623, + 471, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 623, + 471, + 697 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 140, + 623, + 471, + 697 + ], + "spans": [ + { + "bbox": [ + 140, + 623, + 471, + 697 + ], + "score": 0.959, + "type": "image", + "image_path": "51a8af5576a494d4ff89c4d3741816ce78e27e93e3c83213ba9b0e683e37e048.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 140, + 623, + 471, + 647.6666666666666 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 140, + 647.6666666666666, + 471, + 672.3333333333333 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 140, + 672.3333333333333, + 471, + 696.9999999999999 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 706, + 505, + 730 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 706, + 506, + 719 + ], + "spans": [ + { + "bbox": [ + 106, + 706, + 506, + 719 + ], + "score": 1.0, + "content": "Figure 14: Mean filters of VGG-16 trained on ImageNet under different conditions. Most mean", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 717, + 486, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 717, + 298, + 730 + ], + "score": 1.0, + "content": "filters exhibit high symmetry when trained with", + "type": "text" + }, + { + "bbox": [ + 299, + 717, + 338, + 728 + ], + "score": 0.88, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 717, + 486, + 730 + ], + "score": 1.0, + "content": "images where the size violates Eq. 4.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "index": 44.25 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 455, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 457, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 457, + 95 + ], + "score": 1.0, + "content": "C THE IMPACT OF THE PADDING METHOD ON LEARNED WEIGHTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 106, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 104, + 105, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 505, + 120 + ], + "score": 1.0, + "content": "In the presence of uneven application of padding, 0-padding causes skewness in the learned weights", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 506, + 130 + ], + "score": 1.0, + "content": "because the filters are exposed more frequently to feature-map patches with zeros at their top and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "left sides. Redundancy methods such as circular or mirror padding mitigate such skewness because", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 139, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 505, + 151 + ], + "score": 1.0, + "content": "they fill the padding areas with values taken from the feature maps. PartialConv also mitigates such", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 163 + ], + "score": 1.0, + "content": "skewness because it assumes the pixels in the padding area are missing, and rescales the partial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 161, + 504, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 161, + 504, + 172 + ], + "score": 1.0, + "content": "convolutional sum to account for them. Below we show the effectiveness of these alternatives in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 172, + 325, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 325, + 184 + ], + "score": 1.0, + "content": "mitigating the skewness in three ResNet architectures.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4, + "bbox_fs": [ + 104, + 105, + 506, + 184 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 195, + 282, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 195, + 282, + 284 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 110, + 195, + 282, + 284 + ], + "spans": [ + { + "bbox": [ + 110, + 195, + 282, + 284 + ], + "score": 0.755, + "type": "image", + "image_path": "c9b6cd065fd2a6ff0aad0928693bb5a1ef91f3a05a54734b812f2ef68cb1a3cc.jpg" + } + ] + } + ], + "index": 13.0, + "virtual_lines": [ + { + "bbox": [ + 110, + 195, + 282, + 209.83333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 209.83333333333334, + 282, + 224.66666666666669 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 110, + 224.66666666666669, + 282, + 239.50000000000003 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 110, + 239.50000000000003, + 282, + 254.33333333333337 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 254.33333333333337, + 282, + 269.1666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 110, + 269.1666666666667, + 282, + 284.0 + ], + "spans": [], + "index": 18 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "image", + "bbox": [ + 331, + 194, + 498, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 331, + 194, + 498, + 284 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 331, + 194, + 498, + 284 + ], + "spans": [ + { + "bbox": [ + 331, + 194, + 498, + 284 + ], + "score": 0.249, + "type": "image", + "image_path": "28bf70ac734a23168ae18bc3334f51e2d26eca67e7b3e2d076b3dfd9008e4880.jpg" + } + ] + } + ], + "index": 14.0, + "virtual_lines": [ + { + "bbox": [ + 331, + 194, + 498, + 209.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 331, + 209.0, + 498, + 224.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 331, + 224.0, + 498, + 239.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 331, + 239.0, + 498, + 254.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 331, + 254.0, + 498, + 269.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 331, + 269.0, + 498, + 284.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 324, + 289, + 505, + 329 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 288, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 323, + 288, + 468, + 299 + ], + "score": 1.0, + "content": "(b) Mean filters of ResNet-50 trained on", + "type": "text" + }, + { + "bbox": [ + 469, + 288, + 504, + 299 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 323, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 323, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "images under two padding methods, reaching", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 325, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 325, + 308, + 354, + 318 + ], + "score": 0.85, + "content": "7 6 . 1 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "top-1 accuracy under 0-padding and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 324, + 318, + 480, + 330 + ], + "spans": [ + { + "bbox": [ + 324, + 319, + 354, + 329 + ], + "score": 0.85, + "content": "7 6 . 6 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 318, + 480, + 330 + ], + "score": 1.0, + "content": "top-1 accuracy under PartialConv.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24.5 + } + ], + "index": 19.25 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 287, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 286, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 250, + 299 + ], + "score": 1.0, + "content": "(a) Mean filters of ResNet-18 trained on", + "type": "text" + }, + { + "bbox": [ + 250, + 288, + 286, + 299 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 298, + 288, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 288, + 310 + ], + "score": 1.0, + "content": "images under two padding methods, reaching", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 307, + 287, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 137, + 318 + ], + "score": 0.85, + "content": "6 9 . 9 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 307, + 287, + 320 + ], + "score": 1.0, + "content": "top-1 accuracy under 0-padding and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 317, + 279, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 137, + 329 + ], + "score": 0.85, + "content": "7 0 . 2 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 317, + 279, + 331 + ], + "score": 1.0, + "content": "top-1 accuracy under circular padding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 288, + 288, + 331 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 409, + 351 + ], + "score": 1.0, + "content": "Figure 12: Mean filters of two ResNet models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 409, + 339, + 449, + 349 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "images. The", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 505, + 361 + ], + "score": 1.0, + "content": "input size causes uneven application of padding, leading to frequent asymmetries in the mean filters", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 455, + 373 + ], + "score": 1.0, + "content": "under 0 padding. We illustrate how two alternatives, circular padding and PartialConv", + "type": "text" + }, + { + "bbox": [ + 455, + 360, + 473, + 372 + ], + "score": 0.74, + "content": "\\pmb { \\left. \\pmb { \\left. \\bar { 2 3 } \\right. } \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 361, + 505, + 372 + ], + "score": 1.0, + "content": ", enable", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 371, + 435, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 435, + 385 + ], + "score": 1.0, + "content": "learning highly-symmetric mean filters despite the uneven application of padding.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 338, + 505, + 385 + ] + }, + { + "type": "image", + "bbox": [ + 117, + 400, + 496, + 496 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 400, + 496, + 496 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 400, + 496, + 496 + ], + "spans": [ + { + "bbox": [ + 117, + 400, + 496, + 496 + ], + "score": 0.97, + "type": "image", + "image_path": "7bc6aeb0c0d5fa125d64fbf452bdfc569b0089cb7b988e4d3f582824309d0153.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 117, + 400, + 496, + 432.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 117, + 432.0, + 496, + 464.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 117, + 464.0, + 496, + 496.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 505, + 505, + 549 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 374, + 517 + ], + "score": 1.0, + "content": "Figure 13: Mean filters of ResNet-101 trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 375, + 505, + 414, + 515 + ], + "score": 0.88, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "images under both 0-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 211, + 528 + ], + "score": 1.0, + "content": "padding and PartialConv", + "type": "text" + }, + { + "bbox": [ + 212, + 515, + 229, + 527 + ], + "score": 0.32, + "content": "\\mathbb { \\lVert 2 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 515, + 506, + 528 + ], + "score": 1.0, + "content": ". The input size causes uneven application of padding, leading to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 525, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 541 + ], + "score": 1.0, + "content": "frequent asymmetries in the mean filters under 0 padding. In contrast, PartialConv produces highly", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "score": 1.0, + "content": "symmetric mean filters, thanks for its treatment of pixels outside the feature map as missing values.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + } + ], + "index": 34.75 + }, + { + "type": "text", + "bbox": [ + 107, + 568, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "What if no padding is applied during downsampling? VGG models perform downsampling", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 131, + 592 + ], + "score": 1.0, + "content": "using", + "type": "text" + }, + { + "bbox": [ + 132, + 579, + 157, + 590 + ], + "score": 0.88, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "pooling layers that do not apply any padding. Accordingly, the mean filters do not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 589, + 396, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 396, + 604 + ], + "score": 1.0, + "content": "exhibit significant skewness, even if the input size does not satisfy Eq 4:", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 567, + 506, + 604 + ] + }, + { + "type": "image", + "bbox": [ + 140, + 623, + 471, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 623, + 471, + 697 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 140, + 623, + 471, + 697 + ], + "spans": [ + { + "bbox": [ + 140, + 623, + 471, + 697 + ], + "score": 0.959, + "type": "image", + "image_path": "51a8af5576a494d4ff89c4d3741816ce78e27e93e3c83213ba9b0e683e37e048.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 140, + 623, + 471, + 647.6666666666666 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 140, + 647.6666666666666, + 471, + 672.3333333333333 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 140, + 672.3333333333333, + 471, + 696.9999999999999 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 706, + 505, + 730 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 706, + 506, + 719 + ], + "spans": [ + { + "bbox": [ + 106, + 706, + 506, + 719 + ], + "score": 1.0, + "content": "Figure 14: Mean filters of VGG-16 trained on ImageNet under different conditions. Most mean", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 717, + 486, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 717, + 298, + 730 + ], + "score": 1.0, + "content": "filters exhibit high symmetry when trained with", + "type": "text" + }, + { + "bbox": [ + 299, + 717, + 338, + 728 + ], + "score": 0.88, + "content": "2 2 5 \\times 2 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 717, + 486, + 730 + ], + "score": 1.0, + "content": "images where the size violates Eq. 4.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "index": 44.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 105, + 81, + 474, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 475, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 475, + 96 + ], + "score": 1.0, + "content": "D THE IMPACT OF PADDING METHODS ON FEATURE-MAP ARTIFACTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 129 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 120 + ], + "score": 1.0, + "content": "We show per-layer mean feature maps in ResNet-18 under different padding methods. The mean", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 359, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 359, + 130 + ], + "score": 1.0, + "content": "maps are averaged over 20 input samples generated at random.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "image", + "bbox": [ + 109, + 142, + 504, + 218 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 142, + 504, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 142, + 504, + 218 + ], + "spans": [ + { + "bbox": [ + 109, + 142, + 504, + 218 + ], + "score": 0.963, + "type": "image", + "image_path": "4dc8a25d06b953aa2901c2738d70c12d28f3f668e42fe6df8e21854cbd416e19.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 109, + 142, + 504, + 167.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 167.33333333333334, + 504, + 192.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 192.66666666666669, + 504, + 218.00000000000003 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 227, + 504, + 250 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "Figure 15: Feature map artifacts under zero padding. Line artifacts accumulate to become significant", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 238, + 240, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 240, + 251 + ], + "score": 1.0, + "content": "and asymmetric at deeper layers.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + } + ], + "index": 5.25 + }, + { + "type": "image", + "bbox": [ + 107, + 267, + 502, + 344 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 267, + 502, + 344 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 267, + 502, + 344 + ], + "spans": [ + { + "bbox": [ + 107, + 267, + 502, + 344 + ], + "score": 0.965, + "type": "image", + "image_path": "6640c3d2bb4a8864228e5cb9e90262456ad14527bc623e13533134115c042001.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 267, + 502, + 292.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 292.6666666666667, + 502, + 318.33333333333337 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 318.33333333333337, + 502, + 344.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 124, + 352, + 484, + 364 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 125, + 351, + 486, + 365 + ], + "spans": [ + { + "bbox": [ + 125, + 351, + 486, + 365 + ], + "score": 1.0, + "content": "Figure 16: Circular padding largely preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + } + ], + "index": 10.0 + }, + { + "type": "image", + "bbox": [ + 109, + 383, + 504, + 459 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 383, + 504, + 459 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 109, + 383, + 504, + 459 + ], + "spans": [ + { + "bbox": [ + 109, + 383, + 504, + 459 + ], + "score": 0.966, + "type": "image", + "image_path": "f3cd7587862b9845b41afe3205965e60be50ddde1f43d11c94bed84f9ba5ee93.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 109, + 383, + 504, + 408.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 408.3333333333333, + 504, + 433.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 109, + 433.66666666666663, + 504, + 458.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 469, + 502, + 481 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 468, + 504, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 504, + 482 + ], + "score": 1.0, + "content": "Figure 17: SYMMETRIC mirror padding also preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "image", + "bbox": [ + 109, + 499, + 504, + 576 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 499, + 504, + 576 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 109, + 499, + 504, + 576 + ], + "spans": [ + { + "bbox": [ + 109, + 499, + 504, + 576 + ], + "score": 0.964, + "type": "image", + "image_path": "56552f00a0e6677217affe158f48306a5f1d42e3ca61577ae6dcc64f4fc464fc.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 109, + 499, + 504, + 524.6666666666666 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 109, + 524.6666666666666, + 504, + 550.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 109, + 550.3333333333333, + 504, + 575.9999999999999 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 111, + 586, + 497, + 598 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 113, + 584, + 497, + 598 + ], + "spans": [ + { + "bbox": [ + 113, + 584, + 497, + 598 + ], + "score": 1.0, + "content": "Figure 18: REFLECT mirror padding also preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + } + ], + "index": 18.0 + }, + { + "type": "image", + "bbox": [ + 109, + 614, + 504, + 690 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 614, + 504, + 690 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 109, + 614, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 109, + 614, + 504, + 690 + ], + "score": 0.963, + "type": "image", + "image_path": "72d9bfe380a72b51f7220f1be45381b387d02402c6fb51a1ea2bb58ba8ea21f1.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 109, + 614, + 504, + 639.3333333333334 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 109, + 639.3333333333334, + 504, + 664.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 109, + 664.6666666666667, + 504, + 690.0000000000001 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 200, + 713 + ], + "score": 1.0, + "content": "Figure 19: PartialConv", + "type": "text" + }, + { + "bbox": [ + 200, + 699, + 218, + 711 + ], + "score": 0.72, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "highly preserves the symmetry of the feature maps. The scaling factors", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 709, + 306, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 306, + 723 + ], + "score": 1.0, + "content": "it uses can break the randomness at the boundary.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 105, + 81, + 474, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 475, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 475, + 96 + ], + "score": 1.0, + "content": "D THE IMPACT OF PADDING METHODS ON FEATURE-MAP ARTIFACTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 129 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 120 + ], + "score": 1.0, + "content": "We show per-layer mean feature maps in ResNet-18 under different padding methods. The mean", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 359, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 359, + 130 + ], + "score": 1.0, + "content": "maps are averaged over 20 input samples generated at random.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 105, + 505, + 130 + ] + }, + { + "type": "image", + "bbox": [ + 109, + 142, + 504, + 218 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 142, + 504, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 142, + 504, + 218 + ], + "spans": [ + { + "bbox": [ + 109, + 142, + 504, + 218 + ], + "score": 0.963, + "type": "image", + "image_path": "4dc8a25d06b953aa2901c2738d70c12d28f3f668e42fe6df8e21854cbd416e19.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 109, + 142, + 504, + 167.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 167.33333333333334, + 504, + 192.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 192.66666666666669, + 504, + 218.00000000000003 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 227, + 504, + 250 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "Figure 15: Feature map artifacts under zero padding. Line artifacts accumulate to become significant", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 238, + 240, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 240, + 251 + ], + "score": 1.0, + "content": "and asymmetric at deeper layers.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + } + ], + "index": 5.25 + }, + { + "type": "image", + "bbox": [ + 107, + 267, + 502, + 344 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 267, + 502, + 344 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 267, + 502, + 344 + ], + "spans": [ + { + "bbox": [ + 107, + 267, + 502, + 344 + ], + "score": 0.965, + "type": "image", + "image_path": "6640c3d2bb4a8864228e5cb9e90262456ad14527bc623e13533134115c042001.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 267, + 502, + 292.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 292.6666666666667, + 502, + 318.33333333333337 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 318.33333333333337, + 502, + 344.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 124, + 352, + 484, + 364 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 125, + 351, + 486, + 365 + ], + "spans": [ + { + "bbox": [ + 125, + 351, + 486, + 365 + ], + "score": 1.0, + "content": "Figure 16: Circular padding largely preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + } + ], + "index": 10.0 + }, + { + "type": "image", + "bbox": [ + 109, + 383, + 504, + 459 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 383, + 504, + 459 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 109, + 383, + 504, + 459 + ], + "spans": [ + { + "bbox": [ + 109, + 383, + 504, + 459 + ], + "score": 0.966, + "type": "image", + "image_path": "f3cd7587862b9845b41afe3205965e60be50ddde1f43d11c94bed84f9ba5ee93.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 109, + 383, + 504, + 408.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 408.3333333333333, + 504, + 433.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 109, + 433.66666666666663, + 504, + 458.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 469, + 502, + 481 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 468, + 504, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 504, + 482 + ], + "score": 1.0, + "content": "Figure 17: SYMMETRIC mirror padding also preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "image", + "bbox": [ + 109, + 499, + 504, + 576 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 499, + 504, + 576 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 109, + 499, + 504, + 576 + ], + "spans": [ + { + "bbox": [ + 109, + 499, + 504, + 576 + ], + "score": 0.964, + "type": "image", + "image_path": "56552f00a0e6677217affe158f48306a5f1d42e3ca61577ae6dcc64f4fc464fc.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 109, + 499, + 504, + 524.6666666666666 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 109, + 524.6666666666666, + 504, + 550.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 109, + 550.3333333333333, + 504, + 575.9999999999999 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 111, + 586, + 497, + 598 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 113, + 584, + 497, + 598 + ], + "spans": [ + { + "bbox": [ + 113, + 584, + 497, + 598 + ], + "score": 1.0, + "content": "Figure 18: REFLECT mirror padding also preserves the randomness and mitigates line artifacts.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + } + ], + "index": 18.0 + }, + { + "type": "image", + "bbox": [ + 109, + 614, + 504, + 690 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 614, + 504, + 690 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 109, + 614, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 109, + 614, + 504, + 690 + ], + "score": 0.963, + "type": "image", + "image_path": "72d9bfe380a72b51f7220f1be45381b387d02402c6fb51a1ea2bb58ba8ea21f1.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 109, + 614, + 504, + 639.3333333333334 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 109, + 639.3333333333334, + 504, + 664.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 109, + 664.6666666666667, + 504, + 690.0000000000001 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 200, + 713 + ], + "score": 1.0, + "content": "Figure 19: PartialConv", + "type": "text" + }, + { + "bbox": [ + 200, + 699, + 218, + 711 + ], + "score": 0.72, + "content": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "highly preserves the symmetry of the feature maps. The scaling factors", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 709, + 306, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 306, + 723 + ], + "score": 1.0, + "content": "it uses can break the randomness at the boundary.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 698, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 505, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 505, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 505, + 183 + ], + "score": 0.966, + "type": "image", + "image_path": "3813e0f0f363045277b94f3a816f45c0ab5f295c9700bc4537785b52027e2f6d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 505, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 113.66666666666666, + 505, + 148.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 148.33333333333331, + 505, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 191, + 506, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 205 + ], + "score": 1.0, + "content": "Figure 20: Feature map artifacts of a VGG-19 model under Distribution Padding (interpolation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 202, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 134, + 215 + ], + "score": 1.0, + "content": "mode)", + "type": "text" + }, + { + "bbox": [ + 135, + 202, + 152, + 214 + ], + "score": 0.31, + "content": "\\textcircled { \\lvert 3 0 \\rvert }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 203, + 505, + 215 + ], + "score": 1.0, + "content": ". Due to multiple resize operations used to fill the padding area, the artifacts grow from", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 439, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 439, + 226 + ], + "score": 1.0, + "content": "the boundary inwards. We use a saturated constant input to make the effect visible.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 244, + 435, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 436, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 436, + 258 + ], + "score": 1.0, + "content": "E THE IMPACT OF ANTIALIASING ON THE LEARNED WEIGHTS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 268, + 506, + 291 + ], + "lines": [ + { + "bbox": [ + 106, + 267, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 282 + ], + "score": 1.0, + "content": "We demonstrate how antialiasing [43] significantly reduces the asymmetry of mean filters around", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 279, + 418, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 418, + 294 + ], + "score": 1.0, + "content": "downsampling layers, even in the presence of unevenly-applied zero padding.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "image", + "bbox": [ + 112, + 301, + 496, + 465 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 301, + 496, + 465 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 301, + 496, + 465 + ], + "spans": [ + { + "bbox": [ + 112, + 301, + 496, + 465 + ], + "score": 0.972, + "type": "image", + "image_path": "2012189e75de9944003498504b5db09216e0eb5f175070e1ae807135b096f5fd.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 112, + 301, + 496, + 355.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 112, + 355.6666666666667, + 496, + 410.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 112, + 410.33333333333337, + 496, + 465.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 473, + 504, + 496 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 471, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 366, + 487 + ], + "score": 1.0, + "content": "Figure 21: Mean filters of four models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 366, + 473, + 406, + 484 + ], + "score": 0.85, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 471, + 505, + 487 + ], + "score": 1.0, + "content": "images under 0-padding", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 482, + 248, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 248, + 497 + ], + "score": 1.0, + "content": "both without and with antialiasing.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + } + ], + "index": 11.25 + }, + { + "type": "image", + "bbox": [ + 130, + 512, + 477, + 690 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 130, + 512, + 477, + 690 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 130, + 512, + 477, + 690 + ], + "spans": [ + { + "bbox": [ + 130, + 512, + 477, + 690 + ], + "score": 0.976, + "type": "image", + "image_path": "57e3f9aa779c1d5920a44c13b73025feb076f2064e727af4c763cdde601bb09a.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 130, + 512, + 477, + 571.3333333333334 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 130, + 571.3333333333334, + 477, + 630.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 130, + 630.6666666666667, + 477, + 690.0000000000001 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 700, + 505, + 724 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 699, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 365, + 715 + ], + "score": 1.0, + "content": "Figure 22: Mean filters of two models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 366, + 701, + 405, + 712 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 699, + 505, + 715 + ], + "score": 1.0, + "content": "images under 0-padding", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 710, + 248, + 725 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 248, + 725 + ], + "score": 1.0, + "content": "both without and with antialiasing.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 16.25 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 505, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 505, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 505, + 183 + ], + "score": 0.966, + "type": "image", + "image_path": "3813e0f0f363045277b94f3a816f45c0ab5f295c9700bc4537785b52027e2f6d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 505, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 113.66666666666666, + 505, + 148.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 148.33333333333331, + 505, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 191, + 506, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 205 + ], + "score": 1.0, + "content": "Figure 20: Feature map artifacts of a VGG-19 model under Distribution Padding (interpolation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 202, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 134, + 215 + ], + "score": 1.0, + "content": "mode)", + "type": "text" + }, + { + "bbox": [ + 135, + 202, + 152, + 214 + ], + "score": 0.31, + "content": "\\textcircled { \\lvert 3 0 \\rvert }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 203, + 505, + 215 + ], + "score": 1.0, + "content": ". Due to multiple resize operations used to fill the padding area, the artifacts grow from", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 439, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 439, + 226 + ], + "score": 1.0, + "content": "the boundary inwards. We use a saturated constant input to make the effect visible.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 244, + 435, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 436, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 436, + 258 + ], + "score": 1.0, + "content": "E THE IMPACT OF ANTIALIASING ON THE LEARNED WEIGHTS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 268, + 506, + 291 + ], + "lines": [ + { + "bbox": [ + 106, + 267, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 282 + ], + "score": 1.0, + "content": "We demonstrate how antialiasing [43] significantly reduces the asymmetry of mean filters around", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 279, + 418, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 418, + 294 + ], + "score": 1.0, + "content": "downsampling layers, even in the presence of unevenly-applied zero padding.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 267, + 505, + 294 + ] + }, + { + "type": "image", + "bbox": [ + 112, + 301, + 496, + 465 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 301, + 496, + 465 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 301, + 496, + 465 + ], + "spans": [ + { + "bbox": [ + 112, + 301, + 496, + 465 + ], + "score": 0.972, + "type": "image", + "image_path": "2012189e75de9944003498504b5db09216e0eb5f175070e1ae807135b096f5fd.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 112, + 301, + 496, + 355.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 112, + 355.6666666666667, + 496, + 410.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 112, + 410.33333333333337, + 496, + 465.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 473, + 504, + 496 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 471, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 366, + 487 + ], + "score": 1.0, + "content": "Figure 21: Mean filters of four models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 366, + 473, + 406, + 484 + ], + "score": 0.85, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 471, + 505, + 487 + ], + "score": 1.0, + "content": "images under 0-padding", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 482, + 248, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 248, + 497 + ], + "score": 1.0, + "content": "both without and with antialiasing.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + } + ], + "index": 11.25 + }, + { + "type": "image", + "bbox": [ + 130, + 512, + 477, + 690 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 130, + 512, + 477, + 690 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 130, + 512, + 477, + 690 + ], + "spans": [ + { + "bbox": [ + 130, + 512, + 477, + 690 + ], + "score": 0.976, + "type": "image", + "image_path": "57e3f9aa779c1d5920a44c13b73025feb076f2064e727af4c763cdde601bb09a.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 130, + 512, + 477, + 571.3333333333334 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 130, + 571.3333333333334, + 477, + 630.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 130, + 630.6666666666667, + 477, + 690.0000000000001 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 700, + 505, + 724 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 699, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 365, + 715 + ], + "score": 1.0, + "content": "Figure 22: Mean filters of two models trained on ImageNet with", + "type": "text" + }, + { + "bbox": [ + 366, + 701, + 405, + 712 + ], + "score": 0.86, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 699, + 505, + 715 + ], + "score": 1.0, + "content": "images under 0-padding", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 710, + 248, + 725 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 248, + 725 + ], + "score": 1.0, + "content": "both without and with antialiasing.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 16.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 383, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 80, + 384, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 80, + 384, + 96 + ], + "score": 1.0, + "content": "F FOVEATION ANALYSIS OF PADDING ALGORITHMS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 502, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "Refer to http://mind-the-pad.github.io for an interactive and animated visual illustra-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 492, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 492, + 129 + ], + "score": 1.0, + "content": "tion of padding algorithms and their foveation behavior. This appendix serves as a print version.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 505, + 179 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "score": 1.0, + "content": "Among the SAME padding algorithms we discussed in the manuscript, two algorithms warrant that", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 146, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 505, + 158 + ], + "score": 1.0, + "content": "each input pixel is involved in an equal number of convolutional operations, leading to uniform", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 157, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "foveation maps: circular padding and SYMMETRIC mirror padding. In contrast, this number varies", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 167, + 477, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 477, + 180 + ], + "score": 1.0, + "content": "under zero padding, REFLECT mirror padding, replication padding, and partial convolution.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "We illustrate in detail how each padding algorithm treats the input pixels. For this purpose we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "illustrate step by step how each pixel is processed by the convolutional kernel. We choose a set of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 206, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 506, + 218 + ], + "score": 1.0, + "content": "pixels that are sufficient to expose the behavior of the respective algorithm. This set spans an area", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "score": 1.0, + "content": "within two or three pixels from the boundary that encompasses all relevant cases for the analysis and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 228, + 424, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 424, + 240 + ], + "score": 1.0, + "content": "is situated at the top-left corner. The behavior at the other corners is analogous.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 109, + 244, + 486, + 256 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 487, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 487, + 259 + ], + "score": 1.0, + "content": "All illustrations use a stride of 1. Except for VALID, all configurations warrant SAME padding.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 131, + 266, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 132, + 265, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 132, + 265, + 349, + 278 + ], + "score": 1.0, + "content": "• VALID Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 349, + 266, + 373, + 276 + ], + "score": 0.89, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 265, + 505, + 278 + ], + "score": 1.0, + "content": "kernel without dilation. A larger", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 277, + 391, + 288 + ], + "spans": [ + { + "bbox": [ + 142, + 277, + 391, + 288 + ], + "score": 1.0, + "content": "kernel size or dilation factor will increase the foveation effect.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 140, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 140, + 290, + 345, + 304 + ], + "score": 1.0, + "content": "Zero Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 345, + 292, + 370, + 302 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "kernel without dilation. A larger", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 303, + 391, + 314 + ], + "spans": [ + { + "bbox": [ + 142, + 303, + 391, + 314 + ], + "score": 1.0, + "content": "kernel size or dilation factor will increase the foveation effect.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 140, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 140, + 317, + 366, + 329 + ], + "score": 1.0, + "content": "Circular Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 366, + 317, + 392, + 328 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "kernel without dilation. It", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 329, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 505, + 341 + ], + "score": 1.0, + "content": "is straightforward to prove that the algorithm warrants equal treatment of the pixels irre-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "spective of the kernel size or dilation factor. This is because it effectively applies circular", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "convolution: Once the kernel hits one side, it can seamlessly operate on the pixels of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 142, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "other side. Circular convolution hence renders the feature map as infinite to the kernel,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 373, + 448, + 384 + ], + "spans": [ + { + "bbox": [ + 142, + 373, + 448, + 384 + ], + "score": 1.0, + "content": "warranting that edge pixels are treated in the same manner as interior pixels.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 133, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 133, + 390, + 138, + 396 + ], + "score": 0.818, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 144, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC): This algorithm warrants that each pixel is involved in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 142, + 398, + 504, + 410 + ], + "spans": [ + { + "bbox": [ + 142, + 398, + 504, + 410 + ], + "score": 1.0, + "content": "the same number of convolutional operations. It is important to notice that, unlike un-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 410, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 142, + 410, + 505, + 421 + ], + "score": 1.0, + "content": "der circular convolution, these operations do not utilize the kernel pixels uniformly as we", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 420, + 494, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 494, + 433 + ], + "score": 1.0, + "content": "demonstrate in detail. We illustrate the algorithm behavior under the following settings:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 150, + 435, + 504, + 509 + ], + "lines": [ + { + "bbox": [ + 151, + 434, + 311, + 447 + ], + "spans": [ + { + "bbox": [ + 151, + 434, + 162, + 447 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 435, + 186, + 446 + ], + "score": 0.86, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 434, + 311, + 447 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 151, + 447, + 310, + 459 + ], + "spans": [ + { + "bbox": [ + 151, + 447, + 162, + 459 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 448, + 186, + 459 + ], + "score": 0.83, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 447, + 310, + 459 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 152, + 461, + 310, + 472 + ], + "spans": [ + { + "bbox": [ + 152, + 461, + 162, + 471 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 461, + 186, + 472 + ], + "score": 0.8, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 461, + 310, + 471 + ], + "score": 1.0, + "content": "kernel and dilation factor of 2.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 150, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 150, + 473, + 162, + 487 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 474, + 186, + 484 + ], + "score": 0.83, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 473, + 505, + 487 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1, along with a grouped padding strategy to com-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 160, + 484, + 296, + 497 + ], + "spans": [ + { + "bbox": [ + 160, + 486, + 276, + 496 + ], + "score": 1.0, + "content": "pensate for uneven padding", + "type": "text" + }, + { + "bbox": [ + 277, + 484, + 294, + 496 + ], + "score": 0.27, + "content": "\\pm \\amalg", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 484, + 296, + 497 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 151, + 496, + 485, + 511 + ], + "spans": [ + { + "bbox": [ + 151, + 496, + 162, + 511 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 498, + 186, + 508 + ], + "score": 0.86, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 496, + 485, + 511 + ], + "score": 1.0, + "content": "kernel size and dilation factor of 1, along with a grouped padding strategy.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 133, + 513, + 505, + 535 + ], + "lines": [ + { + "bbox": [ + 132, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 132, + 511, + 399, + 525 + ], + "score": 1.0, + "content": "• Mirror Padding (REFLECT): This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 399, + 513, + 423, + 523 + ], + "score": 0.87, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "kernel without dila-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 141, + 523, + 164, + 536 + ], + "spans": [ + { + "bbox": [ + 141, + 523, + 164, + 536 + ], + "score": 1.0, + "content": "tion.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 133, + 538, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 133, + 538, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 133, + 538, + 368, + 550 + ], + "score": 1.0, + "content": "Replication Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 368, + 539, + 392, + 549 + ], + "score": 0.89, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 538, + 505, + 550 + ], + "score": 1.0, + "content": "kernel without dilation. We", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 141, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 141, + 549, + 262, + 562 + ], + "score": 1.0, + "content": "choose this kernel size since a", + "type": "text" + }, + { + "bbox": [ + 262, + 550, + 283, + 560 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "kernel under SAME padding would render the algorithm", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 561, + 317, + 574 + ], + "spans": [ + { + "bbox": [ + 141, + 561, + 317, + 574 + ], + "score": 1.0, + "content": "equivalent to SYMMETRIC mirror padding.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 132, + 575, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 132, + 575, + 368, + 588 + ], + "score": 1.0, + "content": "• Partial Convolution: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 369, + 576, + 393, + 586 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 575, + 506, + 588 + ], + "score": 1.0, + "content": "kernel without dilation. Its", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 141, + 586, + 391, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 586, + 391, + 600 + ], + "score": 1.0, + "content": "foveation behavior is analogous to REFLECT mirror padding.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 383, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 80, + 384, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 80, + 384, + 96 + ], + "score": 1.0, + "content": "F FOVEATION ANALYSIS OF PADDING ALGORITHMS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 502, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "Refer to http://mind-the-pad.github.io for an interactive and animated visual illustra-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 492, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 492, + 129 + ], + "score": 1.0, + "content": "tion of padding algorithms and their foveation behavior. This appendix serves as a print version.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 105, + 505, + 129 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 505, + 179 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "score": 1.0, + "content": "Among the SAME padding algorithms we discussed in the manuscript, two algorithms warrant that", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 146, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 505, + 158 + ], + "score": 1.0, + "content": "each input pixel is involved in an equal number of convolutional operations, leading to uniform", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 157, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "foveation maps: circular padding and SYMMETRIC mirror padding. In contrast, this number varies", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 167, + 477, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 477, + 180 + ], + "score": 1.0, + "content": "under zero padding, REFLECT mirror padding, replication padding, and partial convolution.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 135, + 505, + 180 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "We illustrate in detail how each padding algorithm treats the input pixels. For this purpose we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "illustrate step by step how each pixel is processed by the convolutional kernel. We choose a set of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 206, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 506, + 218 + ], + "score": 1.0, + "content": "pixels that are sufficient to expose the behavior of the respective algorithm. This set spans an area", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "score": 1.0, + "content": "within two or three pixels from the boundary that encompasses all relevant cases for the analysis and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 228, + 424, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 424, + 240 + ], + "score": 1.0, + "content": "is situated at the top-left corner. The behavior at the other corners is analogous.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 183, + 506, + 240 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 244, + 486, + 256 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 487, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 487, + 259 + ], + "score": 1.0, + "content": "All illustrations use a stride of 1. Except for VALID, all configurations warrant SAME padding.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 242, + 487, + 259 + ] + }, + { + "type": "list", + "bbox": [ + 131, + 266, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 132, + 265, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 132, + 265, + 349, + 278 + ], + "score": 1.0, + "content": "• VALID Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 349, + 266, + 373, + 276 + ], + "score": 0.89, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 265, + 505, + 278 + ], + "score": 1.0, + "content": "kernel without dilation. A larger", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 277, + 391, + 288 + ], + "spans": [ + { + "bbox": [ + 142, + 277, + 391, + 288 + ], + "score": 1.0, + "content": "kernel size or dilation factor will increase the foveation effect.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 140, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 140, + 290, + 345, + 304 + ], + "score": 1.0, + "content": "Zero Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 345, + 292, + 370, + 302 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "kernel without dilation. A larger", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 303, + 391, + 314 + ], + "spans": [ + { + "bbox": [ + 142, + 303, + 391, + 314 + ], + "score": 1.0, + "content": "kernel size or dilation factor will increase the foveation effect.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 140, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 140, + 317, + 366, + 329 + ], + "score": 1.0, + "content": "Circular Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 366, + 317, + 392, + 328 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "kernel without dilation. It", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 329, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 505, + 341 + ], + "score": 1.0, + "content": "is straightforward to prove that the algorithm warrants equal treatment of the pixels irre-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "spective of the kernel size or dilation factor. This is because it effectively applies circular", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "convolution: Once the kernel hits one side, it can seamlessly operate on the pixels of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 142, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "other side. Circular convolution hence renders the feature map as infinite to the kernel,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 373, + 448, + 384 + ], + "spans": [ + { + "bbox": [ + 142, + 373, + 448, + 384 + ], + "score": 1.0, + "content": "warranting that edge pixels are treated in the same manner as interior pixels.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 133, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 133, + 390, + 138, + 396 + ], + "score": 0.818, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 144, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC): This algorithm warrants that each pixel is involved in", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 398, + 504, + 410 + ], + "spans": [ + { + "bbox": [ + 142, + 398, + 504, + 410 + ], + "score": 1.0, + "content": "the same number of convolutional operations. It is important to notice that, unlike un-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 410, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 142, + 410, + 505, + 421 + ], + "score": 1.0, + "content": "der circular convolution, these operations do not utilize the kernel pixels uniformly as we", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 420, + 494, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 494, + 433 + ], + "score": 1.0, + "content": "demonstrate in detail. We illustrate the algorithm behavior under the following settings:", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 151, + 434, + 311, + 447 + ], + "spans": [ + { + "bbox": [ + 151, + 434, + 162, + 447 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 435, + 186, + 446 + ], + "score": 0.86, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 434, + 311, + 447 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1.", + "type": "text" + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 151, + 447, + 310, + 459 + ], + "spans": [ + { + "bbox": [ + 151, + 447, + 162, + 459 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 448, + 186, + 459 + ], + "score": 0.83, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 447, + 310, + 459 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1.", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 152, + 461, + 310, + 472 + ], + "spans": [ + { + "bbox": [ + 152, + 461, + 162, + 471 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 461, + 186, + 472 + ], + "score": 0.8, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 461, + 310, + 471 + ], + "score": 1.0, + "content": "kernel and dilation factor of 2.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 150, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 150, + 473, + 162, + 487 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 474, + 186, + 484 + ], + "score": 0.83, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 473, + 505, + 487 + ], + "score": 1.0, + "content": "kernel and dilation factor of 1, along with a grouped padding strategy to com-", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 160, + 484, + 296, + 497 + ], + "spans": [ + { + "bbox": [ + 160, + 486, + 276, + 496 + ], + "score": 1.0, + "content": "pensate for uneven padding", + "type": "text" + }, + { + "bbox": [ + 277, + 484, + 294, + 496 + ], + "score": 0.27, + "content": "\\pm \\amalg", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 484, + 296, + 497 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 151, + 496, + 485, + 511 + ], + "spans": [ + { + "bbox": [ + 151, + 496, + 162, + 511 + ], + "score": 1.0, + "content": "–", + "type": "text" + }, + { + "bbox": [ + 162, + 498, + 186, + 508 + ], + "score": 0.86, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 496, + 485, + 511 + ], + "score": 1.0, + "content": "kernel size and dilation factor of 1, along with a grouped padding strategy.", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 19.5, + "bbox_fs": [ + 132, + 265, + 506, + 433 + ] + }, + { + "type": "list", + "bbox": [ + 150, + 435, + 504, + 509 + ], + "lines": [], + "index": 29.5, + "bbox_fs": [ + 150, + 434, + 505, + 511 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 133, + 513, + 505, + 535 + ], + "lines": [ + { + "bbox": [ + 132, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 132, + 511, + 399, + 525 + ], + "score": 1.0, + "content": "• Mirror Padding (REFLECT): This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 399, + 513, + 423, + 523 + ], + "score": 0.87, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "kernel without dila-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 141, + 523, + 164, + 536 + ], + "spans": [ + { + "bbox": [ + 141, + 523, + 164, + 536 + ], + "score": 1.0, + "content": "tion.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 132, + 511, + 505, + 536 + ] + }, + { + "type": "list", + "bbox": [ + 133, + 538, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 133, + 538, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 133, + 538, + 368, + 550 + ], + "score": 1.0, + "content": "Replication Padding: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 368, + 539, + 392, + 549 + ], + "score": 0.89, + "content": "5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 538, + 505, + 550 + ], + "score": 1.0, + "content": "kernel without dilation. We", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 141, + 549, + 262, + 562 + ], + "score": 1.0, + "content": "choose this kernel size since a", + "type": "text" + }, + { + "bbox": [ + 262, + 550, + 283, + 560 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "kernel under SAME padding would render the algorithm", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 561, + 317, + 574 + ], + "spans": [ + { + "bbox": [ + 141, + 561, + 317, + 574 + ], + "score": 1.0, + "content": "equivalent to SYMMETRIC mirror padding.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 575, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 132, + 575, + 368, + 588 + ], + "score": 1.0, + "content": "• Partial Convolution: This algorithm is illustrated on a", + "type": "text" + }, + { + "bbox": [ + 369, + 576, + 393, + 586 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 575, + 506, + 588 + ], + "score": 1.0, + "content": "kernel without dilation. Its", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 586, + 391, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 586, + 391, + 600 + ], + "score": 1.0, + "content": "foveation behavior is analogous to REFLECT mirror padding.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + } + ], + "index": 37, + "bbox_fs": [ + 132, + 538, + 506, + 600 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 45, + 76, + 342, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 73, + 342, + 99 + ], + "spans": [ + { + "bbox": [ + 44, + 73, + 180, + 99 + ], + "score": 1.0, + "content": "VALID Padding", + "type": "text" + }, + { + "bbox": [ + 190, + 80, + 342, + 96 + ], + "score": 1.0, + "content": "Illustrated on a 3x3 kernel", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 82, + 114, + 110, + 127 + ], + "lines": [ + { + "bbox": [ + 81, + 112, + 111, + 130 + ], + "spans": [ + { + "bbox": [ + 81, + 112, + 111, + 130 + ], + "score": 1.0, + "content": "Input", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 162, + 110, + 255, + 134 + ], + "lines": [ + { + "bbox": [ + 162, + 111, + 255, + 123 + ], + "spans": [ + { + "bbox": [ + 162, + 111, + 255, + 123 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 161, + 121, + 254, + 134 + ], + "spans": [ + { + "bbox": [ + 161, + 121, + 254, + 134 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 63, + 143, + 128, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 63, + 143, + 128, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 63, + 143, + 128, + 211 + ], + "spans": [ + { + "bbox": [ + 63, + 143, + 128, + 211 + ], + "score": 0.152, + "html": "
abC
def
gh
", + "type": "table", + "image_path": "1bd1e81c91dba38729b05ee03816c771ae6ca8a523f03363c9811b15a7413f26.jpg" + } + ] + } + ], + "index": 5.0, + "virtual_lines": [ + { + "bbox": [ + 63, + 143, + 128, + 177.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 63, + 177.0, + 128, + 211.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 167, + 141, + 248, + 210 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 167, + 141, + 248, + 210 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 167, + 141, + 248, + 210 + ], + "spans": [ + { + "bbox": [ + 167, + 141, + 248, + 210 + ], + "score": 0.211, + "html": "
12333
24666
36999
36999
36999
", + "type": "table", + "image_path": "a59efdde31063700c0c4e43e5424da3e21980c2ec1190cc5622cbcddd3171f09.jpg" + } + ] + } + ], + "index": 6.0, + "virtual_lines": [ + { + "bbox": [ + 167, + 141, + 248, + 175.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 167, + 175.5, + 248, + 210.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 50, + 218, + 224, + 231 + ], + "lines": [ + { + "bbox": [ + 48, + 217, + 225, + 233 + ], + "spans": [ + { + "bbox": [ + 48, + 217, + 225, + 233 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 60, + 236, + 509, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 60, + 236, + 509, + 300 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 60, + 236, + 509, + 300 + ], + "spans": [ + { + "bbox": [ + 60, + 236, + 509, + 300 + ], + "score": 0.911, + "type": "image", + "image_path": "b2f7ea17392566a942f1830fef3e08688e93e3711a4ea27135d9a2b9cf7d34e6.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 60, + 236, + 509, + 257.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 60, + 257.3333333333333, + 509, + 278.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 60, + 278.66666666666663, + 509, + 299.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 57, + 344, + 547, + 422 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 48, + 326, + 288, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "spans": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image_body", + "bbox": [ + 57, + 344, + 547, + 422 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 344, + 547, + 422 + ], + "spans": [ + { + "bbox": [ + 57, + 344, + 547, + 422 + ], + "score": 0.396, + "type": "image", + "image_path": "74f48cb87d1f6bdd6666b649c425ea1dfaa5ae5f8b7d6d1487b47797a64dddd6.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 57, + 344, + 547, + 370.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 57, + 370.0, + 547, + 396.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 57, + 396.0, + 547, + 422.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "text", + "bbox": [ + 57, + 445, + 230, + 456 + ], + "lines": [ + { + "bbox": [ + 57, + 444, + 230, + 457 + ], + "spans": [ + { + "bbox": [ + 57, + 444, + 230, + 457 + ], + "score": 1.0, + "content": "Convolutions involving (d): rotated version of (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 57, + 468, + 151, + 478 + ], + "lines": [ + { + "bbox": [ + 57, + 466, + 152, + 480 + ], + "spans": [ + { + "bbox": [ + 57, + 466, + 152, + 480 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "image", + "bbox": [ + 66, + 488, + 520, + 547 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 66, + 488, + 520, + 547 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 66, + 488, + 520, + 547 + ], + "spans": [ + { + "bbox": [ + 66, + 488, + 520, + 547 + ], + "score": 0.093, + "type": "image", + "image_path": "075393ac4f0fb1926036df9540e55b75341726cc0c0ce597fe3930694431dbc7.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 66, + 488, + 520, + 507.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 66, + 507.6666666666667, + 520, + 527.3333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 66, + 527.3333333333334, + 520, + 547.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 57, + 568, + 150, + 578 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 57, + 565, + 150, + 580 + ], + "spans": [ + { + "bbox": [ + 57, + 565, + 150, + 580 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 70, + 590, + 524, + 649 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 590, + 524, + 649 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 70, + 590, + 524, + 649 + ], + "spans": [ + { + "bbox": [ + 70, + 590, + 524, + 649 + ], + "score": 0.881, + "type": "image", + "image_path": "a9884b5f8b96c555974a7c1a3e43a6404f579230ff72e5c6ffb7dbcd48eee7a4.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 70, + 590, + 524, + 609.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 70, + 609.6666666666666, + 524, + 629.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 70, + 629.3333333333333, + 524, + 648.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 59, + 670, + 234, + 681 + ], + "lines": [ + { + "bbox": [ + 58, + 669, + 233, + 681 + ], + "spans": [ + { + "bbox": [ + 58, + 669, + 233, + 681 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 57, + 693, + 231, + 704 + ], + "lines": [ + { + "bbox": [ + 57, + 692, + 231, + 705 + ], + "spans": [ + { + "bbox": [ + 57, + 692, + 231, + 705 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 57, + 716, + 246, + 728 + ], + "lines": [ + { + "bbox": [ + 57, + 717, + 247, + 728 + ], + "spans": [ + { + "bbox": [ + 57, + 717, + 247, + 728 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 45, + 76, + 342, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 73, + 342, + 99 + ], + "spans": [ + { + "bbox": [ + 44, + 73, + 180, + 99 + ], + "score": 1.0, + "content": "VALID Padding", + "type": "text" + }, + { + "bbox": [ + 190, + 80, + 342, + 96 + ], + "score": 1.0, + "content": "Illustrated on a 3x3 kernel", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 82, + 114, + 110, + 127 + ], + "lines": [ + { + "bbox": [ + 81, + 112, + 111, + 130 + ], + "spans": [ + { + "bbox": [ + 81, + 112, + 111, + 130 + ], + "score": 1.0, + "content": "Input", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 162, + 110, + 255, + 134 + ], + "lines": [ + { + "bbox": [ + 162, + 111, + 255, + 123 + ], + "spans": [ + { + "bbox": [ + 162, + 111, + 255, + 123 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 161, + 121, + 254, + 134 + ], + "spans": [ + { + "bbox": [ + 161, + 121, + 254, + 134 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 63, + 143, + 128, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 63, + 143, + 128, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 63, + 143, + 128, + 211 + ], + "spans": [ + { + "bbox": [ + 63, + 143, + 128, + 211 + ], + "score": 0.152, + "html": "
abC
def
gh
", + "type": "table", + "image_path": "1bd1e81c91dba38729b05ee03816c771ae6ca8a523f03363c9811b15a7413f26.jpg" + } + ] + } + ], + "index": 5.0, + "virtual_lines": [ + { + "bbox": [ + 63, + 143, + 128, + 177.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 63, + 177.0, + 128, + 211.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 167, + 141, + 248, + 210 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 167, + 141, + 248, + 210 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 167, + 141, + 248, + 210 + ], + "spans": [ + { + "bbox": [ + 167, + 141, + 248, + 210 + ], + "score": 0.211, + "html": "
12333
24666
36999
36999
36999
", + "type": "table", + "image_path": "a59efdde31063700c0c4e43e5424da3e21980c2ec1190cc5622cbcddd3171f09.jpg" + } + ] + } + ], + "index": 6.0, + "virtual_lines": [ + { + "bbox": [ + 167, + 141, + 248, + 175.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 167, + 175.5, + 248, + 210.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 50, + 218, + 224, + 231 + ], + "lines": [ + { + "bbox": [ + 48, + 217, + 225, + 233 + ], + "spans": [ + { + "bbox": [ + 48, + 217, + 225, + 233 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 60, + 236, + 509, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 60, + 236, + 509, + 300 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 60, + 236, + 509, + 300 + ], + "spans": [ + { + "bbox": [ + 60, + 236, + 509, + 300 + ], + "score": 0.911, + "type": "image", + "image_path": "b2f7ea17392566a942f1830fef3e08688e93e3711a4ea27135d9a2b9cf7d34e6.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 60, + 236, + 509, + 257.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 60, + 257.3333333333333, + 509, + 278.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 60, + 278.66666666666663, + 509, + 299.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 57, + 344, + 547, + 422 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 48, + 326, + 288, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "spans": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image_body", + "bbox": [ + 57, + 344, + 547, + 422 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 344, + 547, + 422 + ], + "spans": [ + { + "bbox": [ + 57, + 344, + 547, + 422 + ], + "score": 0.396, + "type": "image", + "image_path": "74f48cb87d1f6bdd6666b649c425ea1dfaa5ae5f8b7d6d1487b47797a64dddd6.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 57, + 344, + 547, + 370.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 57, + 370.0, + 547, + 396.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 57, + 396.0, + 547, + 422.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "text", + "bbox": [ + 57, + 445, + 230, + 456 + ], + "lines": [ + { + "bbox": [ + 57, + 444, + 230, + 457 + ], + "spans": [ + { + "bbox": [ + 57, + 444, + 230, + 457 + ], + "score": 1.0, + "content": "Convolutions involving (d): rotated version of (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 57, + 444, + 230, + 457 + ] + }, + { + "type": "text", + "bbox": [ + 57, + 468, + 151, + 478 + ], + "lines": [ + { + "bbox": [ + 57, + 466, + 152, + 480 + ], + "spans": [ + { + "bbox": [ + 57, + 466, + 152, + 480 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 57, + 466, + 152, + 480 + ] + }, + { + "type": "image", + "bbox": [ + 66, + 488, + 520, + 547 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 66, + 488, + 520, + 547 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 66, + 488, + 520, + 547 + ], + "spans": [ + { + "bbox": [ + 66, + 488, + 520, + 547 + ], + "score": 0.093, + "type": "image", + "image_path": "075393ac4f0fb1926036df9540e55b75341726cc0c0ce597fe3930694431dbc7.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 66, + 488, + 520, + 507.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 66, + 507.6666666666667, + 520, + 527.3333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 66, + 527.3333333333334, + 520, + 547.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 57, + 568, + 150, + 578 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 57, + 565, + 150, + 580 + ], + "spans": [ + { + "bbox": [ + 57, + 565, + 150, + 580 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 70, + 590, + 524, + 649 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 590, + 524, + 649 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 70, + 590, + 524, + 649 + ], + "spans": [ + { + "bbox": [ + 70, + 590, + 524, + 649 + ], + "score": 0.881, + "type": "image", + "image_path": "a9884b5f8b96c555974a7c1a3e43a6404f579230ff72e5c6ffb7dbcd48eee7a4.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 70, + 590, + 524, + 609.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 70, + 609.6666666666666, + 524, + 629.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 70, + 629.3333333333333, + 524, + 648.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 59, + 670, + 234, + 681 + ], + "lines": [ + { + "bbox": [ + 58, + 669, + 233, + 681 + ], + "spans": [ + { + "bbox": [ + 58, + 669, + 233, + 681 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 58, + 669, + 233, + 681 + ] + }, + { + "type": "text", + "bbox": [ + 57, + 693, + 231, + 704 + ], + "lines": [ + { + "bbox": [ + 57, + 692, + 231, + 705 + ], + "spans": [ + { + "bbox": [ + 57, + 692, + 231, + 705 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 57, + 692, + 231, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 57, + 716, + 246, + 728 + ], + "lines": [ + { + "bbox": [ + 57, + 717, + 247, + 728 + ], + "spans": [ + { + "bbox": [ + 57, + 717, + 247, + 728 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 57, + 717, + 247, + 728 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 162, + 95 + ], + "lines": [ + { + "bbox": [ + 43, + 72, + 164, + 99 + ], + "spans": [ + { + "bbox": [ + 43, + 72, + 164, + 99 + ], + "score": 1.0, + "content": "Zero Padding", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 45, + 183, + 113, + 196 + ], + "lines": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "spans": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "score": 1.0, + "content": "Original Input", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table", + "bbox": [ + 137, + 201, + 218, + 273 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 153, + 184, + 218, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 182, + 219, + 198 + ], + "spans": [ + { + "bbox": [ + 152, + 182, + 219, + 198 + ], + "score": 1.0, + "content": "Padded Input", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_body", + "bbox": [ + 137, + 201, + 218, + 273 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 137, + 201, + 218, + 273 + ], + "spans": [ + { + "bbox": [ + 137, + 201, + 218, + 273 + ], + "score": 0.516, + "html": "
000• ·
0ab
0Cd
0·
• =
", + "type": "table", + "image_path": "556bf414f5fa773bb6e62cc6bf9a5be9d31d1ce35ef4ed652f3e5eb2c42a6fc1.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 137, + 201, + 218, + 237.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 137, + 237.0, + 218, + 273.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "table", + "bbox": [ + 252, + 209, + 322, + 273 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 243, + 178, + 336, + 201 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 242, + 177, + 336, + 191 + ], + "spans": [ + { + "bbox": [ + 242, + 177, + 336, + 191 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 244, + 189, + 335, + 201 + ], + "spans": [ + { + "bbox": [ + 244, + 189, + 335, + 201 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 252, + 209, + 322, + 273 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 252, + 209, + 322, + 273 + ], + "spans": [ + { + "bbox": [ + 252, + 209, + 322, + 273 + ], + "score": 0.546, + "html": "
46666
69999
6999
69999
69999
", + "type": "table", + "image_path": "f558934c361d8576a111002b479ccad6f1b9812730e47cc3b73814ad1b83231e.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 209, + 322, + 241.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 252, + 241.0, + 322, + 273.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 7.0 + }, + { + "type": "table", + "bbox": [ + 44, + 209, + 113, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 44, + 209, + 113, + 272 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 44, + 209, + 113, + 272 + ], + "spans": [ + { + "bbox": [ + 44, + 209, + 113, + 272 + ], + "score": 0.583, + "html": "
ab··
Cd
·
", + "type": "table", + "image_path": "359a2e24fb2010e55e4bae06e157a5c8ccb87b3102cb9f9cc8606187601612e7.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 44, + 209, + 113, + 240.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 44, + 240.5, + 113, + 272.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 45, + 296, + 219, + 308 + ], + "lines": [ + { + "bbox": [ + 44, + 295, + 220, + 310 + ], + "spans": [ + { + "bbox": [ + 44, + 295, + 220, + 310 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 99, + 328, + 405, + 412 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 328, + 405, + 412 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 99, + 328, + 405, + 412 + ], + "spans": [ + { + "bbox": [ + 99, + 328, + 405, + 412 + ], + "score": 0.874, + "type": "image", + "image_path": "05776e978b31b6a3e6f1341bc05f18a6de2a45877c2b627e7689f8e71e5ec1b5.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 99, + 328, + 405, + 356.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 99, + 356.0, + 405, + 384.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 99, + 384.0, + 405, + 412.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 46, + 452, + 285, + 465 + ], + "lines": [ + { + "bbox": [ + 46, + 452, + 285, + 465 + ], + "spans": [ + { + "bbox": [ + 46, + 452, + 285, + 465 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 45, + 487, + 139, + 497 + ], + "lines": [ + { + "bbox": [ + 45, + 485, + 140, + 499 + ], + "spans": [ + { + "bbox": [ + 45, + 485, + 140, + 499 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "image", + "bbox": [ + 70, + 503, + 387, + 567 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 503, + 387, + 567 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 70, + 503, + 387, + 567 + ], + "spans": [ + { + "bbox": [ + 70, + 503, + 387, + 567 + ], + "score": 0.917, + "type": "image", + "image_path": "2db3478b53918d06b972e15db60b69b5650f9e2b06fc4d4bbc8dc175ac75984c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 70, + 503, + 387, + 524.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 70, + 524.3333333333334, + 387, + 545.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 70, + 545.6666666666667, + 387, + 567.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 43, + 595, + 137, + 605 + ], + "lines": [ + { + "bbox": [ + 43, + 593, + 138, + 607 + ], + "spans": [ + { + "bbox": [ + 43, + 593, + 138, + 607 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "image", + "bbox": [ + 69, + 611, + 551, + 675 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 69, + 611, + 551, + 675 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 69, + 611, + 551, + 675 + ], + "spans": [ + { + "bbox": [ + 69, + 611, + 551, + 675 + ], + "score": 0.898, + "type": "image", + "image_path": "63ee4d6ede122d55ac9f5d26ee63c1d6ddf758f4267f0d482b84beda20114155.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 69, + 611, + 551, + 632.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 69, + 632.3333333333334, + 551, + 653.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 69, + 653.6666666666667, + 551, + 675.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 43, + 703, + 249, + 713 + ], + "lines": [ + { + "bbox": [ + 44, + 702, + 248, + 714 + ], + "spans": [ + { + "bbox": [ + 44, + 702, + 248, + 714 + ], + "score": 1.0, + "content": "Other border cases are translation or rotation of (a) or (b)", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 162, + 95 + ], + "lines": [ + { + "bbox": [ + 43, + 72, + 164, + 99 + ], + "spans": [ + { + "bbox": [ + 43, + 72, + 164, + 99 + ], + "score": 1.0, + "content": "Zero Padding", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 45, + 183, + 113, + 196 + ], + "lines": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "spans": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "score": 1.0, + "content": "Original Input", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table", + "bbox": [ + 137, + 201, + 218, + 273 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 153, + 184, + 218, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 182, + 219, + 198 + ], + "spans": [ + { + "bbox": [ + 152, + 182, + 219, + 198 + ], + "score": 1.0, + "content": "Padded Input", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_body", + "bbox": [ + 137, + 201, + 218, + 273 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 137, + 201, + 218, + 273 + ], + "spans": [ + { + "bbox": [ + 137, + 201, + 218, + 273 + ], + "score": 0.516, + "html": "
000• ·
0ab
0Cd
0·
• =
", + "type": "table", + "image_path": "556bf414f5fa773bb6e62cc6bf9a5be9d31d1ce35ef4ed652f3e5eb2c42a6fc1.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 137, + 201, + 218, + 237.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 137, + 237.0, + 218, + 273.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "table", + "bbox": [ + 252, + 209, + 322, + 273 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 243, + 178, + 336, + 201 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 242, + 177, + 336, + 191 + ], + "spans": [ + { + "bbox": [ + 242, + 177, + 336, + 191 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 244, + 189, + 335, + 201 + ], + "spans": [ + { + "bbox": [ + 244, + 189, + 335, + 201 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 252, + 209, + 322, + 273 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 252, + 209, + 322, + 273 + ], + "spans": [ + { + "bbox": [ + 252, + 209, + 322, + 273 + ], + "score": 0.546, + "html": "
46666
69999
6999
69999
69999
", + "type": "table", + "image_path": "f558934c361d8576a111002b479ccad6f1b9812730e47cc3b73814ad1b83231e.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 209, + 322, + 241.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 252, + 241.0, + 322, + 273.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 7.0 + }, + { + "type": "table", + "bbox": [ + 44, + 209, + 113, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 44, + 209, + 113, + 272 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 44, + 209, + 113, + 272 + ], + "spans": [ + { + "bbox": [ + 44, + 209, + 113, + 272 + ], + "score": 0.583, + "html": "
ab··
Cd
·
", + "type": "table", + "image_path": "359a2e24fb2010e55e4bae06e157a5c8ccb87b3102cb9f9cc8606187601612e7.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 44, + 209, + 113, + 240.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 44, + 240.5, + 113, + 272.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 45, + 296, + 219, + 308 + ], + "lines": [ + { + "bbox": [ + 44, + 295, + 220, + 310 + ], + "spans": [ + { + "bbox": [ + 44, + 295, + 220, + 310 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 99, + 328, + 405, + 412 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 328, + 405, + 412 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 99, + 328, + 405, + 412 + ], + "spans": [ + { + "bbox": [ + 99, + 328, + 405, + 412 + ], + "score": 0.874, + "type": "image", + "image_path": "05776e978b31b6a3e6f1341bc05f18a6de2a45877c2b627e7689f8e71e5ec1b5.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 99, + 328, + 405, + 356.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 99, + 356.0, + 405, + 384.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 99, + 384.0, + 405, + 412.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 46, + 452, + 285, + 465 + ], + "lines": [ + { + "bbox": [ + 46, + 452, + 285, + 465 + ], + "spans": [ + { + "bbox": [ + 46, + 452, + 285, + 465 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 45, + 487, + 139, + 497 + ], + "lines": [ + { + "bbox": [ + 45, + 485, + 140, + 499 + ], + "spans": [ + { + "bbox": [ + 45, + 485, + 140, + 499 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 45, + 485, + 140, + 499 + ] + }, + { + "type": "image", + "bbox": [ + 70, + 503, + 387, + 567 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 503, + 387, + 567 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 70, + 503, + 387, + 567 + ], + "spans": [ + { + "bbox": [ + 70, + 503, + 387, + 567 + ], + "score": 0.917, + "type": "image", + "image_path": "2db3478b53918d06b972e15db60b69b5650f9e2b06fc4d4bbc8dc175ac75984c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 70, + 503, + 387, + 524.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 70, + 524.3333333333334, + 387, + 545.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 70, + 545.6666666666667, + 387, + 567.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 43, + 595, + 137, + 605 + ], + "lines": [ + { + "bbox": [ + 43, + 593, + 138, + 607 + ], + "spans": [ + { + "bbox": [ + 43, + 593, + 138, + 607 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 43, + 593, + 138, + 607 + ] + }, + { + "type": "image", + "bbox": [ + 69, + 611, + 551, + 675 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 69, + 611, + 551, + 675 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 69, + 611, + 551, + 675 + ], + "spans": [ + { + "bbox": [ + 69, + 611, + 551, + 675 + ], + "score": 0.898, + "type": "image", + "image_path": "63ee4d6ede122d55ac9f5d26ee63c1d6ddf758f4267f0d482b84beda20114155.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 69, + 611, + 551, + 632.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 69, + 632.3333333333334, + 551, + 653.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 69, + 653.6666666666667, + 551, + 675.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 43, + 703, + 249, + 713 + ], + "lines": [ + { + "bbox": [ + 44, + 702, + 248, + 714 + ], + "spans": [ + { + "bbox": [ + 44, + 702, + 248, + 714 + ], + "score": 1.0, + "content": "Other border cases are translation or rotation of (a) or (b)", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 44, + 702, + 248, + 714 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 192, + 96 + ], + "lines": [ + { + "bbox": [ + 45, + 72, + 193, + 99 + ], + "spans": [ + { + "bbox": [ + 45, + 72, + 193, + 99 + ], + "score": 1.0, + "content": "Circular Padding", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 45, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 44, + 106, + 302, + 127 + ], + "spans": [ + { + "bbox": [ + 44, + 106, + 302, + 127 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "image", + "bbox": [ + 97, + 134, + 497, + 304 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 97, + 134, + 497, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 134, + 497, + 304 + ], + "spans": [ + { + "bbox": [ + 97, + 134, + 497, + 304 + ], + "score": 0.618, + "type": "image", + "image_path": "312ea301340019a6276321b4c734de7b4bca19c3c6e7a0942faf86f2dfa2b9d9.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 97, + 134, + 497, + 190.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 97, + 190.66666666666666, + 497, + 247.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 97, + 247.33333333333331, + 497, + 304.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 101, + 313, + 276, + 326 + ], + "lines": [ + { + "bbox": [ + 100, + 312, + 276, + 327 + ], + "spans": [ + { + "bbox": [ + 100, + 312, + 276, + 327 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 100, + 334, + 244, + 418 + ], + "lines": [ + { + "bbox": [ + 101, + 340, + 238, + 350 + ], + "spans": [ + { + "bbox": [ + 101, + 342, + 108, + 349 + ], + "score": 1.0, + "content": "a", + "type": "text" + }, + { + "bbox": [ + 121, + 341, + 128, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 340, + 144, + 350 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 153, + 341, + 159, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 179, + 340, + 187, + 350 + ], + "score": 1.0, + "content": "b", + "type": "text" + }, + { + "bbox": [ + 199, + 341, + 207, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 215, + 340, + 223, + 350 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 231, + 341, + 238, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 121, + 357, + 239, + 366 + ], + "spans": [ + { + "bbox": [ + 121, + 357, + 128, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 357, + 144, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 153, + 357, + 159, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 199, + 357, + 207, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 214, + 357, + 223, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 231, + 357, + 239, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 121, + 373, + 239, + 383 + ], + "spans": [ + { + "bbox": [ + 121, + 374, + 128, + 382 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 373, + 144, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 152, + 373, + 160, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 199, + 373, + 207, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 214, + 373, + 223, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 230, + 373, + 239, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 127, + 388, + 237, + 400 + ], + "spans": [ + { + "bbox": [ + 127, + 388, + 158, + 399 + ], + "score": 1.0, + "content": "sum = 9", + "type": "text" + }, + { + "bbox": [ + 206, + 388, + 237, + 400 + ], + "score": 1.0, + "content": "sum = 9", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 128, + 403, + 231, + 415 + ], + "spans": [ + { + "bbox": [ + 128, + 403, + 152, + 415 + ], + "score": 1.0, + "content": "uniform", + "type": "text" + }, + { + "bbox": [ + 206, + 403, + 231, + 415 + ], + "score": 1.0, + "content": "uniform", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 96, + 431, + 335, + 444 + ], + "lines": [ + { + "bbox": [ + 96, + 431, + 335, + 444 + ], + "spans": [ + { + "bbox": [ + 96, + 431, + 335, + 444 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 99, + 451, + 193, + 461 + ], + "lines": [ + { + "bbox": [ + 99, + 449, + 194, + 463 + ], + "spans": [ + { + "bbox": [ + 99, + 449, + 194, + 463 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 99, + 457, + 465, + 733 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 457, + 465, + 733 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 99, + 457, + 465, + 733 + ], + "spans": [ + { + "bbox": [ + 99, + 457, + 465, + 733 + ], + "score": 0.664, + "type": "image", + "image_path": "315c801266b735e20fd2dda8208ed2b4da076ddadc7c67b0b3f5105bfd9d73fd.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 99, + 457, + 465, + 549.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 99, + 549.0, + 465, + 641.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 99, + 641.0, + 465, + 733.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 102, + 741, + 307, + 752 + ], + "lines": [ + { + "bbox": [ + 102, + 740, + 307, + 753 + ], + "spans": [ + { + "bbox": [ + 102, + 740, + 307, + 753 + ], + "score": 1.0, + "content": "Other border cases are translation or rotation of (a) or (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 192, + 96 + ], + "lines": [ + { + "bbox": [ + 45, + 72, + 193, + 99 + ], + "spans": [ + { + "bbox": [ + 45, + 72, + 193, + 99 + ], + "score": 1.0, + "content": "Circular Padding", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 45, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 44, + 106, + 302, + 127 + ], + "spans": [ + { + "bbox": [ + 44, + 106, + 302, + 127 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 44, + 106, + 302, + 127 + ] + }, + { + "type": "image", + "bbox": [ + 97, + 134, + 497, + 304 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 97, + 134, + 497, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 134, + 497, + 304 + ], + "spans": [ + { + "bbox": [ + 97, + 134, + 497, + 304 + ], + "score": 0.618, + "type": "image", + "image_path": "312ea301340019a6276321b4c734de7b4bca19c3c6e7a0942faf86f2dfa2b9d9.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 97, + 134, + 497, + 190.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 97, + 190.66666666666666, + 497, + 247.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 97, + 247.33333333333331, + 497, + 304.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 101, + 313, + 276, + 326 + ], + "lines": [ + { + "bbox": [ + 100, + 312, + 276, + 327 + ], + "spans": [ + { + "bbox": [ + 100, + 312, + 276, + 327 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "index", + "bbox": [ + 100, + 334, + 244, + 418 + ], + "lines": [ + { + "bbox": [ + 101, + 340, + 238, + 350 + ], + "spans": [ + { + "bbox": [ + 101, + 342, + 108, + 349 + ], + "score": 1.0, + "content": "a", + "type": "text" + }, + { + "bbox": [ + 121, + 341, + 128, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 340, + 144, + 350 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 153, + 341, + 159, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 179, + 340, + 187, + 350 + ], + "score": 1.0, + "content": "b", + "type": "text" + }, + { + "bbox": [ + 199, + 341, + 207, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 215, + 340, + 223, + 350 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 231, + 341, + 238, + 349 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 357, + 239, + 366 + ], + "spans": [ + { + "bbox": [ + 121, + 357, + 128, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 357, + 144, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 153, + 357, + 159, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 199, + 357, + 207, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 214, + 357, + 223, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 231, + 357, + 239, + 366 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 373, + 239, + 383 + ], + "spans": [ + { + "bbox": [ + 121, + 374, + 128, + 382 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 136, + 373, + 144, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 152, + 373, + 160, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 199, + 373, + 207, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 214, + 373, + 223, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 230, + 373, + 239, + 383 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 388, + 237, + 400 + ], + "spans": [ + { + "bbox": [ + 127, + 388, + 158, + 399 + ], + "score": 1.0, + "content": "sum = 9", + "type": "text" + }, + { + "bbox": [ + 206, + 388, + 237, + 400 + ], + "score": 1.0, + "content": "sum = 9", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 403, + 231, + 415 + ], + "spans": [ + { + "bbox": [ + 128, + 403, + 152, + 415 + ], + "score": 1.0, + "content": "uniform", + "type": "text" + }, + { + "bbox": [ + 206, + 403, + 231, + 415 + ], + "score": 1.0, + "content": "uniform", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + } + ], + "index": 8, + "bbox_fs": [ + 101, + 340, + 239, + 415 + ] + }, + { + "type": "title", + "bbox": [ + 96, + 431, + 335, + 444 + ], + "lines": [ + { + "bbox": [ + 96, + 431, + 335, + 444 + ], + "spans": [ + { + "bbox": [ + 96, + 431, + 335, + 444 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 99, + 451, + 193, + 461 + ], + "lines": [ + { + "bbox": [ + 99, + 449, + 194, + 463 + ], + "spans": [ + { + "bbox": [ + 99, + 449, + 194, + 463 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 99, + 449, + 194, + 463 + ] + }, + { + "type": "image", + "bbox": [ + 99, + 457, + 465, + 733 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 457, + 465, + 733 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 99, + 457, + 465, + 733 + ], + "spans": [ + { + "bbox": [ + 99, + 457, + 465, + 733 + ], + "score": 0.664, + "type": "image", + "image_path": "315c801266b735e20fd2dda8208ed2b4da076ddadc7c67b0b3f5105bfd9d73fd.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 99, + 457, + 465, + 549.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 99, + 549.0, + 465, + 641.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 99, + 641.0, + 465, + 733.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 102, + 741, + 307, + 752 + ], + "lines": [ + { + "bbox": [ + 102, + 740, + 307, + 753 + ], + "spans": [ + { + "bbox": [ + 102, + 740, + 307, + 753 + ], + "score": 1.0, + "content": "Other border cases are translation or rotation of (a) or (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 102, + 740, + 307, + 753 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 301, + 96 + ], + "lines": [ + { + "bbox": [ + 43, + 73, + 302, + 99 + ], + "spans": [ + { + "bbox": [ + 43, + 73, + 302, + 99 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "image", + "bbox": [ + 50, + 159, + 351, + 266 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 50, + 159, + 351, + 266 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 50, + 159, + 351, + 266 + ], + "spans": [ + { + "bbox": [ + 50, + 159, + 351, + 266 + ], + "score": 0.53, + "type": "image", + "image_path": "6ab53447be0e84af51087dd7f95b2e53e57f5b4c85508482b2d0499f14281ec1.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 50, + 159, + 351, + 194.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 50, + 194.66666666666666, + 351, + 230.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 50, + 230.33333333333331, + 351, + 266.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 54, + 283, + 228, + 295 + ], + "lines": [ + { + "bbox": [ + 54, + 283, + 229, + 297 + ], + "spans": [ + { + "bbox": [ + 54, + 283, + 229, + 297 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "image", + "bbox": [ + 52, + 300, + 517, + 374 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 52, + 300, + 517, + 374 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 52, + 300, + 517, + 374 + ], + "spans": [ + { + "bbox": [ + 52, + 300, + 517, + 374 + ], + "score": 0.409, + "type": "image", + "image_path": "6122910bbc215ed57d822669b52bf0365aa06feda5eeb930d346877584b23468.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 52, + 300, + 517, + 324.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 52, + 324.6666666666667, + 517, + 349.33333333333337 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 52, + 349.33333333333337, + 517, + 374.00000000000006 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 54, + 398, + 293, + 412 + ], + "lines": [ + { + "bbox": [ + 54, + 399, + 293, + 412 + ], + "spans": [ + { + "bbox": [ + 54, + 399, + 293, + 412 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 78, + 442, + 172, + 452 + ], + "lines": [ + { + "bbox": [ + 77, + 440, + 172, + 454 + ], + "spans": [ + { + "bbox": [ + 77, + 440, + 172, + 454 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 78, + 459, + 384, + 524 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 78, + 459, + 384, + 524 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 78, + 459, + 384, + 524 + ], + "spans": [ + { + "bbox": [ + 78, + 459, + 384, + 524 + ], + "score": 0.639, + "type": "image", + "image_path": "207ecdbeda64fb86e5c2fb8bd0fca5301fca26b3c149e3e0de0f78f19b80b6fc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 78, + 459, + 384, + 480.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 78, + 480.6666666666667, + 384, + 502.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 78, + 502.33333333333337, + 384, + 524.0 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 78, + 568, + 542, + 634 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 78, + 550, + 172, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 77, + 548, + 172, + 562 + ], + "spans": [ + { + "bbox": [ + 77, + 548, + 172, + 562 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "image_body", + "bbox": [ + 78, + 568, + 542, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 568, + 542, + 634 + ], + "spans": [ + { + "bbox": [ + 78, + 568, + 542, + 634 + ], + "score": 0.762, + "type": "image", + "image_path": "ce9ae3f6f155e9103378337a6607d52884729cba759d261dd7aa7d104d009ab0.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 78, + 568, + 542, + 590.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 78, + 590.0, + 542, + 612.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 78, + 612.0, + 542, + 634.0 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 15.0 + }, + { + "type": "image", + "bbox": [ + 81, + 679, + 552, + 744 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 81, + 679, + 552, + 744 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 81, + 679, + 552, + 744 + ], + "spans": [ + { + "bbox": [ + 81, + 679, + 552, + 744 + ], + "score": 0.743, + "type": "image", + "image_path": "7ff8f52e1f1fa7323a591bc9045a4daa5fbdbf91e09c8b8c0c73e8c62390c4f3.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 81, + 679, + 552, + 700.6666666666666 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 81, + 700.6666666666666, + 552, + 722.3333333333333 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 81, + 722.3333333333333, + 552, + 743.9999999999999 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 19 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 301, + 96 + ], + "lines": [ + { + "bbox": [ + 43, + 73, + 302, + 99 + ], + "spans": [ + { + "bbox": [ + 43, + 73, + 302, + 99 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 45, + 106, + 302, + 126 + ] + }, + { + "type": "image", + "bbox": [ + 50, + 159, + 351, + 266 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 50, + 159, + 351, + 266 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 50, + 159, + 351, + 266 + ], + "spans": [ + { + "bbox": [ + 50, + 159, + 351, + 266 + ], + "score": 0.53, + "type": "image", + "image_path": "6ab53447be0e84af51087dd7f95b2e53e57f5b4c85508482b2d0499f14281ec1.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 50, + 159, + 351, + 194.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 50, + 194.66666666666666, + 351, + 230.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 50, + 230.33333333333331, + 351, + 266.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 54, + 283, + 228, + 295 + ], + "lines": [ + { + "bbox": [ + 54, + 283, + 229, + 297 + ], + "spans": [ + { + "bbox": [ + 54, + 283, + 229, + 297 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "image", + "bbox": [ + 52, + 300, + 517, + 374 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 52, + 300, + 517, + 374 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 52, + 300, + 517, + 374 + ], + "spans": [ + { + "bbox": [ + 52, + 300, + 517, + 374 + ], + "score": 0.409, + "type": "image", + "image_path": "6122910bbc215ed57d822669b52bf0365aa06feda5eeb930d346877584b23468.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 52, + 300, + 517, + 324.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 52, + 324.6666666666667, + 517, + 349.33333333333337 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 52, + 349.33333333333337, + 517, + 374.00000000000006 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 54, + 398, + 293, + 412 + ], + "lines": [ + { + "bbox": [ + 54, + 399, + 293, + 412 + ], + "spans": [ + { + "bbox": [ + 54, + 399, + 293, + 412 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 78, + 442, + 172, + 452 + ], + "lines": [ + { + "bbox": [ + 77, + 440, + 172, + 454 + ], + "spans": [ + { + "bbox": [ + 77, + 440, + 172, + 454 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 77, + 440, + 172, + 454 + ] + }, + { + "type": "image", + "bbox": [ + 78, + 459, + 384, + 524 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 78, + 459, + 384, + 524 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 78, + 459, + 384, + 524 + ], + "spans": [ + { + "bbox": [ + 78, + 459, + 384, + 524 + ], + "score": 0.639, + "type": "image", + "image_path": "207ecdbeda64fb86e5c2fb8bd0fca5301fca26b3c149e3e0de0f78f19b80b6fc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 78, + 459, + 384, + 480.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 78, + 480.6666666666667, + 384, + 502.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 78, + 502.33333333333337, + 384, + 524.0 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 78, + 568, + 542, + 634 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 78, + 550, + 172, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 77, + 548, + 172, + 562 + ], + "spans": [ + { + "bbox": [ + 77, + 548, + 172, + 562 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "image_body", + "bbox": [ + 78, + 568, + 542, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 568, + 542, + 634 + ], + "spans": [ + { + "bbox": [ + 78, + 568, + 542, + 634 + ], + "score": 0.762, + "type": "image", + "image_path": "ce9ae3f6f155e9103378337a6607d52884729cba759d261dd7aa7d104d009ab0.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 78, + 568, + 542, + 590.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 78, + 590.0, + 542, + 612.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 78, + 612.0, + 542, + 634.0 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 15.0 + }, + { + "type": "image", + "bbox": [ + 81, + 679, + 552, + 744 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 81, + 679, + 552, + 744 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 81, + 679, + 552, + 744 + ], + "spans": [ + { + "bbox": [ + 81, + 679, + 552, + 744 + ], + "score": 0.743, + "type": "image", + "image_path": "7ff8f52e1f1fa7323a591bc9045a4daa5fbdbf91e09c8b8c0c73e8c62390c4f3.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 81, + 679, + 552, + 700.6666666666666 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 81, + 700.6666666666666, + 552, + 722.3333333333333 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 81, + 722.3333333333333, + 552, + 743.9999999999999 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 19 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 50, + 129, + 142, + 138 + ], + "lines": [ + { + "bbox": [ + 48, + 127, + 143, + 140 + ], + "spans": [ + { + "bbox": [ + 48, + 127, + 143, + 140 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 77, + 158, + 534, + 302 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 77, + 158, + 534, + 302 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 77, + 158, + 534, + 302 + ], + "spans": [ + { + "bbox": [ + 77, + 158, + 534, + 302 + ], + "score": 0.89, + "type": "image", + "image_path": "635b17420ad7db5958a7c27e978a4071b990979ca04dc4ace8f486033de20cd5.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 77, + 158, + 534, + 206.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 77, + 206.0, + 534, + 254.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 77, + 254.0, + 534, + 302.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 48, + 323, + 140, + 333 + ], + "lines": [ + { + "bbox": [ + 48, + 321, + 141, + 335 + ], + "spans": [ + { + "bbox": [ + 48, + 321, + 141, + 335 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 80, + 346, + 524, + 490 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 80, + 346, + 524, + 490 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 80, + 346, + 524, + 490 + ], + "spans": [ + { + "bbox": [ + 80, + 346, + 524, + 490 + ], + "score": 0.908, + "type": "image", + "image_path": "83b97cfd9b41665e91f8e411ba270b85c9a125588a02eeca4e21f821f73b4bdb.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 80, + 346, + 524, + 394.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 80, + 394.0, + 524, + 442.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 80, + 442.0, + 524, + 490.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 47, + 566, + 222, + 576 + ], + "lines": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "spans": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 46, + 589, + 219, + 600 + ], + "lines": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "spans": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 46, + 613, + 235, + 623 + ], + "lines": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "spans": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 50, + 104, + 224, + 114 + ], + "lines": [ + { + "bbox": [ + 49, + 103, + 225, + 115 + ], + "spans": [ + { + "bbox": [ + 49, + 103, + 225, + 115 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 50, + 129, + 142, + 138 + ], + "lines": [ + { + "bbox": [ + 48, + 127, + 143, + 140 + ], + "spans": [ + { + "bbox": [ + 48, + 127, + 143, + 140 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 48, + 127, + 143, + 140 + ] + }, + { + "type": "image", + "bbox": [ + 77, + 158, + 534, + 302 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 77, + 158, + 534, + 302 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 77, + 158, + 534, + 302 + ], + "spans": [ + { + "bbox": [ + 77, + 158, + 534, + 302 + ], + "score": 0.89, + "type": "image", + "image_path": "635b17420ad7db5958a7c27e978a4071b990979ca04dc4ace8f486033de20cd5.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 77, + 158, + 534, + 206.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 77, + 206.0, + 534, + 254.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 77, + 254.0, + 534, + 302.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 48, + 323, + 140, + 333 + ], + "lines": [ + { + "bbox": [ + 48, + 321, + 141, + 335 + ], + "spans": [ + { + "bbox": [ + 48, + 321, + 141, + 335 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 48, + 321, + 141, + 335 + ] + }, + { + "type": "image", + "bbox": [ + 80, + 346, + 524, + 490 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 80, + 346, + 524, + 490 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 80, + 346, + 524, + 490 + ], + "spans": [ + { + "bbox": [ + 80, + 346, + 524, + 490 + ], + "score": 0.908, + "type": "image", + "image_path": "83b97cfd9b41665e91f8e411ba270b85c9a125588a02eeca4e21f821f73b4bdb.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 80, + 346, + 524, + 394.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 80, + 394.0, + 524, + 442.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 80, + 442.0, + 524, + 490.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 47, + 566, + 222, + 576 + ], + "lines": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "spans": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 46, + 564, + 222, + 578 + ] + }, + { + "type": "text", + "bbox": [ + 46, + 589, + 219, + 600 + ], + "lines": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "spans": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 45, + 588, + 219, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 46, + 613, + 235, + 623 + ], + "lines": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "spans": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 46, + 613, + 235, + 624 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 74, + 80, + 241, + 160 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 74, + 80, + 241, + 160 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 74, + 80, + 241, + 160 + ], + "spans": [ + { + "bbox": [ + 74, + 80, + 241, + 160 + ], + "score": 0.681, + "type": "image", + "image_path": "8a853098ca11735df8a765434a1c5d4f3771546980790927ebd6e4534ec0b669.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 74, + 80, + 241, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 74, + 96.0, + 241, + 112.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 74, + 112.0, + 241, + 128.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 74, + 128.0, + 241, + 144.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 74, + 144.0, + 241, + 160.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 289, + 79, + 545, + 100 + ], + "lines": [ + { + "bbox": [ + 287, + 75, + 547, + 102 + ], + "spans": [ + { + "bbox": [ + 287, + 75, + 547, + 102 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 127 + ], + "lines": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "spans": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "score": 1.0, + "content": "Illustrated on 5x5 kernel and 2-pixel padding", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 71, + 166, + 245, + 178 + ], + "lines": [ + { + "bbox": [ + 70, + 166, + 246, + 180 + ], + "spans": [ + { + "bbox": [ + 70, + 166, + 246, + 180 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 91, + 182, + 509, + 241 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 91, + 182, + 509, + 241 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 91, + 182, + 509, + 241 + ], + "spans": [ + { + "bbox": [ + 91, + 182, + 509, + 241 + ], + "score": 0.937, + "type": "image", + "image_path": "2ed647a62055c626bfc6be0708fddbc84405c35e0aa5605dfe2983f80d1917e0.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 91, + 182, + 509, + 201.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 91, + 201.66666666666666, + 509, + 221.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 91, + 221.33333333333331, + 509, + 240.99999999999997 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 70, + 251, + 309, + 263 + ], + "lines": [ + { + "bbox": [ + 70, + 251, + 309, + 264 + ], + "spans": [ + { + "bbox": [ + 70, + 251, + 309, + 264 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 71, + 269, + 164, + 279 + ], + "lines": [ + { + "bbox": [ + 70, + 267, + 165, + 281 + ], + "spans": [ + { + "bbox": [ + 70, + 267, + 165, + 281 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 106, + 283, + 479, + 395 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 283, + 479, + 395 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 283, + 479, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 479, + 395 + ], + "score": 0.922, + "type": "image", + "image_path": "fb93f2025462e471e18bab29c553707d796f5a5df450e689eb1d800b265197c9.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 283, + 479, + 320.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 320.3333333333333, + 479, + 357.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 357.66666666666663, + 479, + 394.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 69, + 405, + 164, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 69, + 403, + 164, + 417 + ], + "spans": [ + { + "bbox": [ + 69, + 403, + 164, + 417 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + } + ], + "index": 15.0 + }, + { + "type": "image", + "bbox": [ + 105, + 424, + 501, + 537 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 424, + 501, + 537 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 424, + 501, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 501, + 537 + ], + "score": 0.921, + "type": "image", + "image_path": "d6b1ab684ccc9d8b4d0633f56a62178f4ad4cceda1c73f8111bc9eb705c2d5a9.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 105, + 424, + 501, + 461.6666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 105, + 461.6666666666667, + 501, + 499.33333333333337 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 105, + 499.33333333333337, + 501, + 537.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 67, + 548, + 161, + 558 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 67, + 546, + 162, + 560 + ], + "spans": [ + { + "bbox": [ + 67, + 546, + 162, + 560 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 19.0 + }, + { + "type": "image", + "bbox": [ + 105, + 567, + 501, + 739 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 567, + 501, + 739 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 105, + 567, + 501, + 739 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 501, + 739 + ], + "score": 0.946, + "type": "image", + "image_path": "923c01362a6ed3e205d71bfc24d478c52b3681aa29cd39b01495efc03ffbd84f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 105, + 567, + 501, + 624.3333333333334 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 105, + 624.3333333333334, + 501, + 681.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 105, + 681.6666666666667, + 501, + 739.0000000000001 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 74, + 80, + 241, + 160 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 74, + 80, + 241, + 160 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 74, + 80, + 241, + 160 + ], + "spans": [ + { + "bbox": [ + 74, + 80, + 241, + 160 + ], + "score": 0.681, + "type": "image", + "image_path": "8a853098ca11735df8a765434a1c5d4f3771546980790927ebd6e4534ec0b669.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 74, + 80, + 241, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 74, + 96.0, + 241, + 112.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 74, + 112.0, + 241, + 128.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 74, + 128.0, + 241, + 144.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 74, + 144.0, + 241, + 160.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 289, + 79, + 545, + 100 + ], + "lines": [ + { + "bbox": [ + 287, + 75, + 547, + 102 + ], + "spans": [ + { + "bbox": [ + 287, + 75, + 547, + 102 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 127 + ], + "lines": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "spans": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "score": 1.0, + "content": "Illustrated on 5x5 kernel and 2-pixel padding", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 289, + 110, + 545, + 129 + ] + }, + { + "type": "title", + "bbox": [ + 71, + 166, + 245, + 178 + ], + "lines": [ + { + "bbox": [ + 70, + 166, + 246, + 180 + ], + "spans": [ + { + "bbox": [ + 70, + 166, + 246, + 180 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 91, + 182, + 509, + 241 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 91, + 182, + 509, + 241 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 91, + 182, + 509, + 241 + ], + "spans": [ + { + "bbox": [ + 91, + 182, + 509, + 241 + ], + "score": 0.937, + "type": "image", + "image_path": "2ed647a62055c626bfc6be0708fddbc84405c35e0aa5605dfe2983f80d1917e0.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 91, + 182, + 509, + 201.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 91, + 201.66666666666666, + 509, + 221.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 91, + 221.33333333333331, + 509, + 240.99999999999997 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 70, + 251, + 309, + 263 + ], + "lines": [ + { + "bbox": [ + 70, + 251, + 309, + 264 + ], + "spans": [ + { + "bbox": [ + 70, + 251, + 309, + 264 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 71, + 269, + 164, + 279 + ], + "lines": [ + { + "bbox": [ + 70, + 267, + 165, + 281 + ], + "spans": [ + { + "bbox": [ + 70, + 267, + 165, + 281 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 70, + 267, + 165, + 281 + ] + }, + { + "type": "image", + "bbox": [ + 106, + 283, + 479, + 395 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 283, + 479, + 395 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 283, + 479, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 479, + 395 + ], + "score": 0.922, + "type": "image", + "image_path": "fb93f2025462e471e18bab29c553707d796f5a5df450e689eb1d800b265197c9.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 283, + 479, + 320.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 320.3333333333333, + 479, + 357.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 357.66666666666663, + 479, + 394.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 69, + 405, + 164, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 69, + 403, + 164, + 417 + ], + "spans": [ + { + "bbox": [ + 69, + 403, + 164, + 417 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + } + ], + "index": 15.0 + }, + { + "type": "image", + "bbox": [ + 105, + 424, + 501, + 537 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 424, + 501, + 537 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 424, + 501, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 501, + 537 + ], + "score": 0.921, + "type": "image", + "image_path": "d6b1ab684ccc9d8b4d0633f56a62178f4ad4cceda1c73f8111bc9eb705c2d5a9.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 105, + 424, + 501, + 461.6666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 105, + 461.6666666666667, + 501, + 499.33333333333337 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 105, + 499.33333333333337, + 501, + 537.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 67, + 548, + 161, + 558 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 67, + 546, + 162, + 560 + ], + "spans": [ + { + "bbox": [ + 67, + 546, + 162, + 560 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 19.0 + }, + { + "type": "image", + "bbox": [ + 105, + 567, + 501, + 739 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 567, + 501, + 739 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 105, + 567, + 501, + 739 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 501, + 739 + ], + "score": 0.946, + "type": "image", + "image_path": "923c01362a6ed3e205d71bfc24d478c52b3681aa29cd39b01495efc03ffbd84f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 105, + 567, + 501, + 624.3333333333334 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 105, + 624.3333333333334, + 501, + 681.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 105, + 681.6666666666667, + 501, + 739.0000000000001 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 79, + 84, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "spans": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 105, + 111, + 476, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 111, + 476, + 308 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 111, + 476, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 476, + 308 + ], + "score": 0.943, + "type": "image", + "image_path": "7e697dd45efbc10ee0c5668063cefbca79bec51cfe0a5bfd6d8827c821509c0a.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 105, + 111, + 476, + 176.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 105, + 176.66666666666669, + 476, + 242.33333333333337 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 105, + 242.33333333333337, + 476, + 308.00000000000006 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 95, + 354, + 476, + 610 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 336, + 171, + 345 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 79, + 334, + 172, + 348 + ], + "spans": [ + { + "bbox": [ + 79, + 334, + 172, + 348 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 95, + 354, + 476, + 610 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 95, + 354, + 476, + 610 + ], + "spans": [ + { + "bbox": [ + 95, + 354, + 476, + 610 + ], + "score": 0.949, + "type": "image", + "image_path": "9a8f10a070589ef06b9f1a5b0651e3ebe32e891395ebc4fbf96356dc972aba0c.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 95, + 354, + 476, + 439.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 95, + 439.3333333333333, + 476, + 524.6666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 95, + 524.6666666666666, + 476, + 610.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 72, + 650, + 246, + 661 + ], + "lines": [ + { + "bbox": [ + 71, + 649, + 247, + 662 + ], + "spans": [ + { + "bbox": [ + 71, + 649, + 247, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "spans": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 79, + 64, + 253, + 74 + ], + "lines": [ + { + "bbox": [ + 78, + 63, + 254, + 75 + ], + "spans": [ + { + "bbox": [ + 78, + 63, + 254, + 75 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 79, + 84, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "spans": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 78, + 82, + 173, + 96 + ] + }, + { + "type": "image", + "bbox": [ + 105, + 111, + 476, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 105, + 111, + 476, + 308 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 111, + 476, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 476, + 308 + ], + "score": 0.943, + "type": "image", + "image_path": "7e697dd45efbc10ee0c5668063cefbca79bec51cfe0a5bfd6d8827c821509c0a.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 105, + 111, + 476, + 176.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 105, + 176.66666666666669, + 476, + 242.33333333333337 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 105, + 242.33333333333337, + 476, + 308.00000000000006 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 95, + 354, + 476, + 610 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 336, + 171, + 345 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 79, + 334, + 172, + 348 + ], + "spans": [ + { + "bbox": [ + 79, + 334, + 172, + 348 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 95, + 354, + 476, + 610 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 95, + 354, + 476, + 610 + ], + "spans": [ + { + "bbox": [ + 95, + 354, + 476, + 610 + ], + "score": 0.949, + "type": "image", + "image_path": "9a8f10a070589ef06b9f1a5b0651e3ebe32e891395ebc4fbf96356dc972aba0c.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 95, + 354, + 476, + 439.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 95, + 439.3333333333333, + 476, + 524.6666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 95, + 524.6666666666666, + 476, + 610.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 72, + 650, + 246, + 661 + ], + "lines": [ + { + "bbox": [ + 71, + 649, + 247, + 662 + ], + "spans": [ + { + "bbox": [ + 71, + 649, + 247, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 71, + 649, + 247, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "spans": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 70, + 673, + 244, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 70, + 697, + 259, + 708 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 76, + 80, + 253, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 76, + 80, + 253, + 155 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 76, + 80, + 253, + 155 + ], + "spans": [ + { + "bbox": [ + 76, + 80, + 253, + 155 + ], + "score": 0.853, + "type": "image", + "image_path": "6ab1a20295a0fdba167ef4c254d75ebac90a68354c628ae8597a45a489da49ce.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 76, + 80, + 253, + 95.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 76, + 95.0, + 253, + 110.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 76, + 110.0, + 253, + 125.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 76, + 125.0, + 253, + 140.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 76, + 140.0, + 253, + 155.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 289, + 79, + 545, + 100 + ], + "lines": [ + { + "bbox": [ + 287, + 74, + 547, + 103 + ], + "spans": [ + { + "bbox": [ + 287, + 74, + 547, + 103 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 140 + ], + "lines": [ + { + "bbox": [ + 288, + 110, + 545, + 128 + ], + "spans": [ + { + "bbox": [ + 288, + 110, + 367, + 128 + ], + "score": 1.0, + "content": "Illustrated on", + "type": "text" + }, + { + "bbox": [ + 367, + 113, + 390, + 126 + ], + "score": 0.46, + "content": "\\mathbf { 3 \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 110, + 545, + 128 + ], + "score": 1.0, + "content": "kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 350, + 126, + 485, + 139 + ], + "spans": [ + { + "bbox": [ + 350, + 126, + 485, + 139 + ], + "score": 1.0, + "content": "with dilation factor of 2", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 71, + 162, + 244, + 174 + ], + "lines": [ + { + "bbox": [ + 69, + 162, + 245, + 176 + ], + "spans": [ + { + "bbox": [ + 69, + 162, + 245, + 176 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 92, + 177, + 380, + 228 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 92, + 177, + 380, + 228 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 92, + 177, + 380, + 228 + ], + "spans": [ + { + "bbox": [ + 92, + 177, + 380, + 228 + ], + "score": 0.942, + "type": "image", + "image_path": "d99e99f62153187884e0a145ec5dd4f0201fb8255b44b2c45aa57158fc39430c.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 92, + 177, + 380, + 194.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 92, + 194.0, + 380, + 211.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 92, + 211.0, + 380, + 228.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 67, + 237, + 307, + 249 + ], + "lines": [ + { + "bbox": [ + 68, + 236, + 307, + 249 + ], + "spans": [ + { + "bbox": [ + 68, + 236, + 307, + 249 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 78, + 257, + 172, + 268 + ], + "lines": [ + { + "bbox": [ + 77, + 255, + 172, + 270 + ], + "spans": [ + { + "bbox": [ + 77, + 255, + 172, + 270 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "image", + "bbox": [ + 107, + 273, + 492, + 394 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 273, + 492, + 394 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 273, + 492, + 394 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 492, + 394 + ], + "score": 0.948, + "type": "image", + "image_path": "749f38dde001c33c4de5e80e1053df61ebc232047d92d82b0267074b9fdafae1.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 107, + 273, + 492, + 313.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 107, + 313.3333333333333, + 492, + 353.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 353.66666666666663, + 492, + 393.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 78, + 405, + 173, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 403, + 173, + 417 + ], + "spans": [ + { + "bbox": [ + 78, + 403, + 173, + 417 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 16.0 + }, + { + "type": "image", + "bbox": [ + 107, + 424, + 518, + 545 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 424, + 518, + 545 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 424, + 518, + 545 + ], + "spans": [ + { + "bbox": [ + 107, + 424, + 518, + 545 + ], + "score": 0.96, + "type": "image", + "image_path": "3d93f43f0bbad03a7c07d066a148fcedbe3c2528cec4208b76899c07614a04d1.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 107, + 424, + 518, + 464.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 107, + 464.3333333333333, + 518, + 504.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 504.66666666666663, + 518, + 545.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 78, + 550, + 172, + 560 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 78, + 548, + 173, + 562 + ], + "spans": [ + { + "bbox": [ + 78, + 548, + 173, + 562 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 108, + 570, + 518, + 752 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 570, + 518, + 752 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 108, + 570, + 518, + 752 + ], + "spans": [ + { + "bbox": [ + 108, + 570, + 518, + 752 + ], + "score": 0.964, + "type": "image", + "image_path": "0cda8ee3d259cb24f0c45181ae446820fe9b1146d356cd59c256c8ab595dfe26.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 108, + 570, + 518, + 630.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 108, + 630.6666666666666, + 518, + 691.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 108, + 691.3333333333333, + 518, + 751.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 76, + 80, + 253, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 76, + 80, + 253, + 155 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 76, + 80, + 253, + 155 + ], + "spans": [ + { + "bbox": [ + 76, + 80, + 253, + 155 + ], + "score": 0.853, + "type": "image", + "image_path": "6ab1a20295a0fdba167ef4c254d75ebac90a68354c628ae8597a45a489da49ce.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 76, + 80, + 253, + 95.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 76, + 95.0, + 253, + 110.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 76, + 110.0, + 253, + 125.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 76, + 125.0, + 253, + 140.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 76, + 140.0, + 253, + 155.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 289, + 79, + 545, + 100 + ], + "lines": [ + { + "bbox": [ + 287, + 74, + 547, + 103 + ], + "spans": [ + { + "bbox": [ + 287, + 74, + 547, + 103 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC)", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 140 + ], + "lines": [ + { + "bbox": [ + 288, + 110, + 545, + 128 + ], + "spans": [ + { + "bbox": [ + 288, + 110, + 367, + 128 + ], + "score": 1.0, + "content": "Illustrated on", + "type": "text" + }, + { + "bbox": [ + 367, + 113, + 390, + 126 + ], + "score": 0.46, + "content": "\\mathbf { 3 \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 110, + 545, + 128 + ], + "score": 1.0, + "content": "kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 350, + 126, + 485, + 139 + ], + "spans": [ + { + "bbox": [ + 350, + 126, + 485, + 139 + ], + "score": 1.0, + "content": "with dilation factor of 2", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 288, + 110, + 545, + 139 + ] + }, + { + "type": "title", + "bbox": [ + 71, + 162, + 244, + 174 + ], + "lines": [ + { + "bbox": [ + 69, + 162, + 245, + 176 + ], + "spans": [ + { + "bbox": [ + 69, + 162, + 245, + 176 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 92, + 177, + 380, + 228 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 92, + 177, + 380, + 228 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 92, + 177, + 380, + 228 + ], + "spans": [ + { + "bbox": [ + 92, + 177, + 380, + 228 + ], + "score": 0.942, + "type": "image", + "image_path": "d99e99f62153187884e0a145ec5dd4f0201fb8255b44b2c45aa57158fc39430c.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 92, + 177, + 380, + 194.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 92, + 194.0, + 380, + 211.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 92, + 211.0, + 380, + 228.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 67, + 237, + 307, + 249 + ], + "lines": [ + { + "bbox": [ + 68, + 236, + 307, + 249 + ], + "spans": [ + { + "bbox": [ + 68, + 236, + 307, + 249 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 78, + 257, + 172, + 268 + ], + "lines": [ + { + "bbox": [ + 77, + 255, + 172, + 270 + ], + "spans": [ + { + "bbox": [ + 77, + 255, + 172, + 270 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 77, + 255, + 172, + 270 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 273, + 492, + 394 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 273, + 492, + 394 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 273, + 492, + 394 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 492, + 394 + ], + "score": 0.948, + "type": "image", + "image_path": "749f38dde001c33c4de5e80e1053df61ebc232047d92d82b0267074b9fdafae1.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 107, + 273, + 492, + 313.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 107, + 313.3333333333333, + 492, + 353.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 353.66666666666663, + 492, + 393.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 78, + 405, + 173, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 403, + 173, + 417 + ], + "spans": [ + { + "bbox": [ + 78, + 403, + 173, + 417 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 16.0 + }, + { + "type": "image", + "bbox": [ + 107, + 424, + 518, + 545 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 424, + 518, + 545 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 424, + 518, + 545 + ], + "spans": [ + { + "bbox": [ + 107, + 424, + 518, + 545 + ], + "score": 0.96, + "type": "image", + "image_path": "3d93f43f0bbad03a7c07d066a148fcedbe3c2528cec4208b76899c07614a04d1.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 107, + 424, + 518, + 464.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 107, + 464.3333333333333, + 518, + 504.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 504.66666666666663, + 518, + 545.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 78, + 550, + 172, + 560 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 78, + 548, + 173, + 562 + ], + "spans": [ + { + "bbox": [ + 78, + 548, + 173, + 562 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 108, + 570, + 518, + 752 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 570, + 518, + 752 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 108, + 570, + 518, + 752 + ], + "spans": [ + { + "bbox": [ + 108, + 570, + 518, + 752 + ], + "score": 0.964, + "type": "image", + "image_path": "0cda8ee3d259cb24f0c45181ae446820fe9b1146d356cd59c256c8ab595dfe26.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 108, + 570, + 518, + 630.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 108, + 630.6666666666666, + 518, + 691.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 108, + 691.3333333333333, + 518, + 751.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 79, + 84, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "spans": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 107, + 102, + 492, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 102, + 492, + 307 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 102, + 492, + 307 + ], + "spans": [ + { + "bbox": [ + 107, + 102, + 492, + 307 + ], + "score": 0.95, + "type": "image", + "image_path": "bd8b805d273315488c5e4f9d9c491d79b396d892db1a7990f4d7f5ab6b0f6113.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 107, + 102, + 492, + 170.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 170.33333333333331, + 492, + 238.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 107, + 238.66666666666663, + 492, + 306.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 106, + 355, + 493, + 625 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 335, + 171, + 346 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 79, + 333, + 172, + 348 + ], + "spans": [ + { + "bbox": [ + 79, + 333, + 172, + 348 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 106, + 355, + 493, + 625 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 355, + 493, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 493, + 625 + ], + "score": 0.961, + "type": "image", + "image_path": "0c0ebb9c5742e01fd04882c2956ef19826f0f171ae4291c9a9ae0ed4d2866900.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 355, + 493, + 445.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 445.0, + 493, + 535.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 535.0, + 493, + 625.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 72, + 650, + 246, + 661 + ], + "lines": [ + { + "bbox": [ + 71, + 649, + 246, + 662 + ], + "spans": [ + { + "bbox": [ + 71, + 649, + 246, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "spans": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "page_idx": 26, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 79, + 64, + 254, + 74 + ], + "lines": [ + { + "bbox": [ + 78, + 63, + 254, + 75 + ], + "spans": [ + { + "bbox": [ + 78, + 63, + 254, + 75 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 79, + 84, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "spans": [ + { + "bbox": [ + 78, + 82, + 173, + 96 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 78, + 82, + 173, + 96 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 102, + 492, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 102, + 492, + 307 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 102, + 492, + 307 + ], + "spans": [ + { + "bbox": [ + 107, + 102, + 492, + 307 + ], + "score": 0.95, + "type": "image", + "image_path": "bd8b805d273315488c5e4f9d9c491d79b396d892db1a7990f4d7f5ab6b0f6113.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 107, + 102, + 492, + 170.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 170.33333333333331, + 492, + 238.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 107, + 238.66666666666663, + 492, + 306.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 106, + 355, + 493, + 625 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 335, + 171, + 346 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 79, + 333, + 172, + 348 + ], + "spans": [ + { + "bbox": [ + 79, + 333, + 172, + 348 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 106, + 355, + 493, + 625 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 355, + 493, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 493, + 625 + ], + "score": 0.961, + "type": "image", + "image_path": "0c0ebb9c5742e01fd04882c2956ef19826f0f171ae4291c9a9ae0ed4d2866900.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 355, + 493, + 445.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 445.0, + 493, + 535.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 535.0, + 493, + 625.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 72, + 650, + 246, + 661 + ], + "lines": [ + { + "bbox": [ + 71, + 649, + 246, + 662 + ], + "spans": [ + { + "bbox": [ + 71, + 649, + 246, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 71, + 649, + 246, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "spans": [ + { + "bbox": [ + 70, + 673, + 244, + 685 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 70, + 673, + 244, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 70, + 697, + 259, + 708 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 45, + 75, + 428, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 72, + 429, + 99 + ], + "spans": [ + { + "bbox": [ + 44, + 72, + 429, + 99 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC) with Grouping", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 109, + 512, + 137 + ], + "lines": [ + { + "bbox": [ + 44, + 105, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 44, + 105, + 123, + 126 + ], + "score": 1.0, + "content": "Illustrated on", + "type": "text" + }, + { + "bbox": [ + 124, + 109, + 146, + 122 + ], + "score": 0.27, + "content": "\\pmb { 2 \\times 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 105, + 302, + 126 + ], + "score": 1.0, + "content": "kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 44, + 120, + 513, + 139 + ], + "spans": [ + { + "bbox": [ + 44, + 120, + 513, + 139 + ], + "score": 1.0, + "content": "A grouped padding strategy is applied to balance uneven padding (Wu et al 2019)", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "image", + "bbox": [ + 66, + 171, + 482, + 266 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 66, + 171, + 482, + 266 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 66, + 171, + 482, + 266 + ], + "spans": [ + { + "bbox": [ + 66, + 171, + 482, + 266 + ], + "score": 0.601, + "type": "image", + "image_path": "3cb8ab9203b09b8e6b320ed18ac7056ddcb59f83631f0a18b5a139b841c32530.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 66, + 171, + 482, + 202.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 66, + 202.66666666666666, + 482, + 234.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 66, + 234.33333333333331, + 482, + 266.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 57, + 276, + 273, + 289 + ], + "lines": [ + { + "bbox": [ + 57, + 277, + 272, + 290 + ], + "spans": [ + { + "bbox": [ + 57, + 277, + 272, + 290 + ], + "score": 1.0, + "content": "Number of conv ops each pixel is involved in", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 151, + 302, + 571, + 385 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 151, + 302, + 571, + 385 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 151, + 302, + 571, + 385 + ], + "spans": [ + { + "bbox": [ + 151, + 302, + 571, + 385 + ], + "score": 0.649, + "type": "image", + "image_path": "ac51aae79692b49ae5a912d11b076b5c573aa842106e0cc6e722bb121f69aed2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 151, + 302, + 571, + 329.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 151, + 329.6666666666667, + 571, + 357.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 151, + 357.33333333333337, + 571, + 385.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 59, + 399, + 233, + 411 + ], + "lines": [ + { + "bbox": [ + 57, + 397, + 234, + 412 + ], + "spans": [ + { + "bbox": [ + 57, + 397, + 234, + 412 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 48, + 439, + 562, + 744 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 48, + 439, + 562, + 744 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 48, + 439, + 562, + 744 + ], + "spans": [ + { + "bbox": [ + 48, + 439, + 562, + 744 + ], + "score": 0.531, + "type": "image", + "image_path": "5c6f2c7fbee023779650ad2f065adc250d53bccea9b4cedc28b2d3322622dff4.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 48, + 439, + 562, + 540.6666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 48, + 540.6666666666666, + 562, + 642.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 48, + 642.3333333333333, + 562, + 743.9999999999999 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 12 + } + ], + "page_idx": 27, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 45, + 75, + 428, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 72, + 429, + 99 + ], + "spans": [ + { + "bbox": [ + 44, + 72, + 429, + 99 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC) with Grouping", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 109, + 512, + 137 + ], + "lines": [ + { + "bbox": [ + 44, + 105, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 44, + 105, + 123, + 126 + ], + "score": 1.0, + "content": "Illustrated on", + "type": "text" + }, + { + "bbox": [ + 124, + 109, + 146, + 122 + ], + "score": 0.27, + "content": "\\pmb { 2 \\times 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 105, + 302, + 126 + ], + "score": 1.0, + "content": "kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 44, + 120, + 513, + 139 + ], + "spans": [ + { + "bbox": [ + 44, + 120, + 513, + 139 + ], + "score": 1.0, + "content": "A grouped padding strategy is applied to balance uneven padding (Wu et al 2019)", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 44, + 105, + 513, + 139 + ] + }, + { + "type": "image", + "bbox": [ + 66, + 171, + 482, + 266 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 66, + 171, + 482, + 266 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 66, + 171, + 482, + 266 + ], + "spans": [ + { + "bbox": [ + 66, + 171, + 482, + 266 + ], + "score": 0.601, + "type": "image", + "image_path": "3cb8ab9203b09b8e6b320ed18ac7056ddcb59f83631f0a18b5a139b841c32530.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 66, + 171, + 482, + 202.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 66, + 202.66666666666666, + 482, + 234.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 66, + 234.33333333333331, + 482, + 266.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 57, + 276, + 273, + 289 + ], + "lines": [ + { + "bbox": [ + 57, + 277, + 272, + 290 + ], + "spans": [ + { + "bbox": [ + 57, + 277, + 272, + 290 + ], + "score": 1.0, + "content": "Number of conv ops each pixel is involved in", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 57, + 277, + 272, + 290 + ] + }, + { + "type": "image", + "bbox": [ + 151, + 302, + 571, + 385 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 151, + 302, + 571, + 385 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 151, + 302, + 571, + 385 + ], + "spans": [ + { + "bbox": [ + 151, + 302, + 571, + 385 + ], + "score": 0.649, + "type": "image", + "image_path": "ac51aae79692b49ae5a912d11b076b5c573aa842106e0cc6e722bb121f69aed2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 151, + 302, + 571, + 329.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 151, + 329.6666666666667, + 571, + 357.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 151, + 357.33333333333337, + 571, + 385.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 59, + 399, + 233, + 411 + ], + "lines": [ + { + "bbox": [ + 57, + 397, + 234, + 412 + ], + "spans": [ + { + "bbox": [ + 57, + 397, + 234, + 412 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 57, + 397, + 234, + 412 + ] + }, + { + "type": "image", + "bbox": [ + 48, + 439, + 562, + 744 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 48, + 439, + 562, + 744 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 48, + 439, + 562, + 744 + ], + "spans": [ + { + "bbox": [ + 48, + 439, + 562, + 744 + ], + "score": 0.531, + "type": "image", + "image_path": "5c6f2c7fbee023779650ad2f065adc250d53bccea9b4cedc28b2d3322622dff4.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 48, + 439, + 562, + 540.6666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 48, + 540.6666666666666, + 562, + 642.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 48, + 642.3333333333333, + 562, + 743.9999999999999 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 12 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 73, + 127, + 558, + 308 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 53, + 86, + 293, + 98 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 54, + 85, + 292, + 99 + ], + "spans": [ + { + "bbox": [ + 54, + 85, + 292, + 99 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 73, + 127, + 558, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 73, + 127, + 558, + 308 + ], + "spans": [ + { + "bbox": [ + 73, + 127, + 558, + 308 + ], + "score": 0.839, + "type": "image", + "image_path": "2c5c68905e8fca1d112cde596ce4b53cb4b79975e54fc8cea24edcdef086be98.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 73, + 127, + 558, + 187.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 73, + 187.33333333333334, + 558, + 247.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 73, + 247.66666666666669, + 558, + 308.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "image", + "bbox": [ + 74, + 366, + 543, + 595 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 70, + 338, + 164, + 348 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 70, + 336, + 165, + 350 + ], + "spans": [ + { + "bbox": [ + 70, + 336, + 165, + 350 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 74, + 366, + 543, + 595 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 74, + 366, + 543, + 595 + ], + "spans": [ + { + "bbox": [ + 74, + 366, + 543, + 595 + ], + "score": 0.669, + "type": "image", + "image_path": "ecf7ba29a9b7754a971b14400d054da81dbb097f5b1facc5aeaa574edbabe472.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 74, + 366, + 543, + 442.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 74, + 442.3333333333333, + 543, + 518.6666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 74, + 518.6666666666666, + 543, + 595.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + } + ], + "page_idx": 28, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 58, + 640, + 232, + 650 + ], + "lines": [ + { + "bbox": [ + 57, + 639, + 233, + 651 + ], + "spans": [ + { + "bbox": [ + 57, + 639, + 233, + 651 + ], + "score": 1.0, + "content": "Convolutions involving (c): Rotated version of (b)", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 57, + 665, + 233, + 676 + ], + "lines": [ + { + "bbox": [ + 57, + 664, + 233, + 676 + ], + "spans": [ + { + "bbox": [ + 57, + 664, + 233, + 676 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (a)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 73, + 127, + 558, + 308 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 53, + 86, + 293, + 98 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 54, + 85, + 292, + 99 + ], + "spans": [ + { + "bbox": [ + 54, + 85, + 292, + 99 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 73, + 127, + 558, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 73, + 127, + 558, + 308 + ], + "spans": [ + { + "bbox": [ + 73, + 127, + 558, + 308 + ], + "score": 0.839, + "type": "image", + "image_path": "2c5c68905e8fca1d112cde596ce4b53cb4b79975e54fc8cea24edcdef086be98.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 73, + 127, + 558, + 187.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 73, + 187.33333333333334, + 558, + 247.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 73, + 247.66666666666669, + 558, + 308.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "image", + "bbox": [ + 74, + 366, + 543, + 595 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 70, + 338, + 164, + 348 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 70, + 336, + 165, + 350 + ], + "spans": [ + { + "bbox": [ + 70, + 336, + 165, + 350 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 74, + 366, + 543, + 595 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 74, + 366, + 543, + 595 + ], + "spans": [ + { + "bbox": [ + 74, + 366, + 543, + 595 + ], + "score": 0.669, + "type": "image", + "image_path": "ecf7ba29a9b7754a971b14400d054da81dbb097f5b1facc5aeaa574edbabe472.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 74, + 366, + 543, + 442.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 74, + 442.3333333333333, + 543, + 518.6666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 74, + 518.6666666666666, + 543, + 595.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 47, + 68, + 428, + 102 + ], + "lines": [ + { + "bbox": [ + 44, + 65, + 429, + 91 + ], + "spans": [ + { + "bbox": [ + 44, + 65, + 429, + 91 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC) with Grouping", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 45, + 87, + 302, + 104 + ], + "spans": [ + { + "bbox": [ + 45, + 87, + 302, + 104 + ], + "score": 1.0, + "content": "Illustrated on 4x4 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table", + "bbox": [ + 114, + 114, + 502, + 200 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 114, + 114, + 502, + 200 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 114, + 114, + 498, + 200 + ], + "spans": [ + { + "bbox": [ + 114, + 114, + 498, + 200 + ], + "score": 0.5, + "html": "
Original InputPadded at top-leftPadded at bottom-Padded at top-rightPadded at bottom left corner
edde fddefright corner
abCbabaabaC
deedb eC fbC fa da deC faab
hhgd ghie hd gd ge highde
gggh
", + "type": "table", + "image_path": "fb8c094e8c999c997ba4d08ba2ab639d22f53c597775025b2c2401ae4a44b15a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 114, + 114, + 502, + 142.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 114, + 142.66666666666666, + 502, + 171.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 114, + 171.33333333333331, + 502, + 199.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "table", + "bbox": [ + 114, + 226, + 495, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 206, + 321, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 207, + 320, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 320, + 220 + ], + "score": 1.0, + "content": "Number of conv ops each pixel is involved in", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 114, + 226, + 495, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 226, + 495, + 310 + ], + "spans": [ + { + "bbox": [ + 114, + 226, + 495, + 310 + ], + "score": 0.832, + "html": "
Padded at top-leftPadded at bottom-leftPadded at top-rightPadded at bottom-rightAverage (grouped padding strategy)
252520202015151212121515202020991212121616161616
252520202015151212121515202020991212121616161616
20201616162020161616121216161612121616161616161616
20201616162020161616121216161612121616161616161616
20201616 1620201616121216161612121616161616161616
", + "type": "table", + "image_path": "aaa852930f4bed4db64a9515a48e23a8fe7d82f7409fd5f982abdff6de16aa11.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 114, + 226, + 495, + 254.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 114, + 254.0, + 495, + 282.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 114, + 282.0, + 495, + 310.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "image", + "bbox": [ + 133, + 341, + 466, + 763 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 105, + 324, + 280, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 103, + 322, + 281, + 339 + ], + "spans": [ + { + "bbox": [ + 103, + 322, + 281, + 339 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "image_body", + "bbox": [ + 133, + 341, + 466, + 763 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 341, + 466, + 763 + ], + "spans": [ + { + "bbox": [ + 133, + 341, + 466, + 763 + ], + "score": 0.722, + "type": "image", + "image_path": "3c18f875aebd47ed9b36050ea6c6f9e4ee9a4c0390f1a3d6358113ae4f001675.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 133, + 341, + 466, + 481.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 133, + 481.66666666666663, + 466, + 622.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 133, + 622.3333333333333, + 466, + 762.9999999999999 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + } + ], + "page_idx": 29, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 47, + 68, + 428, + 102 + ], + "lines": [ + { + "bbox": [ + 44, + 65, + 429, + 91 + ], + "spans": [ + { + "bbox": [ + 44, + 65, + 429, + 91 + ], + "score": 1.0, + "content": "Mirror Padding (SYMMETRIC) with Grouping", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 45, + 87, + 302, + 104 + ], + "spans": [ + { + "bbox": [ + 45, + 87, + 302, + 104 + ], + "score": 1.0, + "content": "Illustrated on 4x4 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table", + "bbox": [ + 114, + 114, + 502, + 200 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 114, + 114, + 502, + 200 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 114, + 114, + 498, + 200 + ], + "spans": [ + { + "bbox": [ + 114, + 114, + 498, + 200 + ], + "score": 0.5, + "html": "
Original InputPadded at top-leftPadded at bottom-Padded at top-rightPadded at bottom left corner
edde fddefright corner
abCbabaabaC
deedb eC fbC fa da deC faab
hhgd ghie hd gd ge highde
gggh
", + "type": "table", + "image_path": "fb8c094e8c999c997ba4d08ba2ab639d22f53c597775025b2c2401ae4a44b15a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 114, + 114, + 502, + 142.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 114, + 142.66666666666666, + 502, + 171.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 114, + 171.33333333333331, + 502, + 199.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3 + }, + { + "type": "table", + "bbox": [ + 114, + 226, + 495, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 206, + 321, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 207, + 320, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 320, + 220 + ], + "score": 1.0, + "content": "Number of conv ops each pixel is involved in", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 114, + 226, + 495, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 226, + 495, + 310 + ], + "spans": [ + { + "bbox": [ + 114, + 226, + 495, + 310 + ], + "score": 0.832, + "html": "
Padded at top-leftPadded at bottom-leftPadded at top-rightPadded at bottom-rightAverage (grouped padding strategy)
252520202015151212121515202020991212121616161616
252520202015151212121515202020991212121616161616
20201616162020161616121216161612121616161616161616
20201616162020161616121216161612121616161616161616
20201616 1620201616121216161612121616161616161616
", + "type": "table", + "image_path": "aaa852930f4bed4db64a9515a48e23a8fe7d82f7409fd5f982abdff6de16aa11.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 114, + 226, + 495, + 254.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 114, + 254.0, + 495, + 282.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 114, + 282.0, + 495, + 310.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "image", + "bbox": [ + 133, + 341, + 466, + 763 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 105, + 324, + 280, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 103, + 322, + 281, + 339 + ], + "spans": [ + { + "bbox": [ + 103, + 322, + 281, + 339 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "image_body", + "bbox": [ + 133, + 341, + 466, + 763 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 341, + 466, + 763 + ], + "spans": [ + { + "bbox": [ + 133, + 341, + 466, + 763 + ], + "score": 0.722, + "type": "image", + "image_path": "3c18f875aebd47ed9b36050ea6c6f9e4ee9a4c0390f1a3d6358113ae4f001675.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 133, + 341, + 466, + 481.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 133, + 481.66666666666663, + 466, + 622.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 133, + 622.3333333333333, + 466, + 762.9999999999999 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 329, + 79, + 505, + 99 + ], + "lines": [ + { + "bbox": [ + 327, + 76, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 327, + 76, + 506, + 102 + ], + "score": 1.0, + "content": "Replication Padding", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 75, + 80, + 254, + 159 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 75, + 80, + 254, + 159 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 75, + 80, + 254, + 159 + ], + "spans": [ + { + "bbox": [ + 75, + 80, + 254, + 159 + ], + "score": 0.623, + "type": "image", + "image_path": "305042bd7a20cd1eafbe805182fa540e7e2ce9a422379289d9ee15436767c91c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 75, + 80, + 254, + 95.8 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 75, + 95.8, + 254, + 111.6 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 75, + 111.6, + 254, + 127.39999999999999 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 75, + 127.39999999999999, + 254, + 143.2 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 75, + 143.2, + 254, + 159.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 127 + ], + "lines": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "spans": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "score": 1.0, + "content": "Illustrated on 5x5 kernel and 2-pixel padding", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 78, + 168, + 253, + 180 + ], + "lines": [ + { + "bbox": [ + 76, + 166, + 254, + 182 + ], + "spans": [ + { + "bbox": [ + 76, + 166, + 254, + 182 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 95, + 181, + 506, + 241 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 95, + 181, + 506, + 241 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 95, + 181, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 95, + 181, + 506, + 241 + ], + "score": 0.902, + "type": "image", + "image_path": "4a3779ad3fcb3e7be7745952303c7cebdc2bebab9c0d386a100a95ee8c53f573.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 95, + 181, + 506, + 201.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 95, + 201.0, + 506, + 221.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 95, + 221.0, + 506, + 241.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 73, + 255, + 313, + 267 + ], + "lines": [ + { + "bbox": [ + 74, + 255, + 313, + 268 + ], + "spans": [ + { + "bbox": [ + 74, + 255, + 313, + 268 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 81, + 280, + 176, + 291 + ], + "lines": [ + { + "bbox": [ + 81, + 279, + 176, + 293 + ], + "spans": [ + { + "bbox": [ + 81, + 279, + 176, + 293 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 110, + 290, + 478, + 405 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 290, + 478, + 405 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 110, + 290, + 478, + 405 + ], + "spans": [ + { + "bbox": [ + 110, + 290, + 478, + 405 + ], + "score": 0.607, + "type": "image", + "image_path": "bf7eec628f94a9d43e2e8a7d0f391f88d89ee6a81d3ab11c5aa917df496a5a5f.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 110, + 290, + 478, + 328.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 328.3333333333333, + 478, + 366.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 366.66666666666663, + 478, + 404.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 81, + 415, + 175, + 425 + ], + "lines": [ + { + "bbox": [ + 81, + 413, + 176, + 427 + ], + "spans": [ + { + "bbox": [ + 81, + 413, + 176, + 427 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 110, + 432, + 498, + 543 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 432, + 498, + 543 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 432, + 498, + 543 + ], + "spans": [ + { + "bbox": [ + 110, + 432, + 498, + 543 + ], + "score": 0.654, + "type": "image", + "image_path": "9787549e926a66cc5f95e14c83661906e3e54f96ed8952a4ab6af039ad13d70a.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 110, + 432, + 498, + 469.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 110, + 469.0, + 498, + 506.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 110, + 506.0, + 498, + 543.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 82, + 556, + 175, + 565 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 81, + 554, + 175, + 568 + ], + "spans": [ + { + "bbox": [ + 81, + 554, + 175, + 568 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 19.0 + }, + { + "type": "image", + "bbox": [ + 110, + 574, + 500, + 745 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 574, + 500, + 745 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 110, + 574, + 500, + 745 + ], + "spans": [ + { + "bbox": [ + 110, + 574, + 500, + 745 + ], + "score": 0.861, + "type": "image", + "image_path": "f3e1205ae4b631ab25b18d5297cbc480dd1ff5d33e0f8ade664d2ce85347cfe8.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 110, + 574, + 500, + 631.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 110, + 631.0, + 500, + 688.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 110, + 688.0, + 500, + 745.0 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + } + ], + "page_idx": 30, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 329, + 79, + 505, + 99 + ], + "lines": [ + { + "bbox": [ + 327, + 76, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 327, + 76, + 506, + 102 + ], + "score": 1.0, + "content": "Replication Padding", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 75, + 80, + 254, + 159 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 75, + 80, + 254, + 159 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 75, + 80, + 254, + 159 + ], + "spans": [ + { + "bbox": [ + 75, + 80, + 254, + 159 + ], + "score": 0.623, + "type": "image", + "image_path": "305042bd7a20cd1eafbe805182fa540e7e2ce9a422379289d9ee15436767c91c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 75, + 80, + 254, + 95.8 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 75, + 95.8, + 254, + 111.6 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 75, + 111.6, + 254, + 127.39999999999999 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 75, + 127.39999999999999, + 254, + 143.2 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 75, + 143.2, + 254, + 159.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 289, + 112, + 545, + 127 + ], + "lines": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "spans": [ + { + "bbox": [ + 289, + 110, + 545, + 129 + ], + "score": 1.0, + "content": "Illustrated on 5x5 kernel and 2-pixel padding", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 289, + 110, + 545, + 129 + ] + }, + { + "type": "title", + "bbox": [ + 78, + 168, + 253, + 180 + ], + "lines": [ + { + "bbox": [ + 76, + 166, + 254, + 182 + ], + "spans": [ + { + "bbox": [ + 76, + 166, + 254, + 182 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 95, + 181, + 506, + 241 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 95, + 181, + 506, + 241 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 95, + 181, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 95, + 181, + 506, + 241 + ], + "score": 0.902, + "type": "image", + "image_path": "4a3779ad3fcb3e7be7745952303c7cebdc2bebab9c0d386a100a95ee8c53f573.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 95, + 181, + 506, + 201.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 95, + 201.0, + 506, + 221.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 95, + 221.0, + 506, + 241.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 73, + 255, + 313, + 267 + ], + "lines": [ + { + "bbox": [ + 74, + 255, + 313, + 268 + ], + "spans": [ + { + "bbox": [ + 74, + 255, + 313, + 268 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 81, + 280, + 176, + 291 + ], + "lines": [ + { + "bbox": [ + 81, + 279, + 176, + 293 + ], + "spans": [ + { + "bbox": [ + 81, + 279, + 176, + 293 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 81, + 279, + 176, + 293 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 290, + 478, + 405 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 290, + 478, + 405 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 110, + 290, + 478, + 405 + ], + "spans": [ + { + "bbox": [ + 110, + 290, + 478, + 405 + ], + "score": 0.607, + "type": "image", + "image_path": "bf7eec628f94a9d43e2e8a7d0f391f88d89ee6a81d3ab11c5aa917df496a5a5f.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 110, + 290, + 478, + 328.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 110, + 328.3333333333333, + 478, + 366.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 110, + 366.66666666666663, + 478, + 404.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 81, + 415, + 175, + 425 + ], + "lines": [ + { + "bbox": [ + 81, + 413, + 176, + 427 + ], + "spans": [ + { + "bbox": [ + 81, + 413, + 176, + 427 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 81, + 413, + 176, + 427 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 432, + 498, + 543 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 432, + 498, + 543 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 432, + 498, + 543 + ], + "spans": [ + { + "bbox": [ + 110, + 432, + 498, + 543 + ], + "score": 0.654, + "type": "image", + "image_path": "9787549e926a66cc5f95e14c83661906e3e54f96ed8952a4ab6af039ad13d70a.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 110, + 432, + 498, + 469.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 110, + 469.0, + 498, + 506.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 110, + 506.0, + 498, + 543.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 82, + 556, + 175, + 565 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 81, + 554, + 175, + 568 + ], + "spans": [ + { + "bbox": [ + 81, + 554, + 175, + 568 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 19.0 + }, + { + "type": "image", + "bbox": [ + 110, + 574, + 500, + 745 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 574, + 500, + 745 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 110, + 574, + 500, + 745 + ], + "spans": [ + { + "bbox": [ + 110, + 574, + 500, + 745 + ], + "score": 0.861, + "type": "image", + "image_path": "f3e1205ae4b631ab25b18d5297cbc480dd1ff5d33e0f8ade664d2ce85347cfe8.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 110, + 574, + 500, + 631.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 110, + 631.0, + 500, + 688.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 110, + 688.0, + 500, + 745.0 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 122, + 475, + 314 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 101, + 172, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 99, + 173, + 113 + ], + "spans": [ + { + "bbox": [ + 78, + 99, + 173, + 113 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 110, + 122, + 475, + 314 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 122, + 475, + 314 + ], + "spans": [ + { + "bbox": [ + 110, + 122, + 475, + 314 + ], + "score": 0.952, + "type": "image", + "image_path": "80193f24e2b38ecaf141f9d7a2e076285ced3991ba0e63c46aed2978b3d183e8.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 110, + 122, + 475, + 186.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 186.0, + 475, + 250.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 250.0, + 475, + 314.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "image", + "bbox": [ + 110, + 370, + 474, + 624 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 78, + 348, + 169, + 358 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 77, + 346, + 170, + 360 + ], + "spans": [ + { + "bbox": [ + 77, + 346, + 170, + 360 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 110, + 370, + 474, + 624 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 370, + 474, + 624 + ], + "spans": [ + { + "bbox": [ + 110, + 370, + 474, + 624 + ], + "score": 0.956, + "type": "image", + "image_path": "91bfaefdb8c2ebf071ebe5893fed27de19e40373f9ce1eba526fc7fbf1186d9c.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 110, + 370, + 474, + 454.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 454.6666666666667, + 474, + 539.3333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 539.3333333333334, + 474, + 624.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 71, + 650, + 245, + 660 + ], + "lines": [ + { + "bbox": [ + 70, + 649, + 246, + 662 + ], + "spans": [ + { + "bbox": [ + 70, + 649, + 246, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 69, + 672, + 244, + 686 + ], + "spans": [ + { + "bbox": [ + 69, + 672, + 244, + 686 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "page_idx": 31, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 79, + 65, + 253, + 74 + ], + "lines": [ + { + "bbox": [ + 79, + 64, + 254, + 76 + ], + "spans": [ + { + "bbox": [ + 79, + 64, + 254, + 76 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 122, + 475, + 314 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 79, + 101, + 172, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 78, + 99, + 173, + 113 + ], + "spans": [ + { + "bbox": [ + 78, + 99, + 173, + 113 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 110, + 122, + 475, + 314 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 122, + 475, + 314 + ], + "spans": [ + { + "bbox": [ + 110, + 122, + 475, + 314 + ], + "score": 0.952, + "type": "image", + "image_path": "80193f24e2b38ecaf141f9d7a2e076285ced3991ba0e63c46aed2978b3d183e8.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 110, + 122, + 475, + 186.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 186.0, + 475, + 250.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 250.0, + 475, + 314.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "image", + "bbox": [ + 110, + 370, + 474, + 624 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 78, + 348, + 169, + 358 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 77, + 346, + 170, + 360 + ], + "spans": [ + { + "bbox": [ + 77, + 346, + 170, + 360 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image_body", + "bbox": [ + 110, + 370, + 474, + 624 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 370, + 474, + 624 + ], + "spans": [ + { + "bbox": [ + 110, + 370, + 474, + 624 + ], + "score": 0.956, + "type": "image", + "image_path": "91bfaefdb8c2ebf071ebe5893fed27de19e40373f9ce1eba526fc7fbf1186d9c.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 110, + 370, + 474, + 454.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 454.6666666666667, + 474, + 539.3333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 539.3333333333334, + 474, + 624.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 71, + 650, + 245, + 660 + ], + "lines": [ + { + "bbox": [ + 70, + 649, + 246, + 662 + ], + "spans": [ + { + "bbox": [ + 70, + 649, + 246, + 662 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 70, + 649, + 246, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 71, + 673, + 244, + 684 + ], + "lines": [ + { + "bbox": [ + 69, + 672, + 244, + 686 + ], + "spans": [ + { + "bbox": [ + 69, + 672, + 244, + 686 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 69, + 672, + 244, + 686 + ] + }, + { + "type": "text", + "bbox": [ + 69, + 697, + 259, + 708 + ], + "lines": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "spans": [ + { + "bbox": [ + 70, + 697, + 259, + 708 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 70, + 697, + 259, + 708 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 276, + 96 + ], + "lines": [ + { + "bbox": [ + 43, + 70, + 278, + 100 + ], + "spans": [ + { + "bbox": [ + 43, + 70, + 278, + 100 + ], + "score": 1.0, + "content": "Mirror Padding (REFLECT)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 243, + 178, + 336, + 201 + ], + "lines": [ + { + "bbox": [ + 242, + 177, + 336, + 190 + ], + "spans": [ + { + "bbox": [ + 242, + 177, + 336, + 190 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 243, + 189, + 336, + 201 + ], + "spans": [ + { + "bbox": [ + 243, + 189, + 336, + 201 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "table", + "bbox": [ + 44, + 210, + 113, + 272 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 45, + 183, + 113, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "spans": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "score": 1.0, + "content": "Original Input", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 44, + 210, + 113, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 44, + 210, + 113, + 272 + ], + "spans": [ + { + "bbox": [ + 44, + 210, + 113, + 272 + ], + "score": 0.508, + "html": "
abC
def
gh
• ·
", + "type": "table", + "image_path": "66c037e5000c5f15328d5a862d9327138356030ac9ad265f688caf6674843321.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 44, + 210, + 113, + 241.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 44, + 241.0, + 113, + 272.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 4.25 + }, + { + "type": "table", + "bbox": [ + 138, + 201, + 219, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 138, + 201, + 219, + 272 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 138, + 201, + 219, + 272 + ], + "spans": [ + { + "bbox": [ + 138, + 201, + 219, + 272 + ], + "score": 0.396, + "html": "
edef. • . •
babC= =
edef
hgh
= =
", + "type": "table", + "image_path": "52cce5556c6a9fed6b3cfd88ed27294598ef69edd8569c27a9700b92554ad13c.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 138, + 201, + 219, + 236.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 138, + 236.5, + 219, + 272.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "table", + "bbox": [ + 252, + 209, + 322, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 252, + 209, + 322, + 272 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 252, + 209, + 322, + 272 + ], + "spans": [ + { + "bbox": [ + 252, + 209, + 322, + 272 + ], + "score": 0.481, + "html": "
48666
816121212
612999
612999
612999
", + "type": "table", + "image_path": "fb4c72f78124a1cbb79a80404101d7e9d2f6771809c49d980bf98f947cb55cc8.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 209, + 322, + 240.5 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 252, + 240.5, + 322, + 272.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 47, + 280, + 221, + 291 + ], + "lines": [ + { + "bbox": [ + 45, + 278, + 222, + 294 + ], + "spans": [ + { + "bbox": [ + 45, + 278, + 222, + 294 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 44, + 297, + 525, + 359 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 44, + 297, + 525, + 359 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 44, + 297, + 525, + 359 + ], + "spans": [ + { + "bbox": [ + 44, + 297, + 525, + 359 + ], + "score": 0.681, + "type": "image", + "image_path": "d37f8d090f0d4bb551a64bb18b98a508db3428a6185e94f93a10792f876b3330.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 44, + 297, + 525, + 317.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 44, + 317.6666666666667, + 525, + 338.33333333333337 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 44, + 338.33333333333337, + 525, + 359.00000000000006 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 46, + 393, + 285, + 405 + ], + "lines": [ + { + "bbox": [ + 46, + 393, + 285, + 406 + ], + "spans": [ + { + "bbox": [ + 46, + 393, + 285, + 406 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 45, + 427, + 139, + 438 + ], + "lines": [ + { + "bbox": [ + 45, + 425, + 140, + 439 + ], + "spans": [ + { + "bbox": [ + 45, + 425, + 140, + 439 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 71, + 444, + 388, + 507 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 71, + 444, + 388, + 507 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 71, + 444, + 388, + 507 + ], + "spans": [ + { + "bbox": [ + 71, + 444, + 388, + 507 + ], + "score": 0.603, + "type": "image", + "image_path": "a8960c789ef1b038c5adb0e2e1fc943e5a08c712af011e46e3f21af3294ed5ba.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 71, + 444, + 388, + 465.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 71, + 465.0, + 388, + 486.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 71, + 486.0, + 388, + 507.0 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 43, + 535, + 137, + 546 + ], + "lines": [ + { + "bbox": [ + 43, + 533, + 138, + 547 + ], + "spans": [ + { + "bbox": [ + 43, + 533, + 138, + 547 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "image", + "bbox": [ + 70, + 551, + 552, + 615 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 551, + 552, + 615 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 70, + 551, + 552, + 615 + ], + "spans": [ + { + "bbox": [ + 70, + 551, + 552, + 615 + ], + "score": 0.621, + "type": "image", + "image_path": "4d298c0343a75ab39732efadb3565e384d2c110e83cefde748e91d4bfb8226d9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 70, + 551, + 552, + 572.3333333333334 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 70, + 572.3333333333334, + 552, + 593.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 70, + 593.6666666666667, + 552, + 615.0000000000001 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 44, + 644, + 137, + 653 + ], + "lines": [ + { + "bbox": [ + 43, + 641, + 137, + 655 + ], + "spans": [ + { + "bbox": [ + 43, + 641, + 137, + 655 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "image", + "bbox": [ + 74, + 659, + 564, + 723 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 74, + 659, + 564, + 723 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 74, + 659, + 564, + 723 + ], + "spans": [ + { + "bbox": [ + 74, + 659, + 564, + 723 + ], + "score": 0.657, + "type": "image", + "image_path": "9a44ed4932a742181366a04189d4ecc324022b23ef90542a89c1c7c432db1ba0.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 74, + 659, + 564, + 680.3333333333334 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 74, + 680.3333333333334, + 564, + 701.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 74, + 701.6666666666667, + 564, + 723.0000000000001 + ], + "spans": [], + "index": 27 + } + ] + } + ], + "index": 26 + } + ], + "page_idx": 32, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 75, + 276, + 96 + ], + "lines": [ + { + "bbox": [ + 43, + 70, + 278, + 100 + ], + "spans": [ + { + "bbox": [ + 43, + 70, + 278, + 100 + ], + "score": 1.0, + "content": "Mirror Padding (REFLECT)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 46, + 108, + 302, + 123 + ], + "lines": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "spans": [ + { + "bbox": [ + 45, + 106, + 302, + 126 + ], + "score": 1.0, + "content": "Illustrated on 3x3 kernel and 1-pixel padding", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 45, + 106, + 302, + 126 + ] + }, + { + "type": "text", + "bbox": [ + 243, + 178, + 336, + 201 + ], + "lines": [ + { + "bbox": [ + 242, + 177, + 336, + 190 + ], + "spans": [ + { + "bbox": [ + 242, + 177, + 336, + 190 + ], + "score": 1.0, + "content": "# of conv ops each", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 243, + 189, + 336, + 201 + ], + "spans": [ + { + "bbox": [ + 243, + 189, + 336, + 201 + ], + "score": 1.0, + "content": "pixel is involved in", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 242, + 177, + 336, + 201 + ] + }, + { + "type": "table", + "bbox": [ + 44, + 210, + 113, + 272 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 45, + 183, + 113, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "spans": [ + { + "bbox": [ + 45, + 183, + 114, + 197 + ], + "score": 1.0, + "content": "Original Input", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 44, + 210, + 113, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 44, + 210, + 113, + 272 + ], + "spans": [ + { + "bbox": [ + 44, + 210, + 113, + 272 + ], + "score": 0.508, + "html": "
abC
def
gh
• ·
", + "type": "table", + "image_path": "66c037e5000c5f15328d5a862d9327138356030ac9ad265f688caf6674843321.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 44, + 210, + 113, + 241.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 44, + 241.0, + 113, + 272.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 4.25 + }, + { + "type": "table", + "bbox": [ + 138, + 201, + 219, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 138, + 201, + 219, + 272 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 138, + 201, + 219, + 272 + ], + "spans": [ + { + "bbox": [ + 138, + 201, + 219, + 272 + ], + "score": 0.396, + "html": "
edef. • . •
babC= =
edef
hgh
= =
", + "type": "table", + "image_path": "52cce5556c6a9fed6b3cfd88ed27294598ef69edd8569c27a9700b92554ad13c.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 138, + 201, + 219, + 236.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 138, + 236.5, + 219, + 272.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "table", + "bbox": [ + 252, + 209, + 322, + 272 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 252, + 209, + 322, + 272 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 252, + 209, + 322, + 272 + ], + "spans": [ + { + "bbox": [ + 252, + 209, + 322, + 272 + ], + "score": 0.481, + "html": "
48666
816121212
612999
612999
612999
", + "type": "table", + "image_path": "fb4c72f78124a1cbb79a80404101d7e9d2f6771809c49d980bf98f947cb55cc8.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 252, + 209, + 322, + 240.5 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 252, + 240.5, + 322, + 272.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 47, + 280, + 221, + 291 + ], + "lines": [ + { + "bbox": [ + 45, + 278, + 222, + 294 + ], + "spans": [ + { + "bbox": [ + 45, + 278, + 222, + 294 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 44, + 297, + 525, + 359 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 44, + 297, + 525, + 359 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 44, + 297, + 525, + 359 + ], + "spans": [ + { + "bbox": [ + 44, + 297, + 525, + 359 + ], + "score": 0.681, + "type": "image", + "image_path": "d37f8d090f0d4bb551a64bb18b98a508db3428a6185e94f93a10792f876b3330.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 44, + 297, + 525, + 317.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 44, + 317.6666666666667, + 525, + 338.33333333333337 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 44, + 338.33333333333337, + 525, + 359.00000000000006 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 46, + 393, + 285, + 405 + ], + "lines": [ + { + "bbox": [ + 46, + 393, + 285, + 406 + ], + "spans": [ + { + "bbox": [ + 46, + 393, + 285, + 406 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 45, + 427, + 139, + 438 + ], + "lines": [ + { + "bbox": [ + 45, + 425, + 140, + 439 + ], + "spans": [ + { + "bbox": [ + 45, + 425, + 140, + 439 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 45, + 425, + 140, + 439 + ] + }, + { + "type": "image", + "bbox": [ + 71, + 444, + 388, + 507 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 71, + 444, + 388, + 507 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 71, + 444, + 388, + 507 + ], + "spans": [ + { + "bbox": [ + 71, + 444, + 388, + 507 + ], + "score": 0.603, + "type": "image", + "image_path": "a8960c789ef1b038c5adb0e2e1fc943e5a08c712af011e46e3f21af3294ed5ba.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 71, + 444, + 388, + 465.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 71, + 465.0, + 388, + 486.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 71, + 486.0, + 388, + 507.0 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 43, + 535, + 137, + 546 + ], + "lines": [ + { + "bbox": [ + 43, + 533, + 138, + 547 + ], + "spans": [ + { + "bbox": [ + 43, + 533, + 138, + 547 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 43, + 533, + 138, + 547 + ] + }, + { + "type": "image", + "bbox": [ + 70, + 551, + 552, + 615 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 70, + 551, + 552, + 615 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 70, + 551, + 552, + 615 + ], + "spans": [ + { + "bbox": [ + 70, + 551, + 552, + 615 + ], + "score": 0.621, + "type": "image", + "image_path": "4d298c0343a75ab39732efadb3565e384d2c110e83cefde748e91d4bfb8226d9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 70, + 551, + 552, + 572.3333333333334 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 70, + 572.3333333333334, + 552, + 593.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 70, + 593.6666666666667, + 552, + 615.0000000000001 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 44, + 644, + 137, + 653 + ], + "lines": [ + { + "bbox": [ + 43, + 641, + 137, + 655 + ], + "spans": [ + { + "bbox": [ + 43, + 641, + 137, + 655 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 43, + 641, + 137, + 655 + ] + }, + { + "type": "image", + "bbox": [ + 74, + 659, + 564, + 723 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 74, + 659, + 564, + 723 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 74, + 659, + 564, + 723 + ], + "spans": [ + { + "bbox": [ + 74, + 659, + 564, + 723 + ], + "score": 0.657, + "type": "image", + "image_path": "9a44ed4932a742181366a04189d4ecc324022b23ef90542a89c1c7c432db1ba0.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 74, + 659, + 564, + 680.3333333333334 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 74, + 680.3333333333334, + 564, + 701.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 74, + 701.6666666666667, + 564, + 723.0000000000001 + ], + "spans": [], + "index": 27 + } + ] + } + ], + "index": 26 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 51, + 188, + 144, + 198 + ], + "lines": [ + { + "bbox": [ + 50, + 186, + 145, + 200 + ], + "spans": [ + { + "bbox": [ + 50, + 186, + 145, + 200 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 76, + 215, + 543, + 354 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 76, + 215, + 543, + 354 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 76, + 215, + 543, + 354 + ], + "spans": [ + { + "bbox": [ + 76, + 215, + 543, + 354 + ], + "score": 0.881, + "type": "image", + "image_path": "1ab66efe5cac10d4fc9f6ae0a5a8eff9871f5a5a1e76816a55c37644030707da.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 76, + 215, + 543, + 261.3333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 76, + 261.3333333333333, + 543, + 307.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 76, + 307.66666666666663, + 543, + 353.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 53, + 380, + 145, + 390 + ], + "lines": [ + { + "bbox": [ + 52, + 379, + 145, + 393 + ], + "spans": [ + { + "bbox": [ + 52, + 379, + 145, + 393 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 79, + 400, + 537, + 539 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 79, + 400, + 537, + 539 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 79, + 400, + 537, + 539 + ], + "spans": [ + { + "bbox": [ + 79, + 400, + 537, + 539 + ], + "score": 0.915, + "type": "image", + "image_path": "9c8456965691c401ca270510270cca64da0ffbfd29e905a6210b07a3a22f0387.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 79, + 400, + 537, + 446.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 79, + 446.3333333333333, + 537, + 492.66666666666663 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 79, + 492.66666666666663, + 537, + 539.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 48, + 566, + 222, + 576 + ], + "lines": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "spans": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 46, + 589, + 219, + 600 + ], + "lines": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "spans": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 46, + 613, + 235, + 623 + ], + "lines": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "spans": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "page_idx": 33, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 51, + 145, + 226, + 155 + ], + "lines": [ + { + "bbox": [ + 50, + 144, + 227, + 156 + ], + "spans": [ + { + "bbox": [ + 50, + 144, + 227, + 156 + ], + "score": 1.0, + "content": "Convolutions involving (d): Rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 51, + 188, + 144, + 198 + ], + "lines": [ + { + "bbox": [ + 50, + 186, + 145, + 200 + ], + "spans": [ + { + "bbox": [ + 50, + 186, + 145, + 200 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 50, + 186, + 145, + 200 + ] + }, + { + "type": "image", + "bbox": [ + 76, + 215, + 543, + 354 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 76, + 215, + 543, + 354 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 76, + 215, + 543, + 354 + ], + "spans": [ + { + "bbox": [ + 76, + 215, + 543, + 354 + ], + "score": 0.881, + "type": "image", + "image_path": "1ab66efe5cac10d4fc9f6ae0a5a8eff9871f5a5a1e76816a55c37644030707da.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 76, + 215, + 543, + 261.3333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 76, + 261.3333333333333, + 543, + 307.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 76, + 307.66666666666663, + 543, + 353.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 53, + 380, + 145, + 390 + ], + "lines": [ + { + "bbox": [ + 52, + 379, + 145, + 393 + ], + "spans": [ + { + "bbox": [ + 52, + 379, + 145, + 393 + ], + "score": 1.0, + "content": "Convolutions involving (f)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 52, + 379, + 145, + 393 + ] + }, + { + "type": "image", + "bbox": [ + 79, + 400, + 537, + 539 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 79, + 400, + 537, + 539 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 79, + 400, + 537, + 539 + ], + "spans": [ + { + "bbox": [ + 79, + 400, + 537, + 539 + ], + "score": 0.915, + "type": "image", + "image_path": "9c8456965691c401ca270510270cca64da0ffbfd29e905a6210b07a3a22f0387.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 79, + 400, + 537, + 446.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 79, + 446.3333333333333, + 537, + 492.66666666666663 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 79, + 492.66666666666663, + 537, + 539.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 48, + 566, + 222, + 576 + ], + "lines": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "spans": [ + { + "bbox": [ + 46, + 564, + 222, + 578 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 46, + 564, + 222, + 578 + ] + }, + { + "type": "text", + "bbox": [ + 46, + 589, + 219, + 600 + ], + "lines": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "spans": [ + { + "bbox": [ + 45, + 588, + 219, + 601 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 45, + 588, + 219, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 46, + 613, + 235, + 623 + ], + "lines": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "spans": [ + { + "bbox": [ + 46, + 613, + 235, + 624 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 46, + 613, + 235, + 624 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 76, + 376, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 74, + 379, + 98 + ], + "spans": [ + { + "bbox": [ + 44, + 74, + 379, + 98 + ], + "score": 1.0, + "content": "Partial Convolution Illustrated on a 3x3 kernel", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 82, + 115, + 109, + 127 + ], + "lines": [ + { + "bbox": [ + 81, + 113, + 111, + 129 + ], + "spans": [ + { + "bbox": [ + 81, + 113, + 111, + 129 + ], + "score": 1.0, + "content": "Input", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 147, + 110, + 263, + 134 + ], + "lines": [ + { + "bbox": [ + 148, + 110, + 263, + 123 + ], + "spans": [ + { + "bbox": [ + 148, + 110, + 263, + 123 + ], + "score": 1.0, + "content": "Weighted # of conv ops", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 146, + 122, + 263, + 134 + ], + "spans": [ + { + "bbox": [ + 146, + 122, + 263, + 134 + ], + "score": 1.0, + "content": "each pixel is involved in", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 62, + 142, + 128, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 62, + 142, + 128, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 62, + 142, + 128, + 211 + ], + "spans": [ + { + "bbox": [ + 62, + 142, + 128, + 211 + ], + "score": 0.101, + "html": "
abC
def
gh
", + "type": "table", + "image_path": "29f6d4252d6ee1cd5aab6da7481bcee47d1724469c368dd7e3e702b2ce66f4a8.jpg" + } + ] + } + ], + "index": 5.0, + "virtual_lines": [ + { + "bbox": [ + 62, + 142, + 128, + 176.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 62, + 176.5, + 128, + 211.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 168, + 142, + 248, + 210 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 168, + 142, + 248, + 210 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 168, + 142, + 248, + 210 + ], + "spans": [ + { + "bbox": [ + 168, + 142, + 248, + 210 + ], + "score": 0.142, + "html": "
6.258.757.57.57.5
8.75 12.2510.5 10.510.5
7.510.5999
7.510.5999
7.510.5999
", + "type": "table", + "image_path": "234f6997e14ab6a62e25a1c2ec989ecd397bad2952f8a45ffd4fac96eb40a809.jpg" + } + ] + } + ], + "index": 6.0, + "virtual_lines": [ + { + "bbox": [ + 168, + 142, + 248, + 176.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 168, + 176.0, + 248, + 210.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 50, + 219, + 224, + 231 + ], + "lines": [ + { + "bbox": [ + 48, + 217, + 225, + 232 + ], + "spans": [ + { + "bbox": [ + 48, + 217, + 225, + 232 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 59, + 236, + 509, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 59, + 236, + 509, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 59, + 236, + 509, + 307 + ], + "spans": [ + { + "bbox": [ + 59, + 236, + 509, + 307 + ], + "score": 0.845, + "type": "image", + "image_path": "e44405b931feb473f680ba423538de1d9620da5536ae1e8dac14e2710ecdbfba.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 59, + 236, + 509, + 259.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 59, + 259.6666666666667, + 509, + 283.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 59, + 283.33333333333337, + 509, + 307.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 48, + 326, + 288, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "spans": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + } + ], + "index": 11.0 + }, + { + "type": "text", + "bbox": [ + 61, + 344, + 155, + 355 + ], + "lines": [ + { + "bbox": [ + 61, + 343, + 156, + 357 + ], + "spans": [ + { + "bbox": [ + 61, + 343, + 156, + 357 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "image", + "bbox": [ + 61, + 348, + 383, + 456 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 61, + 348, + 383, + 456 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 61, + 348, + 383, + 456 + ], + "spans": [ + { + "bbox": [ + 61, + 348, + 383, + 456 + ], + "score": 0.604, + "type": "image", + "image_path": "b7d84fb1a287aa41e70703547a09d1f42b59c2cdbaeced40b094c93ee2bd4ec3.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 61, + 348, + 383, + 384.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 61, + 384.0, + 383, + 420.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 61, + 420.0, + 383, + 456.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 60, + 474, + 154, + 484 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 59, + 472, + 154, + 486 + ], + "spans": [ + { + "bbox": [ + 59, + 472, + 154, + 486 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 16.0 + }, + { + "type": "image", + "bbox": [ + 60, + 491, + 536, + 585 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 60, + 491, + 536, + 585 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 60, + 491, + 536, + 585 + ], + "spans": [ + { + "bbox": [ + 60, + 491, + 536, + 585 + ], + "score": 0.626, + "type": "image", + "image_path": "d9c752239c01303d3667d6eb1e568e38601275535c9e3d462c560285f3ba00ad.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 60, + 491, + 536, + 522.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 60, + 522.3333333333334, + 536, + 553.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 60, + 553.6666666666667, + 536, + 585.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 60, + 605, + 154, + 615 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 59, + 603, + 154, + 617 + ], + "spans": [ + { + "bbox": [ + 59, + 603, + 154, + 617 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 67, + 622, + 549, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 67, + 622, + 549, + 704 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 67, + 622, + 549, + 704 + ], + "spans": [ + { + "bbox": [ + 67, + 622, + 549, + 704 + ], + "score": 0.798, + "type": "image", + "image_path": "075465da09bf54db467ff6536761903b1e0c7278cd14b3304dfab0d3c9fd8eaa.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 67, + 622, + 549, + 649.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 67, + 649.3333333333334, + 549, + 676.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 67, + 676.6666666666667, + 549, + 704.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + } + ], + "page_idx": 34, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 46, + 76, + 376, + 96 + ], + "lines": [ + { + "bbox": [ + 44, + 74, + 379, + 98 + ], + "spans": [ + { + "bbox": [ + 44, + 74, + 379, + 98 + ], + "score": 1.0, + "content": "Partial Convolution Illustrated on a 3x3 kernel", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 82, + 115, + 109, + 127 + ], + "lines": [ + { + "bbox": [ + 81, + 113, + 111, + 129 + ], + "spans": [ + { + "bbox": [ + 81, + 113, + 111, + 129 + ], + "score": 1.0, + "content": "Input", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 147, + 110, + 263, + 134 + ], + "lines": [ + { + "bbox": [ + 148, + 110, + 263, + 123 + ], + "spans": [ + { + "bbox": [ + 148, + 110, + 263, + 123 + ], + "score": 1.0, + "content": "Weighted # of conv ops", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 146, + 122, + 263, + 134 + ], + "spans": [ + { + "bbox": [ + 146, + 122, + 263, + 134 + ], + "score": 1.0, + "content": "each pixel is involved in", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 62, + 142, + 128, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 62, + 142, + 128, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 62, + 142, + 128, + 211 + ], + "spans": [ + { + "bbox": [ + 62, + 142, + 128, + 211 + ], + "score": 0.101, + "html": "
abC
def
gh
", + "type": "table", + "image_path": "29f6d4252d6ee1cd5aab6da7481bcee47d1724469c368dd7e3e702b2ce66f4a8.jpg" + } + ] + } + ], + "index": 5.0, + "virtual_lines": [ + { + "bbox": [ + 62, + 142, + 128, + 176.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 62, + 176.5, + 128, + 211.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 168, + 142, + 248, + 210 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 168, + 142, + 248, + 210 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 168, + 142, + 248, + 210 + ], + "spans": [ + { + "bbox": [ + 168, + 142, + 248, + 210 + ], + "score": 0.142, + "html": "
6.258.757.57.57.5
8.75 12.2510.5 10.510.5
7.510.5999
7.510.5999
7.510.5999
", + "type": "table", + "image_path": "234f6997e14ab6a62e25a1c2ec989ecd397bad2952f8a45ffd4fac96eb40a809.jpg" + } + ] + } + ], + "index": 6.0, + "virtual_lines": [ + { + "bbox": [ + 168, + 142, + 248, + 176.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 168, + 176.0, + 248, + 210.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 50, + 219, + 224, + 231 + ], + "lines": [ + { + "bbox": [ + 48, + 217, + 225, + 232 + ], + "spans": [ + { + "bbox": [ + 48, + 217, + 225, + 232 + ], + "score": 1.0, + "content": "Which kernel cells these ops utilize?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 59, + 236, + 509, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 59, + 236, + 509, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 59, + 236, + 509, + 307 + ], + "spans": [ + { + "bbox": [ + 59, + 236, + 509, + 307 + ], + "score": 0.845, + "type": "image", + "image_path": "e44405b931feb473f680ba423538de1d9620da5536ae1e8dac14e2710ecdbfba.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 59, + 236, + 509, + 259.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 59, + 259.6666666666667, + 509, + 283.33333333333337 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 59, + 283.33333333333337, + 509, + 307.00000000000006 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 48, + 326, + 288, + 339 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "spans": [ + { + "bbox": [ + 49, + 326, + 288, + 339 + ], + "score": 1.0, + "content": "Detailed Illustration of how the counts are derived", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + } + ], + "index": 11.0 + }, + { + "type": "text", + "bbox": [ + 61, + 344, + 155, + 355 + ], + "lines": [ + { + "bbox": [ + 61, + 343, + 156, + 357 + ], + "spans": [ + { + "bbox": [ + 61, + 343, + 156, + 357 + ], + "score": 1.0, + "content": "Convolutions involving (a)", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 61, + 343, + 156, + 357 + ] + }, + { + "type": "image", + "bbox": [ + 61, + 348, + 383, + 456 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 61, + 348, + 383, + 456 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 61, + 348, + 383, + 456 + ], + "spans": [ + { + "bbox": [ + 61, + 348, + 383, + 456 + ], + "score": 0.604, + "type": "image", + "image_path": "b7d84fb1a287aa41e70703547a09d1f42b59c2cdbaeced40b094c93ee2bd4ec3.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 61, + 348, + 383, + 384.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 61, + 384.0, + 383, + 420.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 61, + 420.0, + 383, + 456.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 60, + 474, + 154, + 484 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 59, + 472, + 154, + 486 + ], + "spans": [ + { + "bbox": [ + 59, + 472, + 154, + 486 + ], + "score": 1.0, + "content": "Convolutions involving (b)", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 16.0 + }, + { + "type": "image", + "bbox": [ + 60, + 491, + 536, + 585 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 60, + 491, + 536, + 585 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 60, + 491, + 536, + 585 + ], + "spans": [ + { + "bbox": [ + 60, + 491, + 536, + 585 + ], + "score": 0.626, + "type": "image", + "image_path": "d9c752239c01303d3667d6eb1e568e38601275535c9e3d462c560285f3ba00ad.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 60, + 491, + 536, + 522.3333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 60, + 522.3333333333334, + 536, + 553.6666666666667 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 60, + 553.6666666666667, + 536, + 585.0000000000001 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 60, + 605, + 154, + 615 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 59, + 603, + 154, + 617 + ], + "spans": [ + { + "bbox": [ + 59, + 603, + 154, + 617 + ], + "score": 1.0, + "content": "Convolutions involving (c)", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 20.0 + }, + { + "type": "image", + "bbox": [ + 67, + 622, + 549, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 67, + 622, + 549, + 704 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 67, + 622, + 549, + 704 + ], + "spans": [ + { + "bbox": [ + 67, + 622, + 549, + 704 + ], + "score": 0.798, + "type": "image", + "image_path": "075465da09bf54db467ff6536761903b1e0c7278cd14b3304dfab0d3c9fd8eaa.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 67, + 622, + 549, + 649.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 67, + 649.3333333333334, + 549, + 676.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 67, + 676.6666666666667, + 549, + 704.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 23 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 57, + 141, + 539, + 557 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 58, + 124, + 151, + 135 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 123, + 152, + 137 + ], + "spans": [ + { + "bbox": [ + 57, + 123, + 152, + 137 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 57, + 141, + 539, + 557 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 141, + 539, + 557 + ], + "spans": [ + { + "bbox": [ + 57, + 141, + 539, + 557 + ], + "score": 0.47, + "type": "image", + "image_path": "c5b335e14940e3ec3c3e7a0b95ef1060da39442c5d1eba43d8d09eeed1a09da0.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 57, + 141, + 539, + 279.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 57, + 279.66666666666663, + 539, + 418.33333333333326 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 57, + 418.33333333333326, + 539, + 556.9999999999999 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 71, + 610, + 245, + 621 + ], + "lines": [ + { + "bbox": [ + 70, + 610, + 246, + 622 + ], + "spans": [ + { + "bbox": [ + 70, + 610, + 246, + 622 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 70, + 634, + 244, + 645 + ], + "lines": [ + { + "bbox": [ + 70, + 633, + 244, + 646 + ], + "spans": [ + { + "bbox": [ + 70, + 633, + 244, + 646 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 69, + 657, + 259, + 668 + ], + "lines": [ + { + "bbox": [ + 70, + 657, + 259, + 668 + ], + "spans": [ + { + "bbox": [ + 70, + 657, + 259, + 668 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + } + ], + "page_idx": 35, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 59, + 103, + 231, + 113 + ], + "lines": [ + { + "bbox": [ + 57, + 101, + 231, + 114 + ], + "spans": [ + { + "bbox": [ + 57, + 101, + 231, + 114 + ], + "score": 1.0, + "content": "Convolutions involving (d): rotated version of (b)", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 57, + 141, + 539, + 557 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 58, + 124, + 151, + 135 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 123, + 152, + 137 + ], + "spans": [ + { + "bbox": [ + 57, + 123, + 152, + 137 + ], + "score": 1.0, + "content": "Convolutions involving (e)", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "image_body", + "bbox": [ + 57, + 141, + 539, + 557 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 57, + 141, + 539, + 557 + ], + "spans": [ + { + "bbox": [ + 57, + 141, + 539, + 557 + ], + "score": 0.47, + "type": "image", + "image_path": "c5b335e14940e3ec3c3e7a0b95ef1060da39442c5d1eba43d8d09eeed1a09da0.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 57, + 141, + 539, + 279.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 57, + 279.66666666666663, + 539, + 418.33333333333326 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 57, + 418.33333333333326, + 539, + 556.9999999999999 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 71, + 610, + 245, + 621 + ], + "lines": [ + { + "bbox": [ + 70, + 610, + 246, + 622 + ], + "spans": [ + { + "bbox": [ + 70, + 610, + 246, + 622 + ], + "score": 1.0, + "content": "Convolutions involving (g): Rotated version of (c)", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 70, + 610, + 246, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 70, + 634, + 244, + 645 + ], + "lines": [ + { + "bbox": [ + 70, + 633, + 244, + 646 + ], + "spans": [ + { + "bbox": [ + 70, + 633, + 244, + 646 + ], + "score": 1.0, + "content": "Convolutions involving (h): Rotated version of (f)", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 70, + 633, + 244, + 646 + ] + }, + { + "type": "text", + "bbox": [ + 69, + 657, + 259, + 668 + ], + "lines": [ + { + "bbox": [ + 70, + 657, + 259, + 668 + ], + "spans": [ + { + "bbox": [ + 70, + 657, + 259, + 668 + ], + "score": 1.0, + "content": "Convolutions involving (i): Regular uniform treatment", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 70, + 657, + 259, + 668 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/m1CD7tPubNy/m1CD7tPubNy_model.json b/parse/train/m1CD7tPubNy/m1CD7tPubNy_model.json new file mode 100644 index 0000000000000000000000000000000000000000..1b61cb094856e71a26d7c13d30b3b7348b10dfcc --- /dev/null +++ b/parse/train/m1CD7tPubNy/m1CD7tPubNy_model.json @@ -0,0 +1,145954 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 638, + 1303, + 638, + 1303, + 914, + 397, + 914 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1819, + 1404, + 1819, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1036, + 1404, + 1036, + 1404, + 1282, + 298, + 1282 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1296, + 1404, + 1296, + 1404, + 1481, + 298, + 1481 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 369, + 1503, + 1400, + 1503, + 1400, + 1709, + 369, + 1709 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 971, + 320, + 1144, + 320, + 1144, + 383, + 971, + 383 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 615, + 320, + 849, + 320, + 849, + 383, + 615, + 383 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 313, + 320, + 493, + 320, + 493, + 383, + 313, + 383 + ], + "score": 0.933 + }, + { + "category_id": 0, + "poly": [ + 297, + 217, + 1398, + 217, + 1398, + 269, + 297, + 269 + ], + "score": 0.929 + }, + { + "category_id": 0, + "poly": [ + 298, + 1751, + 995, + 1751, + 995, + 1787, + 298, + 1787 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 302, + 967, + 535, + 967, + 535, + 1003, + 302, + 1003 + ], + "score": 0.898 + }, + { + "category_id": 1, + "poly": [ + 1265, + 321, + 1386, + 321, + 1386, + 382, + 1265, + 382 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 773, + 573, + 927, + 573, + 927, + 606, + 773, + 606 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 313, + 429, + 627, + 429, + 627, + 490, + 313, + 490 + ], + "score": 0.843 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 857, + 2088, + 857, + 2112, + 841, + 2112 + ], + "score": 0.681 + }, + { + "category_id": 13, + "poly": [ + 804, + 1881, + 853, + 1881, + 853, + 1912, + 804, + 1912 + ], + "score": 0.88, + "latex": "\\left[ \\left[ 2 6 \\right] \\right]" + }, + { + "category_id": 13, + "poly": [ + 788, + 1971, + 835, + 1971, + 835, + 2001, + 788, + 2001 + ], + "score": 0.86, + "latex": "1 1 ^ { \\mathrm { t h } }" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1880, + 1351, + 1880, + 1351, + 1913, + 1301, + 1913 + ], + "score": 0.77, + "latex": "\\mathbb { \\lVert \\lambda \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1881, + 1059, + 1881, + 1059, + 1912, + 1025, + 1912 + ], + "score": 0.73, + "latex": "\\mathbb { I I }" + }, + { + "category_id": 13, + "poly": [ + 1002, + 1911, + 1037, + 1911, + 1037, + 1943, + 1002, + 1943 + ], + "score": 0.64, + "latex": "\\pmb { \\Vert 4 \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 525, + 1941, + 552, + 1941, + 552, + 1977, + 525, + 1977 + ], + "score": 0.5, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1849, + 1393, + 1849, + 1393, + 1881, + 1345, + 1881 + ], + "score": 0.38, + "latex": "\\pmb { \\mathbb { B } } \\pmb { \\ 6 } \\|" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 218.0, + 1403.0, + 218.0, + 1403.0, + 274.0, + 293.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1748.0, + 998.0, + 1748.0, + 998.0, + 1791.0, + 292.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 966.0, + 540.0, + 966.0, + 540.0, + 1010.0, + 294.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 569.0, + 934.0, + 569.0, + 934.0, + 612.0, + 768.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 840.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 638.0, + 1304.0, + 638.0, + 1304.0, + 673.0, + 395.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 671.0, + 1303.0, + 671.0, + 1303.0, + 701.0, + 395.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 702.0, + 1304.0, + 702.0, + 1304.0, + 732.0, + 395.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 731.0, + 1307.0, + 731.0, + 1307.0, + 765.0, + 392.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 762.0, + 1306.0, + 762.0, + 1306.0, + 794.0, + 393.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 792.0, + 1305.0, + 792.0, + 1305.0, + 824.0, + 395.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 823.0, + 1304.0, + 823.0, + 1304.0, + 855.0, + 393.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 854.0, + 1304.0, + 854.0, + 1304.0, + 887.0, + 395.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 882.0, + 1252.0, + 882.0, + 1252.0, + 920.0, + 393.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1404.0, + 1818.0, + 1404.0, + 1854.0, + 293.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1848.0, + 1344.0, + 1848.0, + 1344.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1848.0, + 1401.0, + 1848.0, + 1401.0, + 1886.0, + 1394.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1878.0, + 803.0, + 1878.0, + 803.0, + 1915.0, + 293.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1878.0, + 1024.0, + 1878.0, + 1024.0, + 1915.0, + 854.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1878.0, + 1300.0, + 1878.0, + 1300.0, + 1915.0, + 1060.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1878.0, + 1408.0, + 1878.0, + 1408.0, + 1915.0, + 1352.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1913.0, + 1001.0, + 1913.0, + 1001.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1913.0, + 1404.0, + 1913.0, + 1404.0, + 1944.0, + 1038.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1942.0, + 524.0, + 1942.0, + 524.0, + 1977.0, + 293.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1977.0, + 553.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 787.0, + 1970.0, + 787.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1970.0, + 1406.0, + 1970.0, + 1406.0, + 2008.0, + 836.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2003.0, + 1382.0, + 2003.0, + 1382.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1035.0, + 1403.0, + 1035.0, + 1403.0, + 1071.0, + 295.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1068.0, + 1406.0, + 1068.0, + 1406.0, + 1101.0, + 294.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1098.0, + 1404.0, + 1098.0, + 1404.0, + 1131.0, + 293.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1129.0, + 1405.0, + 1129.0, + 1405.0, + 1163.0, + 294.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1160.0, + 1404.0, + 1160.0, + 1404.0, + 1190.0, + 296.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1189.0, + 1405.0, + 1189.0, + 1405.0, + 1223.0, + 293.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1218.0, + 1406.0, + 1218.0, + 1406.0, + 1253.0, + 293.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1252.0, + 1398.0, + 1252.0, + 1398.0, + 1283.0, + 294.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1296.0, + 1404.0, + 1296.0, + 1404.0, + 1333.0, + 295.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1326.0, + 1404.0, + 1326.0, + 1404.0, + 1363.0, + 293.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1359.0, + 1405.0, + 1359.0, + 1405.0, + 1391.0, + 296.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1388.0, + 1404.0, + 1388.0, + 1404.0, + 1420.0, + 294.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 1405.0, + 1418.0, + 1405.0, + 1452.0, + 293.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1447.0, + 723.0, + 1447.0, + 723.0, + 1485.0, + 295.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1501.0, + 1385.0, + 1501.0, + 1385.0, + 1543.0, + 364.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1541.0, + 1210.0, + 1541.0, + 1210.0, + 1575.0, + 367.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1573.0, + 1353.0, + 1573.0, + 1353.0, + 1613.0, + 365.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1611.0, + 1315.0, + 1611.0, + 1315.0, + 1644.0, + 381.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1644.0, + 1405.0, + 1644.0, + 1405.0, + 1684.0, + 372.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1675.0, + 823.0, + 1675.0, + 823.0, + 1712.0, + 393.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 318.0, + 1148.0, + 318.0, + 1148.0, + 356.0, + 972.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 350.0, + 1125.0, + 350.0, + 1125.0, + 383.0, + 970.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 319.0, + 851.0, + 319.0, + 851.0, + 355.0, + 614.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 353.0, + 767.0, + 353.0, + 767.0, + 382.0, + 615.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 321.0, + 495.0, + 321.0, + 495.0, + 350.0, + 313.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 353.0, + 465.0, + 353.0, + 465.0, + 382.0, + 312.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 318.0, + 1391.0, + 318.0, + 1391.0, + 355.0, + 1262.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 349.0, + 1338.0, + 349.0, + 1338.0, + 384.0, + 1262.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 427.0, + 628.0, + 427.0, + 628.0, + 462.0, + 312.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 461.0, + 466.0, + 461.0, + 466.0, + 490.0, + 312.0, + 490.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1207, + 1406, + 1207, + 1406, + 1454, + 297, + 1454 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 945, + 1404, + 945, + 1404, + 1193, + 297, + 1193 + ], + "score": 0.981 + }, + { + "category_id": 3, + "poly": [ + 331, + 222, + 1361, + 222, + 1361, + 755, + 331, + 755 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 330, + 1498, + 1379, + 1498, + 1379, + 1957, + 330, + 1957 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 299, + 779, + 1400, + 779, + 1400, + 875, + 299, + 875 + ], + "score": 0.955 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.886 + }, + { + "category_id": 4, + "poly": [ + 298, + 1993, + 1397, + 1993, + 1397, + 2027, + 298, + 2027 + ], + "score": 0.805 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.753 + }, + { + "category_id": 4, + "poly": [ + 301, + 1993, + 1396, + 1993, + 1396, + 2027, + 301, + 2027 + ], + "score": 0.273 + }, + { + "category_id": 13, + "poly": [ + 1267, + 1993, + 1386, + 1993, + 1386, + 2024, + 1267, + 2024 + ], + "score": 0.86, + "latex": "\\mathrm { H } { \\times } \\mathrm { W } { \\times } \\mathrm { C } )" + }, + { + "category_id": 13, + "poly": [ + 726, + 1236, + 755, + 1236, + 755, + 1273, + 726, + 1273 + ], + "score": 0.59, + "latex": "2" + }, + { + "category_id": 13, + "poly": [ + 801, + 1099, + 828, + 1099, + 828, + 1135, + 801, + 1135 + ], + "score": 0.45, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 591, + 946, + 618, + 946, + 618, + 982, + 591, + 982 + ], + "score": 0.3, + "latex": "\\bigstar" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 228.0, + 458.0, + 228.0, + 458.0, + 248.0, + 414.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 222.0, + 1007.0, + 222.0, + 1007.0, + 249.0, + 691.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 226.0, + 1333.0, + 226.0, + 1333.0, + 248.0, + 1189.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 252.0, + 1087.0, + 252.0, + 1087.0, + 369.0, + 939.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 263.0, + 1346.0, + 263.0, + 1346.0, + 348.0, + 1176.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 267.0, + 970.0, + 267.0, + 970.0, + 305.0, + 904.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 277.0, + 1017.0, + 277.0, + 1017.0, + 292.0, + 973.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 292.0, + 791.0, + 292.0, + 791.0, + 322.0, + 744.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 305.0, + 701.0, + 305.0, + 701.0, + 324.0, + 665.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 364.0, + 406.0, + 364.0, + 406.0, + 382.0, + 338.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 369.0, + 940.0, + 369.0, + 940.0, + 380.0, + 899.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 359.0, + 1341.0, + 359.0, + 1341.0, + 387.0, + 1166.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 383.0, + 719.0, + 383.0, + 719.0, + 393.0, + 707.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 383.0, + 1272.0, + 383.0, + 1272.0, + 393.0, + 1260.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 425.0, + 717.0, + 425.0, + 717.0, + 434.0, + 707.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 425.0, + 1272.0, + 425.0, + 1272.0, + 434.0, + 1260.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 455.0, + 1086.0, + 455.0, + 1086.0, + 544.0, + 888.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 454.0, + 1366.0, + 454.0, + 1366.0, + 546.0, + 1158.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 564.0, + 433.0, + 564.0, + 433.0, + 586.0, + 343.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 591.0, + 717.0, + 591.0, + 717.0, + 601.0, + 706.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 591.0, + 992.0, + 591.0, + 992.0, + 601.0, + 980.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 592.0, + 1267.0, + 592.0, + 1267.0, + 601.0, + 1253.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 617.0, + 806.0, + 617.0, + 806.0, + 639.0, + 615.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 617.0, + 1082.0, + 617.0, + 1082.0, + 639.0, + 892.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 617.0, + 1359.0, + 617.0, + 1359.0, + 639.0, + 1166.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 680.0, + 601.0, + 680.0, + 601.0, + 703.0, + 499.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 694.0, + 605.0, + 694.0, + 605.0, + 720.0, + 496.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 262.5, + 774.0, + 262.5, + 774.0, + 300.0, + 651.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 363.0, + 679.0, + 363.0, + 679.0, + 385.0, + 619.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1507.0, + 516.0, + 1507.0, + 516.0, + 1526.0, + 357.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1504.0, + 777.0, + 1504.0, + 777.0, + 1527.0, + 615.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1507.0, + 1040.0, + 1507.0, + 1040.0, + 1526.0, + 872.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1507.0, + 1301.0, + 1507.0, + 1301.0, + 1526.0, + 1133.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1531.0, + 851.0, + 1531.0, + 851.0, + 1552.0, + 820.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1540.0, + 1112.0, + 1540.0, + 1112.0, + 1562.0, + 1080.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1535.0, + 1372.0, + 1535.0, + 1372.0, + 1557.0, + 1339.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1549.0, + 590.0, + 1549.0, + 590.0, + 1571.0, + 559.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1576.0, + 591.0, + 1576.0, + 591.0, + 1602.0, + 558.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1587.0, + 851.0, + 1587.0, + 851.0, + 1609.0, + 817.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1574.0, + 1112.0, + 1574.0, + 1112.0, + 1596.0, + 1078.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1577.0, + 1373.0, + 1577.0, + 1373.0, + 1603.0, + 1338.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1605.0, + 591.0, + 1605.0, + 591.0, + 1631.0, + 558.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1608.0, + 1109.0, + 1608.0, + 1109.0, + 1630.0, + 1080.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1623.0, + 1372.0, + 1623.0, + 1372.0, + 1645.0, + 1339.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1635.0, + 591.0, + 1635.0, + 591.0, + 1656.0, + 560.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1655.0, + 516.0, + 1655.0, + 516.0, + 1677.0, + 354.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1656.0, + 777.0, + 1656.0, + 777.0, + 1679.0, + 615.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1656.0, + 1034.0, + 1656.0, + 1034.0, + 1679.0, + 881.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 1656.0, + 1294.0, + 1656.0, + 1294.0, + 1679.0, + 1141.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1695.0, + 598.0, + 1695.0, + 598.0, + 1719.0, + 556.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1679.0, + 850.0, + 1679.0, + 850.0, + 1696.0, + 821.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1684.0, + 1113.0, + 1684.0, + 1113.0, + 1709.0, + 1068.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1687.0, + 1372.0, + 1687.0, + 1372.0, + 1709.0, + 1332.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1725.0, + 598.0, + 1725.0, + 598.0, + 1750.0, + 556.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 1721.0, + 851.0, + 1721.0, + 851.0, + 1743.0, + 816.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1729.0, + 1111.0, + 1729.0, + 1111.0, + 1751.0, + 1080.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1725.0, + 1373.0, + 1725.0, + 1373.0, + 1751.0, + 1337.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1755.0, + 598.0, + 1755.0, + 598.0, + 1780.0, + 558.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1765.0, + 851.0, + 1765.0, + 851.0, + 1787.0, + 820.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1771.0, + 1111.0, + 1771.0, + 1111.0, + 1793.0, + 1080.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1768.0, + 1372.0, + 1768.0, + 1372.0, + 1789.0, + 1339.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1787.0, + 598.0, + 1787.0, + 598.0, + 1808.0, + 560.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1805.0, + 514.0, + 1805.0, + 514.0, + 1831.0, + 359.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1805.0, + 772.0, + 1805.0, + 772.0, + 1831.0, + 619.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1808.0, + 1039.0, + 1808.0, + 1039.0, + 1830.0, + 876.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1808.0, + 1299.0, + 1808.0, + 1299.0, + 1830.0, + 1137.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1854.0, + 590.0, + 1854.0, + 590.0, + 1875.0, + 559.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1845.0, + 1113.0, + 1845.0, + 1113.0, + 1870.0, + 1070.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1867.0, + 851.0, + 1867.0, + 851.0, + 1889.0, + 819.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1878.0, + 1112.0, + 1878.0, + 1112.0, + 1900.0, + 1078.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1867.0, + 1373.0, + 1867.0, + 1373.0, + 1892.0, + 1334.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1896.0, + 591.0, + 1896.0, + 591.0, + 1917.0, + 559.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1913.0, + 851.0, + 1913.0, + 851.0, + 1934.0, + 819.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1910.0, + 1112.0, + 1910.0, + 1112.0, + 1932.0, + 1077.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1925.0, + 1370.0, + 1925.0, + 1370.0, + 1948.0, + 1339.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1939.0, + 589.0, + 1939.0, + 589.0, + 1957.0, + 561.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 779.0, + 1405.0, + 779.0, + 1405.0, + 817.0, + 294.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 810.0, + 1404.0, + 810.0, + 1404.0, + 847.0, + 294.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 842.0, + 1404.0, + 842.0, + 1404.0, + 876.0, + 295.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1989.0, + 1266.0, + 1989.0, + 1266.0, + 2031.0, + 298.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1387.0, + 1989.0, + 1400.0, + 1989.0, + 1400.0, + 2031.0, + 1387.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1989.0, + 1266.0, + 1989.0, + 1266.0, + 2031.0, + 298.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1387.0, + 1989.0, + 1400.0, + 1989.0, + 1400.0, + 2031.0, + 1387.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1206.0, + 1404.0, + 1206.0, + 1404.0, + 1241.0, + 294.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1237.0, + 725.0, + 1237.0, + 725.0, + 1273.0, + 292.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1237.0, + 1406.0, + 1237.0, + 1406.0, + 1273.0, + 756.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1269.0, + 1404.0, + 1269.0, + 1404.0, + 1303.0, + 295.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1299.0, + 1404.0, + 1299.0, + 1404.0, + 1333.0, + 294.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1328.0, + 1404.0, + 1328.0, + 1404.0, + 1363.0, + 292.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1357.0, + 1404.0, + 1357.0, + 1404.0, + 1396.0, + 294.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1391.0, + 1406.0, + 1391.0, + 1406.0, + 1424.0, + 295.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1422.0, + 1008.0, + 1422.0, + 1008.0, + 1456.0, + 295.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 946.0, + 590.0, + 946.0, + 590.0, + 982.0, + 294.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 946.0, + 1405.0, + 946.0, + 1405.0, + 982.0, + 619.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 975.0, + 1406.0, + 975.0, + 1406.0, + 1015.0, + 294.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1009.0, + 1406.0, + 1009.0, + 1406.0, + 1042.0, + 294.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1041.0, + 1405.0, + 1041.0, + 1405.0, + 1071.0, + 296.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1068.0, + 1404.0, + 1068.0, + 1404.0, + 1105.0, + 292.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1097.0, + 800.0, + 1097.0, + 800.0, + 1137.0, + 291.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1097.0, + 1406.0, + 1097.0, + 1406.0, + 1137.0, + 829.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1131.0, + 1405.0, + 1131.0, + 1405.0, + 1164.0, + 294.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1162.0, + 1317.0, + 1162.0, + 1317.0, + 1196.0, + 295.0, + 1196.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 398, + 1405, + 398, + 1405, + 617, + 297, + 617 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1403, + 229, + 1403, + 385, + 297, + 385 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 976, + 1403, + 976, + 1403, + 1069, + 298, + 1069 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 296, + 1083, + 1404, + 1083, + 1404, + 1237, + 296, + 1237 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 306, + 1273, + 1395, + 1273, + 1395, + 1843, + 306, + 1843 + ], + "score": 0.965 + }, + { + "category_id": 3, + "poly": [ + 316, + 652, + 1389, + 652, + 1389, + 750, + 316, + 750 + ], + "score": 0.945 + }, + { + "category_id": 4, + "poly": [ + 295, + 1870, + 1403, + 1870, + 1403, + 2025, + 295, + 2025 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 301, + 905, + 810, + 905, + 810, + 940, + 301, + 940 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 813, + 76, + 813, + 104, + 300, + 104 + ], + "score": 0.897 + }, + { + "category_id": 4, + "poly": [ + 302, + 788, + 1386, + 788, + 1386, + 821, + 302, + 821 + ], + "score": 0.774 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.654 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.363 + }, + { + "category_id": 1, + "poly": [ + 302, + 788, + 1386, + 788, + 1386, + 821, + 302, + 821 + ], + "score": 0.189 + }, + { + "category_id": 0, + "poly": [ + 960, + 1276, + 1395, + 1276, + 1395, + 1300, + 960, + 1300 + ], + "score": 0.134 + }, + { + "category_id": 13, + "poly": [ + 297, + 1933, + 393, + 1933, + 393, + 1963, + 297, + 1963 + ], + "score": 0.89, + "latex": "4 5 \\times 8 0" + }, + { + "category_id": 13, + "poly": [ + 1005, + 790, + 1070, + 790, + 1070, + 818, + 1005, + 818 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1334, + 1121, + 1334, + 1121, + 1354, + 1055, + 1354 + ], + "score": 0.76, + "latex": "4 5 x 8 0" + }, + { + "category_id": 13, + "poly": [ + 375, + 1082, + 404, + 1082, + 404, + 1120, + 375, + 1120 + ], + "score": 0.57, + "latex": "\\sharp" + }, + { + "category_id": 13, + "poly": [ + 955, + 461, + 1001, + 461, + 1001, + 489, + 955, + 489 + ], + "score": 0.35, + "latex": "3 { \\tt X } 3" + }, + { + "category_id": 13, + "poly": [ + 997, + 293, + 1043, + 293, + 1043, + 321, + 997, + 321 + ], + "score": 0.29, + "latex": "3 { \\tt X } 3" + }, + { + "category_id": 13, + "poly": [ + 761, + 1314, + 828, + 1314, + 828, + 1336, + 761, + 1336 + ], + "score": 0.29, + "latex": "( 4 5 \\times 8 0 )" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1272.0, + 498.0, + 1272.0, + 498.0, + 1303.0, + 298.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1272.0, + 847.0, + 1272.0, + 847.0, + 1303.0, + 638.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1272.0, + 1398.0, + 1272.0, + 1398.0, + 1304.0, + 958.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1310.0, + 481.0, + 1310.0, + 481.0, + 1338.0, + 314.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1308.0, + 760.0, + 1308.0, + 760.0, + 1339.0, + 651.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1308.0, + 833.0, + 1308.0, + 833.0, + 1339.0, + 829.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1310.0, + 1387.0, + 1310.0, + 1387.0, + 1336.0, + 956.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1331.0, + 1054.0, + 1331.0, + 1054.0, + 1358.0, + 952.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1331.0, + 1392.0, + 1331.0, + 1392.0, + 1358.0, + 1122.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1354.0, + 919.0, + 1354.0, + 919.0, + 1377.0, + 891.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1386.0, + 576.0, + 1386.0, + 576.0, + 1427.0, + 532.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1379.0, + 919.0, + 1379.0, + 919.0, + 1429.0, + 890.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1437.0, + 911.0, + 1437.0, + 911.0, + 1453.0, + 896.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1478.0, + 828.0, + 1478.0, + 828.0, + 1510.0, + 656.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1506.0, + 916.0, + 1506.0, + 916.0, + 1528.0, + 886.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1554.0, + 580.0, + 1554.0, + 580.0, + 1607.0, + 527.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1553.0, + 922.0, + 1553.0, + 922.0, + 1576.0, + 891.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1602.0, + 922.0, + 1602.0, + 922.0, + 1625.0, + 889.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 1652.0, + 1179.0, + 1652.0, + 1179.0, + 1685.0, + 867.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1817.0, + 867.0, + 1817.0, + 867.0, + 1844.0, + 525.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1817.0, + 1081.0, + 1817.0, + 1081.0, + 1845.0, + 964.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1817.0, + 1388.0, + 1817.0, + 1388.0, + 1844.0, + 1179.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 1351.0, + 525.0, + 1351.0, + 525.0, + 1387.0, + 474.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1660.5, + 514.0, + 1660.5, + 514.0, + 1702.5, + 501.0, + 1702.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1749.0, + 537.0, + 1749.0, + 537.0, + 1786.0, + 467.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 654.0, + 378.0, + 654.0, + 378.0, + 680.0, + 321.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 655.0, + 468.0, + 655.0, + 468.0, + 680.0, + 410.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 654.0, + 557.0, + 654.0, + 557.0, + 680.0, + 499.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 655.0, + 650.0, + 655.0, + 650.0, + 680.0, + 582.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 655.0, + 739.0, + 655.0, + 739.0, + 680.0, + 672.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 655.0, + 829.0, + 655.0, + 829.0, + 680.0, + 762.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 655.0, + 918.0, + 655.0, + 918.0, + 680.0, + 851.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 655.0, + 1005.0, + 655.0, + 1005.0, + 680.0, + 940.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 655.0, + 1096.0, + 655.0, + 1096.0, + 680.0, + 1030.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 655.0, + 1184.0, + 655.0, + 1184.0, + 680.0, + 1118.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 655.0, + 1279.0, + 655.0, + 1279.0, + 680.0, + 1204.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 655.0, + 1368.0, + 655.0, + 1368.0, + 680.0, + 1294.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 689.0, + 405.0, + 689.0, + 405.0, + 707.0, + 383.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 684.0, + 486.0, + 684.0, + 486.0, + 695.0, + 474.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 685.0, + 578.0, + 685.0, + 578.0, + 700.0, + 559.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 682.0, + 667.0, + 682.0, + 667.0, + 697.0, + 650.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 682.0, + 756.0, + 682.0, + 756.0, + 696.0, + 739.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 690.0, + 851.0, + 690.0, + 851.0, + 708.0, + 828.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 684.0, + 940.0, + 684.0, + 940.0, + 701.0, + 916.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 687.0, + 1030.0, + 687.0, + 1030.0, + 706.0, + 1006.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 684.0, + 1112.0, + 684.0, + 1112.0, + 697.0, + 1096.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 690.0, + 1206.0, + 690.0, + 1206.0, + 705.0, + 1186.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 687.0, + 1301.0, + 687.0, + 1301.0, + 706.0, + 1273.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 682.0, + 1385.0, + 682.0, + 1385.0, + 700.0, + 1361.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 703.0, + 405.0, + 703.0, + 405.0, + 723.0, + 383.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 708.0, + 487.0, + 708.0, + 487.0, + 718.0, + 474.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 703.0, + 580.0, + 703.0, + 580.0, + 722.0, + 559.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 703.0, + 669.0, + 703.0, + 669.0, + 722.0, + 648.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 706.0, + 757.0, + 706.0, + 757.0, + 721.0, + 740.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 703.0, + 851.0, + 703.0, + 851.0, + 723.0, + 828.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 703.0, + 940.0, + 703.0, + 940.0, + 722.0, + 916.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 703.0, + 1030.0, + 703.0, + 1030.0, + 723.0, + 1006.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 705.0, + 1113.0, + 705.0, + 1113.0, + 721.0, + 1097.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 703.0, + 1208.0, + 703.0, + 1208.0, + 723.0, + 1185.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 703.0, + 1301.0, + 703.0, + 1301.0, + 722.0, + 1273.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 703.0, + 1386.0, + 703.0, + 1386.0, + 722.0, + 1362.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 722.0, + 407.0, + 722.0, + 407.0, + 737.0, + 387.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 733.0, + 492.0, + 733.0, + 492.0, + 743.0, + 479.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 727.0, + 582.0, + 727.0, + 582.0, + 741.0, + 565.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 733.0, + 670.0, + 733.0, + 670.0, + 742.0, + 655.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 733.0, + 760.0, + 733.0, + 760.0, + 742.0, + 744.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 718.0, + 855.0, + 718.0, + 855.0, + 737.0, + 832.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 726.0, + 945.0, + 726.0, + 945.0, + 743.0, + 921.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 721.0, + 1033.0, + 721.0, + 1033.0, + 739.0, + 1009.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 732.0, + 1115.0, + 732.0, + 1115.0, + 742.0, + 1103.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 722.0, + 1209.0, + 722.0, + 1209.0, + 737.0, + 1191.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 722.0, + 1303.0, + 722.0, + 1303.0, + 739.0, + 1276.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 727.0, + 1391.0, + 727.0, + 1391.0, + 744.0, + 1366.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1872.0, + 1406.0, + 1872.0, + 1406.0, + 1905.0, + 295.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1904.0, + 1405.0, + 1904.0, + 1405.0, + 1937.0, + 295.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1932.0, + 1404.0, + 1932.0, + 1404.0, + 1967.0, + 394.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1962.0, + 1405.0, + 1962.0, + 1405.0, + 1999.0, + 294.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1994.0, + 1403.0, + 1994.0, + 1403.0, + 2028.0, + 295.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 900.0, + 815.0, + 900.0, + 815.0, + 946.0, + 291.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 72.0, + 816.0, + 72.0, + 816.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 784.0, + 1004.0, + 784.0, + 1004.0, + 825.0, + 306.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 784.0, + 1390.0, + 784.0, + 1390.0, + 825.0, + 1071.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1272.0, + 1397.0, + 1272.0, + 1397.0, + 1304.0, + 958.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 399.0, + 1405.0, + 399.0, + 1405.0, + 434.0, + 294.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 429.0, + 1405.0, + 429.0, + 1405.0, + 464.0, + 295.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 459.0, + 954.0, + 459.0, + 954.0, + 498.0, + 294.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 459.0, + 1406.0, + 459.0, + 1406.0, + 498.0, + 1002.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 490.0, + 1405.0, + 490.0, + 1405.0, + 525.0, + 295.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 519.0, + 1403.0, + 519.0, + 1403.0, + 559.0, + 292.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 550.0, + 1405.0, + 550.0, + 1405.0, + 587.0, + 295.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 584.0, + 937.0, + 584.0, + 937.0, + 618.0, + 296.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 231.0, + 1401.0, + 231.0, + 1401.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 295.0, + 295.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 292.0, + 996.0, + 292.0, + 996.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 292.0, + 1404.0, + 292.0, + 1404.0, + 326.0, + 1044.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 318.0, + 1405.0, + 318.0, + 1405.0, + 359.0, + 294.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 352.0, + 1375.0, + 352.0, + 1375.0, + 390.0, + 295.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 977.0, + 1404.0, + 977.0, + 1404.0, + 1010.0, + 297.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1007.0, + 1403.0, + 1007.0, + 1403.0, + 1041.0, + 296.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1038.0, + 1255.0, + 1038.0, + 1255.0, + 1072.0, + 296.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1085.0, + 374.0, + 1085.0, + 374.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1085.0, + 1404.0, + 1085.0, + 1404.0, + 1119.0, + 405.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1114.0, + 1401.0, + 1114.0, + 1401.0, + 1147.0, + 295.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1143.0, + 1405.0, + 1143.0, + 1405.0, + 1183.0, + 293.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1173.0, + 1405.0, + 1173.0, + 1405.0, + 1212.0, + 293.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1205.0, + 1405.0, + 1205.0, + 1405.0, + 1242.0, + 294.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 784.0, + 1004.0, + 784.0, + 1004.0, + 825.0, + 306.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 784.0, + 1390.0, + 784.0, + 1390.0, + 825.0, + 1071.0, + 825.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1079, + 1404, + 1079, + 1404, + 1416, + 298, + 1416 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 299, + 696, + 1404, + 696, + 1404, + 1064, + 299, + 1064 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1577, + 1403, + 1577, + 1403, + 1761, + 298, + 1761 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1773, + 1403, + 1773, + 1403, + 1987, + 298, + 1987 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 311, + 226, + 1387, + 226, + 1387, + 488, + 311, + 488 + ], + "score": 0.963 + }, + { + "category_id": 4, + "poly": [ + 298, + 514, + 1404, + 514, + 1404, + 637, + 298, + 637 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 298, + 2002, + 1189, + 2002, + 1189, + 2034, + 298, + 2034 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 812, + 76, + 812, + 104, + 299, + 104 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.756 + }, + { + "category_id": 0, + "poly": [ + 299, + 1459, + 1065, + 1459, + 1065, + 1495, + 299, + 1495 + ], + "score": 0.644 + }, + { + "category_id": 0, + "poly": [ + 295, + 1529, + 1346, + 1529, + 1346, + 1562, + 295, + 1562 + ], + "score": 0.592 + }, + { + "category_id": 1, + "poly": [ + 295, + 1529, + 1346, + 1529, + 1346, + 1562, + 295, + 1562 + ], + "score": 0.365 + }, + { + "category_id": 1, + "poly": [ + 299, + 1459, + 1065, + 1459, + 1065, + 1495, + 299, + 1495 + ], + "score": 0.293 + }, + { + "category_id": 13, + "poly": [ + 996, + 1141, + 1038, + 1141, + 1038, + 1170, + 996, + 1170 + ], + "score": 0.87, + "latex": "7 \\%" + }, + { + "category_id": 13, + "poly": [ + 1023, + 1774, + 1078, + 1774, + 1078, + 1803, + 1023, + 1803 + ], + "score": 0.86, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 957, + 1728, + 1005, + 1728, + 1005, + 1761, + 957, + 1761 + ], + "score": 0.55, + "latex": "\\mathbf { \\bar { \\rho } }" + }, + { + "category_id": 13, + "poly": [ + 999, + 1231, + 1027, + 1231, + 1027, + 1267, + 999, + 1267 + ], + "score": 0.38, + "latex": "\\mathbf { \\overline { { \\mathbb { D } } } }" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 226.0, + 805.0, + 226.0, + 805.0, + 266.0, + 632.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 222.0, + 1009.0, + 222.0, + 1009.0, + 258.0, + 908.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 230.0, + 1346.0, + 230.0, + 1346.0, + 261.0, + 1034.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 250.0, + 990.0, + 250.0, + 990.0, + 279.0, + 925.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 289.0, + 448.0, + 289.0, + 448.0, + 314.0, + 359.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 299.0, + 990.0, + 299.0, + 990.0, + 323.0, + 957.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 341.0, + 990.0, + 341.0, + 990.0, + 363.0, + 957.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 354.0, + 450.0, + 354.0, + 450.0, + 383.0, + 361.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 382.0, + 989.0, + 382.0, + 989.0, + 405.0, + 957.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 423.0, + 454.0, + 423.0, + 454.0, + 448.0, + 351.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 424.0, + 987.0, + 424.0, + 987.0, + 446.0, + 957.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 455.0, + 354.0, + 455.0, + 354.0, + 469.0, + 344.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 514.0, + 1405.0, + 514.0, + 1405.0, + 550.0, + 293.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 544.0, + 1405.0, + 544.0, + 1405.0, + 580.0, + 292.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 574.0, + 1405.0, + 574.0, + 1405.0, + 610.0, + 293.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 606.0, + 1335.0, + 606.0, + 1335.0, + 638.0, + 294.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1457.0, + 1069.0, + 1457.0, + 1069.0, + 1500.0, + 293.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1526.0, + 1347.0, + 1526.0, + 1347.0, + 1568.0, + 294.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1079.0, + 1405.0, + 1079.0, + 1405.0, + 1113.0, + 296.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1110.0, + 1402.0, + 1110.0, + 1402.0, + 1144.0, + 294.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1139.0, + 995.0, + 1139.0, + 995.0, + 1174.0, + 294.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1139.0, + 1405.0, + 1139.0, + 1405.0, + 1174.0, + 1039.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1167.0, + 1405.0, + 1167.0, + 1405.0, + 1210.0, + 292.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1199.0, + 1405.0, + 1199.0, + 1405.0, + 1236.0, + 292.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1231.0, + 998.0, + 1231.0, + 998.0, + 1266.0, + 294.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1231.0, + 1406.0, + 1231.0, + 1406.0, + 1266.0, + 1028.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1261.0, + 1405.0, + 1261.0, + 1405.0, + 1297.0, + 292.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1291.0, + 1409.0, + 1291.0, + 1409.0, + 1329.0, + 293.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1324.0, + 1404.0, + 1324.0, + 1404.0, + 1356.0, + 296.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1354.0, + 1405.0, + 1354.0, + 1405.0, + 1385.0, + 296.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1387.0, + 1376.0, + 1387.0, + 1376.0, + 1418.0, + 296.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 696.0, + 1405.0, + 696.0, + 1405.0, + 734.0, + 294.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 729.0, + 1404.0, + 729.0, + 1404.0, + 763.0, + 294.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 759.0, + 1404.0, + 759.0, + 1404.0, + 793.0, + 295.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 790.0, + 1404.0, + 790.0, + 1404.0, + 824.0, + 294.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 820.0, + 1404.0, + 820.0, + 1404.0, + 854.0, + 294.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 847.0, + 1405.0, + 847.0, + 1405.0, + 890.0, + 292.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 915.0, + 295.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 910.0, + 1405.0, + 910.0, + 1405.0, + 945.0, + 294.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 942.0, + 1405.0, + 942.0, + 1405.0, + 976.0, + 294.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 971.0, + 1404.0, + 971.0, + 1404.0, + 1006.0, + 294.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1001.0, + 1406.0, + 1001.0, + 1406.0, + 1037.0, + 292.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1034.0, + 1065.0, + 1034.0, + 1065.0, + 1066.0, + 294.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1576.0, + 1407.0, + 1576.0, + 1407.0, + 1612.0, + 294.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1609.0, + 1404.0, + 1609.0, + 1404.0, + 1641.0, + 297.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1639.0, + 1405.0, + 1639.0, + 1405.0, + 1674.0, + 294.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1670.0, + 1404.0, + 1670.0, + 1404.0, + 1702.0, + 296.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1697.0, + 1405.0, + 1697.0, + 1405.0, + 1734.0, + 293.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1730.0, + 956.0, + 1730.0, + 956.0, + 1763.0, + 293.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 1730.0, + 1105.0, + 1730.0, + 1105.0, + 1763.0, + 1006.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1773.0, + 1022.0, + 1773.0, + 1022.0, + 1808.0, + 296.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1773.0, + 1402.0, + 1773.0, + 1402.0, + 1808.0, + 1079.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1804.0, + 1405.0, + 1804.0, + 1405.0, + 1838.0, + 296.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1835.0, + 1407.0, + 1835.0, + 1407.0, + 1870.0, + 292.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1865.0, + 1405.0, + 1865.0, + 1405.0, + 1900.0, + 294.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1892.0, + 1407.0, + 1892.0, + 1407.0, + 1932.0, + 292.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1925.0, + 1405.0, + 1925.0, + 1405.0, + 1960.0, + 293.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1956.0, + 1260.0, + 1956.0, + 1260.0, + 1989.0, + 293.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1997.0, + 1194.0, + 1997.0, + 1194.0, + 2041.0, + 294.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1526.0, + 1347.0, + 1526.0, + 1347.0, + 1568.0, + 294.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1457.0, + 1069.0, + 1457.0, + 1069.0, + 1500.0, + 293.0, + 1500.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 941, + 1404, + 941, + 1404, + 1160, + 297, + 1160 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1477, + 1405, + 1477, + 1405, + 1695, + 297, + 1695 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1233, + 1404, + 1233, + 1404, + 1388, + 298, + 1388 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 301, + 218, + 1404, + 218, + 1404, + 650, + 301, + 650 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1702, + 1404, + 1702, + 1404, + 1920, + 297, + 1920 + ], + "score": 0.974 + }, + { + "category_id": 4, + "poly": [ + 296, + 669, + 1405, + 669, + 1405, + 825, + 296, + 825 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 298, + 1947, + 1403, + 1947, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 298, + 883, + 1068, + 883, + 1068, + 920, + 298, + 920 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.881 + }, + { + "category_id": 8, + "poly": [ + 303, + 1175, + 1328, + 1175, + 1328, + 1218, + 303, + 1218 + ], + "score": 0.88 + }, + { + "category_id": 8, + "poly": [ + 298, + 1406, + 1387, + 1406, + 1387, + 1446, + 298, + 1446 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.748 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1412, + 1399, + 1412, + 1399, + 1442, + 1369, + 1442 + ], + "score": 0.729 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1183, + 1400, + 1183, + 1400, + 1213, + 1368, + 1213 + ], + "score": 0.568 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1183, + 1400, + 1183, + 1400, + 1213, + 1368, + 1213 + ], + "score": 0.2 + }, + { + "category_id": 13, + "poly": [ + 784, + 1094, + 877, + 1094, + 877, + 1129, + 784, + 1129 + ], + "score": 0.92, + "latex": "( s _ { i } ^ { h } , s _ { i } ^ { w } )" + }, + { + "category_id": 13, + "poly": [ + 508, + 1094, + 609, + 1094, + 609, + 1128, + 508, + 1128 + ], + "score": 0.92, + "latex": "\\overline { { k } } _ { i } ^ { h } \\times k _ { i } ^ { w }" + }, + { + "category_id": 14, + "poly": [ + 306, + 1174, + 1361, + 1174, + 1361, + 1217, + 306, + 1217 + ], + "score": 0.92, + "latex": "\\forall i \\in \\{ 1 , \\ldots , d \\} : h _ { i - 1 } = s _ { i } ^ { h } \\cdot ( h _ { i } - 1 ) + k _ { i } ^ { h } - 2 \\cdot p _ { i } ^ { h } \\quad \\wedge \\quad w _ { i - 1 } = s _ { i } ^ { w } \\cdot ( w _ { i } - 1 ) + k _ { i } ^ { w } - 2 \\cdot p _ { i } ^ { w } \\quad ." + }, + { + "category_id": 13, + "poly": [ + 833, + 1573, + 898, + 1573, + 898, + 1601, + 833, + 1601 + ], + "score": 0.91, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1049, + 1096, + 1179, + 1096, + 1179, + 1129, + 1049, + 1129 + ], + "score": 0.91, + "latex": "\\mathbf { \\Sigma } = ( p _ { i } ^ { h } , p _ { i } ^ { w } )" + }, + { + "category_id": 13, + "poly": [ + 734, + 1858, + 856, + 1858, + 856, + 1886, + 734, + 1886 + ], + "score": 0.89, + "latex": "2 2 6 \\times 2 2 6" + }, + { + "category_id": 13, + "poly": [ + 831, + 1633, + 941, + 1633, + 941, + 1662, + 831, + 1662 + ], + "score": 0.89, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 14, + "poly": [ + 307, + 1402, + 1362, + 1402, + 1362, + 1445, + 307, + 1445 + ], + "score": 0.89, + "latex": "h _ { 0 } = a _ { 1 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 1 } + 1 \\quad \\mathrm { a n d } \\quad w _ { 0 } = a _ { 2 } \\times 2 ^ { d } + 1 = 3 2 \\cdot a _ { 2 } + 1 \\quad \\mathrm { w h e r e } \\quad a _ { 1 } , a _ { 2 } \\in \\mathbb { N } ^ { + }" + }, + { + "category_id": 13, + "poly": [ + 644, + 1704, + 753, + 1704, + 753, + 1734, + 644, + 1734 + ], + "score": 0.89, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 13, + "poly": [ + 1162, + 1480, + 1270, + 1480, + 1270, + 1510, + 1162, + 1510 + ], + "score": 0.89, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1066, + 1156, + 1066, + 1156, + 1096, + 1066, + 1096 + ], + "score": 0.88, + "latex": "( h _ { i } , w _ { i } )" + }, + { + "category_id": 13, + "poly": [ + 427, + 1236, + 459, + 1236, + 459, + 1266, + 427, + 1266 + ], + "score": 0.88, + "latex": "h _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1006, + 702, + 1064, + 702, + 1064, + 731, + 1006, + 731 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1266, + 1295, + 1343, + 1295, + 1343, + 1324, + 1266, + 1324 + ], + "score": 0.88, + "latex": "\\mathit { a } = 5" + }, + { + "category_id": 13, + "poly": [ + 512, + 1240, + 547, + 1240, + 547, + 1265, + 512, + 1265 + ], + "score": 0.86, + "latex": "w _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 471, + 1474, + 496, + 1474, + 496, + 1515, + 471, + 1515 + ], + "score": 0.8, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 578, + 1293, + 628, + 1293, + 628, + 1326, + 578, + 1326 + ], + "score": 0.79, + "latex": "\\pmb { \\mathbb { I } } \\pmb { \\mathcal { 2 } } \\Vert" + }, + { + "category_id": 13, + "poly": [ + 781, + 1601, + 817, + 1601, + 817, + 1636, + 781, + 1636 + ], + "score": 0.79, + "latex": "6 { \\mathsf { b } }" + }, + { + "category_id": 13, + "poly": [ + 708, + 1067, + 726, + 1067, + 726, + 1093, + 708, + 1093 + ], + "score": 0.78, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 810, + 1294, + 859, + 1294, + 859, + 1326, + 810, + 1326 + ], + "score": 0.78, + "latex": "\\pmb { \\mathbb { I } } \\pmb { \\overbrace { 3 } } \\|" + }, + { + "category_id": 13, + "poly": [ + 429, + 1631, + 478, + 1631, + 478, + 1664, + 429, + 1664 + ], + "score": 0.69, + "latex": "\\mathbb { \\lVert 3 3 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 499, + 1128, + 531, + 1128, + 531, + 1162, + 499, + 1162 + ], + "score": 0.58, + "latex": "\\boxed { \\mathrm { A } }" + }, + { + "category_id": 13, + "poly": [ + 784, + 973, + 818, + 973, + 818, + 1009, + 784, + 1009 + ], + "score": 0.53, + "latex": "6 \\mathrm { a }" + }, + { + "category_id": 13, + "poly": [ + 572, + 1795, + 606, + 1795, + 606, + 1830, + 572, + 1830 + ], + "score": 0.39, + "latex": "\\bar { 6 } 6" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 223.0, + 330.0, + 223.0, + 330.0, + 245.0, + 306.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 223.0, + 452.0, + 223.0, + 452.0, + 247.0, + 355.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 222.0, + 568.0, + 222.0, + 568.0, + 247.0, + 540.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 220.0, + 1216.0, + 220.0, + 1216.0, + 256.0, + 805.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 242.0, + 513.0, + 242.0, + 513.0, + 266.0, + 354.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 287.0, + 588.0, + 287.0, + 588.0, + 416.0, + 561.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 310.0, + 570.0, + 310.0, + 570.0, + 397.0, + 545.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 451.0, + 529.0, + 451.0, + 529.0, + 475.0, + 301.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 447.0, + 713.0, + 447.0, + 713.0, + 471.0, + 633.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 449.0, + 934.0, + 449.0, + 934.0, + 470.0, + 858.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 449.0, + 1238.0, + 449.0, + 1238.0, + 470.0, + 1156.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 471.0, + 509.0, + 471.0, + 509.0, + 492.0, + 324.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 485.0, + 487.0, + 485.0, + 487.0, + 513.0, + 348.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 499.0, + 588.0, + 499.0, + 588.0, + 627.0, + 561.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 511.0, + 430.0, + 511.0, + 430.0, + 538.0, + 411.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 522.0, + 568.0, + 522.0, + 568.0, + 610.0, + 545.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 543.0, + 472.0, + 543.0, + 472.0, + 567.0, + 371.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 247.0, + 806.0, + 247.0, + 806.0, + 264.0, + 756.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 671.0, + 1405.0, + 671.0, + 1405.0, + 704.0, + 296.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 700.0, + 1005.0, + 700.0, + 1005.0, + 736.0, + 294.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 700.0, + 1405.0, + 700.0, + 1405.0, + 736.0, + 1065.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 731.0, + 1405.0, + 731.0, + 1405.0, + 768.0, + 294.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 756.0, + 1405.0, + 756.0, + 1405.0, + 800.0, + 291.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 792.0, + 1358.0, + 792.0, + 1358.0, + 830.0, + 295.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1940.0, + 1358.0, + 1940.0, + 1358.0, + 1982.0, + 327.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1969.0, + 1408.0, + 1969.0, + 1408.0, + 2016.0, + 325.0, + 2016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2004.0, + 1371.0, + 2004.0, + 1371.0, + 2039.0, + 293.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 880.0, + 1071.0, + 880.0, + 1071.0, + 924.0, + 292.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 938.0, + 1405.0, + 938.0, + 1405.0, + 981.0, + 292.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 972.0, + 783.0, + 972.0, + 783.0, + 1011.0, + 295.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 972.0, + 1406.0, + 972.0, + 1406.0, + 1011.0, + 819.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1004.0, + 1404.0, + 1004.0, + 1404.0, + 1039.0, + 294.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1034.0, + 1405.0, + 1034.0, + 1405.0, + 1069.0, + 294.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1063.0, + 707.0, + 1063.0, + 707.0, + 1100.0, + 294.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1063.0, + 1065.0, + 1063.0, + 1065.0, + 1100.0, + 727.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1063.0, + 1406.0, + 1063.0, + 1406.0, + 1100.0, + 1157.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1087.0, + 507.0, + 1087.0, + 507.0, + 1138.0, + 290.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1087.0, + 783.0, + 1087.0, + 783.0, + 1138.0, + 610.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1087.0, + 1048.0, + 1087.0, + 1048.0, + 1138.0, + 878.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1087.0, + 1410.0, + 1087.0, + 1410.0, + 1138.0, + 1180.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1124.0, + 498.0, + 1124.0, + 498.0, + 1163.0, + 293.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1124.0, + 673.0, + 1124.0, + 673.0, + 1163.0, + 532.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1475.0, + 470.0, + 1475.0, + 470.0, + 1516.0, + 297.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1475.0, + 1161.0, + 1475.0, + 1161.0, + 1516.0, + 497.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1516.0, + 1271.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1510.0, + 1406.0, + 1510.0, + 1406.0, + 1545.0, + 296.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1541.0, + 1405.0, + 1541.0, + 1405.0, + 1576.0, + 295.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1571.0, + 832.0, + 1571.0, + 832.0, + 1606.0, + 294.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1571.0, + 1406.0, + 1571.0, + 1406.0, + 1606.0, + 899.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1602.0, + 780.0, + 1602.0, + 780.0, + 1637.0, + 295.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1602.0, + 1405.0, + 1602.0, + 1405.0, + 1637.0, + 818.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1632.0, + 428.0, + 1632.0, + 428.0, + 1667.0, + 294.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1632.0, + 830.0, + 1632.0, + 830.0, + 1667.0, + 479.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1632.0, + 1406.0, + 1632.0, + 1406.0, + 1667.0, + 942.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1663.0, + 1359.0, + 1663.0, + 1359.0, + 1698.0, + 295.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1235.0, + 426.0, + 1235.0, + 426.0, + 1268.0, + 297.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1235.0, + 511.0, + 1235.0, + 511.0, + 1268.0, + 460.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1235.0, + 1404.0, + 1235.0, + 1404.0, + 1268.0, + 548.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1261.0, + 1403.0, + 1261.0, + 1403.0, + 1300.0, + 295.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 577.0, + 1295.0, + 577.0, + 1328.0, + 296.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1295.0, + 809.0, + 1295.0, + 809.0, + 1328.0, + 629.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1295.0, + 1265.0, + 1295.0, + 1265.0, + 1328.0, + 860.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1295.0, + 1405.0, + 1295.0, + 1405.0, + 1328.0, + 1344.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1322.0, + 1409.0, + 1322.0, + 1409.0, + 1361.0, + 292.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1354.0, + 1287.0, + 1354.0, + 1287.0, + 1393.0, + 293.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1702.0, + 643.0, + 1702.0, + 643.0, + 1738.0, + 293.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1702.0, + 1404.0, + 1702.0, + 1404.0, + 1738.0, + 754.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1734.0, + 1404.0, + 1734.0, + 1404.0, + 1770.0, + 295.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1764.0, + 1405.0, + 1764.0, + 1405.0, + 1800.0, + 294.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1797.0, + 571.0, + 1797.0, + 571.0, + 1828.0, + 296.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1797.0, + 1404.0, + 1797.0, + 1404.0, + 1828.0, + 607.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1826.0, + 1406.0, + 1826.0, + 1406.0, + 1861.0, + 295.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1856.0, + 733.0, + 1856.0, + 733.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1856.0, + 1406.0, + 1856.0, + 1406.0, + 1890.0, + 857.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1888.0, + 1216.0, + 1888.0, + 1216.0, + 1923.0, + 295.0, + 1923.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 920, + 1405, + 920, + 1405, + 1136, + 296, + 1136 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1786, + 1404, + 1786, + 1404, + 2035, + 296, + 2035 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 299, + 1158, + 1399, + 1158, + 1399, + 1533, + 299, + 1533 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 623, + 1405, + 623, + 1405, + 810, + 297, + 810 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1403, + 229, + 1403, + 388, + 298, + 388 + ], + "score": 0.972 + }, + { + "category_id": 5, + "poly": [ + 315, + 466, + 1383, + 466, + 1383, + 596, + 315, + 596 + ], + "score": 0.97, + "html": "
Input Size ²MobileNetResNet-18ResNet-34ResNet-50ResNet-101
224×22468.19 (88.44)69.93 (89.22)73.30 (91.42)75.65 (92.47)77.37 (93.56)
225×22568.80 (88.78)70.27 (89.52)73.72 (91.58)76.01 (92.90)77.67 (93.81)
" + }, + { + "category_id": 4, + "poly": [ + 295, + 1556, + 1405, + 1556, + 1405, + 1742, + 295, + 1742 + ], + "score": 0.968 + }, + { + "category_id": 0, + "poly": [ + 298, + 852, + 922, + 852, + 922, + 888, + 298, + 888 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.785 + }, + { + "category_id": 6, + "poly": [ + 296, + 408, + 1393, + 408, + 1393, + 443, + 296, + 443 + ], + "score": 0.689 + }, + { + "category_id": 1, + "poly": [ + 296, + 408, + 1393, + 408, + 1393, + 443, + 296, + 443 + ], + "score": 0.2 + }, + { + "category_id": 13, + "poly": [ + 682, + 623, + 773, + 623, + 773, + 660, + 682, + 660 + ], + "score": 0.91, + "latex": "( p _ { i } ^ { h } = 0" + }, + { + "category_id": 13, + "poly": [ + 808, + 626, + 900, + 626, + 900, + 660, + 808, + 660 + ], + "score": 0.91, + "latex": "p _ { i } ^ { w } = 0 ," + }, + { + "category_id": 13, + "poly": [ + 297, + 716, + 392, + 716, + 392, + 746, + 297, + 746 + ], + "score": 0.9, + "latex": "2 ^ { d } = 3 2" + }, + { + "category_id": 13, + "poly": [ + 348, + 1589, + 472, + 1589, + 472, + 1618, + 348, + 1618 + ], + "score": 0.89, + "latex": "5 1 2 \\times 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 781, + 718, + 894, + 718, + 894, + 748, + 781, + 748 + ], + "score": 0.82, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 13, + "poly": [ + 655, + 718, + 770, + 718, + 770, + 748, + 655, + 748 + ], + "score": 0.79, + "latex": "{ \\mathrm { 1 9 3 \\times 1 9 3 } }" + }, + { + "category_id": 13, + "poly": [ + 904, + 718, + 1019, + 718, + 1019, + 748, + 904, + 748 + ], + "score": 0.7, + "latex": "2 5 7 \\times 2 5 7" + }, + { + "category_id": 13, + "poly": [ + 1282, + 291, + 1313, + 291, + 1313, + 326, + 1282, + 326 + ], + "score": 0.66, + "latex": "\\boxed { \\dot { \\mathbf { C } } }" + }, + { + "category_id": 13, + "poly": [ + 584, + 685, + 646, + 685, + 646, + 722, + 584, + 722 + ], + "score": 0.61, + "latex": "\\mathrm { E q } \\ 1" + }, + { + "category_id": 13, + "poly": [ + 727, + 1042, + 775, + 1042, + 775, + 1074, + 727, + 1074 + ], + "score": 0.46, + "latex": "\\pmb { \\pmb { 2 8 } }" + }, + { + "category_id": 13, + "poly": [ + 1293, + 1879, + 1323, + 1879, + 1323, + 1917, + 1293, + 1917 + ], + "score": 0.45, + "latex": "\\mathbb { H }" + }, + { + "category_id": 13, + "poly": [ + 1076, + 320, + 1126, + 320, + 1126, + 353, + 1076, + 353 + ], + "score": 0.42, + "latex": "\\mathbb { \\lVert \\boldsymbol { 4 3 } \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 364, + 685, + 445, + 685, + 445, + 721, + 364, + 721 + ], + "score": 0.39, + "latex": "( { \\mathrm { F i g ~ } } 7 { \\mathrm { \\rho } }" + }, + { + "category_id": 13, + "poly": [ + 362, + 560, + 463, + 560, + 463, + 591, + 362, + 591 + ], + "score": 0.36, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 13, + "poly": [ + 363, + 519, + 464, + 519, + 464, + 547, + 363, + 547 + ], + "score": 0.31, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 374, + 1788, + 409, + 1788, + 409, + 1824, + 374, + 1824 + ], + "score": 0.29, + "latex": "7 \\mathrm { a }" + }, + { + "category_id": 13, + "poly": [ + 484, + 1849, + 532, + 1849, + 532, + 1882, + 484, + 1882 + ], + "score": 0.29, + "latex": "\\mathbb { \\lVert \\boldsymbol { 1 6 } \\rVert }" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1181.0, + 330.0, + 1181.0, + 330.0, + 1207.0, + 303.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1178.0, + 463.0, + 1178.0, + 463.0, + 1210.0, + 361.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1177.0, + 649.0, + 1177.0, + 649.0, + 1210.0, + 515.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1178.0, + 823.0, + 1178.0, + 823.0, + 1210.0, + 677.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1156.0, + 869.0, + 1156.0, + 869.0, + 1217.0, + 833.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1177.0, + 998.0, + 1177.0, + 998.0, + 1209.0, + 913.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1181.0, + 1397.0, + 1181.0, + 1397.0, + 1208.0, + 1032.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1202.0, + 484.0, + 1202.0, + 484.0, + 1356.0, + 328.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1215.0, + 882.0, + 1215.0, + 882.0, + 1331.0, + 848.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1224.0, + 1064.0, + 1224.0, + 1064.0, + 1332.0, + 1032.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1223.0, + 1248.0, + 1223.0, + 1248.0, + 1332.0, + 1216.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1350.0, + 332.0, + 1350.0, + 332.0, + 1381.0, + 299.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1367.0, + 492.0, + 1367.0, + 492.0, + 1395.0, + 334.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1368.0, + 663.0, + 1368.0, + 663.0, + 1396.0, + 527.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1365.0, + 853.0, + 1365.0, + 853.0, + 1398.0, + 695.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1367.0, + 991.0, + 1367.0, + 991.0, + 1395.0, + 919.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 1365.0, + 1186.0, + 1365.0, + 1186.0, + 1396.0, + 1082.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1365.0, + 1371.0, + 1365.0, + 1371.0, + 1396.0, + 1268.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1556.0, + 1406.0, + 1556.0, + 1406.0, + 1592.0, + 294.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1588.0, + 347.0, + 1588.0, + 347.0, + 1623.0, + 294.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1588.0, + 1406.0, + 1588.0, + 1406.0, + 1623.0, + 473.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1614.0, + 1406.0, + 1614.0, + 1406.0, + 1657.0, + 293.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1649.0, + 1405.0, + 1649.0, + 1405.0, + 1685.0, + 295.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1680.0, + 1405.0, + 1680.0, + 1405.0, + 1715.0, + 295.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1710.0, + 854.0, + 1710.0, + 854.0, + 1746.0, + 294.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 849.0, + 924.0, + 849.0, + 924.0, + 893.0, + 294.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 403.0, + 1394.0, + 403.0, + 1394.0, + 448.0, + 302.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 922.0, + 1403.0, + 922.0, + 1403.0, + 956.0, + 295.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 951.0, + 1406.0, + 951.0, + 1406.0, + 986.0, + 294.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 980.0, + 1406.0, + 980.0, + 1406.0, + 1017.0, + 294.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1012.0, + 1405.0, + 1012.0, + 1405.0, + 1049.0, + 293.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1042.0, + 726.0, + 1042.0, + 726.0, + 1080.0, + 293.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 1042.0, + 1405.0, + 1042.0, + 1405.0, + 1080.0, + 776.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1071.0, + 1406.0, + 1071.0, + 1406.0, + 1109.0, + 294.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1103.0, + 1182.0, + 1103.0, + 1182.0, + 1138.0, + 295.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1788.0, + 373.0, + 1788.0, + 373.0, + 1824.0, + 296.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1788.0, + 1405.0, + 1788.0, + 1405.0, + 1824.0, + 410.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1820.0, + 1404.0, + 1820.0, + 1404.0, + 1854.0, + 295.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1851.0, + 483.0, + 1851.0, + 483.0, + 1885.0, + 296.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1885.0, + 533.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 1292.0, + 1881.0, + 1292.0, + 1914.0, + 293.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 1884.0, + 1404.0, + 1884.0, + 1404.0, + 1913.0, + 1328.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1402.0, + 1910.0, + 1402.0, + 1944.0, + 294.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 1404.0, + 1941.0, + 1404.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 415.0, + 2003.0, + 415.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 622.0, + 681.0, + 622.0, + 681.0, + 661.0, + 294.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 622.0, + 807.0, + 622.0, + 807.0, + 661.0, + 774.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 622.0, + 1406.0, + 622.0, + 1406.0, + 661.0, + 901.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 656.0, + 1406.0, + 656.0, + 1406.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 682.0, + 363.0, + 682.0, + 363.0, + 723.0, + 291.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 682.0, + 583.0, + 682.0, + 583.0, + 723.0, + 446.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 682.0, + 1407.0, + 682.0, + 1407.0, + 723.0, + 647.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 710.0, + 296.0, + 710.0, + 296.0, + 756.0, + 292.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 710.0, + 654.0, + 710.0, + 654.0, + 756.0, + 393.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 710.0, + 780.0, + 710.0, + 780.0, + 756.0, + 771.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 710.0, + 903.0, + 710.0, + 903.0, + 756.0, + 895.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 710.0, + 1406.0, + 710.0, + 1406.0, + 756.0, + 1020.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 745.0, + 1406.0, + 745.0, + 1406.0, + 783.0, + 294.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 779.0, + 647.0, + 779.0, + 647.0, + 811.0, + 296.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1403.0, + 229.0, + 1403.0, + 263.0, + 296.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 259.0, + 1404.0, + 259.0, + 1404.0, + 296.0, + 294.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 291.0, + 1281.0, + 291.0, + 1281.0, + 325.0, + 293.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 291.0, + 1403.0, + 291.0, + 1403.0, + 325.0, + 1314.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 319.0, + 1075.0, + 319.0, + 1075.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 319.0, + 1404.0, + 319.0, + 1404.0, + 356.0, + 1127.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 350.0, + 962.0, + 350.0, + 962.0, + 389.0, + 292.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 403.0, + 1394.0, + 403.0, + 1394.0, + 448.0, + 302.0, + 448.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1316, + 1404, + 1316, + 1404, + 1593, + 297, + 1593 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1706, + 1404, + 1706, + 1404, + 1924, + 297, + 1924 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1088, + 1403, + 1088, + 1403, + 1303, + 298, + 1303 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 913, + 1403, + 913, + 1403, + 1067, + 298, + 1067 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 709, + 1404, + 709, + 1404, + 893, + 298, + 893 + ], + "score": 0.976 + }, + { + "category_id": 2, + "poly": [ + 298, + 1940, + 1402, + 1940, + 1402, + 2035, + 298, + 2035 + ], + "score": 0.955 + }, + { + "category_id": 0, + "poly": [ + 300, + 1638, + 1093, + 1638, + 1093, + 1674, + 300, + 1674 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.883 + }, + { + "category_id": 1, + "poly": [ + 292, + 228, + 1404, + 228, + 1404, + 690, + 292, + 690 + ], + "score": 0.707 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.674 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.244 + }, + { + "category_id": 13, + "poly": [ + 1091, + 771, + 1198, + 771, + 1198, + 800, + 1091, + 800 + ], + "score": 0.89, + "latex": "1 2 7 \\times 1 2 7" + }, + { + "category_id": 13, + "poly": [ + 930, + 1439, + 965, + 1439, + 965, + 1474, + 930, + 1474 + ], + "score": 0.8, + "latex": "\\bar { 5 } 6" + }, + { + "category_id": 13, + "poly": [ + 893, + 800, + 928, + 800, + 928, + 836, + 893, + 836 + ], + "score": 0.8, + "latex": "\\mathbf { A } )" + }, + { + "category_id": 13, + "poly": [ + 1304, + 564, + 1353, + 564, + 1353, + 597, + 1304, + 597 + ], + "score": 0.78, + "latex": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1408, + 1260, + 1408, + 1260, + 1441, + 1211, + 1441 + ], + "score": 0.78, + "latex": "\\left[ \\left[ 2 6 \\right] \\right]" + }, + { + "category_id": 13, + "poly": [ + 1130, + 595, + 1179, + 595, + 1179, + 627, + 1130, + 627 + ], + "score": 0.76, + "latex": "\\pmb { \\mathbb { B } } 0 \\|" + }, + { + "category_id": 13, + "poly": [ + 830, + 1707, + 865, + 1707, + 865, + 1743, + 830, + 1743 + ], + "score": 0.73, + "latex": "5 \\mathsf { b }" + }, + { + "category_id": 13, + "poly": [ + 743, + 1560, + 792, + 1560, + 792, + 1593, + 743, + 1593 + ], + "score": 0.7, + "latex": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }" + }, + { + "category_id": 13, + "poly": [ + 345, + 1347, + 376, + 1347, + 376, + 1383, + 345, + 1383 + ], + "score": 0.67, + "latex": "\\bar { 7 } )" + }, + { + "category_id": 13, + "poly": [ + 1210, + 1707, + 1244, + 1707, + 1244, + 1744, + 1210, + 1744 + ], + "score": 0.61, + "latex": "\\textcircled { 5 } \\textcircled { \\times }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 1269, + 1201, + 1269, + 1201, + 1302, + 1152, + 1302 + ], + "score": 0.61, + "latex": "\\mathbb { B } 2 \\mathbb { I }" + }, + { + "category_id": 13, + "poly": [ + 1118, + 943, + 1166, + 943, + 1166, + 976, + 1118, + 976 + ], + "score": 0.55, + "latex": "\\pm 2 \\|" + }, + { + "category_id": 13, + "poly": [ + 1092, + 1768, + 1118, + 1768, + 1118, + 1803, + 1092, + 1803 + ], + "score": 0.54, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 534, + 769, + 568, + 769, + 568, + 805, + 534, + 805 + ], + "score": 0.48, + "latex": "\\textcircled { 7 } \\textcircled { \\times }" + }, + { + "category_id": 13, + "poly": [ + 610, + 533, + 659, + 533, + 659, + 566, + 610, + 566 + ], + "score": 0.46, + "latex": "\\lVert 2 2 \\rVert" + }, + { + "category_id": 13, + "poly": [ + 414, + 1497, + 441, + 1497, + 441, + 1536, + 414, + 1536 + ], + "score": 0.38, + "latex": "^ 4 \\cdot" + }, + { + "category_id": 13, + "poly": [ + 1143, + 1347, + 1172, + 1347, + 1172, + 1383, + 1143, + 1383 + ], + "score": 0.37, + "latex": "^ { 8 ) }" + }, + { + "category_id": 13, + "poly": [ + 708, + 259, + 757, + 259, + 757, + 293, + 708, + 293 + ], + "score": 0.33, + "latex": " { \\mathbb { E } } { \\ b { \\mathbb { Z } } } ] " + }, + { + "category_id": 13, + "poly": [ + 1361, + 973, + 1394, + 973, + 1394, + 1009, + 1361, + 1009 + ], + "score": 0.25, + "latex": "\\textcircled { 7 } \\textcircled { < }" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1937.0, + 1406.0, + 1937.0, + 1406.0, + 1982.0, + 324.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 757.0, + 1972.0, + 757.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1996.0, + 1406.0, + 1996.0, + 1406.0, + 2041.0, + 327.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1635.0, + 1097.0, + 1635.0, + 1097.0, + 1678.0, + 293.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1317.0, + 1402.0, + 1317.0, + 1402.0, + 1350.0, + 297.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1346.0, + 344.0, + 1346.0, + 344.0, + 1384.0, + 296.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1346.0, + 1142.0, + 1346.0, + 1142.0, + 1384.0, + 377.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1346.0, + 1406.0, + 1346.0, + 1406.0, + 1384.0, + 1173.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1377.0, + 1405.0, + 1377.0, + 1405.0, + 1414.0, + 295.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1409.0, + 1210.0, + 1409.0, + 1210.0, + 1443.0, + 293.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1409.0, + 1404.0, + 1409.0, + 1404.0, + 1443.0, + 1261.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1437.0, + 929.0, + 1437.0, + 929.0, + 1478.0, + 292.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 1437.0, + 1406.0, + 1437.0, + 1406.0, + 1478.0, + 966.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1470.0, + 1402.0, + 1470.0, + 1402.0, + 1503.0, + 295.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1498.0, + 413.0, + 1498.0, + 413.0, + 1538.0, + 292.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1498.0, + 1406.0, + 1498.0, + 1406.0, + 1538.0, + 442.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1530.0, + 1406.0, + 1530.0, + 1406.0, + 1566.0, + 294.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1558.0, + 742.0, + 1558.0, + 742.0, + 1597.0, + 292.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1558.0, + 1283.0, + 1558.0, + 1283.0, + 1597.0, + 793.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1708.0, + 829.0, + 1708.0, + 829.0, + 1743.0, + 294.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1708.0, + 1209.0, + 1708.0, + 1209.0, + 1743.0, + 866.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1708.0, + 1404.0, + 1708.0, + 1404.0, + 1743.0, + 1245.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1772.0, + 296.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1768.0, + 1091.0, + 1768.0, + 1091.0, + 1806.0, + 294.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1768.0, + 1407.0, + 1768.0, + 1407.0, + 1806.0, + 1119.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1799.0, + 1402.0, + 1799.0, + 1402.0, + 1835.0, + 295.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1831.0, + 1401.0, + 1831.0, + 1401.0, + 1862.0, + 296.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1861.0, + 1404.0, + 1861.0, + 1404.0, + 1896.0, + 295.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1890.0, + 1016.0, + 1890.0, + 1016.0, + 1929.0, + 292.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1124.0, + 293.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1152.0, + 294.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1150.0, + 1404.0, + 1150.0, + 1404.0, + 1183.0, + 293.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1178.0, + 1405.0, + 1178.0, + 1405.0, + 1214.0, + 293.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1209.0, + 1405.0, + 1209.0, + 1405.0, + 1246.0, + 296.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1239.0, + 1405.0, + 1239.0, + 1405.0, + 1275.0, + 292.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1268.0, + 1151.0, + 1268.0, + 1151.0, + 1306.0, + 293.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1268.0, + 1207.0, + 1268.0, + 1207.0, + 1306.0, + 1202.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 913.0, + 1404.0, + 913.0, + 1404.0, + 946.0, + 297.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 944.0, + 1117.0, + 944.0, + 1117.0, + 978.0, + 293.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 944.0, + 1404.0, + 944.0, + 1404.0, + 978.0, + 1167.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 974.0, + 1360.0, + 974.0, + 1360.0, + 1011.0, + 293.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1002.0, + 1364.0, + 1002.0, + 1364.0, + 1042.0, + 292.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1010.0, + 1403.0, + 1010.0, + 1403.0, + 1035.0, + 1374.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1034.0, + 1365.0, + 1034.0, + 1365.0, + 1072.0, + 293.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 710.0, + 1404.0, + 710.0, + 1404.0, + 741.0, + 296.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 1406.0, + 739.0, + 1406.0, + 774.0, + 294.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 767.0, + 533.0, + 767.0, + 533.0, + 805.0, + 295.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 767.0, + 1090.0, + 767.0, + 1090.0, + 805.0, + 569.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 767.0, + 1406.0, + 767.0, + 1406.0, + 805.0, + 1199.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 798.0, + 892.0, + 798.0, + 892.0, + 837.0, + 293.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 798.0, + 1405.0, + 798.0, + 1405.0, + 837.0, + 929.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 830.0, + 1405.0, + 830.0, + 1405.0, + 866.0, + 294.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 861.0, + 1341.0, + 861.0, + 1341.0, + 896.0, + 294.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1406.0, + 228.0, + 1406.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 258.0, + 707.0, + 258.0, + 707.0, + 297.0, + 292.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 258.0, + 1405.0, + 258.0, + 1405.0, + 297.0, + 758.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 291.0, + 1406.0, + 291.0, + 1406.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 321.0, + 827.0, + 321.0, + 827.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 322.0, + 1403.0, + 322.0, + 1403.0, + 354.0, + 832.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 351.0, + 1403.0, + 351.0, + 1403.0, + 386.0, + 294.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 382.0, + 1405.0, + 382.0, + 1405.0, + 416.0, + 292.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 448.0, + 295.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 441.0, + 1406.0, + 441.0, + 1406.0, + 477.0, + 292.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 470.0, + 1406.0, + 470.0, + 1406.0, + 513.0, + 293.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 503.0, + 1406.0, + 503.0, + 1406.0, + 539.0, + 294.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 534.0, + 609.0, + 534.0, + 609.0, + 569.0, + 295.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 534.0, + 1406.0, + 534.0, + 1406.0, + 569.0, + 660.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 567.0, + 1303.0, + 567.0, + 1303.0, + 598.0, + 296.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 567.0, + 1405.0, + 567.0, + 1405.0, + 598.0, + 1354.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 594.0, + 1129.0, + 594.0, + 1129.0, + 630.0, + 295.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 594.0, + 1405.0, + 594.0, + 1405.0, + 630.0, + 1180.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 1407.0, + 625.0, + 1407.0, + 663.0, + 294.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 657.0, + 1141.0, + 657.0, + 1141.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1160, + 1404, + 1160, + 1404, + 1496, + 297, + 1496 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1618, + 1403, + 1618, + 1403, + 1867, + 297, + 1867 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1880, + 1402, + 1880, + 1402, + 2035, + 298, + 2035 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 594, + 1404, + 594, + 1404, + 718, + 299, + 718 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 967, + 1404, + 967, + 1404, + 1122, + 298, + 1122 + ], + "score": 0.968 + }, + { + "category_id": 5, + "poly": [ + 361, + 813, + 1342, + 813, + 1342, + 931, + 361, + 931 + ], + "score": 0.968, + "html": "
Average Precision (AP)AP@.20I0UAP@ .50I0UAP@.75I0UAP@.90I0U
Zero Padding80.24%49.58%3.7%0.007%
Mirror Padding83.20%57%8.44%0.02%
" + }, + { + "category_id": 3, + "poly": [ + 301, + 221, + 1402, + 221, + 1402, + 441, + 301, + 441 + ], + "score": 0.964 + }, + { + "category_id": 0, + "poly": [ + 299, + 1547, + 890, + 1547, + 890, + 1583, + 299, + 1583 + ], + "score": 0.924 + }, + { + "category_id": 4, + "poly": [ + 299, + 463, + 1403, + 463, + 1403, + 530, + 299, + 530 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 812, + 76, + 812, + 104, + 299, + 104 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.796 + }, + { + "category_id": 6, + "poly": [ + 292, + 750, + 1401, + 750, + 1401, + 785, + 292, + 785 + ], + "score": 0.777 + }, + { + "category_id": 13, + "poly": [ + 1214, + 1743, + 1273, + 1743, + 1273, + 1772, + 1214, + 1772 + ], + "score": 0.89, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 996, + 1910, + 1030, + 1910, + 1030, + 1943, + 996, + 1943 + ], + "score": 0.7, + "latex": "\\mathbb { \\left[ \\bigcirc \\right] }" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1741, + 1394, + 1741, + 1394, + 1774, + 1345, + 1774 + ], + "score": 0.64, + "latex": "\\mathbf { \\bar { \\textmu } }" + }, + { + "category_id": 13, + "poly": [ + 770, + 967, + 799, + 967, + 799, + 1003, + 770, + 1003 + ], + "score": 0.61, + "latex": "6 ," + }, + { + "category_id": 13, + "poly": [ + 862, + 1803, + 893, + 1803, + 893, + 1839, + 862, + 1839 + ], + "score": 0.6, + "latex": "\\mathbf { B }" + }, + { + "category_id": 13, + "poly": [ + 351, + 1711, + 400, + 1711, + 400, + 1744, + 351, + 1744 + ], + "score": 0.54, + "latex": "\\dot { [ \\lVert { 4 } \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1911, + 1394, + 1911, + 1394, + 1943, + 1345, + 1943 + ], + "score": 0.44, + "latex": "\\begin{array} { r l } { { \\bigl \\| \\overline { { 3 5 } } \\bigr \\| } } & { { } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 361, + 654, + 388, + 654, + 388, + 691, + 361, + 691 + ], + "score": 0.32, + "latex": "2" + }, + { + "category_id": 13, + "poly": [ + 457, + 1680, + 506, + 1680, + 506, + 1713, + 457, + 1713 + ], + "score": 0.29, + "latex": "\\mathbb { \\lVert 2 4 \\rVert }" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 230.0, + 442.0, + 230.0, + 442.0, + 246.0, + 304.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 230.0, + 621.0, + 230.0, + 621.0, + 246.0, + 484.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 230.0, + 805.0, + 230.0, + 805.0, + 246.0, + 661.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 230.0, + 988.0, + 230.0, + 988.0, + 246.0, + 844.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 230.0, + 1171.0, + 230.0, + 1171.0, + 246.0, + 1034.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 230.0, + 1351.0, + 230.0, + 1351.0, + 246.0, + 1216.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 245.0, + 1401.0, + 245.0, + 1401.0, + 262.0, + 1367.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 252.0, + 480.0, + 252.0, + 480.0, + 271.0, + 456.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 252.0, + 661.0, + 252.0, + 661.0, + 271.0, + 636.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 253.0, + 841.0, + 253.0, + 841.0, + 269.0, + 821.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 257.0, + 1023.0, + 257.0, + 1023.0, + 272.0, + 1002.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 260.0, + 1054.0, + 260.0, + 1054.0, + 269.0, + 1044.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 257.0, + 1214.0, + 257.0, + 1214.0, + 275.0, + 1186.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 262.0, + 1401.0, + 262.0, + 1401.0, + 280.0, + 1366.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 276.0, + 480.0, + 276.0, + 480.0, + 295.0, + 456.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 277.0, + 661.0, + 277.0, + 661.0, + 296.0, + 636.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 273.0, + 841.0, + 273.0, + 841.0, + 292.0, + 819.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 285.0, + 1023.0, + 285.0, + 1023.0, + 303.0, + 999.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 279.0, + 1214.0, + 279.0, + 1214.0, + 297.0, + 1186.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 281.0, + 1401.0, + 281.0, + 1401.0, + 299.0, + 1366.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 299.0, + 480.0, + 299.0, + 480.0, + 317.0, + 456.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 302.0, + 661.0, + 302.0, + 661.0, + 321.0, + 635.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 296.0, + 841.0, + 296.0, + 841.0, + 316.0, + 819.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 301.0, + 1214.0, + 301.0, + 1214.0, + 319.0, + 1186.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 300.0, + 1401.0, + 300.0, + 1401.0, + 317.0, + 1366.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 323.0, + 477.0, + 323.0, + 477.0, + 337.0, + 457.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 323.0, + 839.0, + 323.0, + 839.0, + 333.0, + 824.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 315.0, + 1022.0, + 315.0, + 1022.0, + 330.0, + 1002.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 317.0, + 1401.0, + 317.0, + 1401.0, + 335.0, + 1363.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 336.0, + 439.0, + 336.0, + 439.0, + 352.0, + 307.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 335.0, + 620.0, + 335.0, + 620.0, + 351.0, + 489.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 337.0, + 800.0, + 337.0, + 800.0, + 353.0, + 670.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 337.0, + 982.0, + 337.0, + 982.0, + 353.0, + 851.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 336.0, + 1169.0, + 336.0, + 1169.0, + 352.0, + 1032.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 333.0, + 1352.0, + 333.0, + 1352.0, + 355.0, + 1212.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 346.0, + 668.0, + 346.0, + 668.0, + 360.0, + 639.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 349.0, + 1027.0, + 349.0, + 1027.0, + 362.0, + 1002.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 349.0, + 1213.0, + 349.0, + 1213.0, + 362.0, + 1188.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 361.0, + 419.0, + 361.0, + 419.0, + 370.0, + 399.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 359.0, + 484.0, + 359.0, + 484.0, + 432.0, + 456.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 367.0, + 670.0, + 367.0, + 670.0, + 421.0, + 638.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 357.0, + 800.0, + 357.0, + 800.0, + 366.0, + 774.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 362.0, + 848.0, + 362.0, + 848.0, + 410.0, + 819.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 373.0, + 1029.0, + 373.0, + 1029.0, + 417.0, + 999.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 369.0, + 1214.0, + 369.0, + 1214.0, + 430.0, + 1186.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 359.0, + 1396.0, + 359.0, + 1396.0, + 435.0, + 1366.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 427.0, + 669.0, + 427.0, + 669.0, + 440.0, + 639.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 425.0, + 846.0, + 425.0, + 846.0, + 439.0, + 821.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 427.0, + 1028.0, + 427.0, + 1028.0, + 441.0, + 999.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 386.0, + 486.0, + 386.0, + 486.0, + 404.0, + 454.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1544.0, + 894.0, + 1544.0, + 894.0, + 1588.0, + 292.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 465.0, + 1404.0, + 465.0, + 1404.0, + 503.0, + 295.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 497.0, + 1380.0, + 497.0, + 1380.0, + 533.0, + 295.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 745.0, + 1403.0, + 745.0, + 1403.0, + 789.0, + 294.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1157.0, + 1405.0, + 1157.0, + 1405.0, + 1199.0, + 292.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1193.0, + 1405.0, + 1193.0, + 1405.0, + 1228.0, + 294.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1223.0, + 1405.0, + 1223.0, + 1405.0, + 1257.0, + 295.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1252.0, + 1404.0, + 1252.0, + 1404.0, + 1287.0, + 294.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1281.0, + 1405.0, + 1281.0, + 1405.0, + 1321.0, + 292.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1313.0, + 1405.0, + 1313.0, + 1405.0, + 1348.0, + 295.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1344.0, + 1404.0, + 1344.0, + 1404.0, + 1379.0, + 295.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1375.0, + 1404.0, + 1375.0, + 1404.0, + 1410.0, + 295.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1406.0, + 1404.0, + 1406.0, + 1404.0, + 1441.0, + 294.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 1404.0, + 1436.0, + 1404.0, + 1470.0, + 294.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1467.0, + 712.0, + 1467.0, + 712.0, + 1499.0, + 293.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1619.0, + 1407.0, + 1619.0, + 1407.0, + 1656.0, + 295.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1652.0, + 1405.0, + 1652.0, + 1405.0, + 1686.0, + 295.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1678.0, + 456.0, + 1678.0, + 456.0, + 1716.0, + 296.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1678.0, + 1404.0, + 1678.0, + 1404.0, + 1716.0, + 507.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1711.0, + 350.0, + 1711.0, + 350.0, + 1749.0, + 293.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1711.0, + 1407.0, + 1711.0, + 1407.0, + 1749.0, + 401.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1741.0, + 1213.0, + 1741.0, + 1213.0, + 1779.0, + 295.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 1741.0, + 1344.0, + 1741.0, + 1344.0, + 1779.0, + 1274.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1741.0, + 1400.0, + 1741.0, + 1400.0, + 1779.0, + 1395.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1808.0, + 295.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1804.0, + 861.0, + 1804.0, + 861.0, + 1838.0, + 293.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1804.0, + 1404.0, + 1804.0, + 1404.0, + 1838.0, + 894.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1833.0, + 1140.0, + 1833.0, + 1140.0, + 1870.0, + 295.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1878.0, + 1402.0, + 1878.0, + 1402.0, + 1914.0, + 293.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1911.0, + 995.0, + 1911.0, + 995.0, + 1944.0, + 297.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1911.0, + 1344.0, + 1911.0, + 1344.0, + 1944.0, + 1031.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1911.0, + 1398.0, + 1911.0, + 1398.0, + 1944.0, + 1395.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1936.0, + 1407.0, + 1936.0, + 1407.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 1392.0, + 2000.0, + 1392.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 629.0, + 293.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 627.0, + 1404.0, + 627.0, + 1404.0, + 660.0, + 295.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 652.0, + 360.0, + 652.0, + 360.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 652.0, + 1404.0, + 652.0, + 1404.0, + 692.0, + 389.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 689.0, + 805.0, + 689.0, + 805.0, + 720.0, + 294.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 965.0, + 769.0, + 965.0, + 769.0, + 1003.0, + 295.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 965.0, + 1405.0, + 965.0, + 1405.0, + 1003.0, + 800.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 995.0, + 1404.0, + 995.0, + 1404.0, + 1034.0, + 293.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1025.0, + 1406.0, + 1025.0, + 1406.0, + 1068.0, + 292.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1059.0, + 1402.0, + 1059.0, + 1402.0, + 1095.0, + 293.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1090.0, + 680.0, + 1090.0, + 680.0, + 1126.0, + 295.0, + 1126.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 230, + 1403, + 230, + 1403, + 658, + 298, + 658 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1523, + 1405, + 1523, + 1405, + 1827, + 298, + 1827 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 673, + 1403, + 673, + 1403, + 826, + 298, + 826 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 300, + 859, + 1404, + 859, + 1404, + 1013, + 300, + 1013 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 301, + 1942, + 1401, + 1942, + 1401, + 2034, + 301, + 2034 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 368, + 1037, + 1404, + 1037, + 1404, + 1345, + 368, + 1345 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 1368, + 1403, + 1368, + 1403, + 1493, + 298, + 1493 + ], + "score": 0.971 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 301, + 1876, + 608, + 1876, + 608, + 1908, + 301, + 1908 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.786 + }, + { + "category_id": 13, + "poly": [ + 906, + 383, + 1014, + 383, + 1014, + 412, + 906, + 412 + ], + "score": 0.9, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 455, + 474, + 565, + 474, + 565, + 504, + 455, + 504 + ], + "score": 0.88, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 814, + 890, + 869, + 890, + 869, + 919, + 814, + 919 + ], + "score": 0.86, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1116, + 1211, + 1165, + 1211, + 1165, + 1240, + 1116, + 1240 + ], + "score": 0.84, + "latex": "> 1" + }, + { + "category_id": 13, + "poly": [ + 978, + 763, + 1027, + 763, + 1027, + 796, + 978, + 796 + ], + "score": 0.68, + "latex": "\\boxed { 1 0 }" + }, + { + "category_id": 13, + "poly": [ + 402, + 671, + 451, + 671, + 451, + 705, + 402, + 705 + ], + "score": 0.66, + "latex": "\\pmb { \\Vert 2 8 \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 789, + 1397, + 819, + 1397, + 819, + 1432, + 789, + 1432 + ], + "score": 0.65, + "latex": "\\boxed { 5 }" + }, + { + "category_id": 13, + "poly": [ + 1342, + 1239, + 1391, + 1239, + 1391, + 1272, + 1342, + 1272 + ], + "score": 0.64, + "latex": "| \\widehat { \\mathsf { B } } \\widehat { \\mathsf { S } } \\|" + }, + { + "category_id": 13, + "poly": [ + 556, + 290, + 586, + 290, + 586, + 327, + 556, + 327 + ], + "score": 0.62, + "latex": "^ { 1 , }" + }, + { + "category_id": 13, + "poly": [ + 924, + 1310, + 973, + 1310, + 973, + 1343, + 924, + 1343 + ], + "score": 0.54, + "latex": "\\textcircled { \\lVert { 3 0 } \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 1361, + 381, + 1393, + 381, + 1393, + 417, + 1361, + 417 + ], + "score": 0.48, + "latex": "\\mathrm { E } )" + }, + { + "category_id": 13, + "poly": [ + 1132, + 1310, + 1168, + 1310, + 1168, + 1347, + 1132, + 1347 + ], + "score": 0.39, + "latex": "{ \\bf D } )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1876.0, + 613.0, + 1876.0, + 613.0, + 1915.0, + 298.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 259.0, + 1405.0, + 259.0, + 1405.0, + 297.0, + 293.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 289.0, + 555.0, + 289.0, + 555.0, + 327.0, + 292.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 289.0, + 1405.0, + 289.0, + 1405.0, + 327.0, + 587.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 321.0, + 1405.0, + 321.0, + 1405.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 353.0, + 1407.0, + 353.0, + 1407.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 381.0, + 905.0, + 381.0, + 905.0, + 420.0, + 293.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 381.0, + 1360.0, + 381.0, + 1360.0, + 420.0, + 1015.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 381.0, + 1404.0, + 381.0, + 1404.0, + 420.0, + 1394.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 411.0, + 1405.0, + 411.0, + 1405.0, + 449.0, + 294.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 443.0, + 1405.0, + 443.0, + 1405.0, + 477.0, + 293.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 474.0, + 454.0, + 474.0, + 454.0, + 510.0, + 293.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 474.0, + 1405.0, + 474.0, + 1405.0, + 510.0, + 566.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 501.0, + 1405.0, + 501.0, + 1405.0, + 539.0, + 293.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 533.0, + 1402.0, + 533.0, + 1402.0, + 568.0, + 294.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 567.0, + 1404.0, + 567.0, + 1404.0, + 599.0, + 296.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 594.0, + 1407.0, + 594.0, + 1407.0, + 632.0, + 293.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 626.0, + 558.0, + 626.0, + 558.0, + 662.0, + 294.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1522.0, + 1405.0, + 1522.0, + 1405.0, + 1558.0, + 295.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1553.0, + 1406.0, + 1553.0, + 1406.0, + 1589.0, + 293.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1584.0, + 1405.0, + 1584.0, + 1405.0, + 1619.0, + 296.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1616.0, + 1405.0, + 1616.0, + 1405.0, + 1651.0, + 295.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1645.0, + 1406.0, + 1645.0, + 1406.0, + 1680.0, + 295.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1675.0, + 1405.0, + 1675.0, + 1405.0, + 1713.0, + 295.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1708.0, + 1402.0, + 1708.0, + 1402.0, + 1739.0, + 296.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1736.0, + 1405.0, + 1736.0, + 1405.0, + 1771.0, + 296.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1763.0, + 1405.0, + 1763.0, + 1405.0, + 1806.0, + 292.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1799.0, + 787.0, + 1799.0, + 787.0, + 1829.0, + 294.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 401.0, + 673.0, + 401.0, + 706.0, + 296.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 673.0, + 1404.0, + 673.0, + 1404.0, + 706.0, + 452.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 703.0, + 1404.0, + 703.0, + 1404.0, + 736.0, + 296.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 732.0, + 1400.0, + 732.0, + 1400.0, + 769.0, + 294.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 761.0, + 977.0, + 761.0, + 977.0, + 801.0, + 293.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 761.0, + 1407.0, + 761.0, + 1407.0, + 801.0, + 1028.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 794.0, + 675.0, + 794.0, + 675.0, + 830.0, + 295.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 859.0, + 1404.0, + 859.0, + 1404.0, + 892.0, + 298.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 890.0, + 813.0, + 890.0, + 813.0, + 923.0, + 298.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 890.0, + 1403.0, + 890.0, + 1403.0, + 923.0, + 870.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 915.0, + 1405.0, + 915.0, + 1405.0, + 960.0, + 291.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 951.0, + 1406.0, + 951.0, + 1406.0, + 988.0, + 295.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 980.0, + 836.0, + 980.0, + 836.0, + 1015.0, + 295.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1940.0, + 1407.0, + 1940.0, + 1407.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 1275.0, + 2000.0, + 1275.0, + 2041.0, + 295.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1036.0, + 1405.0, + 1036.0, + 1405.0, + 1073.0, + 367.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1069.0, + 1404.0, + 1069.0, + 1404.0, + 1102.0, + 395.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1099.0, + 1387.0, + 1099.0, + 1387.0, + 1133.0, + 396.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1138.0, + 1406.0, + 1138.0, + 1406.0, + 1174.0, + 388.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1168.0, + 1107.0, + 1168.0, + 1107.0, + 1206.0, + 393.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1210.0, + 1115.0, + 1210.0, + 1115.0, + 1244.0, + 368.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 1210.0, + 1404.0, + 1210.0, + 1404.0, + 1244.0, + 1166.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1241.0, + 1341.0, + 1241.0, + 1341.0, + 1273.0, + 396.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1392.0, + 1241.0, + 1397.0, + 1241.0, + 1397.0, + 1273.0, + 1392.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1283.0, + 1402.0, + 1283.0, + 1402.0, + 1315.0, + 368.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1313.0, + 923.0, + 1313.0, + 923.0, + 1346.0, + 394.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1313.0, + 1131.0, + 1313.0, + 1131.0, + 1346.0, + 974.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1366.0, + 1407.0, + 1366.0, + 1407.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1399.0, + 788.0, + 1399.0, + 788.0, + 1432.0, + 294.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1399.0, + 1405.0, + 1399.0, + 1405.0, + 1432.0, + 820.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1428.0, + 1405.0, + 1428.0, + 1405.0, + 1462.0, + 293.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1459.0, + 1371.0, + 1459.0, + 1371.0, + 1495.0, + 293.0, + 1495.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 300, + 228, + 488, + 228, + 488, + 261, + 300, + 261 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 866, + 2088, + 866, + 2113, + 836, + 2113 + ], + "score": 0.838 + }, + { + "category_id": 1, + "poly": [ + 295, + 164, + 1410, + 164, + 1410, + 2042, + 295, + 2042 + ], + "score": 0.806 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 491.0, + 227.0, + 491.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 225.0, + 494.0, + 225.0, + 494.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 279.0, + 1405.0, + 279.0, + 1405.0, + 316.0, + 304.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 310.0, + 1403.0, + 310.0, + 1403.0, + 349.0, + 350.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 343.0, + 876.0, + 343.0, + 876.0, + 376.0, + 355.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 388.0, + 1407.0, + 388.0, + 1407.0, + 431.0, + 302.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 425.0, + 964.0, + 425.0, + 964.0, + 458.0, + 355.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 473.0, + 1407.0, + 473.0, + 1407.0, + 512.0, + 306.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 503.0, + 1405.0, + 503.0, + 1405.0, + 543.0, + 353.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 553.0, + 1409.0, + 553.0, + 1409.0, + 598.0, + 306.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 586.0, + 1407.0, + 586.0, + 1407.0, + 625.0, + 353.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 617.0, + 857.0, + 617.0, + 857.0, + 656.0, + 355.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 668.0, + 1407.0, + 668.0, + 1407.0, + 707.0, + 308.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 699.0, + 1407.0, + 699.0, + 1407.0, + 736.0, + 353.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 728.0, + 1405.0, + 728.0, + 1405.0, + 767.0, + 355.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 761.0, + 552.0, + 761.0, + 552.0, + 796.0, + 348.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 812.0, + 1407.0, + 812.0, + 1407.0, + 852.0, + 306.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 845.0, + 1403.0, + 845.0, + 1403.0, + 878.0, + 355.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 895.0, + 1407.0, + 895.0, + 1407.0, + 934.0, + 306.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 924.0, + 1213.0, + 924.0, + 1213.0, + 963.0, + 353.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 975.0, + 1407.0, + 975.0, + 1407.0, + 1014.0, + 306.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1006.0, + 954.0, + 1006.0, + 954.0, + 1045.0, + 348.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1054.0, + 1409.0, + 1054.0, + 1409.0, + 1097.0, + 302.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1087.0, + 1403.0, + 1087.0, + 1403.0, + 1128.0, + 348.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1140.0, + 1405.0, + 1140.0, + 1405.0, + 1179.0, + 294.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1171.0, + 853.0, + 1171.0, + 853.0, + 1210.0, + 350.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1223.0, + 1407.0, + 1223.0, + 1407.0, + 1262.0, + 294.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1253.0, + 1325.0, + 1253.0, + 1325.0, + 1293.0, + 353.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1305.0, + 1407.0, + 1305.0, + 1407.0, + 1344.0, + 294.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1336.0, + 1312.0, + 1336.0, + 1312.0, + 1375.0, + 353.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1389.0, + 1405.0, + 1389.0, + 1405.0, + 1422.0, + 296.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1457.0, + 351.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1449.0, + 808.0, + 1449.0, + 808.0, + 1488.0, + 353.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1499.0, + 1405.0, + 1499.0, + 1405.0, + 1540.0, + 291.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1532.0, + 1392.0, + 1532.0, + 1392.0, + 1571.0, + 353.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1581.0, + 1405.0, + 1581.0, + 1405.0, + 1622.0, + 291.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1614.0, + 1405.0, + 1614.0, + 1405.0, + 1653.0, + 353.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1645.0, + 485.0, + 1645.0, + 485.0, + 1678.0, + 353.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1694.0, + 1405.0, + 1694.0, + 1405.0, + 1733.0, + 294.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1723.0, + 1376.0, + 1723.0, + 1376.0, + 1764.0, + 351.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1777.0, + 1407.0, + 1777.0, + 1407.0, + 1816.0, + 294.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1808.0, + 1198.0, + 1808.0, + 1198.0, + 1845.0, + 350.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1861.0, + 1405.0, + 1861.0, + 1405.0, + 1894.0, + 296.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1890.0, + 1407.0, + 1890.0, + 1407.0, + 1929.0, + 353.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1923.0, + 667.0, + 1923.0, + 667.0, + 1956.0, + 355.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2012.0, + 294.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 2003.0, + 1013.0, + 2003.0, + 1013.0, + 2043.0, + 350.0, + 2043.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 814, + 75, + 814, + 105, + 299, + 105 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 293, + 136, + 1410, + 136, + 1410, + 2051, + 293, + 2051 + ], + "score": 0.769 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2114, + 835, + 2114 + ], + "score": 0.746 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2114, + 835, + 2114 + ], + "score": 0.232 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1407.0, + 229.0, + 1407.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 262.0, + 1160.0, + 262.0, + 1160.0, + 298.0, + 349.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 319.0, + 1405.0, + 319.0, + 1405.0, + 353.0, + 298.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 344.0, + 1405.0, + 344.0, + 1405.0, + 388.0, + 351.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 380.0, + 566.0, + 380.0, + 566.0, + 413.0, + 357.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 430.0, + 1409.0, + 430.0, + 1409.0, + 474.0, + 292.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 464.0, + 1405.0, + 464.0, + 1405.0, + 504.0, + 353.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 518.0, + 1405.0, + 518.0, + 1405.0, + 558.0, + 296.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 550.0, + 1154.0, + 550.0, + 1154.0, + 588.0, + 351.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 607.0, + 1407.0, + 607.0, + 1407.0, + 644.0, + 292.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 640.0, + 1037.0, + 640.0, + 1037.0, + 674.0, + 357.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 688.0, + 1409.0, + 688.0, + 1409.0, + 737.0, + 292.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 724.0, + 1403.0, + 724.0, + 1403.0, + 758.0, + 355.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 756.0, + 1092.0, + 756.0, + 1092.0, + 789.0, + 355.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 808.0, + 1407.0, + 808.0, + 1407.0, + 850.0, + 294.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 842.0, + 1327.0, + 842.0, + 1327.0, + 882.0, + 355.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 896.0, + 1405.0, + 896.0, + 1405.0, + 936.0, + 296.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 930.0, + 818.0, + 930.0, + 818.0, + 963.0, + 355.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 982.0, + 1407.0, + 982.0, + 1407.0, + 1022.0, + 296.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1014.0, + 1407.0, + 1014.0, + 1407.0, + 1054.0, + 353.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1045.0, + 735.0, + 1045.0, + 735.0, + 1085.0, + 353.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1100.0, + 1407.0, + 1100.0, + 1407.0, + 1140.0, + 296.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1131.0, + 813.0, + 1131.0, + 813.0, + 1169.0, + 349.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1190.0, + 1403.0, + 1190.0, + 1403.0, + 1224.0, + 298.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1215.0, + 1403.0, + 1215.0, + 1403.0, + 1259.0, + 351.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1251.0, + 564.0, + 1251.0, + 564.0, + 1285.0, + 355.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1299.0, + 1409.0, + 1299.0, + 1409.0, + 1348.0, + 290.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1335.0, + 1340.0, + 1335.0, + 1340.0, + 1375.0, + 353.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1389.0, + 1407.0, + 1389.0, + 1407.0, + 1429.0, + 296.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1421.0, + 524.0, + 1421.0, + 524.0, + 1461.0, + 353.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1478.0, + 1407.0, + 1478.0, + 1407.0, + 1517.0, + 294.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1499.0, + 1409.0, + 1499.0, + 1409.0, + 1555.0, + 347.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1538.0, + 1350.0, + 1538.0, + 1350.0, + 1578.0, + 353.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1597.0, + 1405.0, + 1597.0, + 1405.0, + 1631.0, + 298.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1625.0, + 1130.0, + 1625.0, + 1130.0, + 1664.0, + 353.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 1407.0, + 1681.0, + 1407.0, + 1719.0, + 294.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1713.0, + 1405.0, + 1713.0, + 1405.0, + 1753.0, + 353.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1742.0, + 488.0, + 1742.0, + 488.0, + 1782.0, + 353.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1797.0, + 1403.0, + 1797.0, + 1403.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1830.0, + 925.0, + 1830.0, + 925.0, + 1864.0, + 353.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1878.0, + 1405.0, + 1878.0, + 1405.0, + 1929.0, + 290.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1916.0, + 1244.0, + 1916.0, + 1244.0, + 1956.0, + 353.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2011.0, + 296.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 2000.0, + 642.0, + 2000.0, + 642.0, + 2038.0, + 353.0, + 2038.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 294, + 228, + 1408, + 228, + 1408, + 718, + 294, + 718 + ], + "score": 0.656 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 230.0, + 1407.0, + 230.0, + 1407.0, + 268.0, + 296.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 300.0, + 353.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 294.0, + 854.0, + 294.0, + 854.0, + 328.0, + 355.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 343.0, + 1406.0, + 343.0, + 1406.0, + 383.0, + 293.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 376.0, + 1406.0, + 376.0, + 1406.0, + 410.0, + 354.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 407.0, + 828.0, + 407.0, + 828.0, + 441.0, + 355.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 458.0, + 1405.0, + 458.0, + 1405.0, + 493.0, + 295.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 485.0, + 1409.0, + 485.0, + 1409.0, + 531.0, + 351.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 520.0, + 564.0, + 520.0, + 564.0, + 552.0, + 358.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 569.0, + 1407.0, + 569.0, + 1407.0, + 611.0, + 293.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 602.0, + 980.0, + 602.0, + 980.0, + 640.0, + 355.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 652.0, + 1406.0, + 652.0, + 1406.0, + 694.0, + 295.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 687.0, + 771.0, + 687.0, + 771.0, + 720.0, + 354.0, + 720.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 970, + 1405, + 970, + 1405, + 1184, + 297, + 1184 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 294, + 1404, + 294, + 1404, + 481, + 298, + 481 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 1574, + 1405, + 1574, + 1405, + 1669, + 296, + 1669 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 288, + 639, + 1404, + 639, + 1404, + 709, + 288, + 709 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 296, + 1477, + 1402, + 1477, + 1402, + 1543, + 296, + 1543 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 290, + 1329, + 1401, + 1329, + 1401, + 1393, + 290, + 1393 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 296, + 1838, + 1415, + 1838, + 1415, + 1903, + 296, + 1903 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 297, + 825, + 1401, + 825, + 1401, + 888, + 297, + 888 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 654, + 1785, + 1047, + 1785, + 1047, + 1824, + 654, + 1824 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 596, + 1275, + 1100, + 1275, + 1100, + 1316, + 596, + 1316 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 716, + 580, + 980, + 580, + 980, + 621, + 716, + 621 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 297, + 498, + 1401, + 498, + 1401, + 562, + 297, + 562 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 565, + 903, + 1130, + 903, + 1130, + 943, + 565, + 943 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 741, + 783, + 955, + 783, + 955, + 818, + 741, + 818 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 293, + 722, + 1402, + 722, + 1402, + 782, + 293, + 782 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 713, + 1409, + 985, + 1409, + 985, + 1449, + 713, + 1449 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 298, + 1735, + 852, + 1735, + 852, + 1769, + 298, + 1769 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 549, + 1184, + 1147, + 1184, + 1147, + 1220, + 549, + 1220 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 299, + 1226, + 793, + 1226, + 793, + 1260, + 299, + 1260 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 814, + 74, + 814, + 105, + 299, + 105 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 299, + 225, + 1075, + 225, + 1075, + 263, + 299, + 263 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1366, + 907, + 1400, + 907, + 1400, + 937, + 1366, + 937 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1791, + 1400, + 1791, + 1400, + 1820, + 1366, + 1820 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.865 + }, + { + "category_id": 8, + "poly": [ + 573, + 1683, + 1121, + 1683, + 1121, + 1723, + 573, + 1723 + ], + "score": 0.738 + }, + { + "category_id": 13, + "poly": [ + 600, + 635, + 694, + 635, + 694, + 673, + 600, + 673 + ], + "score": 0.93, + "latex": "\\hat { h } _ { i } = \\bar { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 855, + 671, + 1000, + 671, + 1000, + 708, + 855, + 708 + ], + "score": 0.93, + "latex": "\\bar { h } _ { i } - \\hat { h } _ { i } < s _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1141, + 495, + 1233, + 495, + 1233, + 532, + 1141, + 532 + ], + "score": 0.93, + "latex": "\\hat { h } _ { i } \\leq \\bar { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 820, + 297, + 977, + 297, + 977, + 328, + 820, + 328 + ], + "score": 0.93, + "latex": "L _ { 1 } , L _ { 2 } , . . . , L _ { d }" + }, + { + "category_id": 14, + "poly": [ + 718, + 577, + 980, + 577, + 980, + 620, + 718, + 620 + ], + "score": 0.93, + "latex": "\\hat { h } _ { i } = s _ { i } \\cdot \\left( h _ { i } - 1 \\right) + k _ { i }" + }, + { + "category_id": 14, + "poly": [ + 595, + 1272, + 1101, + 1272, + 1101, + 1316, + 595, + 1316 + ], + "score": 0.93, + "latex": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } - ( 2 ^ { d } - 1 ) = 2 ^ { d } \\cdot ( h _ { d } - 1 ) + 1" + }, + { + "category_id": 14, + "poly": [ + 743, + 777, + 956, + 777, + 956, + 819, + 743, + 819 + ], + "score": 0.92, + "latex": "\\bar { h } _ { i } = h _ { i - 1 } + 2 \\cdot p _ { i }" + }, + { + "category_id": 14, + "poly": [ + 713, + 1405, + 987, + 1405, + 987, + 1448, + 713, + 1448 + ], + "score": 0.92, + "latex": "w _ { 0 } = 2 ^ { d } \\cdot ( w _ { d } - 1 ) + 1" + }, + { + "category_id": 13, + "poly": [ + 847, + 1478, + 1032, + 1478, + 1032, + 1509, + 847, + 1509 + ], + "score": 0.91, + "latex": "2 2 5 = 2 ^ { 5 } \\cdot 7 + 1" + }, + { + "category_id": 14, + "poly": [ + 652, + 1781, + 1048, + 1781, + 1048, + 1822, + 652, + 1822 + ], + "score": 0.91, + "latex": "h _ { 0 } = 2 ^ { d } \\cdot h _ { d } \\quad \\mathrm { a n d } \\quad w _ { 0 } = 2 ^ { d } \\cdot w _ { d }" + }, + { + "category_id": 13, + "poly": [ + 685, + 1096, + 805, + 1096, + 805, + 1126, + 685, + 1126 + ], + "score": 0.91, + "latex": "k _ { i } - 2 \\cdot p _ { i }" + }, + { + "category_id": 13, + "poly": [ + 706, + 1065, + 784, + 1065, + 784, + 1095, + 706, + 1095 + ], + "score": 0.9, + "latex": "p _ { i } = 1" + }, + { + "category_id": 13, + "poly": [ + 546, + 1362, + 611, + 1362, + 611, + 1389, + 546, + 1389 + ], + "score": 0.9, + "latex": "1 \\times 1" + }, + { + "category_id": 13, + "poly": [ + 857, + 1837, + 1006, + 1837, + 1006, + 1869, + 857, + 1869 + ], + "score": 0.9, + "latex": "2 2 4 = 2 ^ { 5 } \\cdot 7" + }, + { + "category_id": 13, + "poly": [ + 1061, + 1871, + 1126, + 1871, + 1126, + 1899, + 1061, + 1899 + ], + "score": 0.9, + "latex": "7 \\times 7" + }, + { + "category_id": 14, + "poly": [ + 567, + 903, + 1131, + 903, + 1131, + 942, + 567, + 942 + ], + "score": 0.89, + "latex": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = s _ { i } \\cdot ( h _ { i } - 1 ) + k _ { i } - 2 \\cdot p _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1079, + 1065, + 1145, + 1065, + 1145, + 1093, + 1079, + 1093 + ], + "score": 0.89, + "latex": "7 \\times 7" + }, + { + "category_id": 13, + "poly": [ + 1261, + 827, + 1291, + 827, + 1291, + 856, + 1261, + 856 + ], + "score": 0.89, + "latex": "L _ { i }" + }, + { + "category_id": 14, + "poly": [ + 549, + 1182, + 1147, + 1182, + 1147, + 1220, + 549, + 1220 + ], + "score": 0.89, + "latex": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 1 = 2 \\cdot h _ { i } - 1" + }, + { + "category_id": 13, + "poly": [ + 456, + 498, + 485, + 498, + 485, + 531, + 456, + 531 + ], + "score": 0.88, + "latex": "\\bar { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1331, + 404, + 1331, + 404, + 1361, + 371, + 1361 + ], + "score": 0.88, + "latex": "h _ { d }" + }, + { + "category_id": 13, + "poly": [ + 745, + 532, + 775, + 532, + 775, + 561, + 745, + 561 + ], + "score": 0.88, + "latex": "L _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1282, + 389, + 1314, + 389, + 1314, + 419, + 1282, + 419 + ], + "score": 0.88, + "latex": "h _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 900, + 359, + 928, + 359, + 928, + 388, + 900, + 388 + ], + "score": 0.88, + "latex": "k _ { i }" + }, + { + "category_id": 13, + "poly": [ + 893, + 1513, + 958, + 1513, + 958, + 1540, + 893, + 1540 + ], + "score": 0.88, + "latex": "8 \\times 8" + }, + { + "category_id": 13, + "poly": [ + 413, + 389, + 442, + 389, + 442, + 418, + 413, + 418 + ], + "score": 0.88, + "latex": "h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 325, + 1839, + 434, + 1839, + 434, + 1869, + 325, + 1869 + ], + "score": 0.87, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 1314, + 359, + 1345, + 359, + 1345, + 388, + 1314, + 388 + ], + "score": 0.87, + "latex": "L _ { i }" + }, + { + "category_id": 13, + "poly": [ + 858, + 502, + 888, + 502, + 888, + 531, + 858, + 531 + ], + "score": 0.87, + "latex": "L _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1077, + 389, + 1108, + 389, + 1108, + 418, + 1077, + 418 + ], + "score": 0.87, + "latex": "L _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1344, + 1034, + 1402, + 1034, + 1402, + 1064, + 1344, + 1064 + ], + "score": 0.87, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 392, + 1065, + 474, + 1065, + 474, + 1094, + 392, + 1094 + ], + "score": 0.87, + "latex": "k _ { i } = 3 ," + }, + { + "category_id": 13, + "poly": [ + 324, + 1479, + 432, + 1479, + 432, + 1510, + 324, + 1510 + ], + "score": 0.87, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 14, + "poly": [ + 574, + 1683, + 1123, + 1683, + 1123, + 1722, + 574, + 1722 + ], + "score": 0.87, + "latex": "\\forall i \\in [ 1 . . d ] : \\quad h _ { i - 1 } = 2 \\cdot ( h _ { i } - 1 ) + 2 = 2 \\cdot h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1140, + 974, + 1210, + 974, + 1210, + 1002, + 1140, + 1002 + ], + "score": 0.86, + "latex": "\\mathrm { : } d = 5" + }, + { + "category_id": 13, + "poly": [ + 493, + 391, + 526, + 391, + 526, + 418, + 493, + 418 + ], + "score": 0.86, + "latex": "w _ { i }" + }, + { + "category_id": 13, + "poly": [ + 372, + 830, + 399, + 830, + 399, + 858, + 372, + 858 + ], + "score": 0.85, + "latex": "p _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1108, + 1577, + 1184, + 1577, + 1184, + 1604, + 1108, + 1604 + ], + "score": 0.85, + "latex": "( d = 5" + }, + { + "category_id": 13, + "poly": [ + 823, + 362, + 849, + 362, + 849, + 388, + 823, + 388 + ], + "score": 0.85, + "latex": "s _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1365, + 392, + 1400, + 392, + 1400, + 419, + 1365, + 419 + ], + "score": 0.83, + "latex": "w _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 879, + 1092, + 906, + 1092, + 906, + 1129, + 879, + 1129 + ], + "score": 0.83, + "latex": "3" + }, + { + "category_id": 13, + "poly": [ + 547, + 298, + 566, + 298, + 566, + 324, + 547, + 324 + ], + "score": 0.72, + "latex": "d" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 224.0, + 1078.0, + 224.0, + 1078.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 972.0, + 1139.0, + 972.0, + 1139.0, + 1006.0, + 295.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 972.0, + 1405.0, + 972.0, + 1405.0, + 1006.0, + 1211.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1001.0, + 1407.0, + 1001.0, + 1407.0, + 1039.0, + 291.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1033.0, + 1343.0, + 1033.0, + 1343.0, + 1067.0, + 294.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1033.0, + 1406.0, + 1033.0, + 1406.0, + 1067.0, + 1403.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1064.0, + 391.0, + 1064.0, + 391.0, + 1098.0, + 295.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1064.0, + 705.0, + 1064.0, + 705.0, + 1098.0, + 475.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1064.0, + 1078.0, + 1064.0, + 1078.0, + 1098.0, + 785.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1064.0, + 1405.0, + 1064.0, + 1405.0, + 1098.0, + 1146.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1093.0, + 684.0, + 1093.0, + 684.0, + 1131.0, + 294.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1093.0, + 878.0, + 1093.0, + 878.0, + 1131.0, + 806.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1093.0, + 1405.0, + 1093.0, + 1405.0, + 1131.0, + 907.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1159.0, + 294.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1156.0, + 385.0, + 1156.0, + 385.0, + 1189.0, + 294.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 295.0, + 546.0, + 295.0, + 546.0, + 331.0, + 296.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 295.0, + 819.0, + 295.0, + 819.0, + 331.0, + 567.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 295.0, + 1406.0, + 295.0, + 1406.0, + 331.0, + 978.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 327.0, + 1405.0, + 327.0, + 1405.0, + 359.0, + 296.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 358.0, + 822.0, + 358.0, + 822.0, + 390.0, + 294.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 358.0, + 899.0, + 358.0, + 899.0, + 390.0, + 850.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 358.0, + 1313.0, + 358.0, + 1313.0, + 390.0, + 929.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 358.0, + 1405.0, + 358.0, + 1405.0, + 390.0, + 1346.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 388.0, + 412.0, + 388.0, + 412.0, + 423.0, + 294.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 388.0, + 492.0, + 388.0, + 492.0, + 423.0, + 443.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 388.0, + 1076.0, + 388.0, + 1076.0, + 423.0, + 527.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 388.0, + 1281.0, + 388.0, + 1281.0, + 423.0, + 1109.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 388.0, + 1364.0, + 388.0, + 1364.0, + 423.0, + 1315.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 388.0, + 1404.0, + 388.0, + 1404.0, + 423.0, + 1401.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 457.0, + 292.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 450.0, + 1164.0, + 450.0, + 1164.0, + 482.0, + 296.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1572.0, + 1107.0, + 1572.0, + 1107.0, + 1612.0, + 293.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 1572.0, + 1406.0, + 1572.0, + 1406.0, + 1612.0, + 1185.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1606.0, + 1405.0, + 1606.0, + 1405.0, + 1642.0, + 294.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1637.0, + 1330.0, + 1637.0, + 1330.0, + 1671.0, + 296.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 631.0, + 599.0, + 631.0, + 599.0, + 683.0, + 293.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 631.0, + 1406.0, + 631.0, + 1406.0, + 683.0, + 695.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 673.0, + 854.0, + 673.0, + 854.0, + 710.0, + 295.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 673.0, + 1206.0, + 673.0, + 1206.0, + 710.0, + 1001.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1475.0, + 323.0, + 1475.0, + 323.0, + 1515.0, + 294.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1475.0, + 846.0, + 1475.0, + 846.0, + 1515.0, + 433.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1475.0, + 1404.0, + 1475.0, + 1404.0, + 1515.0, + 1033.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1509.0, + 892.0, + 1509.0, + 892.0, + 1545.0, + 296.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 1509.0, + 970.0, + 1509.0, + 970.0, + 1545.0, + 959.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1328.0, + 370.0, + 1328.0, + 370.0, + 1364.0, + 297.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1328.0, + 1404.0, + 1328.0, + 1404.0, + 1364.0, + 405.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1360.0, + 545.0, + 1360.0, + 545.0, + 1394.0, + 293.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 1360.0, + 1090.0, + 1360.0, + 1090.0, + 1394.0, + 612.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1833.0, + 324.0, + 1833.0, + 324.0, + 1876.0, + 292.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1833.0, + 856.0, + 1833.0, + 856.0, + 1876.0, + 435.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1833.0, + 1408.0, + 1833.0, + 1408.0, + 1876.0, + 1007.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1868.0, + 1060.0, + 1868.0, + 1060.0, + 1904.0, + 294.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1868.0, + 1139.0, + 1868.0, + 1139.0, + 1904.0, + 1127.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 825.0, + 371.0, + 825.0, + 371.0, + 860.0, + 296.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 825.0, + 1260.0, + 825.0, + 1260.0, + 860.0, + 400.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 825.0, + 1403.0, + 825.0, + 1403.0, + 860.0, + 1292.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 855.0, + 1012.0, + 855.0, + 1012.0, + 891.0, + 293.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 498.0, + 455.0, + 498.0, + 455.0, + 534.0, + 296.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 498.0, + 857.0, + 498.0, + 857.0, + 534.0, + 486.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 498.0, + 1140.0, + 498.0, + 1140.0, + 534.0, + 889.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 498.0, + 1403.0, + 498.0, + 1403.0, + 534.0, + 1234.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 529.0, + 744.0, + 529.0, + 744.0, + 566.0, + 292.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 529.0, + 909.0, + 529.0, + 909.0, + 566.0, + 776.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 721.0, + 1405.0, + 721.0, + 1405.0, + 760.0, + 293.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 753.0, + 575.0, + 753.0, + 575.0, + 788.0, + 295.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1733.0, + 852.0, + 1733.0, + 852.0, + 1773.0, + 296.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1225.0, + 795.0, + 1225.0, + 795.0, + 1265.0, + 297.0, + 1265.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 479, + 1502, + 1215, + 1502, + 1215, + 1827, + 479, + 1827 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 374, + 1405, + 374, + 1405, + 465, + 299, + 465 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 489, + 493, + 1214, + 493, + 1214, + 822, + 489, + 822 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 301, + 991, + 1400, + 991, + 1400, + 1217, + 301, + 1217 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 300, + 296, + 1399, + 296, + 1399, + 359, + 300, + 359 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 294, + 1349, + 1407, + 1349, + 1407, + 1473, + 294, + 1473 + ], + "score": 0.939 + }, + { + "category_id": 4, + "poly": [ + 296, + 848, + 1403, + 848, + 1403, + 943, + 296, + 943 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.902 + }, + { + "category_id": 4, + "poly": [ + 295, + 1853, + 1406, + 1853, + 1406, + 2009, + 295, + 2009 + ], + "score": 0.877 + }, + { + "category_id": 0, + "poly": [ + 298, + 226, + 1128, + 226, + 1128, + 262, + 298, + 262 + ], + "score": 0.86 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.854 + }, + { + "category_id": 4, + "poly": [ + 295, + 1244, + 1402, + 1244, + 1402, + 1309, + 295, + 1309 + ], + "score": 0.809 + }, + { + "category_id": 1, + "poly": [ + 295, + 1244, + 1402, + 1244, + 1402, + 1309, + 295, + 1309 + ], + "score": 0.166 + }, + { + "category_id": 1, + "poly": [ + 295, + 1853, + 1406, + 1853, + 1406, + 2009, + 295, + 2009 + ], + "score": 0.164 + }, + { + "category_id": 13, + "poly": [ + 548, + 1916, + 661, + 1916, + 661, + 1945, + 548, + 1945 + ], + "score": 0.9, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 316, + 1885, + 435, + 1885, + 435, + 1914, + 316, + 1914 + ], + "score": 0.9, + "latex": "2 5 7 \\times 2 5 7" + }, + { + "category_id": 13, + "poly": [ + 1250, + 406, + 1314, + 406, + 1314, + 434, + 1250, + 434 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 297, + 879, + 406, + 879, + 406, + 909, + 297, + 909 + ], + "score": 0.87, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1502.0, + 700.0, + 1502.0, + 700.0, + 1528.0, + 589.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1504.0, + 1104.0, + 1504.0, + 1104.0, + 1528.0, + 994.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1530.0, + 1198.0, + 1530.0, + 1198.0, + 1537.0, + 1187.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1551.0, + 813.0, + 1551.0, + 813.0, + 1558.0, + 802.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1545.0, + 1212.0, + 1545.0, + 1212.0, + 1556.0, + 1198.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1587.0, + 816.0, + 1587.0, + 816.0, + 1598.0, + 798.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1586.0, + 898.0, + 1586.0, + 898.0, + 1597.0, + 883.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1585.0, + 1211.0, + 1585.0, + 1211.0, + 1592.0, + 1200.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1625.0, + 816.0, + 1625.0, + 816.0, + 1635.0, + 799.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1621.0, + 1212.0, + 1621.0, + 1212.0, + 1632.0, + 1199.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1641.0, + 496.0, + 1641.0, + 496.0, + 1648.0, + 483.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1641.0, + 897.0, + 1641.0, + 897.0, + 1651.0, + 881.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1662.0, + 816.0, + 1662.0, + 816.0, + 1673.0, + 799.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1660.0, + 1211.0, + 1660.0, + 1211.0, + 1668.0, + 1200.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1695.0, + 495.0, + 1695.0, + 495.0, + 1703.0, + 482.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1700.0, + 816.0, + 1700.0, + 816.0, + 1710.0, + 798.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1697.0, + 1210.0, + 1697.0, + 1210.0, + 1706.0, + 1200.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1737.0, + 815.0, + 1737.0, + 815.0, + 1748.0, + 799.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1736.0, + 1210.0, + 1736.0, + 1210.0, + 1744.0, + 1199.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1749.0, + 896.0, + 1749.0, + 896.0, + 1762.0, + 881.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1776.0, + 814.0, + 1776.0, + 814.0, + 1784.0, + 802.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1774.0, + 1211.0, + 1774.0, + 1211.0, + 1781.0, + 1200.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1817.0, + 614.0, + 1817.0, + 614.0, + 1825.0, + 601.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1817.0, + 668.0, + 1817.0, + 668.0, + 1825.0, + 655.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1817.0, + 777.0, + 1817.0, + 777.0, + 1824.0, + 766.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1815.0, + 1015.0, + 1815.0, + 1015.0, + 1826.0, + 998.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1815.0, + 1067.0, + 1815.0, + 1067.0, + 1826.0, + 1052.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1817.0, + 1121.0, + 1817.0, + 1121.0, + 1825.0, + 1108.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1817.0, + 1176.0, + 1817.0, + 1176.0, + 1825.0, + 1165.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1695.0, + 898.0, + 1695.0, + 898.0, + 1706.5, + 879.0, + 1706.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 496.0, + 696.0, + 496.0, + 696.0, + 519.0, + 613.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 496.0, + 1089.0, + 496.0, + 1089.0, + 519.0, + 999.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 529.0, + 804.0, + 529.0, + 804.0, + 536.0, + 793.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 545.0, + 819.0, + 545.0, + 819.0, + 553.0, + 808.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 559.0, + 1208.0, + 559.0, + 1208.0, + 567.0, + 1197.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 563.0, + 897.0, + 563.0, + 897.0, + 570.0, + 885.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 582.0, + 822.0, + 582.0, + 822.0, + 593.0, + 803.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 594.0, + 896.0, + 594.0, + 896.0, + 601.0, + 886.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 608.0, + 1209.0, + 608.0, + 1209.0, + 616.0, + 1197.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 619.0, + 821.0, + 619.0, + 821.0, + 630.0, + 804.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 622.0, + 899.0, + 622.0, + 899.0, + 633.0, + 883.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 654.0, + 505.0, + 654.0, + 505.0, + 662.0, + 493.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 656.0, + 821.0, + 656.0, + 821.0, + 667.0, + 804.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 652.0, + 897.0, + 652.0, + 897.0, + 664.0, + 882.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 657.0, + 1208.0, + 657.0, + 1208.0, + 666.0, + 1198.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 684.0, + 503.0, + 684.0, + 503.0, + 693.0, + 493.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 683.0, + 897.0, + 683.0, + 897.0, + 694.0, + 881.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 695.0, + 821.0, + 695.0, + 821.0, + 706.0, + 804.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 708.0, + 1208.0, + 708.0, + 1208.0, + 715.0, + 1197.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 715.0, + 504.0, + 715.0, + 504.0, + 722.0, + 493.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 713.0, + 897.0, + 713.0, + 897.0, + 725.0, + 881.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 732.0, + 820.0, + 732.0, + 820.0, + 743.0, + 804.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 744.0, + 506.0, + 744.0, + 506.0, + 754.0, + 491.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 743.0, + 897.0, + 743.0, + 897.0, + 755.0, + 881.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 758.0, + 1208.0, + 758.0, + 1208.0, + 766.0, + 1197.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 771.0, + 819.0, + 771.0, + 819.0, + 779.0, + 807.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 776.0, + 504.0, + 776.0, + 504.0, + 783.0, + 494.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 776.0, + 895.0, + 776.0, + 895.0, + 783.0, + 883.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 811.0, + 544.0, + 811.0, + 544.0, + 818.0, + 533.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 810.0, + 575.0, + 810.0, + 575.0, + 818.0, + 564.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 811.0, + 607.0, + 811.0, + 607.0, + 818.0, + 595.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 809.0, + 639.0, + 809.0, + 639.0, + 820.0, + 622.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 809.0, + 669.0, + 809.0, + 669.0, + 820.0, + 652.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 809.0, + 701.0, + 809.0, + 701.0, + 820.0, + 683.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 809.0, + 731.0, + 809.0, + 731.0, + 819.0, + 713.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 811.0, + 758.0, + 811.0, + 758.0, + 818.0, + 746.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 811.0, + 933.0, + 811.0, + 933.0, + 818.0, + 925.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 811.0, + 965.0, + 811.0, + 965.0, + 818.0, + 954.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 810.0, + 995.0, + 810.0, + 995.0, + 818.0, + 984.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 809.0, + 1029.0, + 809.0, + 1029.0, + 820.0, + 1012.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 809.0, + 1060.0, + 809.0, + 1060.0, + 820.0, + 1041.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 809.0, + 1091.0, + 809.0, + 1091.0, + 819.0, + 1072.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 809.0, + 1120.0, + 809.0, + 1120.0, + 820.0, + 1101.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 809.0, + 1151.0, + 809.0, + 1151.0, + 820.0, + 1134.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 998.0, + 457.0, + 998.0, + 457.0, + 1028.0, + 353.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 998.0, + 679.0, + 998.0, + 679.0, + 1028.0, + 577.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 998.0, + 903.0, + 998.0, + 903.0, + 1028.0, + 801.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 998.0, + 1113.0, + 998.0, + 1113.0, + 1028.0, + 1032.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 998.0, + 1335.0, + 998.0, + 1335.0, + 1028.0, + 1255.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 849.0, + 1406.0, + 849.0, + 1406.0, + 883.0, + 295.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 875.0, + 296.0, + 875.0, + 296.0, + 914.0, + 293.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 875.0, + 1405.0, + 875.0, + 1405.0, + 914.0, + 407.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 909.0, + 1336.0, + 909.0, + 1336.0, + 945.0, + 293.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1854.0, + 1402.0, + 1854.0, + 1402.0, + 1887.0, + 295.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 315.0, + 1883.0, + 315.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1883.0, + 1404.0, + 1883.0, + 1404.0, + 1920.0, + 436.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 547.0, + 1912.0, + 547.0, + 1952.0, + 293.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1912.0, + 1406.0, + 1912.0, + 1406.0, + 1952.0, + 662.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1982.0, + 294.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1976.0, + 1139.0, + 1976.0, + 1139.0, + 2014.0, + 295.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 225.0, + 1130.0, + 225.0, + 1130.0, + 265.0, + 293.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1244.0, + 1404.0, + 1244.0, + 1404.0, + 1280.0, + 295.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1278.0, + 1069.0, + 1278.0, + 1069.0, + 1310.0, + 297.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 374.0, + 1402.0, + 374.0, + 1402.0, + 408.0, + 294.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 403.0, + 1249.0, + 403.0, + 1249.0, + 439.0, + 293.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 403.0, + 1404.0, + 403.0, + 1404.0, + 439.0, + 1315.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 436.0, + 490.0, + 436.0, + 490.0, + 467.0, + 295.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 293.0, + 1404.0, + 293.0, + 1404.0, + 331.0, + 296.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 326.0, + 1123.0, + 326.0, + 1123.0, + 362.0, + 295.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1349.0, + 1355.0, + 1349.0, + 1355.0, + 1385.0, + 293.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1381.0, + 1405.0, + 1381.0, + 1405.0, + 1413.0, + 295.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1411.0, + 1406.0, + 1411.0, + 1406.0, + 1447.0, + 295.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1443.0, + 984.0, + 1443.0, + 984.0, + 1475.0, + 296.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1244.0, + 1404.0, + 1244.0, + 1404.0, + 1280.0, + 295.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1278.0, + 1069.0, + 1278.0, + 1069.0, + 1310.0, + 297.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1854.0, + 1402.0, + 1854.0, + 1402.0, + 1887.0, + 295.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 315.0, + 1883.0, + 315.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1883.0, + 1404.0, + 1883.0, + 1404.0, + 1920.0, + 436.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 547.0, + 1912.0, + 547.0, + 1952.0, + 293.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1912.0, + 1406.0, + 1912.0, + 1406.0, + 1952.0, + 662.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1982.0, + 294.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1976.0, + 1139.0, + 1976.0, + 1139.0, + 2014.0, + 295.0, + 2014.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 295, + 1405, + 295, + 1405, + 510, + 296, + 510 + ], + "score": 0.981 + }, + { + "category_id": 3, + "poly": [ + 326, + 1113, + 1378, + 1113, + 1378, + 1378, + 326, + 1378 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 1579, + 1404, + 1579, + 1404, + 1674, + 298, + 1674 + ], + "score": 0.965 + }, + { + "category_id": 3, + "poly": [ + 389, + 1733, + 1309, + 1733, + 1309, + 1937, + 389, + 1937 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 297, + 804, + 798, + 804, + 798, + 916, + 297, + 916 + ], + "score": 0.954 + }, + { + "category_id": 4, + "poly": [ + 297, + 1403, + 1404, + 1403, + 1404, + 1527, + 297, + 1527 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 299, + 225, + 1266, + 225, + 1266, + 263, + 299, + 263 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.895 + }, + { + "category_id": 4, + "poly": [ + 299, + 1963, + 1404, + 1963, + 1404, + 2028, + 299, + 2028 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.856 + }, + { + "category_id": 1, + "poly": [ + 296, + 941, + 1405, + 941, + 1405, + 1065, + 296, + 1065 + ], + "score": 0.781 + }, + { + "category_id": 4, + "poly": [ + 902, + 804, + 1403, + 804, + 1403, + 916, + 902, + 916 + ], + "score": 0.766 + }, + { + "category_id": 3, + "poly": [ + 308, + 543, + 786, + 543, + 786, + 789, + 308, + 789 + ], + "score": 0.755 + }, + { + "category_id": 4, + "poly": [ + 296, + 941, + 1405, + 941, + 1405, + 1065, + 296, + 1065 + ], + "score": 0.388 + }, + { + "category_id": 3, + "poly": [ + 920, + 540, + 1385, + 540, + 1385, + 789, + 920, + 789 + ], + "score": 0.249 + }, + { + "category_id": 1, + "poly": [ + 902, + 804, + 1403, + 804, + 1403, + 916, + 902, + 916 + ], + "score": 0.206 + }, + { + "category_id": 13, + "poly": [ + 1138, + 942, + 1248, + 942, + 1248, + 971, + 1138, + 971 + ], + "score": 0.88, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 831, + 1994, + 940, + 1994, + 940, + 2023, + 831, + 2023 + ], + "score": 0.88, + "latex": "2 2 5 \\times 2 2 5" + }, + { + "category_id": 13, + "poly": [ + 1042, + 1403, + 1152, + 1403, + 1152, + 1433, + 1042, + 1433 + ], + "score": 0.88, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 367, + 1611, + 437, + 1611, + 437, + 1640, + 367, + 1640 + ], + "score": 0.88, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 1303, + 802, + 1402, + 802, + 1402, + 831, + 1303, + 831 + ], + "score": 0.86, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 697, + 802, + 797, + 802, + 797, + 831, + 697, + 831 + ], + "score": 0.86, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 903, + 858, + 986, + 858, + 986, + 885, + 903, + 885 + ], + "score": 0.85, + "latex": "7 6 . 1 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 297, + 858, + 381, + 858, + 381, + 885, + 297, + 885 + ], + "score": 0.85, + "latex": "6 9 . 9 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 297, + 887, + 381, + 887, + 381, + 914, + 297, + 914 + ], + "score": 0.85, + "latex": "7 0 . 2 8 \\%" + }, + { + "category_id": 13, + "poly": [ + 902, + 887, + 986, + 887, + 986, + 914, + 902, + 914 + ], + "score": 0.85, + "latex": "7 6 . 6 1 \\%" + }, + { + "category_id": 13, + "poly": [ + 1265, + 1001, + 1314, + 1001, + 1314, + 1034, + 1265, + 1034 + ], + "score": 0.74, + "latex": "\\pmb { \\left. \\pmb { \\left. \\bar { 2 3 } \\right. } \\right. }" + }, + { + "category_id": 13, + "poly": [ + 589, + 1432, + 638, + 1432, + 638, + 1465, + 589, + 1465 + ], + "score": 0.32, + "latex": "\\mathbb { \\lVert 2 3 \\rVert }" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1112.0, + 615.0, + 1112.0, + 615.0, + 1147.0, + 515.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1113.0, + 1173.0, + 1113.0, + 1173.0, + 1145.0, + 1063.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1734.0, + 658.0, + 1734.0, + 658.0, + 1763.0, + 394.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1733.0, + 978.0, + 1733.0, + 978.0, + 1763.0, + 716.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1731.0, + 1303.0, + 1731.0, + 1303.0, + 1763.0, + 1031.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1402.0, + 1041.0, + 1402.0, + 1041.0, + 1438.0, + 295.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1402.0, + 1405.0, + 1402.0, + 1405.0, + 1438.0, + 1153.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1432.0, + 588.0, + 1432.0, + 588.0, + 1469.0, + 294.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1432.0, + 1406.0, + 1432.0, + 1406.0, + 1469.0, + 639.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1461.0, + 1405.0, + 1461.0, + 1405.0, + 1504.0, + 292.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1494.0, + 1404.0, + 1494.0, + 1404.0, + 1530.0, + 292.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 223.0, + 1272.0, + 223.0, + 1272.0, + 266.0, + 292.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1962.0, + 1406.0, + 1962.0, + 1406.0, + 1998.0, + 295.0, + 1998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1992.0, + 830.0, + 1992.0, + 830.0, + 2028.0, + 295.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1992.0, + 1352.0, + 1992.0, + 1352.0, + 2028.0, + 941.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 802.0, + 1302.0, + 802.0, + 1302.0, + 832.0, + 899.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 829.0, + 1404.0, + 829.0, + 1404.0, + 863.0, + 899.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 855.0, + 1405.0, + 855.0, + 1405.0, + 890.0, + 987.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 885.0, + 1334.0, + 885.0, + 1334.0, + 918.0, + 987.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 542.0, + 473.0, + 542.0, + 473.0, + 572.0, + 379.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 540.0, + 747.0, + 540.0, + 747.0, + 573.0, + 602.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 940.0, + 1137.0, + 940.0, + 1137.0, + 976.0, + 294.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 940.0, + 1405.0, + 940.0, + 1405.0, + 976.0, + 1249.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1005.0, + 295.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1004.0, + 1264.0, + 1004.0, + 1264.0, + 1037.0, + 296.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1035.0, + 1315.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1031.0, + 1209.0, + 1031.0, + 1209.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 536.0, + 1075.0, + 536.0, + 1075.0, + 571.0, + 979.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 541.0, + 1329.0, + 541.0, + 1329.0, + 566.0, + 1225.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 293.0, + 1405.0, + 293.0, + 1405.0, + 334.0, + 291.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 325.0, + 1406.0, + 325.0, + 1406.0, + 363.0, + 293.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 356.0, + 1405.0, + 356.0, + 1405.0, + 393.0, + 293.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 388.0, + 1405.0, + 388.0, + 1405.0, + 422.0, + 295.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 416.0, + 1405.0, + 416.0, + 1405.0, + 453.0, + 294.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 449.0, + 1402.0, + 449.0, + 1402.0, + 480.0, + 296.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 479.0, + 905.0, + 479.0, + 905.0, + 513.0, + 296.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1577.0, + 1406.0, + 1577.0, + 1406.0, + 1618.0, + 294.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1611.0, + 366.0, + 1611.0, + 366.0, + 1645.0, + 294.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1611.0, + 1405.0, + 1611.0, + 1405.0, + 1645.0, + 438.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1637.0, + 1102.0, + 1637.0, + 1102.0, + 1680.0, + 293.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 802.0, + 696.0, + 802.0, + 696.0, + 832.0, + 294.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 829.0, + 800.0, + 829.0, + 800.0, + 863.0, + 294.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 855.0, + 799.0, + 855.0, + 799.0, + 890.0, + 382.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 882.0, + 776.0, + 882.0, + 776.0, + 921.0, + 382.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 940.0, + 1137.0, + 940.0, + 1137.0, + 976.0, + 294.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 940.0, + 1405.0, + 940.0, + 1405.0, + 976.0, + 1249.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1005.0, + 295.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1004.0, + 1264.0, + 1004.0, + 1264.0, + 1037.0, + 296.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1035.0, + 1315.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1031.0, + 1209.0, + 1031.0, + 1209.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 802.0, + 1302.0, + 802.0, + 1302.0, + 832.0, + 899.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 829.0, + 1404.0, + 829.0, + 1404.0, + 863.0, + 899.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 855.0, + 1405.0, + 855.0, + 1405.0, + 890.0, + 987.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 885.0, + 1334.0, + 885.0, + 1334.0, + 918.0, + 987.0, + 918.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 305, + 1064, + 1401, + 1064, + 1401, + 1277, + 305, + 1277 + ], + "score": 0.966 + }, + { + "category_id": 3, + "poly": [ + 299, + 744, + 1397, + 744, + 1397, + 956, + 299, + 956 + ], + "score": 0.965 + }, + { + "category_id": 3, + "poly": [ + 305, + 1387, + 1400, + 1387, + 1400, + 1600, + 305, + 1600 + ], + "score": 0.964 + }, + { + "category_id": 3, + "poly": [ + 303, + 1708, + 1400, + 1708, + 1400, + 1919, + 303, + 1919 + ], + "score": 0.963 + }, + { + "category_id": 3, + "poly": [ + 303, + 397, + 1402, + 397, + 1402, + 607, + 303, + 607 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 299, + 296, + 1403, + 296, + 1403, + 361, + 299, + 361 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 298, + 632, + 1401, + 632, + 1401, + 695, + 298, + 695 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 292, + 225, + 1317, + 225, + 1317, + 263, + 292, + 263 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.902 + }, + { + "category_id": 4, + "poly": [ + 309, + 1628, + 1381, + 1628, + 1381, + 1662, + 309, + 1662 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.866 + }, + { + "category_id": 4, + "poly": [ + 297, + 1304, + 1397, + 1304, + 1397, + 1338, + 297, + 1338 + ], + "score": 0.851 + }, + { + "category_id": 4, + "poly": [ + 346, + 979, + 1347, + 979, + 1347, + 1013, + 346, + 1013 + ], + "score": 0.846 + }, + { + "category_id": 1, + "poly": [ + 297, + 1943, + 1400, + 1943, + 1400, + 2007, + 297, + 2007 + ], + "score": 0.644 + }, + { + "category_id": 4, + "poly": [ + 297, + 1943, + 1400, + 1943, + 1400, + 2007, + 297, + 2007 + ], + "score": 0.356 + }, + { + "category_id": 4, + "poly": [ + 339, + 979, + 1349, + 979, + 1349, + 1013, + 339, + 1013 + ], + "score": 0.131 + }, + { + "category_id": 13, + "poly": [ + 558, + 1943, + 608, + 1943, + 608, + 1975, + 558, + 1975 + ], + "score": 0.72, + "latex": "\\mathbb { \\left[ \\left. 2 3 \\right] \\right. }" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1069.0, + 402.0, + 1069.0, + 402.0, + 1087.0, + 302.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1069.0, + 539.0, + 1069.0, + 539.0, + 1087.0, + 441.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1066.0, + 679.0, + 1066.0, + 679.0, + 1089.0, + 577.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1069.0, + 817.0, + 1069.0, + 817.0, + 1087.0, + 717.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1070.0, + 953.0, + 1070.0, + 953.0, + 1086.0, + 855.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1069.0, + 1092.0, + 1069.0, + 1092.0, + 1087.0, + 993.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1069.0, + 1229.0, + 1069.0, + 1229.0, + 1087.0, + 1130.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1070.0, + 1366.0, + 1070.0, + 1366.0, + 1086.0, + 1270.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1098.0, + 431.0, + 1098.0, + 431.0, + 1113.0, + 411.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1092.0, + 574.0, + 1092.0, + 574.0, + 1123.0, + 547.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1082.0, + 703.0, + 1082.0, + 703.0, + 1118.0, + 685.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1093.0, + 841.0, + 1093.0, + 841.0, + 1125.0, + 822.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1095.0, + 976.0, + 1095.0, + 976.0, + 1124.0, + 965.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1090.0, + 1121.0, + 1090.0, + 1121.0, + 1123.0, + 1100.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1087.0, + 1259.0, + 1087.0, + 1259.0, + 1122.0, + 1238.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1090.0, + 1395.0, + 1090.0, + 1395.0, + 1119.0, + 1375.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1118.0, + 431.0, + 1118.0, + 431.0, + 1131.0, + 411.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1122.0, + 574.0, + 1122.0, + 574.0, + 1139.0, + 547.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1124.0, + 705.0, + 1124.0, + 705.0, + 1139.0, + 686.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1129.0, + 841.0, + 1129.0, + 841.0, + 1144.0, + 822.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1125.0, + 1121.0, + 1125.0, + 1121.0, + 1144.0, + 1097.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1129.0, + 1258.0, + 1129.0, + 1258.0, + 1144.0, + 1238.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1122.0, + 1395.0, + 1122.0, + 1395.0, + 1135.0, + 1376.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1135.0, + 432.0, + 1135.0, + 432.0, + 1150.0, + 411.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1137.0, + 574.0, + 1137.0, + 574.0, + 1155.0, + 547.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1135.0, + 978.0, + 1135.0, + 978.0, + 1144.0, + 965.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1139.0, + 1396.0, + 1139.0, + 1396.0, + 1152.0, + 1376.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1155.0, + 431.0, + 1155.0, + 431.0, + 1169.0, + 411.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1152.0, + 573.0, + 1152.0, + 573.0, + 1169.0, + 547.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1146.0, + 703.0, + 1146.0, + 703.0, + 1159.0, + 685.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1147.0, + 844.0, + 1147.0, + 844.0, + 1162.0, + 826.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1153.0, + 978.0, + 1153.0, + 978.0, + 1164.0, + 965.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1146.0, + 1120.0, + 1146.0, + 1120.0, + 1159.0, + 1098.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1150.0, + 1258.0, + 1150.0, + 1258.0, + 1164.0, + 1237.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1156.0, + 1395.0, + 1156.0, + 1395.0, + 1169.0, + 1377.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1169.0, + 401.0, + 1169.0, + 401.0, + 1185.0, + 305.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1165.0, + 541.0, + 1165.0, + 541.0, + 1188.0, + 441.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1167.0, + 679.0, + 1167.0, + 679.0, + 1186.0, + 579.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1167.0, + 817.0, + 1167.0, + 817.0, + 1186.0, + 717.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1165.0, + 950.0, + 1165.0, + 950.0, + 1188.0, + 860.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1165.0, + 1087.0, + 1165.0, + 1087.0, + 1188.0, + 997.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 1165.0, + 1226.0, + 1165.0, + 1226.0, + 1188.0, + 1135.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 1169.0, + 1361.0, + 1169.0, + 1361.0, + 1185.0, + 1275.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1190.0, + 427.0, + 1190.0, + 427.0, + 1204.0, + 411.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1186.0, + 565.0, + 1186.0, + 565.0, + 1200.0, + 547.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1194.0, + 702.0, + 1194.0, + 702.0, + 1204.0, + 687.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1189.0, + 844.0, + 1189.0, + 844.0, + 1201.0, + 824.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1189.0, + 1120.0, + 1189.0, + 1120.0, + 1202.0, + 1100.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1191.0, + 1254.0, + 1191.0, + 1254.0, + 1206.0, + 1238.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1211.0, + 428.0, + 1211.0, + 428.0, + 1245.0, + 409.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1202.0, + 570.0, + 1202.0, + 570.0, + 1252.0, + 545.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1217.0, + 703.0, + 1217.0, + 703.0, + 1254.0, + 686.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1204.0, + 844.0, + 1204.0, + 844.0, + 1246.0, + 824.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1200.0, + 979.0, + 1200.0, + 979.0, + 1235.0, + 963.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1204.0, + 1120.0, + 1204.0, + 1120.0, + 1248.0, + 1098.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1208.0, + 1254.0, + 1208.0, + 1254.0, + 1257.0, + 1238.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1200.0, + 1396.0, + 1200.0, + 1396.0, + 1255.0, + 1376.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1254.0, + 426.0, + 1254.0, + 426.0, + 1263.0, + 412.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1257.0, + 569.0, + 1257.0, + 569.0, + 1270.0, + 547.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1250.0, + 976.0, + 1250.0, + 976.0, + 1260.0, + 965.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1250.0, + 1120.0, + 1250.0, + 1120.0, + 1263.0, + 1100.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1265.0, + 1252.0, + 1265.0, + 1252.0, + 1273.0, + 1239.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 752.0, + 409.0, + 752.0, + 409.0, + 768.0, + 300.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 752.0, + 545.0, + 752.0, + 545.0, + 768.0, + 437.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 752.0, + 682.0, + 752.0, + 682.0, + 768.0, + 573.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 752.0, + 818.0, + 752.0, + 818.0, + 768.0, + 711.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 752.0, + 954.0, + 752.0, + 954.0, + 768.0, + 847.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 752.0, + 1091.0, + 752.0, + 1091.0, + 768.0, + 984.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 752.0, + 1227.0, + 752.0, + 1227.0, + 768.0, + 1120.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 750.0, + 1366.0, + 750.0, + 1366.0, + 768.0, + 1255.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 762.0, + 437.0, + 762.0, + 437.0, + 851.0, + 409.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 773.0, + 580.0, + 773.0, + 580.0, + 851.0, + 544.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 766.0, + 706.0, + 766.0, + 706.0, + 841.0, + 678.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 774.0, + 847.0, + 774.0, + 847.0, + 844.0, + 817.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 772.0, + 979.0, + 772.0, + 979.0, + 850.0, + 954.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 771.0, + 1121.0, + 771.0, + 1121.0, + 843.0, + 1091.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 767.0, + 1256.0, + 767.0, + 1256.0, + 845.0, + 1227.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 767.0, + 1393.0, + 767.0, + 1393.0, + 846.0, + 1362.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 850.0, + 409.0, + 850.0, + 409.0, + 866.0, + 301.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 849.0, + 546.0, + 849.0, + 546.0, + 868.0, + 436.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 849.0, + 683.0, + 849.0, + 683.0, + 868.0, + 572.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 849.0, + 819.0, + 849.0, + 819.0, + 868.0, + 709.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 849.0, + 949.0, + 849.0, + 949.0, + 868.0, + 853.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 849.0, + 1086.0, + 849.0, + 1086.0, + 868.0, + 989.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 843.0, + 1119.0, + 843.0, + 1119.0, + 860.0, + 1093.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 849.0, + 1222.0, + 849.0, + 1222.0, + 868.0, + 1125.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 849.0, + 1358.0, + 849.0, + 1358.0, + 868.0, + 1264.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 871.0, + 432.0, + 871.0, + 432.0, + 888.0, + 407.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 866.0, + 570.0, + 866.0, + 570.0, + 883.0, + 542.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 870.0, + 706.0, + 870.0, + 706.0, + 887.0, + 681.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 868.0, + 847.0, + 868.0, + 847.0, + 885.0, + 819.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 870.0, + 1120.0, + 870.0, + 1120.0, + 887.0, + 1093.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 871.0, + 1252.0, + 871.0, + 1252.0, + 885.0, + 1231.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 882.0, + 395.0, + 882.0, + 395.0, + 949.0, + 315.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 891.0, + 434.0, + 891.0, + 434.0, + 908.0, + 409.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 884.0, + 575.0, + 884.0, + 575.0, + 919.0, + 545.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 895.0, + 706.0, + 895.0, + 706.0, + 913.0, + 681.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 882.0, + 847.0, + 882.0, + 847.0, + 913.0, + 819.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 880.0, + 980.0, + 880.0, + 980.0, + 921.0, + 955.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 884.0, + 1120.0, + 884.0, + 1120.0, + 917.0, + 1091.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 889.0, + 1251.0, + 889.0, + 1251.0, + 921.0, + 1231.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 879.0, + 1393.0, + 879.0, + 1393.0, + 916.0, + 1365.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 913.0, + 431.0, + 913.0, + 431.0, + 945.0, + 410.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 921.0, + 575.0, + 921.0, + 575.0, + 938.0, + 545.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 921.0, + 706.0, + 921.0, + 706.0, + 938.0, + 680.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 913.0, + 847.0, + 913.0, + 847.0, + 943.0, + 818.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 929.0, + 979.0, + 929.0, + 979.0, + 943.0, + 957.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 917.0, + 1120.0, + 917.0, + 1120.0, + 948.0, + 1093.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 923.0, + 1251.0, + 923.0, + 1251.0, + 938.0, + 1231.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 922.0, + 1393.0, + 922.0, + 1393.0, + 939.0, + 1362.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 940.0, + 572.0, + 940.0, + 572.0, + 954.0, + 545.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 941.0, + 1251.0, + 941.0, + 1251.0, + 955.0, + 1230.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1396.0, + 402.0, + 1396.0, + 402.0, + 1412.0, + 302.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1396.0, + 539.0, + 1396.0, + 539.0, + 1412.0, + 440.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1396.0, + 678.0, + 1396.0, + 678.0, + 1412.0, + 580.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1396.0, + 816.0, + 1396.0, + 816.0, + 1412.0, + 718.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1396.0, + 955.0, + 1396.0, + 955.0, + 1412.0, + 856.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1396.0, + 1091.0, + 1396.0, + 1091.0, + 1412.0, + 993.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1396.0, + 1230.0, + 1396.0, + 1230.0, + 1412.0, + 1131.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1397.0, + 1365.0, + 1397.0, + 1365.0, + 1409.0, + 1271.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1418.0, + 431.0, + 1418.0, + 431.0, + 1431.0, + 411.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1415.0, + 574.0, + 1415.0, + 574.0, + 1432.0, + 549.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1412.0, + 702.0, + 1412.0, + 702.0, + 1421.0, + 686.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1423.0, + 840.0, + 1423.0, + 840.0, + 1432.0, + 824.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1419.0, + 976.0, + 1419.0, + 976.0, + 1432.0, + 960.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1418.0, + 1120.0, + 1418.0, + 1120.0, + 1431.0, + 1099.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1412.0, + 1258.0, + 1412.0, + 1258.0, + 1425.0, + 1237.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1414.0, + 1389.0, + 1414.0, + 1389.0, + 1424.0, + 1379.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1432.0, + 431.0, + 1432.0, + 431.0, + 1447.0, + 412.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1430.0, + 574.0, + 1430.0, + 574.0, + 1447.0, + 546.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1430.0, + 703.0, + 1430.0, + 703.0, + 1443.0, + 686.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1437.0, + 841.0, + 1437.0, + 841.0, + 1451.0, + 824.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1438.0, + 978.0, + 1438.0, + 978.0, + 1452.0, + 961.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1435.0, + 1120.0, + 1435.0, + 1120.0, + 1448.0, + 1100.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1432.0, + 1258.0, + 1432.0, + 1258.0, + 1447.0, + 1238.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1429.0, + 1394.0, + 1429.0, + 1394.0, + 1442.0, + 1377.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1448.0, + 433.0, + 1448.0, + 433.0, + 1479.0, + 409.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1445.0, + 574.0, + 1445.0, + 574.0, + 1476.0, + 546.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1449.0, + 703.0, + 1449.0, + 703.0, + 1463.0, + 686.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1457.0, + 839.0, + 1457.0, + 839.0, + 1467.0, + 825.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1458.0, + 979.0, + 1458.0, + 979.0, + 1473.0, + 961.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1452.0, + 1120.0, + 1452.0, + 1120.0, + 1467.0, + 1100.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1453.0, + 1258.0, + 1453.0, + 1258.0, + 1468.0, + 1237.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1446.0, + 1395.0, + 1446.0, + 1395.0, + 1476.0, + 1377.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1478.0, + 431.0, + 1478.0, + 431.0, + 1492.0, + 411.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1476.0, + 572.0, + 1476.0, + 572.0, + 1490.0, + 548.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1469.0, + 703.0, + 1469.0, + 703.0, + 1482.0, + 686.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1473.0, + 845.0, + 1473.0, + 845.0, + 1486.0, + 825.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1478.0, + 979.0, + 1478.0, + 979.0, + 1492.0, + 962.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1470.0, + 1120.0, + 1470.0, + 1120.0, + 1485.0, + 1099.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1475.0, + 1258.0, + 1475.0, + 1258.0, + 1490.0, + 1237.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1480.0, + 1395.0, + 1480.0, + 1395.0, + 1493.0, + 1377.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1492.0, + 399.0, + 1492.0, + 399.0, + 1508.0, + 304.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1493.0, + 539.0, + 1493.0, + 539.0, + 1509.0, + 443.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1490.0, + 678.0, + 1490.0, + 678.0, + 1513.0, + 578.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1492.0, + 816.0, + 1492.0, + 816.0, + 1512.0, + 717.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1492.0, + 950.0, + 1492.0, + 950.0, + 1512.0, + 860.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1492.0, + 1088.0, + 1492.0, + 1088.0, + 1512.0, + 998.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 1491.0, + 1225.0, + 1491.0, + 1225.0, + 1513.0, + 1135.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 1491.0, + 1362.0, + 1491.0, + 1362.0, + 1513.0, + 1272.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1511.0, + 427.0, + 1511.0, + 427.0, + 1525.0, + 411.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1509.0, + 565.0, + 1509.0, + 565.0, + 1524.0, + 546.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1518.0, + 701.0, + 1518.0, + 701.0, + 1528.0, + 687.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1511.0, + 844.0, + 1511.0, + 844.0, + 1524.0, + 824.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1516.0, + 1120.0, + 1516.0, + 1120.0, + 1530.0, + 1099.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1516.0, + 1255.0, + 1516.0, + 1255.0, + 1530.0, + 1237.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1512.0, + 1395.0, + 1512.0, + 1395.0, + 1525.0, + 1377.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1533.0, + 427.0, + 1533.0, + 427.0, + 1546.0, + 409.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1529.0, + 567.0, + 1529.0, + 567.0, + 1542.0, + 548.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1525.0, + 845.0, + 1525.0, + 845.0, + 1540.0, + 824.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1525.0, + 977.0, + 1525.0, + 977.0, + 1536.0, + 965.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1530.0, + 1120.0, + 1530.0, + 1120.0, + 1545.0, + 1099.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1535.0, + 1252.0, + 1535.0, + 1252.0, + 1545.0, + 1238.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1530.0, + 1395.0, + 1530.0, + 1395.0, + 1544.0, + 1377.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1553.0, + 427.0, + 1553.0, + 427.0, + 1568.0, + 411.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1549.0, + 569.0, + 1549.0, + 569.0, + 1561.0, + 549.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1545.0, + 702.0, + 1545.0, + 702.0, + 1555.0, + 688.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1541.0, + 845.0, + 1541.0, + 845.0, + 1571.0, + 824.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1547.0, + 979.0, + 1547.0, + 979.0, + 1561.0, + 962.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1545.0, + 1120.0, + 1545.0, + 1120.0, + 1573.0, + 1099.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1551.0, + 1255.0, + 1551.0, + 1255.0, + 1564.0, + 1237.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1550.0, + 1394.0, + 1550.0, + 1394.0, + 1560.0, + 1378.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1577.0, + 426.0, + 1577.0, + 426.0, + 1586.0, + 412.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1567.0, + 569.0, + 1567.0, + 569.0, + 1580.0, + 548.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1571.0, + 701.0, + 1571.0, + 701.0, + 1579.0, + 691.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1574.0, + 843.0, + 1574.0, + 843.0, + 1583.0, + 829.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1571.0, + 978.0, + 1571.0, + 978.0, + 1584.0, + 962.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1574.0, + 1120.0, + 1574.0, + 1120.0, + 1588.0, + 1100.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1569.0, + 1255.0, + 1569.0, + 1255.0, + 1583.0, + 1237.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1566.0, + 1395.0, + 1566.0, + 1395.0, + 1579.0, + 1375.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1588.0, + 567.0, + 1588.0, + 567.0, + 1596.0, + 551.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1584.0, + 1395.0, + 1584.0, + 1395.0, + 1597.0, + 1377.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1716.0, + 413.0, + 1716.0, + 413.0, + 1732.0, + 304.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1716.0, + 549.0, + 1716.0, + 549.0, + 1732.0, + 441.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1716.0, + 686.0, + 1716.0, + 686.0, + 1732.0, + 576.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1716.0, + 822.0, + 1716.0, + 822.0, + 1732.0, + 714.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1716.0, + 957.0, + 1716.0, + 957.0, + 1732.0, + 850.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1716.0, + 1095.0, + 1716.0, + 1095.0, + 1732.0, + 986.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1716.0, + 1231.0, + 1716.0, + 1231.0, + 1732.0, + 1122.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1716.0, + 1366.0, + 1716.0, + 1366.0, + 1732.0, + 1259.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1737.0, + 440.0, + 1737.0, + 440.0, + 1754.0, + 414.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1738.0, + 572.0, + 1738.0, + 572.0, + 1755.0, + 546.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1733.0, + 709.0, + 1733.0, + 709.0, + 1750.0, + 682.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1739.0, + 845.0, + 1739.0, + 845.0, + 1756.0, + 817.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1742.0, + 985.0, + 1742.0, + 985.0, + 1755.0, + 960.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1744.0, + 1122.0, + 1744.0, + 1122.0, + 1758.0, + 1097.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1735.0, + 1254.0, + 1735.0, + 1254.0, + 1752.0, + 1229.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 1736.0, + 1391.0, + 1736.0, + 1391.0, + 1753.0, + 1365.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1752.0, + 441.0, + 1752.0, + 441.0, + 1814.0, + 413.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1762.0, + 576.0, + 1762.0, + 576.0, + 1798.0, + 547.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1755.0, + 713.0, + 1755.0, + 713.0, + 1815.0, + 682.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1760.0, + 848.0, + 1760.0, + 848.0, + 1815.0, + 817.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1765.0, + 987.0, + 1765.0, + 987.0, + 1803.0, + 957.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1767.0, + 1122.0, + 1767.0, + 1122.0, + 1804.0, + 1097.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1761.0, + 1254.0, + 1761.0, + 1254.0, + 1804.0, + 1229.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 1754.0, + 1396.0, + 1754.0, + 1396.0, + 1807.0, + 1366.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1814.0, + 411.0, + 1814.0, + 411.0, + 1829.0, + 305.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1814.0, + 549.0, + 1814.0, + 549.0, + 1829.0, + 440.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1807.0, + 576.0, + 1807.0, + 576.0, + 1821.0, + 553.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1814.0, + 684.0, + 1814.0, + 684.0, + 1829.0, + 577.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1814.0, + 822.0, + 1814.0, + 822.0, + 1829.0, + 713.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1814.0, + 957.0, + 1814.0, + 957.0, + 1829.0, + 850.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1812.0, + 1085.0, + 1812.0, + 1085.0, + 1828.0, + 992.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1812.0, + 1224.0, + 1812.0, + 1224.0, + 1832.0, + 1128.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1814.0, + 1360.0, + 1814.0, + 1360.0, + 1829.0, + 1265.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1840.0, + 439.0, + 1840.0, + 439.0, + 1900.0, + 409.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1826.0, + 576.0, + 1826.0, + 576.0, + 1911.0, + 545.0, + 1911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1837.0, + 718.0, + 1837.0, + 718.0, + 1908.0, + 686.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1831.0, + 850.0, + 1831.0, + 850.0, + 1918.0, + 822.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1826.0, + 990.0, + 1826.0, + 990.0, + 1913.0, + 956.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1837.0, + 1127.0, + 1837.0, + 1127.0, + 1914.0, + 1096.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1834.0, + 1254.0, + 1834.0, + 1254.0, + 1906.0, + 1233.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 1826.0, + 1400.0, + 1826.0, + 1400.0, + 1906.0, + 1366.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 404.0, + 413.0, + 404.0, + 413.0, + 420.0, + 305.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 404.0, + 550.0, + 404.0, + 550.0, + 420.0, + 443.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 404.0, + 686.0, + 404.0, + 686.0, + 420.0, + 579.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 404.0, + 824.0, + 404.0, + 824.0, + 420.0, + 716.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 404.0, + 960.0, + 404.0, + 960.0, + 420.0, + 853.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 404.0, + 1097.0, + 404.0, + 1097.0, + 420.0, + 990.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 404.0, + 1233.0, + 404.0, + 1233.0, + 420.0, + 1127.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 404.0, + 1370.0, + 404.0, + 1370.0, + 420.0, + 1263.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 415.0, + 444.0, + 415.0, + 444.0, + 502.0, + 413.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 425.0, + 584.0, + 425.0, + 584.0, + 502.0, + 552.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 417.0, + 713.0, + 417.0, + 713.0, + 492.0, + 685.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 426.0, + 851.0, + 426.0, + 851.0, + 494.0, + 823.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 423.0, + 985.0, + 423.0, + 985.0, + 501.0, + 961.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 422.0, + 1127.0, + 422.0, + 1127.0, + 492.0, + 1099.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 418.0, + 1263.0, + 418.0, + 1263.0, + 497.0, + 1236.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 421.0, + 1400.0, + 421.0, + 1400.0, + 500.0, + 1368.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 500.0, + 415.0, + 500.0, + 415.0, + 519.0, + 305.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 500.0, + 552.0, + 500.0, + 552.0, + 519.0, + 442.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 500.0, + 689.0, + 500.0, + 689.0, + 519.0, + 579.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 500.0, + 825.0, + 500.0, + 825.0, + 519.0, + 716.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 500.0, + 954.0, + 500.0, + 954.0, + 519.0, + 859.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 500.0, + 1092.0, + 500.0, + 1092.0, + 519.0, + 996.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 500.0, + 1228.0, + 500.0, + 1228.0, + 519.0, + 1131.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 498.0, + 1365.0, + 498.0, + 1365.0, + 520.0, + 1268.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 523.0, + 437.0, + 523.0, + 437.0, + 537.0, + 417.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 518.0, + 468.0, + 518.0, + 468.0, + 526.0, + 455.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 514.0, + 576.0, + 514.0, + 576.0, + 535.0, + 540.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 515.0, + 606.0, + 515.0, + 606.0, + 529.0, + 590.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 520.0, + 711.0, + 520.0, + 711.0, + 537.0, + 686.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 519.0, + 853.0, + 519.0, + 853.0, + 536.0, + 826.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 520.0, + 1126.0, + 520.0, + 1126.0, + 537.0, + 1100.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 523.0, + 1258.0, + 523.0, + 1258.0, + 536.0, + 1237.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 543.0, + 437.0, + 543.0, + 437.0, + 558.0, + 415.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 535.0, + 580.0, + 535.0, + 580.0, + 570.0, + 550.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 546.0, + 711.0, + 546.0, + 711.0, + 564.0, + 686.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 533.0, + 853.0, + 533.0, + 853.0, + 564.0, + 825.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 531.0, + 986.0, + 531.0, + 986.0, + 572.0, + 961.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 535.0, + 1126.0, + 535.0, + 1126.0, + 567.0, + 1099.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 540.0, + 1258.0, + 540.0, + 1258.0, + 571.0, + 1238.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 531.0, + 1398.0, + 531.0, + 1398.0, + 566.0, + 1373.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 564.0, + 437.0, + 564.0, + 437.0, + 596.0, + 417.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 571.0, + 580.0, + 571.0, + 580.0, + 604.0, + 553.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 573.0, + 710.0, + 573.0, + 710.0, + 587.0, + 686.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 565.0, + 851.0, + 565.0, + 851.0, + 593.0, + 828.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 579.0, + 985.0, + 579.0, + 985.0, + 593.0, + 964.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 566.0, + 1127.0, + 566.0, + 1127.0, + 598.0, + 1097.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 575.0, + 1258.0, + 575.0, + 1258.0, + 603.0, + 1238.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 575.0, + 1398.0, + 575.0, + 1398.0, + 588.0, + 1373.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.75, + 456.0, + 1125.75, + 456.0, + 1125.75, + 474.0, + 1097.75, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 493.0, + 1124.0, + 493.0, + 1124.0, + 509.0, + 1099.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 632.0, + 1403.0, + 632.0, + 1403.0, + 667.0, + 296.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 662.0, + 667.0, + 662.0, + 667.0, + 698.0, + 296.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 1321.0, + 221.0, + 1321.0, + 268.0, + 291.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1624.0, + 1383.0, + 1624.0, + 1383.0, + 1663.0, + 314.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 871.0, + 2085.0, + 871.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1300.0, + 1400.0, + 1300.0, + 1400.0, + 1339.0, + 297.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 977.0, + 1350.0, + 977.0, + 1350.0, + 1016.0, + 349.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 557.0, + 1939.0, + 557.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1939.0, + 1405.0, + 1939.0, + 1405.0, + 1981.0, + 609.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1972.0, + 852.0, + 1972.0, + 852.0, + 2010.0, + 292.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 977.0, + 1350.0, + 977.0, + 1350.0, + 1016.0, + 349.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 1404.0, + 292.0, + 1404.0, + 335.0, + 294.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 329.0, + 998.0, + 329.0, + 998.0, + 363.0, + 294.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 557.0, + 1939.0, + 557.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1939.0, + 1405.0, + 1939.0, + 1405.0, + 1981.0, + 609.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1972.0, + 852.0, + 1972.0, + 852.0, + 2010.0, + 292.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 362, + 1423, + 1326, + 1423, + 1326, + 1919, + 362, + 1919 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 312, + 838, + 1379, + 838, + 1379, + 1293, + 312, + 1293 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 298, + 220, + 1403, + 220, + 1403, + 511, + 298, + 511 + ], + "score": 0.966 + }, + { + "category_id": 4, + "poly": [ + 296, + 533, + 1406, + 533, + 1406, + 628, + 296, + 628 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 296, + 746, + 1407, + 746, + 1407, + 811, + 296, + 811 + ], + "score": 0.929 + }, + { + "category_id": 4, + "poly": [ + 293, + 1947, + 1405, + 1947, + 1405, + 2013, + 293, + 2013 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 298, + 680, + 1209, + 680, + 1209, + 715, + 298, + 715 + ], + "score": 0.914 + }, + { + "category_id": 4, + "poly": [ + 294, + 1314, + 1402, + 1314, + 1402, + 1380, + 294, + 1380 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.862 + }, + { + "category_id": 4, + "poly": [ + 292, + 1314, + 1403, + 1314, + 1403, + 1380, + 292, + 1380 + ], + "score": 0.153 + }, + { + "category_id": 13, + "poly": [ + 1017, + 1949, + 1127, + 1949, + 1127, + 1978, + 1017, + 1978 + ], + "score": 0.86, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1315, + 1128, + 1315, + 1128, + 1345, + 1018, + 1345 + ], + "score": 0.85, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 375, + 563, + 424, + 563, + 424, + 596, + 375, + 596 + ], + "score": 0.31, + "latex": "\\textcircled { \\lvert 3 0 \\rvert }" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1427.0, + 1123.0, + 1427.0, + 1123.0, + 1457.0, + 715.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1521.0, + 388.0, + 1521.0, + 388.0, + 1651.0, + 364.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1670.0, + 608.0, + 1670.0, + 608.0, + 1692.0, + 526.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1670.0, + 1083.0, + 1670.0, + 1083.0, + 1691.0, + 994.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1705.0, + 388.0, + 1705.0, + 388.0, + 1858.0, + 364.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1888.0, + 1141.0, + 1888.0, + 1141.0, + 1921.0, + 732.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1885.5, + 711.0, + 1885.5, + 711.0, + 1895.5, + 703.0, + 1895.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 838.0, + 1012.0, + 838.0, + 1012.0, + 869.0, + 622.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 899.0, + 337.0, + 899.0, + 337.0, + 1022.0, + 318.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1046.0, + 493.0, + 1046.0, + 493.0, + 1069.0, + 395.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1047.0, + 720.0, + 1047.0, + 720.0, + 1067.0, + 642.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1044.0, + 929.0, + 1044.0, + 929.0, + 1068.0, + 850.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1044.0, + 1219.0, + 1044.0, + 1219.0, + 1068.0, + 1134.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1090.0, + 337.0, + 1090.0, + 337.0, + 1234.0, + 318.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1264.0, + 890.0, + 1264.0, + 890.0, + 1295.0, + 504.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 849.5, + 614.0, + 849.5, + 614.0, + 865.5, + 575.0, + 865.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.75, + 1266.0, + 920.75, + 1266.0, + 920.75, + 1274.0, + 896.75, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 229.0, + 409.0, + 229.0, + 409.0, + 245.0, + 301.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 229.0, + 547.0, + 229.0, + 547.0, + 245.0, + 438.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 229.0, + 832.0, + 229.0, + 832.0, + 245.0, + 709.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 226.0, + 961.0, + 226.0, + 961.0, + 246.0, + 843.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 229.0, + 1091.0, + 229.0, + 1091.0, + 245.0, + 987.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 226.0, + 1231.0, + 226.0, + 1231.0, + 246.0, + 1124.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 232.0, + 1255.0, + 232.0, + 1255.0, + 241.0, + 1240.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 229.0, + 1365.0, + 229.0, + 1365.0, + 245.0, + 1260.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 234.0, + 1376.0, + 234.0, + 1376.0, + 240.0, + 1370.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 231.0, + 1402.0, + 231.0, + 1402.0, + 243.0, + 1379.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 247.0, + 1123.0, + 247.0, + 1123.0, + 262.0, + 1108.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 253.0, + 1255.0, + 253.0, + 1255.0, + 262.0, + 1245.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 249.0, + 1402.0, + 249.0, + 1402.0, + 262.0, + 1379.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 265.0, + 576.0, + 265.0, + 576.0, + 275.0, + 562.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 263.0, + 714.0, + 263.0, + 714.0, + 271.0, + 699.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 258.0, + 1124.0, + 258.0, + 1124.0, + 284.0, + 1107.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 268.0, + 1402.0, + 268.0, + 1402.0, + 281.0, + 1379.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 280.0, + 576.0, + 280.0, + 576.0, + 289.0, + 563.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 286.0, + 1400.0, + 286.0, + 1400.0, + 298.0, + 1379.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 295.0, + 576.0, + 295.0, + 576.0, + 304.0, + 563.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 295.0, + 1122.0, + 295.0, + 1122.0, + 306.0, + 1109.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 308.0, + 576.0, + 308.0, + 576.0, + 319.0, + 563.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 304.0, + 1400.0, + 304.0, + 1400.0, + 317.0, + 1379.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 320.0, + 577.0, + 320.0, + 577.0, + 335.0, + 562.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 314.0, + 715.0, + 314.0, + 715.0, + 339.0, + 700.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 313.0, + 1123.0, + 313.0, + 1123.0, + 331.0, + 1109.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 317.0, + 1255.0, + 317.0, + 1255.0, + 325.0, + 1243.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 323.0, + 1398.0, + 323.0, + 1398.0, + 336.0, + 1379.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 337.0, + 1255.0, + 337.0, + 1255.0, + 347.0, + 1243.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 344.0, + 733.0, + 344.0, + 733.0, + 359.0, + 696.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 345.0, + 1119.0, + 345.0, + 1119.0, + 351.0, + 1113.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 348.0, + 597.0, + 348.0, + 597.0, + 364.0, + 561.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 376.0, + 406.0, + 376.0, + 406.0, + 392.0, + 304.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 375.0, + 547.0, + 375.0, + 547.0, + 395.0, + 427.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 376.0, + 682.0, + 376.0, + 682.0, + 392.0, + 574.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 376.0, + 819.0, + 376.0, + 819.0, + 392.0, + 712.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 373.0, + 960.0, + 373.0, + 960.0, + 396.0, + 846.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 376.0, + 1094.0, + 376.0, + 1094.0, + 392.0, + 985.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 374.0, + 1233.0, + 374.0, + 1233.0, + 393.0, + 1100.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 376.0, + 1365.0, + 376.0, + 1365.0, + 392.0, + 1260.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 387.0, + 707.0, + 387.0, + 707.0, + 397.0, + 696.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 398.0, + 440.0, + 398.0, + 440.0, + 413.0, + 424.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 394.0, + 466.0, + 394.0, + 466.0, + 406.0, + 443.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 394.0, + 715.0, + 394.0, + 715.0, + 473.0, + 690.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 394.0, + 1078.0, + 394.0, + 1078.0, + 416.0, + 1004.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 403.0, + 1129.0, + 403.0, + 1129.0, + 412.0, + 1108.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 407.0, + 405.0, + 407.0, + 405.0, + 419.0, + 395.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 408.0, + 441.0, + 408.0, + 441.0, + 434.0, + 422.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 411.0, + 1395.0, + 411.0, + 1395.0, + 425.0, + 1379.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 420.0, + 406.0, + 420.0, + 406.0, + 444.0, + 394.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 422.0, + 569.0, + 422.0, + 569.0, + 430.0, + 559.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 423.0, + 1125.0, + 423.0, + 1125.0, + 431.0, + 1110.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 430.0, + 439.0, + 430.0, + 439.0, + 449.0, + 425.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 434.0, + 1393.0, + 434.0, + 1393.0, + 445.0, + 1380.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 441.0, + 982.0, + 441.0, + 982.0, + 451.0, + 972.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 442.0, + 1123.0, + 442.0, + 1123.0, + 453.0, + 1110.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 453.0, + 436.0, + 453.0, + 436.0, + 464.0, + 427.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1381.0, + 457.0, + 1393.0, + 457.0, + 1393.0, + 467.0, + 1381.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 472.0, + 439.0, + 472.0, + 439.0, + 480.0, + 427.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 466.0, + 707.0, + 466.0, + 707.0, + 478.0, + 697.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 479.0, + 1393.0, + 479.0, + 1393.0, + 489.0, + 1380.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 225.0, + 686.0, + 225.0, + 686.0, + 248.0, + 555.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 391.5, + 403.0, + 391.5, + 403.0, + 403.5, + 369.0, + 403.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 1404.0, + 532.0, + 1404.0, + 570.0, + 294.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 565.0, + 374.0, + 565.0, + 374.0, + 599.0, + 294.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 565.0, + 1404.0, + 565.0, + 1404.0, + 599.0, + 425.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 595.0, + 1222.0, + 595.0, + 1222.0, + 629.0, + 294.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1016.0, + 1942.0, + 1016.0, + 1987.0, + 292.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1987.0, + 1128.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 689.0, + 1973.0, + 689.0, + 2016.0, + 294.0, + 2016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 677.0, + 1213.0, + 677.0, + 1213.0, + 719.0, + 292.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1309.0, + 1017.0, + 1309.0, + 1017.0, + 1354.0, + 293.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1309.0, + 1405.0, + 1309.0, + 1405.0, + 1354.0, + 1129.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1340.0, + 690.0, + 1340.0, + 690.0, + 1383.0, + 293.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1308.0, + 1017.0, + 1308.0, + 1017.0, + 1354.0, + 293.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1308.0, + 1405.0, + 1308.0, + 1405.0, + 1354.0, + 1129.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 687.0, + 1344.0, + 687.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 744.0, + 1405.0, + 744.0, + 1405.0, + 784.0, + 295.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 775.0, + 1163.0, + 775.0, + 1163.0, + 817.0, + 295.0, + 817.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 512, + 1405, + 512, + 1405, + 665, + 299, + 665 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 298, + 373, + 1404, + 373, + 1404, + 498, + 298, + 498 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 298, + 296, + 1397, + 296, + 1397, + 358, + 298, + 358 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 419, + 1209, + 1401, + 1209, + 1401, + 1416, + 419, + 1416 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 304, + 680, + 1351, + 680, + 1351, + 712, + 304, + 712 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 371, + 1425, + 1403, + 1425, + 1403, + 1487, + 371, + 1487 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.847 + }, + { + "category_id": 0, + "poly": [ + 298, + 227, + 1065, + 227, + 1065, + 262, + 298, + 262 + ], + "score": 0.793 + }, + { + "category_id": 1, + "poly": [ + 366, + 739, + 1405, + 739, + 1405, + 1201, + 366, + 1201 + ], + "score": 0.629 + }, + { + "category_id": 1, + "poly": [ + 371, + 1497, + 1404, + 1497, + 1404, + 1663, + 371, + 1663 + ], + "score": 0.456 + }, + { + "category_id": 13, + "poly": [ + 1024, + 1499, + 1090, + 1499, + 1090, + 1527, + 1024, + 1527 + ], + "score": 0.89, + "latex": "5 \\times 5" + }, + { + "category_id": 13, + "poly": [ + 972, + 741, + 1037, + 741, + 1037, + 769, + 972, + 769 + ], + "score": 0.89, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1600, + 1092, + 1600, + 1092, + 1629, + 1025, + 1629 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1019, + 883, + 1090, + 883, + 1090, + 912, + 1019, + 912 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 730, + 1529, + 787, + 1529, + 787, + 1557, + 730, + 1557 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 961, + 812, + 1029, + 812, + 1029, + 841, + 961, + 841 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1111, + 1426, + 1176, + 1426, + 1176, + 1455, + 1111, + 1455 + ], + "score": 0.87, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 451, + 1210, + 518, + 1210, + 518, + 1240, + 451, + 1240 + ], + "score": 0.86, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 451, + 1384, + 518, + 1384, + 518, + 1413, + 451, + 1413 + ], + "score": 0.86, + "latex": "4 \\times 4" + }, + { + "category_id": 13, + "poly": [ + 452, + 1318, + 519, + 1318, + 519, + 1347, + 452, + 1347 + ], + "score": 0.83, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 451, + 1246, + 518, + 1246, + 518, + 1276, + 451, + 1276 + ], + "score": 0.83, + "latex": "5 \\times 5" + }, + { + "category_id": 13, + "poly": [ + 451, + 1282, + 517, + 1282, + 517, + 1312, + 451, + 1312 + ], + "score": 0.8, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 770, + 1346, + 819, + 1346, + 819, + 1379, + 770, + 1379 + ], + "score": 0.27, + "latex": "\\pm \\amalg" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 223.0, + 1068.0, + 223.0, + 1068.0, + 268.0, + 291.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 511.0, + 1405.0, + 511.0, + 1405.0, + 546.0, + 297.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 540.0, + 1407.0, + 540.0, + 1407.0, + 578.0, + 294.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 573.0, + 1406.0, + 573.0, + 1406.0, + 608.0, + 293.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 604.0, + 1405.0, + 604.0, + 1405.0, + 637.0, + 297.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 635.0, + 1180.0, + 635.0, + 1180.0, + 668.0, + 294.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 375.0, + 1405.0, + 375.0, + 1405.0, + 407.0, + 296.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 406.0, + 1404.0, + 406.0, + 1404.0, + 439.0, + 294.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 437.0, + 1404.0, + 437.0, + 1404.0, + 469.0, + 294.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 466.0, + 1325.0, + 466.0, + 1325.0, + 502.0, + 294.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 293.0, + 1403.0, + 293.0, + 1403.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 324.0, + 1367.0, + 324.0, + 1367.0, + 361.0, + 294.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1208.0, + 450.0, + 1208.0, + 450.0, + 1242.0, + 422.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1208.0, + 864.0, + 1208.0, + 864.0, + 1242.0, + 519.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1244.0, + 450.0, + 1244.0, + 450.0, + 1277.0, + 421.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1244.0, + 863.0, + 1244.0, + 863.0, + 1277.0, + 519.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1281.0, + 450.0, + 1281.0, + 450.0, + 1311.0, + 423.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1281.0, + 863.0, + 1281.0, + 863.0, + 1311.0, + 518.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 1314.0, + 451.0, + 1314.0, + 451.0, + 1355.0, + 419.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1314.0, + 1403.0, + 1314.0, + 1403.0, + 1355.0, + 520.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1350.0, + 767.0, + 1350.0, + 767.0, + 1380.0, + 447.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1347.0, + 824.0, + 1347.0, + 824.0, + 1381.0, + 820.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1378.0, + 450.0, + 1378.0, + 450.0, + 1422.0, + 420.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1378.0, + 1348.0, + 1378.0, + 1348.0, + 1422.0, + 519.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 1354.0, + 673.0, + 1354.0, + 721.0, + 296.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1422.0, + 1110.0, + 1422.0, + 1110.0, + 1461.0, + 367.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 1422.0, + 1405.0, + 1422.0, + 1405.0, + 1461.0, + 1177.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1453.0, + 457.0, + 1453.0, + 457.0, + 1491.0, + 392.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 737.0, + 971.0, + 737.0, + 971.0, + 774.0, + 368.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 737.0, + 1404.0, + 737.0, + 1404.0, + 774.0, + 1038.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 771.0, + 1088.0, + 771.0, + 1088.0, + 800.0, + 395.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 807.0, + 960.0, + 807.0, + 960.0, + 847.0, + 390.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 807.0, + 1406.0, + 807.0, + 1406.0, + 847.0, + 1030.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 844.0, + 1088.0, + 844.0, + 1088.0, + 873.0, + 395.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 881.0, + 1018.0, + 881.0, + 1018.0, + 915.0, + 391.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 915.0, + 1091.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 915.0, + 1404.0, + 915.0, + 1404.0, + 948.0, + 393.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 943.0, + 1406.0, + 943.0, + 1406.0, + 978.0, + 392.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 976.0, + 1405.0, + 976.0, + 1405.0, + 1009.0, + 394.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1005.0, + 1404.0, + 1005.0, + 1404.0, + 1038.0, + 397.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1038.0, + 1247.0, + 1038.0, + 1247.0, + 1068.0, + 395.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1086.0, + 386.0, + 1086.0, + 386.0, + 1101.0, + 372.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1077.0, + 1404.0, + 1077.0, + 1404.0, + 1109.0, + 400.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1108.0, + 1402.0, + 1108.0, + 1402.0, + 1140.0, + 396.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1139.0, + 1403.0, + 1139.0, + 1403.0, + 1170.0, + 396.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1167.0, + 1373.0, + 1167.0, + 1373.0, + 1204.0, + 394.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1497.0, + 1023.0, + 1497.0, + 1023.0, + 1530.0, + 371.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 1497.0, + 1404.0, + 1497.0, + 1404.0, + 1530.0, + 1091.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1527.0, + 729.0, + 1527.0, + 729.0, + 1562.0, + 394.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1527.0, + 1403.0, + 1527.0, + 1403.0, + 1562.0, + 788.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1559.0, + 882.0, + 1559.0, + 882.0, + 1595.0, + 393.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1598.0, + 1024.0, + 1598.0, + 1024.0, + 1635.0, + 368.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1598.0, + 1406.0, + 1598.0, + 1406.0, + 1635.0, + 1093.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1629.0, + 1087.0, + 1629.0, + 1087.0, + 1668.0, + 392.0, + 1668.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 168, + 656, + 1415, + 656, + 1415, + 835, + 168, + 835 + ], + "score": 0.911 + }, + { + "category_id": 3, + "poly": [ + 196, + 1640, + 1458, + 1640, + 1458, + 1803, + 196, + 1803 + ], + "score": 0.881 + }, + { + "category_id": 3, + "poly": [ + 199, + 1358, + 971, + 1358, + 971, + 1522, + 199, + 1522 + ], + "score": 0.82 + }, + { + "category_id": 0, + "poly": [ + 126, + 213, + 950, + 213, + 950, + 268, + 126, + 268 + ], + "score": 0.818 + }, + { + "category_id": 1, + "poly": [ + 161, + 1927, + 644, + 1927, + 644, + 1957, + 161, + 1957 + ], + "score": 0.805 + }, + { + "category_id": 1, + "poly": [ + 160, + 1991, + 686, + 1991, + 686, + 2023, + 160, + 2023 + ], + "score": 0.763 + }, + { + "category_id": 1, + "poly": [ + 161, + 1238, + 640, + 1238, + 640, + 1268, + 161, + 1268 + ], + "score": 0.737 + }, + { + "category_id": 1, + "poly": [ + 164, + 1862, + 650, + 1862, + 650, + 1892, + 164, + 1892 + ], + "score": 0.728 + }, + { + "category_id": 0, + "poly": [ + 140, + 608, + 624, + 608, + 624, + 643, + 140, + 643 + ], + "score": 0.693 + }, + { + "category_id": 4, + "poly": [ + 160, + 1578, + 417, + 1578, + 417, + 1607, + 160, + 1607 + ], + "score": 0.512 + }, + { + "category_id": 4, + "poly": [ + 136, + 907, + 802, + 907, + 802, + 942, + 136, + 942 + ], + "score": 0.496 + }, + { + "category_id": 0, + "poly": [ + 229, + 319, + 306, + 319, + 306, + 353, + 229, + 353 + ], + "score": 0.433 + }, + { + "category_id": 3, + "poly": [ + 161, + 957, + 1521, + 957, + 1521, + 1173, + 161, + 1173 + ], + "score": 0.396 + }, + { + "category_id": 0, + "poly": [ + 136, + 907, + 802, + 907, + 802, + 942, + 136, + 942 + ], + "score": 0.37 + }, + { + "category_id": 0, + "poly": [ + 451, + 308, + 710, + 308, + 710, + 374, + 451, + 374 + ], + "score": 0.357 + }, + { + "category_id": 1, + "poly": [ + 161, + 1301, + 422, + 1301, + 422, + 1330, + 161, + 1330 + ], + "score": 0.346 + }, + { + "category_id": 4, + "poly": [ + 161, + 1301, + 422, + 1301, + 422, + 1330, + 161, + 1330 + ], + "score": 0.281 + }, + { + "category_id": 3, + "poly": [ + 1055, + 1641, + 1453, + 1641, + 1453, + 1800, + 1055, + 1800 + ], + "score": 0.225 + }, + { + "category_id": 5, + "poly": [ + 466, + 394, + 690, + 394, + 690, + 585, + 466, + 585 + ], + "score": 0.211, + "html": "
12333
24666
36999
36999
36999
" + }, + { + "category_id": 5, + "poly": [ + 175, + 398, + 357, + 398, + 357, + 587, + 175, + 587 + ], + "score": 0.152, + "html": "
abC
def
gh
" + }, + { + "category_id": 4, + "poly": [ + 140, + 608, + 624, + 608, + 624, + 643, + 140, + 643 + ], + "score": 0.143 + }, + { + "category_id": 3, + "poly": [ + 185, + 1358, + 1445, + 1358, + 1445, + 1522, + 185, + 1522 + ], + "score": 0.093 + }, + { + "category_id": 13, + "poly": [ + 460, + 393, + 693, + 393, + 693, + 583, + 460, + 583 + ], + "score": 0.76, + "latex": "\\begin{array} { c c c c c c c c c } { { 1 } } & { { } } & { { 2 } } & { { 3 } } & { { 3 } } & { { 3 } } & { { } } & { { } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 2 } } & { { 4 } } & { { 6 } } & { { 6 } } & { { 6 } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 3 } } & { { 6 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } & { { } } & { { } } \\\\ { { 3 } } & { { 6 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } & { { 9 } } & { { } } \\\\ { { 3 } } & { { 6 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } & { { } } & { { } } & { { } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 188.0, + 663.0, + 206.0, + 663.0, + 206.0, + 680.0, + 188.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 660.0, + 350.0, + 660.0, + 350.0, + 682.0, + 326.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 663.0, + 488.0, + 663.0, + 488.0, + 680.0, + 469.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 663.0, + 629.0, + 663.0, + 629.0, + 680.0, + 609.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 663.0, + 769.0, + 663.0, + 769.0, + 680.0, + 749.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 661.0, + 908.0, + 661.0, + 908.0, + 680.0, + 891.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 661.0, + 1051.0, + 661.0, + 1051.0, + 684.0, + 1029.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 661.0, + 1190.0, + 661.0, + 1190.0, + 680.0, + 1170.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 661.0, + 1330.0, + 661.0, + 1330.0, + 680.0, + 1312.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 191.0, + 702.0, + 202.0, + 702.0, + 202.0, + 715.0, + 191.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 702.0, + 343.0, + 702.0, + 343.0, + 716.0, + 331.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 702.0, + 379.0, + 702.0, + 379.0, + 716.0, + 366.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 702.0, + 483.0, + 702.0, + 483.0, + 715.0, + 472.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 702.0, + 519.0, + 702.0, + 519.0, + 716.0, + 507.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 703.0, + 553.0, + 703.0, + 553.0, + 715.0, + 542.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 702.0, + 623.0, + 702.0, + 623.0, + 715.0, + 613.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 702.0, + 763.0, + 702.0, + 763.0, + 714.0, + 754.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 702.0, + 800.0, + 702.0, + 800.0, + 716.0, + 787.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 702.0, + 905.0, + 702.0, + 905.0, + 716.0, + 893.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 702.0, + 940.0, + 702.0, + 940.0, + 715.0, + 929.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 702.0, + 975.0, + 702.0, + 975.0, + 715.0, + 963.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 702.0, + 1045.0, + 702.0, + 1045.0, + 715.0, + 1034.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 702.0, + 1186.0, + 702.0, + 1186.0, + 715.0, + 1175.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 702.0, + 1221.0, + 702.0, + 1221.0, + 715.0, + 1210.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 702.0, + 1327.0, + 702.0, + 1327.0, + 715.0, + 1315.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 702.0, + 1361.0, + 702.0, + 1361.0, + 715.0, + 1350.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 702.0, + 1398.0, + 702.0, + 1398.0, + 715.0, + 1384.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 737.0, + 625.0, + 737.0, + 625.0, + 751.0, + 612.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 737.0, + 765.0, + 737.0, + 765.0, + 751.0, + 752.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 735.0, + 801.0, + 735.0, + 801.0, + 754.0, + 786.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 737.0, + 905.0, + 737.0, + 905.0, + 751.0, + 893.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 735.0, + 943.0, + 735.0, + 943.0, + 753.0, + 926.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 735.0, + 978.0, + 735.0, + 978.0, + 753.0, + 961.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 737.0, + 1045.0, + 737.0, + 1045.0, + 750.0, + 1034.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 737.0, + 1186.0, + 737.0, + 1186.0, + 751.0, + 1175.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 737.0, + 1221.0, + 737.0, + 1221.0, + 751.0, + 1208.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 737.0, + 1327.0, + 737.0, + 1327.0, + 751.0, + 1315.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 737.0, + 1363.0, + 737.0, + 1363.0, + 751.0, + 1350.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 735.0, + 1399.0, + 735.0, + 1399.0, + 753.0, + 1382.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 772.0, + 1045.0, + 772.0, + 1045.0, + 786.0, + 1034.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 773.0, + 1186.0, + 773.0, + 1186.0, + 786.0, + 1175.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 772.0, + 1222.0, + 772.0, + 1222.0, + 789.0, + 1205.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 773.0, + 1327.0, + 773.0, + 1327.0, + 788.0, + 1315.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 770.0, + 1363.0, + 770.0, + 1363.0, + 789.0, + 1347.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 770.0, + 1399.0, + 770.0, + 1399.0, + 789.0, + 1382.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 805.0, + 271.0, + 805.0, + 271.0, + 828.0, + 195.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 804.0, + 411.0, + 804.0, + 411.0, + 827.0, + 334.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 805.0, + 549.0, + 805.0, + 549.0, + 828.0, + 480.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 805.0, + 693.0, + 805.0, + 693.0, + 828.0, + 616.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 805.0, + 825.0, + 805.0, + 825.0, + 828.0, + 766.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 805.0, + 974.0, + 805.0, + 974.0, + 828.0, + 898.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 805.0, + 1111.0, + 805.0, + 1111.0, + 828.0, + 1043.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 805.0, + 1252.0, + 805.0, + 1252.0, + 825.0, + 1177.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 805.0, + 1386.0, + 805.0, + 1386.0, + 828.0, + 1327.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1663.0, + 235.0, + 1663.0, + 235.0, + 1680.0, + 218.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1663.0, + 271.0, + 1663.0, + 271.0, + 1679.0, + 252.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1663.0, + 305.0, + 1663.0, + 305.0, + 1680.0, + 289.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1669.0, + 340.0, + 1669.0, + 340.0, + 1679.0, + 326.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1663.0, + 445.0, + 1663.0, + 445.0, + 1679.0, + 428.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1661.0, + 481.0, + 1661.0, + 481.0, + 1679.0, + 464.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1663.0, + 516.0, + 1663.0, + 516.0, + 1679.0, + 499.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1660.0, + 650.0, + 1660.0, + 650.0, + 1676.0, + 633.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1658.0, + 684.0, + 1658.0, + 684.0, + 1676.0, + 667.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1660.0, + 719.0, + 1660.0, + 719.0, + 1676.0, + 702.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1667.0, + 750.0, + 1667.0, + 750.0, + 1676.0, + 739.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1660.0, + 848.0, + 1660.0, + 848.0, + 1676.0, + 831.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1658.0, + 882.0, + 1658.0, + 882.0, + 1676.0, + 865.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1660.0, + 919.0, + 1660.0, + 919.0, + 1676.0, + 902.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1663.0, + 1088.0, + 1663.0, + 1088.0, + 1675.0, + 1074.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1658.0, + 1124.0, + 1658.0, + 1124.0, + 1676.0, + 1106.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1661.0, + 1158.0, + 1661.0, + 1158.0, + 1675.0, + 1145.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1665.0, + 1193.0, + 1665.0, + 1193.0, + 1676.0, + 1180.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1667.0, + 1227.0, + 1667.0, + 1227.0, + 1676.0, + 1216.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1663.0, + 1298.0, + 1663.0, + 1298.0, + 1675.0, + 1284.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1658.0, + 1335.0, + 1658.0, + 1335.0, + 1676.0, + 1318.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1660.0, + 1370.0, + 1660.0, + 1370.0, + 1676.0, + 1353.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1698.0, + 235.0, + 1698.0, + 235.0, + 1715.0, + 218.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1698.0, + 269.0, + 1698.0, + 269.0, + 1715.0, + 252.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1700.0, + 305.0, + 1700.0, + 305.0, + 1713.0, + 294.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1705.0, + 340.0, + 1705.0, + 340.0, + 1716.0, + 325.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1698.0, + 445.0, + 1698.0, + 445.0, + 1715.0, + 428.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1698.0, + 481.0, + 1698.0, + 481.0, + 1715.0, + 464.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1700.0, + 515.0, + 1700.0, + 515.0, + 1712.0, + 504.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1695.0, + 650.0, + 1695.0, + 650.0, + 1712.0, + 633.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1695.0, + 684.0, + 1695.0, + 684.0, + 1712.0, + 667.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1695.0, + 848.0, + 1695.0, + 848.0, + 1712.0, + 831.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 1695.0, + 884.0, + 1695.0, + 884.0, + 1712.0, + 867.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1697.0, + 919.0, + 1697.0, + 919.0, + 1711.0, + 906.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1695.0, + 1089.0, + 1695.0, + 1089.0, + 1712.0, + 1072.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1695.0, + 1124.0, + 1695.0, + 1124.0, + 1712.0, + 1106.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1695.0, + 1299.0, + 1695.0, + 1299.0, + 1712.0, + 1282.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1698.0, + 1332.0, + 1698.0, + 1332.0, + 1711.0, + 1319.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 1697.0, + 1369.0, + 1697.0, + 1369.0, + 1709.0, + 1358.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1426.0, + 1702.0, + 1437.0, + 1702.0, + 1437.0, + 1712.0, + 1426.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1735.0, + 235.0, + 1735.0, + 235.0, + 1753.0, + 218.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1734.0, + 269.0, + 1734.0, + 269.0, + 1752.0, + 252.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1737.0, + 302.0, + 1737.0, + 302.0, + 1748.0, + 291.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1735.0, + 445.0, + 1735.0, + 445.0, + 1753.0, + 428.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1734.0, + 481.0, + 1734.0, + 481.0, + 1752.0, + 462.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1732.0, + 648.0, + 1732.0, + 648.0, + 1750.0, + 633.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1731.0, + 684.0, + 1731.0, + 684.0, + 1748.0, + 667.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1732.0, + 847.0, + 1732.0, + 847.0, + 1750.0, + 831.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1732.0, + 881.0, + 1732.0, + 881.0, + 1746.0, + 868.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1734.0, + 915.0, + 1734.0, + 915.0, + 1746.0, + 904.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1734.0, + 1089.0, + 1734.0, + 1089.0, + 1750.0, + 1072.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1734.0, + 1121.0, + 1734.0, + 1121.0, + 1746.0, + 1108.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1734.0, + 1299.0, + 1734.0, + 1299.0, + 1750.0, + 1282.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1734.0, + 1332.0, + 1734.0, + 1332.0, + 1746.0, + 1319.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 1738.0, + 1438.0, + 1738.0, + 1438.0, + 1749.0, + 1424.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 223.0, + 1776.0, + 234.0, + 1776.0, + 234.0, + 1786.0, + 223.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1774.0, + 882.0, + 1774.0, + 882.0, + 1785.0, + 868.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1774.0, + 918.0, + 1774.0, + 918.0, + 1785.0, + 904.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1774.0, + 952.0, + 1774.0, + 952.0, + 1783.0, + 938.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1390.0, + 1774.0, + 1404.0, + 1774.0, + 1404.0, + 1785.0, + 1390.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1426.0, + 1774.0, + 1437.0, + 1774.0, + 1437.0, + 1783.0, + 1426.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1390.25, + 1734.5, + 1404.25, + 1734.5, + 1404.25, + 1750.5, + 1390.25, + 1750.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1382.0, + 233.0, + 1382.0, + 233.0, + 1396.0, + 221.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 256.0, + 1380.0, + 268.0, + 1380.0, + 268.0, + 1395.0, + 256.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1381.0, + 303.0, + 1381.0, + 303.0, + 1396.0, + 291.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1385.0, + 342.0, + 1385.0, + 342.0, + 1397.0, + 327.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1381.0, + 445.0, + 1381.0, + 445.0, + 1396.0, + 432.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1380.0, + 479.0, + 1380.0, + 479.0, + 1395.0, + 467.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1381.0, + 514.0, + 1381.0, + 514.0, + 1396.0, + 502.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1387.0, + 550.0, + 1387.0, + 550.0, + 1395.0, + 539.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1381.0, + 649.0, + 1381.0, + 649.0, + 1396.0, + 636.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1380.0, + 684.0, + 1380.0, + 684.0, + 1395.0, + 672.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1381.0, + 719.0, + 1381.0, + 719.0, + 1396.0, + 707.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1385.0, + 758.0, + 1385.0, + 758.0, + 1397.0, + 741.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1382.0, + 849.0, + 1382.0, + 849.0, + 1397.0, + 836.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1381.0, + 884.0, + 1381.0, + 884.0, + 1396.0, + 871.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1383.0, + 919.0, + 1383.0, + 919.0, + 1397.0, + 907.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1388.0, + 954.0, + 1388.0, + 954.0, + 1396.0, + 942.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1416.0, + 233.0, + 1416.0, + 233.0, + 1431.0, + 221.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 256.0, + 1418.0, + 269.0, + 1418.0, + 269.0, + 1431.0, + 256.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1417.0, + 302.0, + 1417.0, + 302.0, + 1428.0, + 295.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1421.0, + 342.0, + 1421.0, + 342.0, + 1433.0, + 327.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1416.0, + 445.0, + 1416.0, + 445.0, + 1431.0, + 432.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1417.0, + 479.0, + 1417.0, + 479.0, + 1431.0, + 467.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1418.0, + 514.0, + 1418.0, + 514.0, + 1429.0, + 506.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1420.0, + 553.0, + 1420.0, + 553.0, + 1433.0, + 537.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1416.0, + 650.0, + 1416.0, + 650.0, + 1431.0, + 636.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1417.0, + 685.0, + 1417.0, + 685.0, + 1431.0, + 672.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1417.0, + 719.0, + 1417.0, + 719.0, + 1428.0, + 710.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1420.0, + 758.0, + 1420.0, + 758.0, + 1433.0, + 741.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1415.0, + 850.0, + 1415.0, + 850.0, + 1434.0, + 834.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1418.0, + 884.0, + 1418.0, + 884.0, + 1432.0, + 872.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 1419.0, + 919.0, + 1419.0, + 919.0, + 1430.0, + 911.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1425.0, + 954.0, + 1425.0, + 954.0, + 1432.0, + 942.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1453.0, + 233.0, + 1453.0, + 233.0, + 1469.0, + 221.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 256.0, + 1452.0, + 269.0, + 1452.0, + 269.0, + 1467.0, + 256.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1459.0, + 340.0, + 1459.0, + 340.0, + 1467.0, + 328.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1454.0, + 444.0, + 1454.0, + 444.0, + 1470.0, + 432.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1452.0, + 479.0, + 1452.0, + 479.0, + 1467.0, + 467.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1454.0, + 512.0, + 1454.0, + 512.0, + 1466.0, + 504.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1455.0, + 555.0, + 1455.0, + 555.0, + 1470.0, + 535.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1452.0, + 651.0, + 1452.0, + 651.0, + 1471.0, + 634.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1450.0, + 686.0, + 1450.0, + 686.0, + 1468.0, + 670.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 1454.0, + 716.0, + 1454.0, + 716.0, + 1464.0, + 709.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1459.0, + 756.0, + 1459.0, + 756.0, + 1467.0, + 742.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1453.0, + 849.0, + 1453.0, + 849.0, + 1472.0, + 834.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1453.0, + 884.0, + 1453.0, + 884.0, + 1468.0, + 871.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1461.0, + 954.0, + 1461.0, + 954.0, + 1468.0, + 943.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 224.0, + 1496.0, + 234.0, + 1496.0, + 234.0, + 1504.0, + 224.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1495.0, + 271.0, + 1495.0, + 271.0, + 1505.0, + 255.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1496.0, + 445.0, + 1496.0, + 445.0, + 1504.0, + 433.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1496.0, + 480.0, + 1496.0, + 480.0, + 1504.0, + 468.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1496.0, + 515.0, + 1496.0, + 515.0, + 1504.0, + 502.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1495.0, + 553.0, + 1495.0, + 553.0, + 1505.0, + 537.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1496.0, + 651.0, + 1496.0, + 651.0, + 1504.0, + 637.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1496.0, + 685.0, + 1496.0, + 685.0, + 1504.0, + 672.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1496.0, + 720.0, + 1496.0, + 720.0, + 1504.0, + 707.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 1496.0, + 850.0, + 1496.0, + 850.0, + 1504.0, + 837.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1496.0, + 884.0, + 1496.0, + 884.0, + 1503.0, + 873.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1496.0, + 920.0, + 1496.0, + 920.0, + 1504.0, + 908.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 203.0, + 501.0, + 203.0, + 501.0, + 277.0, + 123.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 223.0, + 951.0, + 223.0, + 951.0, + 269.0, + 529.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 134.0, + 604.0, + 625.0, + 604.0, + 625.0, + 648.0, + 134.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1572.0, + 418.0, + 1572.0, + 418.0, + 1612.0, + 160.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 908.0, + 800.0, + 908.0, + 800.0, + 943.0, + 137.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 225.0, + 313.0, + 311.0, + 313.0, + 311.0, + 363.0, + 225.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 158.0, + 955.0, + 426.0, + 955.0, + 426.0, + 991.0, + 158.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 953.0, + 747.0, + 953.0, + 747.0, + 994.0, + 479.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 953.0, + 1200.0, + 953.0, + 1200.0, + 996.0, + 933.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 1032.0, + 268.0, + 1032.0, + 268.0, + 1047.0, + 254.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1034.0, + 306.0, + 1034.0, + 306.0, + 1047.0, + 290.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 1030.0, + 543.0, + 1030.0, + 543.0, + 1042.0, + 529.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1030.0, + 576.0, + 1030.0, + 576.0, + 1042.0, + 564.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1030.0, + 613.0, + 1030.0, + 613.0, + 1044.0, + 601.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1032.0, + 732.0, + 1032.0, + 732.0, + 1044.0, + 718.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1034.0, + 996.0, + 1034.0, + 996.0, + 1045.0, + 984.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1029.0, + 1179.0, + 1029.0, + 1179.0, + 1044.0, + 1165.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1029.0, + 1212.0, + 1029.0, + 1212.0, + 1044.0, + 1200.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1029.0, + 1361.0, + 1029.0, + 1361.0, + 1044.0, + 1347.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 1029.0, + 1396.0, + 1029.0, + 1396.0, + 1042.0, + 1382.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 222.0, + 1068.0, + 233.0, + 1068.0, + 233.0, + 1081.0, + 222.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1070.0, + 268.0, + 1070.0, + 268.0, + 1081.0, + 255.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1068.0, + 732.0, + 1068.0, + 732.0, + 1080.0, + 718.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1068.0, + 996.0, + 1068.0, + 996.0, + 1081.0, + 984.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1067.0, + 1179.0, + 1067.0, + 1179.0, + 1080.0, + 1165.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1065.0, + 1361.0, + 1065.0, + 1361.0, + 1080.0, + 1346.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 1068.0, + 1395.0, + 1068.0, + 1395.0, + 1080.0, + 1382.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1108.0, + 234.0, + 1108.0, + 234.0, + 1121.0, + 221.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 1105.0, + 268.0, + 1105.0, + 268.0, + 1119.0, + 254.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 1103.0, + 543.0, + 1103.0, + 543.0, + 1118.0, + 529.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1101.0, + 578.0, + 1101.0, + 578.0, + 1114.0, + 564.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1103.0, + 765.0, + 1103.0, + 765.0, + 1116.0, + 753.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1103.0, + 998.0, + 1103.0, + 998.0, + 1123.0, + 979.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1103.0, + 1031.0, + 1103.0, + 1031.0, + 1118.0, + 1017.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1101.0, + 1179.0, + 1101.0, + 1179.0, + 1118.0, + 1165.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1101.0, + 1212.0, + 1101.0, + 1212.0, + 1114.0, + 1200.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1100.0, + 1363.0, + 1100.0, + 1363.0, + 1121.0, + 1344.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 1101.0, + 1396.0, + 1101.0, + 1396.0, + 1114.0, + 1382.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.25, + 1105.0, + 735.25, + 1105.0, + 735.25, + 1116.5, + 713.25, + 1116.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 908.0, + 800.0, + 908.0, + 800.0, + 943.0, + 137.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 309.0, + 711.0, + 309.0, + 711.0, + 342.0, + 451.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 338.0, + 708.0, + 338.0, + 708.0, + 374.0, + 448.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1296.0, + 424.0, + 1296.0, + 424.0, + 1335.0, + 160.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1660.0, + 1088.0, + 1660.0, + 1088.0, + 1676.0, + 1072.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1659.0, + 1123.0, + 1659.0, + 1123.0, + 1676.0, + 1108.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1661.0, + 1158.0, + 1661.0, + 1158.0, + 1675.0, + 1145.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1667.0, + 1193.0, + 1667.0, + 1193.0, + 1675.0, + 1180.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1665.0, + 1230.0, + 1665.0, + 1230.0, + 1676.0, + 1213.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1660.0, + 1299.0, + 1660.0, + 1299.0, + 1676.0, + 1283.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1659.0, + 1334.0, + 1659.0, + 1334.0, + 1676.0, + 1318.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1660.0, + 1369.0, + 1660.0, + 1369.0, + 1676.0, + 1353.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 1667.0, + 1404.0, + 1667.0, + 1404.0, + 1675.0, + 1391.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1695.0, + 1089.0, + 1695.0, + 1089.0, + 1712.0, + 1072.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1697.0, + 1122.0, + 1697.0, + 1122.0, + 1711.0, + 1109.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1695.0, + 1159.0, + 1695.0, + 1159.0, + 1710.0, + 1147.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1701.0, + 1195.0, + 1701.0, + 1195.0, + 1712.0, + 1178.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1700.0, + 1230.0, + 1700.0, + 1230.0, + 1712.0, + 1213.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1695.0, + 1299.0, + 1695.0, + 1299.0, + 1712.0, + 1283.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1696.0, + 1334.0, + 1696.0, + 1334.0, + 1712.0, + 1318.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 1696.0, + 1369.0, + 1696.0, + 1369.0, + 1710.0, + 1358.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1389.0, + 1701.0, + 1406.0, + 1701.0, + 1406.0, + 1712.0, + 1389.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 1701.0, + 1441.0, + 1701.0, + 1441.0, + 1712.0, + 1424.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1733.0, + 1088.0, + 1733.0, + 1088.0, + 1751.0, + 1073.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 1731.0, + 1124.0, + 1731.0, + 1124.0, + 1748.0, + 1107.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1734.0, + 1155.0, + 1734.0, + 1155.0, + 1745.0, + 1147.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1737.0, + 1194.0, + 1737.0, + 1194.0, + 1748.0, + 1179.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1737.0, + 1230.0, + 1737.0, + 1230.0, + 1748.0, + 1213.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1733.0, + 1299.0, + 1733.0, + 1299.0, + 1751.0, + 1283.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1731.0, + 1334.0, + 1731.0, + 1334.0, + 1748.0, + 1318.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 1734.0, + 1365.0, + 1734.0, + 1365.0, + 1745.0, + 1357.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 1739.0, + 1404.0, + 1739.0, + 1404.0, + 1747.0, + 1391.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 1737.0, + 1440.0, + 1737.0, + 1440.0, + 1749.0, + 1424.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1773.0, + 1089.0, + 1773.0, + 1089.0, + 1784.0, + 1073.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1773.0, + 1124.0, + 1773.0, + 1124.0, + 1784.0, + 1108.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1773.0, + 1160.0, + 1773.0, + 1160.0, + 1784.0, + 1142.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1773.0, + 1195.0, + 1773.0, + 1195.0, + 1784.0, + 1178.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1773.0, + 1300.0, + 1773.0, + 1300.0, + 1784.0, + 1283.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1773.0, + 1335.0, + 1773.0, + 1335.0, + 1784.0, + 1318.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1773.0, + 1370.0, + 1773.0, + 1370.0, + 1784.0, + 1354.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1389.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1784.0, + 1389.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 1773.0, + 1441.0, + 1773.0, + 1441.0, + 1784.0, + 1424.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 134.0, + 604.0, + 625.0, + 604.0, + 625.0, + 648.0, + 134.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1382.0, + 234.0, + 1382.0, + 234.0, + 1396.0, + 221.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 1380.0, + 271.0, + 1380.0, + 271.0, + 1396.0, + 254.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1382.0, + 305.0, + 1382.0, + 305.0, + 1395.0, + 291.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1381.0, + 448.0, + 1381.0, + 448.0, + 1397.0, + 430.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1380.0, + 482.0, + 1380.0, + 482.0, + 1396.0, + 465.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1382.0, + 515.0, + 1382.0, + 515.0, + 1395.0, + 502.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1382.0, + 651.0, + 1382.0, + 651.0, + 1395.0, + 637.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 1381.0, + 685.0, + 1381.0, + 685.0, + 1393.0, + 671.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1382.0, + 720.0, + 1382.0, + 720.0, + 1395.0, + 707.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1381.0, + 852.0, + 1381.0, + 852.0, + 1397.0, + 835.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1380.0, + 886.0, + 1380.0, + 886.0, + 1397.0, + 869.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1381.0, + 921.0, + 1381.0, + 921.0, + 1397.0, + 904.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 1417.0, + 234.0, + 1417.0, + 234.0, + 1430.0, + 221.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 1418.0, + 268.0, + 1418.0, + 268.0, + 1430.0, + 257.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1417.0, + 305.0, + 1417.0, + 305.0, + 1428.0, + 294.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1417.0, + 444.0, + 1417.0, + 444.0, + 1430.0, + 433.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1419.0, + 478.0, + 1419.0, + 478.0, + 1429.0, + 467.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1418.0, + 649.0, + 1418.0, + 649.0, + 1430.0, + 637.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1417.0, + 722.0, + 1417.0, + 722.0, + 1428.0, + 710.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1424.0, + 753.0, + 1424.0, + 753.0, + 1430.0, + 746.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1418.0, + 849.0, + 1418.0, + 849.0, + 1432.0, + 836.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1419.0, + 883.0, + 1419.0, + 883.0, + 1432.0, + 872.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1418.0, + 920.0, + 1418.0, + 920.0, + 1430.0, + 909.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 219.0, + 1452.0, + 236.0, + 1452.0, + 236.0, + 1470.0, + 219.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1452.0, + 268.0, + 1452.0, + 268.0, + 1466.0, + 255.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1454.0, + 444.0, + 1454.0, + 444.0, + 1468.0, + 433.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1454.0, + 481.0, + 1454.0, + 481.0, + 1466.0, + 467.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1455.0, + 649.0, + 1455.0, + 649.0, + 1468.0, + 637.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 1454.0, + 685.0, + 1454.0, + 685.0, + 1465.0, + 671.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1456.0, + 848.0, + 1456.0, + 848.0, + 1469.0, + 836.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1454.0, + 884.0, + 1454.0, + 884.0, + 1468.0, + 872.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1495.0, + 920.0, + 1495.0, + 920.0, + 1505.0, + 907.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1924.0, + 642.0, + 1924.0, + 642.0, + 1960.0, + 160.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 161.0, + 1992.0, + 687.0, + 1992.0, + 687.0, + 2023.0, + 161.0, + 2023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1236.0, + 641.0, + 1236.0, + 641.0, + 1270.0, + 160.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 162.0, + 1860.0, + 649.0, + 1860.0, + 649.0, + 1894.0, + 162.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1296.0, + 424.0, + 1296.0, + 424.0, + 1335.0, + 160.0, + 1335.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 197, + 1399, + 1077, + 1399, + 1077, + 1577, + 197, + 1577 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 129, + 211, + 451, + 211, + 451, + 266, + 129, + 266 + ], + "score": 0.907 + }, + { + "category_id": 3, + "poly": [ + 194, + 1699, + 1532, + 1699, + 1532, + 1876, + 194, + 1876 + ], + "score": 0.898 + }, + { + "category_id": 3, + "poly": [ + 277, + 913, + 1127, + 913, + 1127, + 1145, + 277, + 1145 + ], + "score": 0.874 + }, + { + "category_id": 0, + "poly": [ + 128, + 1257, + 794, + 1257, + 794, + 1292, + 128, + 1292 + ], + "score": 0.719 + }, + { + "category_id": 0, + "poly": [ + 126, + 824, + 611, + 824, + 611, + 858, + 126, + 858 + ], + "score": 0.702 + }, + { + "category_id": 0, + "poly": [ + 128, + 302, + 839, + 302, + 839, + 344, + 128, + 344 + ], + "score": 0.698 + }, + { + "category_id": 1, + "poly": [ + 127, + 1354, + 388, + 1354, + 388, + 1383, + 127, + 1383 + ], + "score": 0.617 + }, + { + "category_id": 5, + "poly": [ + 123, + 583, + 315, + 583, + 315, + 758, + 123, + 758 + ], + "score": 0.583, + "html": "
ab··
Cd
·
" + }, + { + "category_id": 5, + "poly": [ + 700, + 582, + 895, + 582, + 895, + 759, + 700, + 759 + ], + "score": 0.546, + "html": "
46666
69999
6999
69999
69999
" + }, + { + "category_id": 6, + "poly": [ + 677, + 495, + 934, + 495, + 934, + 560, + 677, + 560 + ], + "score": 0.519 + }, + { + "category_id": 5, + "poly": [ + 383, + 559, + 608, + 559, + 608, + 759, + 383, + 759 + ], + "score": 0.516, + "html": "
000• ·
0ab
0Cd
0·
• =
" + }, + { + "category_id": 1, + "poly": [ + 122, + 1655, + 383, + 1655, + 383, + 1682, + 122, + 1682 + ], + "score": 0.495 + }, + { + "category_id": 1, + "poly": [ + 122, + 1953, + 693, + 1953, + 693, + 1983, + 122, + 1983 + ], + "score": 0.486 + }, + { + "category_id": 0, + "poly": [ + 127, + 511, + 314, + 511, + 314, + 545, + 127, + 545 + ], + "score": 0.348 + }, + { + "category_id": 6, + "poly": [ + 127, + 511, + 314, + 511, + 314, + 545, + 127, + 545 + ], + "score": 0.297 + }, + { + "category_id": 3, + "poly": [ + 700, + 582, + 895, + 582, + 895, + 759, + 700, + 759 + ], + "score": 0.278 + }, + { + "category_id": 6, + "poly": [ + 426, + 512, + 608, + 512, + 608, + 545, + 426, + 545 + ], + "score": 0.242 + }, + { + "category_id": 1, + "poly": [ + 128, + 302, + 839, + 302, + 839, + 344, + 128, + 344 + ], + "score": 0.217 + }, + { + "category_id": 1, + "poly": [ + 126, + 824, + 611, + 824, + 611, + 858, + 126, + 858 + ], + "score": 0.212 + }, + { + "category_id": 1, + "poly": [ + 128, + 1257, + 794, + 1257, + 794, + 1292, + 128, + 1292 + ], + "score": 0.179 + }, + { + "category_id": 3, + "poly": [ + 123, + 583, + 315, + 583, + 315, + 758, + 123, + 758 + ], + "score": 0.145 + }, + { + "category_id": 3, + "poly": [ + 383, + 559, + 608, + 559, + 608, + 759, + 383, + 759 + ], + "score": 0.107 + }, + { + "category_id": 0, + "poly": [ + 127, + 1354, + 388, + 1354, + 388, + 1383, + 127, + 1383 + ], + "score": 0.102 + }, + { + "category_id": 13, + "poly": [ + 694, + 580, + 898, + 580, + 898, + 758, + 694, + 758 + ], + "score": 0.86, + "latex": "{ \\left[ \\begin{array} { l l l l l l } { 4 } & { 6 } & { 6 } & { 6 } & { 6 } \\\\ { 6 } & { 9 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 9 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 9 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 9 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 9 } & { 9 } & { 9 } & { 9 } \\end{array} \\right] }" + }, + { + "category_id": 13, + "poly": [ + 1068, + 1068, + 1102, + 1068, + 1102, + 1091, + 1068, + 1091 + ], + "score": 0.25, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 15, + "poly": [ + 208.0, + 1405.0, + 314.0, + 1405.0, + 314.0, + 1438.0, + 208.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1406.0, + 541.0, + 1406.0, + 541.0, + 1439.0, + 434.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1411.0, + 687.0, + 1411.0, + 687.0, + 1435.0, + 668.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1411.0, + 727.0, + 1411.0, + 727.0, + 1433.0, + 704.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1412.0, + 764.0, + 1412.0, + 764.0, + 1434.0, + 741.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1410.0, + 918.0, + 1410.0, + 918.0, + 1437.0, + 893.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1407.0, + 995.0, + 1407.0, + 995.0, + 1439.0, + 925.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 214.0, + 1448.0, + 228.0, + 1448.0, + 228.0, + 1463.0, + 214.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1449.0, + 269.0, + 1449.0, + 269.0, + 1466.0, + 251.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1447.0, + 305.0, + 1447.0, + 305.0, + 1466.0, + 290.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1446.0, + 457.0, + 1446.0, + 457.0, + 1466.0, + 440.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1448.0, + 497.0, + 1448.0, + 497.0, + 1466.0, + 478.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1447.0, + 533.0, + 1447.0, + 533.0, + 1466.0, + 517.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1443.0, + 686.0, + 1443.0, + 686.0, + 1468.0, + 666.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1446.0, + 725.0, + 1446.0, + 725.0, + 1467.0, + 703.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1447.0, + 759.0, + 1447.0, + 759.0, + 1466.0, + 743.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1443.0, + 914.0, + 1443.0, + 914.0, + 1468.0, + 893.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1445.0, + 952.0, + 1445.0, + 952.0, + 1468.0, + 930.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1443.0, + 988.0, + 1443.0, + 988.0, + 1468.0, + 969.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 215.0, + 1480.0, + 228.0, + 1480.0, + 228.0, + 1496.0, + 215.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1483.0, + 265.0, + 1483.0, + 265.0, + 1498.0, + 253.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1481.0, + 304.0, + 1481.0, + 304.0, + 1497.0, + 291.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1480.0, + 457.0, + 1480.0, + 457.0, + 1500.0, + 440.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1481.0, + 493.0, + 1481.0, + 493.0, + 1496.0, + 481.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1480.0, + 533.0, + 1480.0, + 533.0, + 1500.0, + 516.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1478.0, + 686.0, + 1478.0, + 686.0, + 1502.0, + 666.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1479.0, + 723.0, + 1479.0, + 723.0, + 1499.0, + 706.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1480.0, + 759.0, + 1480.0, + 759.0, + 1499.0, + 743.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1478.0, + 914.0, + 1478.0, + 914.0, + 1502.0, + 893.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 1477.0, + 951.0, + 1477.0, + 951.0, + 1501.0, + 931.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1480.0, + 987.0, + 1480.0, + 987.0, + 1501.0, + 970.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 122.0, + 201.0, + 458.0, + 201.0, + 458.0, + 276.0, + 122.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 211.0, + 1708.0, + 231.0, + 1708.0, + 231.0, + 1728.0, + 211.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1707.0, + 270.0, + 1707.0, + 270.0, + 1730.0, + 246.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1705.0, + 309.0, + 1705.0, + 309.0, + 1731.0, + 283.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1705.0, + 462.0, + 1705.0, + 462.0, + 1731.0, + 438.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1701.0, + 541.0, + 1701.0, + 541.0, + 1736.0, + 468.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1707.0, + 689.0, + 1707.0, + 689.0, + 1731.0, + 667.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1707.0, + 728.0, + 1707.0, + 728.0, + 1731.0, + 704.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1707.0, + 764.0, + 1707.0, + 764.0, + 1731.0, + 740.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1705.0, + 915.0, + 1705.0, + 915.0, + 1731.0, + 891.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 1701.0, + 993.0, + 1701.0, + 993.0, + 1736.0, + 923.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1707.0, + 1145.0, + 1707.0, + 1145.0, + 1731.0, + 1121.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1707.0, + 1181.0, + 1707.0, + 1181.0, + 1731.0, + 1157.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1705.0, + 1220.0, + 1705.0, + 1220.0, + 1731.0, + 1194.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1705.0, + 1371.0, + 1705.0, + 1371.0, + 1731.0, + 1347.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1704.0, + 1411.0, + 1704.0, + 1411.0, + 1733.0, + 1380.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1707.0, + 1444.0, + 1707.0, + 1444.0, + 1731.0, + 1419.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 210.0, + 1741.0, + 232.0, + 1741.0, + 232.0, + 1767.0, + 210.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1743.0, + 273.0, + 1743.0, + 273.0, + 1769.0, + 249.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1741.0, + 307.0, + 1741.0, + 307.0, + 1767.0, + 285.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1741.0, + 460.0, + 1741.0, + 460.0, + 1769.0, + 438.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1743.0, + 501.0, + 1743.0, + 501.0, + 1769.0, + 475.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1743.0, + 536.0, + 1743.0, + 536.0, + 1767.0, + 512.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1744.0, + 574.0, + 1744.0, + 574.0, + 1769.0, + 551.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1741.0, + 688.0, + 1741.0, + 688.0, + 1769.0, + 665.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1744.0, + 727.0, + 1744.0, + 727.0, + 1769.0, + 703.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1743.0, + 764.0, + 1743.0, + 764.0, + 1767.0, + 740.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1741.0, + 915.0, + 1741.0, + 915.0, + 1769.0, + 891.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1743.0, + 953.0, + 1743.0, + 953.0, + 1769.0, + 930.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 1741.0, + 990.0, + 1741.0, + 990.0, + 1767.0, + 966.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1741.0, + 1142.0, + 1741.0, + 1142.0, + 1769.0, + 1119.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1744.0, + 1181.0, + 1744.0, + 1181.0, + 1769.0, + 1157.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1741.0, + 1218.0, + 1741.0, + 1218.0, + 1767.0, + 1194.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1741.0, + 1369.0, + 1741.0, + 1369.0, + 1769.0, + 1345.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1743.0, + 1407.0, + 1743.0, + 1407.0, + 1769.0, + 1384.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1420.0, + 1743.0, + 1444.0, + 1743.0, + 1444.0, + 1767.0, + 1420.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 211.0, + 1776.0, + 234.0, + 1776.0, + 234.0, + 1800.0, + 211.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1779.0, + 270.0, + 1779.0, + 270.0, + 1799.0, + 250.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1776.0, + 309.0, + 1776.0, + 309.0, + 1800.0, + 286.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1777.0, + 459.0, + 1777.0, + 459.0, + 1798.0, + 439.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1777.0, + 499.0, + 1777.0, + 499.0, + 1800.0, + 475.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1776.0, + 536.0, + 1776.0, + 536.0, + 1800.0, + 514.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1777.0, + 574.0, + 1777.0, + 574.0, + 1796.0, + 557.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1777.0, + 686.0, + 1777.0, + 686.0, + 1798.0, + 667.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1779.0, + 724.0, + 1779.0, + 724.0, + 1799.0, + 704.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1776.0, + 763.0, + 1776.0, + 763.0, + 1800.0, + 740.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1776.0, + 914.0, + 1776.0, + 914.0, + 1800.0, + 891.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1779.0, + 951.0, + 1779.0, + 951.0, + 1799.0, + 932.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 1776.0, + 990.0, + 1776.0, + 990.0, + 1800.0, + 968.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1775.0, + 1142.0, + 1775.0, + 1142.0, + 1800.0, + 1119.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 1779.0, + 1178.0, + 1779.0, + 1178.0, + 1799.0, + 1160.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1777.0, + 1215.0, + 1777.0, + 1215.0, + 1798.0, + 1196.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1775.0, + 1369.0, + 1775.0, + 1369.0, + 1800.0, + 1345.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1776.0, + 1407.0, + 1776.0, + 1407.0, + 1802.0, + 1384.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1422.0, + 1776.0, + 1444.0, + 1776.0, + 1444.0, + 1800.0, + 1422.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 938.0, + 299.0, + 938.0, + 299.0, + 956.0, + 282.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 940.0, + 352.0, + 940.0, + 352.0, + 948.0, + 342.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 936.0, + 397.0, + 936.0, + 397.0, + 953.0, + 381.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 934.0, + 520.0, + 934.0, + 520.0, + 956.0, + 500.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 937.0, + 573.0, + 937.0, + 573.0, + 949.0, + 559.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 936.0, + 619.0, + 936.0, + 619.0, + 952.0, + 601.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 936.0, + 662.0, + 936.0, + 662.0, + 952.0, + 646.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 937.0, + 748.0, + 937.0, + 748.0, + 956.0, + 731.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 936.0, + 805.0, + 936.0, + 805.0, + 954.0, + 788.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 935.0, + 848.0, + 935.0, + 848.0, + 954.0, + 832.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 934.0, + 968.0, + 934.0, + 968.0, + 956.0, + 949.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 934.0, + 1028.0, + 934.0, + 1028.0, + 954.0, + 1004.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 934.0, + 1069.0, + 934.0, + 1069.0, + 954.0, + 1047.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 937.0, + 1109.0, + 937.0, + 1109.0, + 949.0, + 1095.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 981.0, + 353.0, + 981.0, + 353.0, + 996.0, + 340.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 979.0, + 401.0, + 979.0, + 401.0, + 1000.0, + 378.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 974.0, + 663.0, + 974.0, + 663.0, + 1005.0, + 553.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 977.0, + 810.0, + 977.0, + 810.0, + 1003.0, + 785.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 977.0, + 851.0, + 977.0, + 851.0, + 1003.0, + 827.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 971.0, + 1079.0, + 971.0, + 1079.0, + 1009.0, + 999.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 978.0, + 1112.0, + 978.0, + 1112.0, + 1001.0, + 1091.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1024.0, + 808.0, + 1024.0, + 808.0, + 1047.0, + 786.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1024.0, + 849.0, + 1024.0, + 849.0, + 1047.0, + 828.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1023.0, + 1028.0, + 1023.0, + 1028.0, + 1048.0, + 1002.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1022.0, + 1074.0, + 1022.0, + 1074.0, + 1050.0, + 1035.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1024.0, + 1113.0, + 1024.0, + 1113.0, + 1046.0, + 1092.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1065.0, + 439.0, + 1065.0, + 439.0, + 1093.0, + 354.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1065.0, + 658.0, + 1065.0, + 658.0, + 1093.0, + 573.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1062.0, + 890.0, + 1062.0, + 890.0, + 1096.0, + 804.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1062.0, + 1067.0, + 1062.0, + 1067.0, + 1096.0, + 1023.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1062.0, + 1110.0, + 1062.0, + 1110.0, + 1096.0, + 1103.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1105.0, + 1092.0, + 1105.0, + 1092.0, + 1137.0, + 1023.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 129.0, + 1258.0, + 793.0, + 1258.0, + 793.0, + 1293.0, + 129.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 820.0, + 613.0, + 820.0, + 613.0, + 862.0, + 123.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 125.0, + 295.0, + 839.0, + 295.0, + 839.0, + 352.0, + 125.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 492.0, + 935.0, + 492.0, + 935.0, + 531.0, + 674.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 526.0, + 933.0, + 526.0, + 933.0, + 559.0, + 680.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 509.0, + 317.0, + 509.0, + 317.0, + 549.0, + 126.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 509.0, + 317.0, + 509.0, + 317.0, + 549.0, + 126.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 508.0, + 611.0, + 508.0, + 611.0, + 550.0, + 424.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 131.0, + 591.0, + 184.0, + 591.0, + 184.0, + 623.0, + 131.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 170.0, + 594.0, + 196.0, + 594.0, + 196.0, + 619.0, + 170.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 133.0, + 627.0, + 170.0, + 627.0, + 170.0, + 655.0, + 133.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 172.0, + 628.0, + 197.0, + 628.0, + 197.0, + 652.0, + 172.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 139.0, + 672.0, + 153.0, + 672.0, + 153.0, + 684.0, + 139.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 554.0, + 490.0, + 554.0, + 490.0, + 591.0, + 381.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 592.0, + 407.0, + 592.0, + 407.0, + 618.0, + 387.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 595.0, + 445.0, + 595.0, + 445.0, + 619.0, + 423.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 595.0, + 480.0, + 595.0, + 480.0, + 618.0, + 461.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 627.0, + 407.0, + 627.0, + 407.0, + 653.0, + 387.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 629.0, + 481.0, + 629.0, + 481.0, + 652.0, + 461.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 630.0, + 444.0, + 630.0, + 444.0, + 653.0, + 423.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 660.0, + 407.0, + 660.0, + 407.0, + 685.0, + 387.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 675.0, + 437.0, + 675.0, + 437.0, + 683.0, + 428.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 1349.0, + 390.0, + 1349.0, + 390.0, + 1388.0, + 126.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 1349.0, + 390.0, + 1349.0, + 390.0, + 1388.0, + 126.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 121.0, + 1649.0, + 384.0, + 1649.0, + 384.0, + 1688.0, + 121.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 1950.0, + 691.0, + 1950.0, + 691.0, + 1985.0, + 123.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 125.0, + 295.0, + 839.0, + 295.0, + 839.0, + 352.0, + 125.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 820.0, + 613.0, + 820.0, + 613.0, + 862.0, + 123.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 129.0, + 1258.0, + 793.0, + 1258.0, + 793.0, + 1293.0, + 129.0, + 1293.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 130, + 211, + 534, + 211, + 534, + 267, + 130, + 267 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 285, + 2061, + 855, + 2061, + 855, + 2091, + 285, + 2091 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 126, + 302, + 840, + 302, + 840, + 344, + 126, + 344 + ], + "score": 0.851 + }, + { + "category_id": 0, + "poly": [ + 268, + 1198, + 933, + 1198, + 933, + 1234, + 268, + 1234 + ], + "score": 0.811 + }, + { + "category_id": 3, + "poly": [ + 275, + 1271, + 1294, + 1271, + 1294, + 2038, + 275, + 2038 + ], + "score": 0.664 + }, + { + "category_id": 0, + "poly": [ + 283, + 872, + 767, + 872, + 767, + 906, + 283, + 906 + ], + "score": 0.629 + }, + { + "category_id": 3, + "poly": [ + 270, + 373, + 1381, + 373, + 1381, + 845, + 270, + 845 + ], + "score": 0.618 + }, + { + "category_id": 1, + "poly": [ + 276, + 1254, + 538, + 1254, + 538, + 1282, + 276, + 1282 + ], + "score": 0.612 + }, + { + "category_id": 1, + "poly": [ + 279, + 930, + 678, + 930, + 678, + 1162, + 279, + 1162 + ], + "score": 0.279 + }, + { + "category_id": 1, + "poly": [ + 283, + 872, + 767, + 872, + 767, + 906, + 283, + 906 + ], + "score": 0.138 + }, + { + "category_id": 5, + "poly": [ + 279, + 930, + 678, + 930, + 678, + 1162, + 279, + 1162 + ], + "score": 0.123, + "html": "
a111b111
111111
111111
sum=9sum=9
uniformuniform
" + }, + { + "category_id": 13, + "poly": [ + 1066, + 468, + 1385, + 468, + 1385, + 800, + 1066, + 800 + ], + "score": 0.61, + "latex": "\\begin{array} { c c c c c c c c c c } { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 125.0, + 201.0, + 537.0, + 201.0, + 537.0, + 276.0, + 125.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 269.0, + 1198.0, + 933.0, + 1198.0, + 933.0, + 1234.0, + 269.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1259.0, + 540.0, + 1259.0, + 540.0, + 1285.0, + 276.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1309.0, + 317.0, + 1309.0, + 317.0, + 1333.0, + 290.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1308.0, + 359.0, + 1308.0, + 359.0, + 1334.0, + 337.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1312.0, + 399.0, + 1312.0, + 399.0, + 1329.0, + 385.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1309.0, + 490.0, + 1309.0, + 490.0, + 1332.0, + 470.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1311.0, + 532.0, + 1311.0, + 532.0, + 1331.0, + 514.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1309.0, + 578.0, + 1309.0, + 578.0, + 1335.0, + 555.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1309.0, + 664.0, + 1309.0, + 664.0, + 1336.0, + 642.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1311.0, + 709.0, + 1311.0, + 709.0, + 1333.0, + 687.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1307.0, + 752.0, + 1307.0, + 752.0, + 1333.0, + 730.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1311.0, + 840.0, + 1311.0, + 840.0, + 1333.0, + 818.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1308.0, + 881.0, + 1308.0, + 881.0, + 1332.0, + 863.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1311.0, + 927.0, + 1311.0, + 927.0, + 1333.0, + 907.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1312.0, + 1008.0, + 1312.0, + 1008.0, + 1322.0, + 998.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1312.0, + 1056.0, + 1312.0, + 1056.0, + 1334.0, + 1037.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1309.0, + 1103.0, + 1309.0, + 1103.0, + 1334.0, + 1079.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1311.0, + 1189.0, + 1311.0, + 1189.0, + 1332.0, + 1169.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1309.0, + 1236.0, + 1309.0, + 1236.0, + 1333.0, + 1208.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1311.0, + 1276.0, + 1311.0, + 1276.0, + 1333.0, + 1257.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1356.0, + 315.0, + 1356.0, + 315.0, + 1384.0, + 293.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1355.0, + 360.0, + 1355.0, + 360.0, + 1379.0, + 336.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1354.0, + 401.0, + 1354.0, + 401.0, + 1376.0, + 381.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1356.0, + 491.0, + 1356.0, + 491.0, + 1379.0, + 469.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1355.0, + 532.0, + 1355.0, + 532.0, + 1376.0, + 512.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1356.0, + 577.0, + 1356.0, + 577.0, + 1378.0, + 557.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1353.0, + 664.0, + 1353.0, + 664.0, + 1379.0, + 642.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1356.0, + 709.0, + 1356.0, + 709.0, + 1379.0, + 687.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1358.0, + 838.0, + 1358.0, + 838.0, + 1378.0, + 819.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1354.0, + 1056.0, + 1354.0, + 1056.0, + 1376.0, + 1037.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1359.0, + 1099.0, + 1359.0, + 1099.0, + 1375.0, + 1084.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1355.0, + 1185.0, + 1355.0, + 1185.0, + 1373.0, + 1173.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1358.0, + 1231.0, + 1358.0, + 1231.0, + 1381.0, + 1213.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1358.0, + 1276.0, + 1358.0, + 1276.0, + 1376.0, + 1255.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1397.0, + 315.0, + 1397.0, + 315.0, + 1424.0, + 293.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1399.0, + 359.0, + 1399.0, + 359.0, + 1424.0, + 336.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1400.0, + 488.0, + 1400.0, + 488.0, + 1422.0, + 469.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1412.0, + 529.0, + 1412.0, + 529.0, + 1420.0, + 518.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1400.0, + 711.0, + 1400.0, + 711.0, + 1424.0, + 685.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1401.0, + 840.0, + 1401.0, + 840.0, + 1421.0, + 818.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1401.0, + 1102.0, + 1401.0, + 1102.0, + 1421.0, + 1081.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1400.0, + 1231.0, + 1400.0, + 1231.0, + 1420.0, + 1213.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1401.0, + 1276.0, + 1401.0, + 1276.0, + 1422.0, + 1257.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1490.0, + 356.0, + 1490.0, + 356.0, + 1511.0, + 338.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1492.0, + 485.0, + 1492.0, + 485.0, + 1510.0, + 471.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1497.0, + 1271.0, + 1497.0, + 1271.0, + 1507.0, + 1261.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1533.0, + 318.0, + 1533.0, + 318.0, + 1561.0, + 288.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1537.0, + 361.0, + 1537.0, + 361.0, + 1559.0, + 342.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1537.0, + 401.0, + 1537.0, + 401.0, + 1557.0, + 384.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1537.0, + 493.0, + 1537.0, + 493.0, + 1559.0, + 473.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1541.0, + 526.0, + 1541.0, + 526.0, + 1550.0, + 516.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1538.0, + 576.0, + 1538.0, + 576.0, + 1560.0, + 556.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1537.0, + 1188.0, + 1537.0, + 1188.0, + 1558.0, + 1169.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1538.0, + 1233.0, + 1538.0, + 1233.0, + 1557.0, + 1211.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1537.0, + 1276.0, + 1537.0, + 1276.0, + 1559.0, + 1258.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1581.0, + 312.0, + 1581.0, + 312.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1581.0, + 358.0, + 1581.0, + 358.0, + 1601.0, + 337.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1577.0, + 402.0, + 1577.0, + 402.0, + 1603.0, + 380.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1581.0, + 488.0, + 1581.0, + 488.0, + 1601.0, + 469.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1580.0, + 532.0, + 1580.0, + 532.0, + 1600.0, + 513.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1581.0, + 577.0, + 1581.0, + 577.0, + 1601.0, + 557.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1581.0, + 1183.0, + 1581.0, + 1183.0, + 1596.0, + 1174.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1581.0, + 1231.0, + 1581.0, + 1231.0, + 1604.0, + 1213.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1581.0, + 1276.0, + 1581.0, + 1276.0, + 1601.0, + 1257.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 1663.0, + 547.0, + 1663.0, + 547.0, + 1702.0, + 279.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1732.0, + 319.0, + 1732.0, + 319.0, + 1756.0, + 293.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1733.0, + 360.0, + 1733.0, + 360.0, + 1756.0, + 340.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1735.0, + 402.0, + 1735.0, + 402.0, + 1753.0, + 384.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1733.0, + 491.0, + 1733.0, + 491.0, + 1759.0, + 469.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1733.0, + 534.0, + 1733.0, + 534.0, + 1755.0, + 514.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1732.0, + 577.0, + 1732.0, + 577.0, + 1753.0, + 558.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1733.0, + 667.0, + 1733.0, + 667.0, + 1756.0, + 647.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1736.0, + 708.0, + 1736.0, + 708.0, + 1752.0, + 692.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1735.0, + 752.0, + 1735.0, + 752.0, + 1757.0, + 733.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1733.0, + 840.0, + 1733.0, + 840.0, + 1755.0, + 821.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1732.0, + 884.0, + 1732.0, + 884.0, + 1753.0, + 864.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 1736.0, + 926.0, + 1736.0, + 926.0, + 1752.0, + 912.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1737.0, + 1011.0, + 1737.0, + 1011.0, + 1749.0, + 1001.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1735.0, + 1058.0, + 1735.0, + 1058.0, + 1756.0, + 1040.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1735.0, + 1102.0, + 1735.0, + 1102.0, + 1757.0, + 1083.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1733.0, + 1190.0, + 1733.0, + 1190.0, + 1752.0, + 1171.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1735.0, + 1234.0, + 1735.0, + 1234.0, + 1755.0, + 1216.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1733.0, + 1278.0, + 1733.0, + 1278.0, + 1753.0, + 1260.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1777.0, + 316.0, + 1777.0, + 316.0, + 1805.0, + 294.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1777.0, + 361.0, + 1777.0, + 361.0, + 1802.0, + 338.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1775.0, + 405.0, + 1775.0, + 405.0, + 1801.0, + 381.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1775.0, + 491.0, + 1775.0, + 491.0, + 1801.0, + 469.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1778.0, + 534.0, + 1778.0, + 534.0, + 1799.0, + 514.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1779.0, + 666.0, + 1779.0, + 666.0, + 1799.0, + 646.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 1778.0, + 709.0, + 1778.0, + 709.0, + 1798.0, + 689.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1779.0, + 753.0, + 1779.0, + 753.0, + 1799.0, + 733.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1778.0, + 840.0, + 1778.0, + 840.0, + 1799.0, + 821.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1777.0, + 1015.0, + 1777.0, + 1015.0, + 1798.0, + 995.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1782.0, + 1057.0, + 1782.0, + 1057.0, + 1797.0, + 1042.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1777.0, + 1102.0, + 1777.0, + 1102.0, + 1799.0, + 1083.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 315.0, + 1822.0, + 315.0, + 1844.0, + 295.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1823.0, + 359.0, + 1823.0, + 359.0, + 1845.0, + 339.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 1824.0, + 484.0, + 1824.0, + 484.0, + 1839.0, + 474.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1823.0, + 537.0, + 1823.0, + 537.0, + 1847.0, + 512.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1824.0, + 667.0, + 1824.0, + 667.0, + 1847.0, + 645.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1824.0, + 842.0, + 1824.0, + 842.0, + 1844.0, + 819.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1914.0, + 356.0, + 1914.0, + 356.0, + 1934.0, + 342.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1916.0, + 660.0, + 1916.0, + 660.0, + 1930.0, + 650.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1957.0, + 319.0, + 1957.0, + 319.0, + 1981.0, + 291.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1958.0, + 364.0, + 1958.0, + 364.0, + 1981.0, + 344.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1962.0, + 400.0, + 1962.0, + 400.0, + 1977.0, + 386.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1958.0, + 670.0, + 1958.0, + 670.0, + 1980.0, + 650.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1963.0, + 704.0, + 1963.0, + 704.0, + 1974.0, + 694.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1960.0, + 752.0, + 1960.0, + 752.0, + 1982.0, + 733.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1961.0, + 1058.0, + 1961.0, + 1058.0, + 1980.0, + 1040.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1961.0, + 1103.0, + 1961.0, + 1103.0, + 1982.0, + 1083.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 316.0, + 2003.0, + 316.0, + 2029.0, + 294.0, + 2029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 2003.0, + 359.0, + 2003.0, + 359.0, + 2026.0, + 339.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 2002.0, + 403.0, + 2002.0, + 403.0, + 2024.0, + 384.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 2004.0, + 666.0, + 2004.0, + 666.0, + 2026.0, + 646.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 2003.0, + 709.0, + 2003.0, + 709.0, + 2024.0, + 690.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 2004.0, + 753.0, + 2004.0, + 753.0, + 2026.0, + 733.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 2002.0, + 1015.0, + 2002.0, + 1015.0, + 2023.0, + 995.0, + 2023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 2004.0, + 1060.0, + 2004.0, + 1060.0, + 2024.0, + 1041.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 2002.0, + 1103.0, + 2002.0, + 1103.0, + 2024.0, + 1083.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 867.0, + 769.0, + 867.0, + 769.0, + 910.0, + 280.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 379.0, + 540.0, + 379.0, + 540.0, + 429.0, + 341.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 379.0, + 920.0, + 379.0, + 920.0, + 428.0, + 725.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 365.0, + 1355.0, + 365.0, + 1355.0, + 439.0, + 1087.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 446.0, + 1024.0, + 446.0, + 1024.0, + 486.0, + 638.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 490.0, + 544.0, + 490.0, + 544.0, + 531.0, + 284.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 497.0, + 578.0, + 497.0, + 578.0, + 526.0, + 552.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 498.0, + 665.0, + 498.0, + 665.0, + 529.0, + 641.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 494.0, + 804.0, + 494.0, + 804.0, + 527.0, + 682.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 498.0, + 843.0, + 498.0, + 843.0, + 522.0, + 813.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 494.0, + 931.0, + 494.0, + 931.0, + 526.0, + 851.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 497.0, + 974.0, + 497.0, + 974.0, + 527.0, + 943.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 497.0, + 1020.0, + 497.0, + 1020.0, + 527.0, + 988.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 541.0, + 318.0, + 541.0, + 318.0, + 573.0, + 288.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 539.0, + 581.0, + 539.0, + 581.0, + 572.0, + 552.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 541.0, + 665.0, + 541.0, + 665.0, + 569.0, + 640.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 544.0, + 711.0, + 544.0, + 711.0, + 572.0, + 683.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 541.0, + 973.0, + 541.0, + 973.0, + 569.0, + 947.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 541.0, + 1019.0, + 541.0, + 1019.0, + 573.0, + 987.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 586.0, + 321.0, + 586.0, + 321.0, + 618.0, + 288.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 587.0, + 577.0, + 587.0, + 577.0, + 612.0, + 556.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 587.0, + 663.0, + 587.0, + 663.0, + 610.0, + 644.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 586.0, + 714.0, + 586.0, + 714.0, + 617.0, + 682.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 587.0, + 971.0, + 587.0, + 971.0, + 612.0, + 949.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 585.0, + 1020.0, + 585.0, + 1020.0, + 617.0, + 988.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 632.0, + 318.0, + 632.0, + 318.0, + 664.0, + 288.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 633.0, + 577.0, + 633.0, + 577.0, + 661.0, + 553.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 635.0, + 662.0, + 635.0, + 662.0, + 661.0, + 644.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 633.0, + 709.0, + 633.0, + 709.0, + 661.0, + 684.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 633.0, + 972.0, + 633.0, + 972.0, + 661.0, + 947.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 632.0, + 1017.0, + 632.0, + 1017.0, + 660.0, + 991.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 678.0, + 317.0, + 678.0, + 317.0, + 707.0, + 291.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 675.0, + 580.0, + 675.0, + 580.0, + 703.0, + 553.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 675.0, + 665.0, + 675.0, + 665.0, + 705.0, + 641.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 678.0, + 711.0, + 678.0, + 711.0, + 705.0, + 684.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 675.0, + 974.0, + 675.0, + 974.0, + 703.0, + 947.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 678.0, + 1019.0, + 678.0, + 1019.0, + 705.0, + 989.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 723.0, + 317.0, + 723.0, + 317.0, + 751.0, + 291.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 725.0, + 574.0, + 725.0, + 574.0, + 744.0, + 559.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 724.0, + 663.0, + 724.0, + 663.0, + 747.0, + 643.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 723.0, + 709.0, + 723.0, + 709.0, + 749.0, + 684.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 724.0, + 971.0, + 724.0, + 971.0, + 746.0, + 950.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 724.0, + 1013.0, + 724.0, + 1013.0, + 748.0, + 992.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 761.0, + 587.0, + 761.0, + 587.0, + 798.0, + 288.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 760.0, + 1026.0, + 760.0, + 1026.0, + 799.0, + 635.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 806.0, + 1028.0, + 806.0, + 1028.0, + 848.0, + 631.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 2058.0, + 854.0, + 2058.0, + 854.0, + 2093.0, + 285.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 124.0, + 295.0, + 840.0, + 295.0, + 840.0, + 353.0, + 124.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1249.0, + 540.0, + 1249.0, + 540.0, + 1287.0, + 276.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 951.0, + 301.0, + 951.0, + 301.0, + 972.0, + 281.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 948.0, + 357.0, + 948.0, + 357.0, + 971.0, + 337.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 946.0, + 402.0, + 946.0, + 402.0, + 973.0, + 379.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 948.0, + 444.0, + 948.0, + 444.0, + 971.0, + 425.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 946.0, + 521.0, + 946.0, + 521.0, + 973.0, + 498.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 948.0, + 576.0, + 948.0, + 576.0, + 971.0, + 555.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 946.0, + 621.0, + 946.0, + 621.0, + 973.0, + 598.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 948.0, + 663.0, + 948.0, + 663.0, + 971.0, + 644.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 994.0, + 358.0, + 994.0, + 358.0, + 1018.0, + 337.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 993.0, + 402.0, + 993.0, + 402.0, + 1019.0, + 379.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 994.0, + 444.0, + 994.0, + 444.0, + 1018.0, + 425.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 994.0, + 576.0, + 994.0, + 576.0, + 1018.0, + 555.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 992.0, + 621.0, + 992.0, + 621.0, + 1019.0, + 597.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 994.0, + 664.0, + 994.0, + 664.0, + 1018.0, + 643.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1039.0, + 357.0, + 1039.0, + 357.0, + 1062.0, + 337.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1037.0, + 402.0, + 1037.0, + 402.0, + 1064.0, + 379.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1037.0, + 446.0, + 1037.0, + 446.0, + 1064.0, + 424.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1037.0, + 577.0, + 1037.0, + 577.0, + 1064.0, + 555.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1037.0, + 622.0, + 1037.0, + 622.0, + 1065.0, + 597.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1037.0, + 666.0, + 1037.0, + 666.0, + 1064.0, + 641.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1080.0, + 440.0, + 1080.0, + 440.0, + 1111.0, + 353.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1078.0, + 660.0, + 1078.0, + 660.0, + 1113.0, + 573.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1122.0, + 424.0, + 1122.0, + 424.0, + 1154.0, + 356.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1122.0, + 642.0, + 1122.0, + 642.0, + 1154.0, + 574.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 867.0, + 769.0, + 867.0, + 769.0, + 910.0, + 280.0, + 910.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 128, + 210, + 838, + 210, + 838, + 268, + 128, + 268 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 128, + 302, + 839, + 302, + 839, + 344, + 128, + 344 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 150, + 1108, + 815, + 1108, + 815, + 1145, + 150, + 1145 + ], + "score": 0.83 + }, + { + "category_id": 0, + "poly": [ + 152, + 788, + 636, + 788, + 636, + 821, + 152, + 821 + ], + "score": 0.801 + }, + { + "category_id": 3, + "poly": [ + 219, + 1580, + 1507, + 1580, + 1507, + 1763, + 219, + 1763 + ], + "score": 0.762 + }, + { + "category_id": 3, + "poly": [ + 225, + 1887, + 1535, + 1887, + 1535, + 2067, + 225, + 2067 + ], + "score": 0.743 + }, + { + "category_id": 3, + "poly": [ + 218, + 1276, + 1067, + 1276, + 1067, + 1456, + 218, + 1456 + ], + "score": 0.639 + }, + { + "category_id": 3, + "poly": [ + 141, + 444, + 977, + 444, + 977, + 741, + 141, + 741 + ], + "score": 0.53 + }, + { + "category_id": 1, + "poly": [ + 218, + 1228, + 478, + 1228, + 478, + 1257, + 218, + 1257 + ], + "score": 0.502 + }, + { + "category_id": 3, + "poly": [ + 145, + 836, + 1437, + 836, + 1437, + 1040, + 145, + 1040 + ], + "score": 0.409 + }, + { + "category_id": 3, + "poly": [ + 145, + 836, + 1434, + 836, + 1434, + 1040, + 145, + 1040 + ], + "score": 0.33 + }, + { + "category_id": 5, + "poly": [ + 218, + 1276, + 1067, + 1276, + 1067, + 1456, + 218, + 1456 + ], + "score": 0.276, + "html": "
aabCaabCaabCaa bC
aabCaab Ca dabCaabC f
ddefddefdefdde
gghigghigghiggh :
" + }, + { + "category_id": 5, + "poly": [ + 225, + 1887, + 1535, + 1887, + 1535, + 2067, + 225, + 2067 + ], + "score": 0.234, + "html": "
aaba abCa abCa abCaab CaabC
aabaaCaa bCaa bCaabCaa bC
ddedefdedefd defddf
gghghighghgg hggh
" + }, + { + "category_id": 6, + "poly": [ + 222, + 1854, + 482, + 1854, + 482, + 1881, + 222, + 1881 + ], + "score": 0.224 + }, + { + "category_id": 5, + "poly": [ + 219, + 1580, + 1507, + 1580, + 1507, + 1763, + 219, + 1763 + ], + "score": 0.189, + "html": "
aab Caab CaabCade faabCaab C
aab CaabCaab CaabCa abCaabC
ddedde fddefd defdd efddef
ggh iggh igghg ghggh igh
g
" + }, + { + "category_id": 4, + "poly": [ + 218, + 1529, + 478, + 1529, + 478, + 1557, + 218, + 1557 + ], + "score": 0.143 + }, + { + "category_id": 1, + "poly": [ + 218, + 1529, + 478, + 1529, + 478, + 1557, + 218, + 1557 + ], + "score": 0.137 + }, + { + "category_id": 13, + "poly": [ + 746, + 554, + 940, + 554, + 940, + 732, + 746, + 732 + ], + "score": 0.9, + "latex": "\\left| \\begin{array} { c c c c c c c } { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\\\ { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } & { { 9 } } \\end{array} \\right." + }, + { + "category_id": 13, + "poly": [ + 427, + 534, + 641, + 534, + 641, + 736, + 427, + 736 + ], + "score": 0.58, + "latex": "{ \\begin{array} { l } { \\mathrm { ~ a ~ } \\quad \\mathrm { ~ a ~ } \\quad \\mathrm { ~ b ~ } \\quad \\mathrm { ~ c ~ } \\quad \\dots \\quad \\dots } \\\\ { \\mathrm { ~ a ~ } { \\Biggl [ } \\mathrm { ~ a ~ } \\quad \\mathrm { ~ b ~ } \\quad \\mathrm { ~ c ~ } \\quad \\dots \\quad \\dots } \\\\ { \\mathrm { ~ d ~ } \\quad \\mathrm { ~ e ~ } \\quad \\mathrm { ~ f ~ } \\quad \\dots \\quad \\dots } \\\\ { \\mathrm { ~ g ~ } { \\Biggl ] } \\quad \\mathrm { ~ { \\dot { \\Gamma } } ~ { \\dot { \\mathrm { ~ h ~ } } } ~ { \\dot { \\mathrm { ~ i ~ } } } \\quad \\dots \\quad \\dots } } \\\\ { \\dots \\quad \\dots \\quad \\dots \\quad \\dots \\quad \\dots } \\\\ { \\dots \\quad \\dots \\quad \\dots \\quad \\dots \\quad \\dots } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1591, + 1505, + 1591, + 1505, + 1760, + 1322, + 1760 + ], + "score": 0.49, + "latex": "\\begin{array} { r } { \\begin{array} { c c l } { \\mathrm { a } } & { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { d } } & { \\mathrm { e } } & { \\mathrm { f } } \\\\ { \\mathrm { g } } & { \\mathrm { h } } & { \\mathrm { i } } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 885, + 1896, + 1051, + 1896, + 1051, + 2065, + 885, + 2065 + ], + "score": 0.44, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ \\in ~ { \\tt ~ b } ~ \\Gamma ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ | ~ a ~ \\mathrm { ~ b ~ \\gamma ~ c ~ } ~ ~ } } \\\\ & { \\mathrm { ~ d ~ | ~ { \\tt ~ d } ~ \\mathrm { ~ e ~ } ~ { \\mathrm { ~ f ~ } } ~ } } \\\\ & { \\mathrm { ~ g ~ | ~ { \\tt ~ p ~ \\mathrm { ~ h ~ } ~ \\mathrm { ~ i ~ } ~ } ~ ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1172, + 869, + 1283, + 869, + 1283, + 967, + 1172, + 967 + ], + "score": 0.42, + "latex": "\\begin{array} { c c c } { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 669, + 1895, + 845, + 1895, + 845, + 2064, + 669, + 2064 + ], + "score": 0.42, + "latex": "\\begin{array} { r } { \\begin{array} { c c c } { \\mathrm { a } } & { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { d } } & { \\mathrm { e } } & { \\mathrm { f } } \\\\ { \\mathrm { g } } & { \\mathrm { h } } & { \\mathrm { i } } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 670, + 1591, + 835, + 1591, + 835, + 1759, + 670, + 1759 + ], + "score": 0.39, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { ~ b ~ } } & { \\mathrm { ~ c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { ~ e ~ } } & { \\mathrm { ~ f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { ~ h ~ } } & { \\mathrm { ~ i ~ } } & { } \\end{array} \\right. } \\\\ & { \\mathrm { ~ \\gamma ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 648, + 975, + 682, + 975, + 682, + 996, + 648, + 996 + ], + "score": 0.38, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 886, + 1590, + 1053, + 1590, + 1053, + 1760, + 886, + 1760 + ], + "score": 0.35, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ ~ \\gamma ~ d ~ ~ e ~ ~ f ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { ~ b ~ } } & { \\mathrm { ~ c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { ~ e ~ } } & { \\mathrm { ~ f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { ~ h ~ } } & { \\mathrm { ~ i ~ } } \\end{array} \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 794, + 974, + 828, + 974, + 828, + 997, + 794, + 997 + ], + "score": 0.32, + "latex": "= 9" + }, + { + "category_id": 13, + "poly": [ + 668, + 1285, + 844, + 1285, + 844, + 1454, + 668, + 1454 + ], + "score": 0.32, + "latex": "\\begin{array} { r } { \\begin{array} { c c c } { \\mathrm { a } } & { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { d } } & { \\mathrm { e } } & { \\mathrm { f } } \\\\ { \\mathrm { g } } & { \\mathrm { h } } & { \\mathrm { i } } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 874, + 870, + 992, + 870, + 992, + 967, + 874, + 967 + ], + "score": 0.31, + "latex": "\\begin{array} { c c c } { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1086, + 975, + 1119, + 975, + 1119, + 996, + 1086, + 996 + ], + "score": 0.29, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1590, + 1273, + 1590, + 1273, + 1760, + 1103, + 1760 + ], + "score": 0.28, + "latex": "\\begin{array} { r } { \\begin{array} { c c l } { \\mathbf { a } } & { \\mathbf { a } } & { \\mathbf { b } } & { \\mathbf { c } } \\\\ { \\mathbf { a } } & { \\mathbf { b } } & { \\mathbf { c } } \\\\ { \\mathbf { d } } & { \\mathbf { e } } & { \\mathbf { f } } \\\\ { \\mathbf { g } } & { \\mathbf { h } } & { \\mathbf { i } } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 445, + 1896, + 627, + 1896, + 627, + 2066, + 445, + 2066 + ], + "score": 0.26, + "latex": "\\begin{array} { r } { \\begin{array} { c c l } { \\mathrm { a } } & { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { d } } & { \\mathrm { e } } & { \\mathrm { f } } \\\\ { \\mathrm { g } } & { \\mathrm { h } } & { \\mathrm { i } } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1102, + 1895, + 1255, + 1895, + 1255, + 2065, + 1102, + 2065 + ], + "score": 0.25, + "latex": "\\begin{array} { l } { \\mathrm { ~ a ~ } \\mathrm { ~ a ~ } \\mathrm { ~ b ~ } \\mathrm { ~ c ~ } } \\\\ { \\mathrm { ~ a ~ } \\Bigg [ \\mathrm { ~ a ~ } \\mathrm { ~ b ~ } \\mathrm { ~ c ~ } } \\\\ { \\mathrm { ~ d ~ } \\mathrm { ~ e ~ } \\mathrm { ~ f ~ } } \\\\ { \\mathrm { ~ g ~ } \\Bigg | \\mathrm { ~ g ~ } \\mathrm { ~ h ~ } \\mathrm { ~ i ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 940, + 975, + 973, + 975, + 973, + 996, + 940, + 996 + ], + "score": 0.25, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 15, + "poly": [ + 121.0, + 204.0, + 840.0, + 204.0, + 840.0, + 276.0, + 121.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 151.0, + 1109.0, + 814.0, + 1109.0, + 814.0, + 1146.0, + 151.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 150.0, + 787.0, + 638.0, + 787.0, + 638.0, + 826.0, + 150.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 230.0, + 1593.0, + 253.0, + 1593.0, + 253.0, + 1618.0, + 230.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1593.0, + 290.0, + 1593.0, + 290.0, + 1618.0, + 266.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1592.0, + 325.0, + 1592.0, + 325.0, + 1617.0, + 302.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1593.0, + 363.0, + 1593.0, + 363.0, + 1618.0, + 340.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1593.0, + 473.0, + 1593.0, + 473.0, + 1618.0, + 448.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1593.0, + 509.0, + 1593.0, + 509.0, + 1618.0, + 484.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1592.0, + 545.0, + 1592.0, + 545.0, + 1618.0, + 522.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1593.0, + 580.0, + 1593.0, + 580.0, + 1618.0, + 556.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1595.0, + 669.0, + 1595.0, + 669.0, + 1618.0, + 666.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 1627.0, + 253.0, + 1627.0, + 253.0, + 1652.0, + 228.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1627.0, + 290.0, + 1627.0, + 290.0, + 1652.0, + 266.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1624.0, + 325.0, + 1624.0, + 325.0, + 1651.0, + 302.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1627.0, + 363.0, + 1627.0, + 363.0, + 1654.0, + 338.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1629.0, + 473.0, + 1629.0, + 473.0, + 1652.0, + 448.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1627.0, + 509.0, + 1627.0, + 509.0, + 1652.0, + 484.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1624.0, + 545.0, + 1624.0, + 545.0, + 1651.0, + 522.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1627.0, + 581.0, + 1627.0, + 581.0, + 1654.0, + 556.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1627.0, + 669.0, + 1627.0, + 669.0, + 1652.0, + 666.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 1660.0, + 253.0, + 1660.0, + 253.0, + 1688.0, + 228.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1660.0, + 289.0, + 1660.0, + 289.0, + 1688.0, + 266.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1661.0, + 327.0, + 1661.0, + 327.0, + 1688.0, + 302.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1658.0, + 364.0, + 1658.0, + 364.0, + 1685.0, + 343.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1660.0, + 471.0, + 1660.0, + 471.0, + 1686.0, + 448.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1660.0, + 509.0, + 1660.0, + 509.0, + 1686.0, + 484.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1661.0, + 543.0, + 1661.0, + 543.0, + 1686.0, + 522.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1661.0, + 581.0, + 1661.0, + 581.0, + 1683.0, + 564.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 1697.0, + 253.0, + 1697.0, + 253.0, + 1725.0, + 228.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1697.0, + 289.0, + 1697.0, + 289.0, + 1723.0, + 266.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1694.0, + 325.0, + 1694.0, + 325.0, + 1720.0, + 302.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1698.0, + 356.0, + 1698.0, + 356.0, + 1716.0, + 344.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1697.0, + 470.0, + 1697.0, + 470.0, + 1723.0, + 448.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1697.0, + 507.0, + 1697.0, + 507.0, + 1723.0, + 484.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1695.0, + 545.0, + 1695.0, + 545.0, + 1720.0, + 522.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1697.0, + 577.0, + 1697.0, + 577.0, + 1717.0, + 561.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 1901.0, + 250.0, + 1901.0, + 250.0, + 1922.0, + 231.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1898.0, + 290.0, + 1898.0, + 290.0, + 1923.0, + 266.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1897.0, + 325.0, + 1897.0, + 325.0, + 1922.0, + 301.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1898.0, + 363.0, + 1898.0, + 363.0, + 1922.0, + 340.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1898.0, + 1347.0, + 1898.0, + 1347.0, + 1923.0, + 1322.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1900.0, + 1384.0, + 1900.0, + 1384.0, + 1923.0, + 1359.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1897.0, + 1419.0, + 1897.0, + 1419.0, + 1922.0, + 1394.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1432.0, + 1898.0, + 1454.0, + 1898.0, + 1454.0, + 1923.0, + 1432.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 229.0, + 1933.0, + 253.0, + 1933.0, + 253.0, + 1958.0, + 229.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1932.0, + 290.0, + 1932.0, + 290.0, + 1958.0, + 266.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1930.0, + 325.0, + 1930.0, + 325.0, + 1957.0, + 301.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1935.0, + 360.0, + 1935.0, + 360.0, + 1955.0, + 341.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1932.0, + 1347.0, + 1932.0, + 1347.0, + 1958.0, + 1322.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1932.0, + 1384.0, + 1932.0, + 1384.0, + 1958.0, + 1359.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 1930.0, + 1419.0, + 1930.0, + 1419.0, + 1957.0, + 1396.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1432.0, + 1933.0, + 1456.0, + 1933.0, + 1456.0, + 1957.0, + 1432.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 229.0, + 1964.0, + 253.0, + 1964.0, + 253.0, + 1990.0, + 229.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 1964.0, + 290.0, + 1964.0, + 290.0, + 1990.0, + 266.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1965.0, + 325.0, + 1965.0, + 325.0, + 1990.0, + 301.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1968.0, + 359.0, + 1968.0, + 359.0, + 1980.0, + 348.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1964.0, + 1347.0, + 1964.0, + 1347.0, + 1990.0, + 1322.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1964.0, + 1384.0, + 1964.0, + 1384.0, + 1990.0, + 1359.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1965.0, + 1419.0, + 1965.0, + 1419.0, + 1990.0, + 1394.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1441.0, + 1967.0, + 1453.0, + 1967.0, + 1453.0, + 1984.0, + 1441.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 229.0, + 2000.0, + 251.0, + 2000.0, + 251.0, + 2028.0, + 229.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 2000.0, + 290.0, + 2000.0, + 290.0, + 2027.0, + 266.0, + 2027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1997.0, + 325.0, + 1997.0, + 325.0, + 2022.0, + 301.0, + 2022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 2000.0, + 1347.0, + 2000.0, + 1347.0, + 2028.0, + 1324.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 2000.0, + 1384.0, + 2000.0, + 1384.0, + 2028.0, + 1359.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 1997.0, + 1419.0, + 1997.0, + 1419.0, + 2024.0, + 1396.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1438.0, + 2003.0, + 1448.0, + 2003.0, + 1448.0, + 2015.0, + 1438.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 230.0, + 1287.0, + 326.0, + 1287.0, + 326.0, + 1314.0, + 230.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1288.0, + 363.0, + 1288.0, + 363.0, + 1314.0, + 338.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1283.0, + 585.0, + 1283.0, + 585.0, + 1318.0, + 443.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1283.0, + 1021.0, + 1283.0, + 1021.0, + 1315.0, + 886.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 1323.0, + 253.0, + 1323.0, + 253.0, + 1348.0, + 228.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 1323.0, + 291.0, + 1323.0, + 291.0, + 1347.0, + 265.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1322.0, + 323.0, + 1322.0, + 323.0, + 1345.0, + 304.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1322.0, + 362.0, + 1322.0, + 362.0, + 1348.0, + 339.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1323.0, + 471.0, + 1323.0, + 471.0, + 1348.0, + 447.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1323.0, + 509.0, + 1323.0, + 509.0, + 1347.0, + 484.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1322.0, + 543.0, + 1322.0, + 543.0, + 1346.0, + 523.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1322.0, + 581.0, + 1322.0, + 581.0, + 1348.0, + 557.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1323.0, + 909.0, + 1323.0, + 909.0, + 1348.0, + 886.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 1323.0, + 946.0, + 1323.0, + 946.0, + 1347.0, + 921.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1323.0, + 980.0, + 1323.0, + 980.0, + 1346.0, + 960.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1324.0, + 1017.0, + 1324.0, + 1017.0, + 1347.0, + 997.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 229.0, + 1354.0, + 252.0, + 1354.0, + 252.0, + 1381.0, + 229.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 1355.0, + 289.0, + 1355.0, + 289.0, + 1381.0, + 265.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1358.0, + 324.0, + 1358.0, + 324.0, + 1381.0, + 303.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1355.0, + 363.0, + 1355.0, + 363.0, + 1380.0, + 344.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1354.0, + 470.0, + 1354.0, + 470.0, + 1382.0, + 449.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1357.0, + 507.0, + 1357.0, + 507.0, + 1379.0, + 485.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1358.0, + 543.0, + 1358.0, + 543.0, + 1381.0, + 523.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1355.0, + 580.0, + 1355.0, + 580.0, + 1380.0, + 562.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1355.0, + 909.0, + 1355.0, + 909.0, + 1382.0, + 886.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1357.0, + 944.0, + 1357.0, + 944.0, + 1380.0, + 924.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1359.0, + 980.0, + 1359.0, + 980.0, + 1380.0, + 960.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1357.0, + 1017.0, + 1357.0, + 1017.0, + 1377.0, + 1002.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 229.0, + 1391.0, + 252.0, + 1391.0, + 252.0, + 1418.0, + 229.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 1386.0, + 335.0, + 1386.0, + 335.0, + 1418.0, + 265.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1390.0, + 357.0, + 1390.0, + 357.0, + 1410.0, + 342.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1392.0, + 471.0, + 1392.0, + 471.0, + 1418.0, + 447.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1392.0, + 507.0, + 1392.0, + 507.0, + 1418.0, + 483.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1389.0, + 546.0, + 1389.0, + 546.0, + 1415.0, + 521.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1390.0, + 576.0, + 1390.0, + 576.0, + 1413.0, + 561.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1391.0, + 909.0, + 1391.0, + 909.0, + 1418.0, + 886.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 1392.0, + 947.0, + 1392.0, + 947.0, + 1418.0, + 921.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1389.0, + 981.0, + 1389.0, + 981.0, + 1415.0, + 958.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1392.0, + 1014.0, + 1392.0, + 1014.0, + 1410.0, + 998.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 142.0, + 453.0, + 340.0, + 453.0, + 340.0, + 499.0, + 142.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 454.0, + 659.0, + 454.0, + 659.0, + 497.0, + 465.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 441.0, + 981.0, + 441.0, + 981.0, + 509.0, + 715.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 154.0, + 565.0, + 185.0, + 565.0, + 185.0, + 595.0, + 154.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 566.0, + 217.0, + 566.0, + 217.0, + 592.0, + 193.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 230.0, + 567.0, + 253.0, + 567.0, + 253.0, + 594.0, + 230.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 157.0, + 599.0, + 182.0, + 599.0, + 182.0, + 626.0, + 157.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 602.0, + 215.0, + 602.0, + 215.0, + 624.0, + 195.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 603.0, + 250.0, + 603.0, + 250.0, + 617.0, + 237.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 155.0, + 635.0, + 183.0, + 635.0, + 183.0, + 665.0, + 155.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 192.0, + 634.0, + 218.0, + 634.0, + 218.0, + 660.0, + 192.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 153.0, + 839.0, + 183.0, + 839.0, + 183.0, + 866.0, + 153.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 836.0, + 330.0, + 836.0, + 330.0, + 865.0, + 298.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 840.0, + 475.0, + 840.0, + 475.0, + 863.0, + 449.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 836.0, + 621.0, + 836.0, + 621.0, + 865.0, + 591.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 840.0, + 766.0, + 840.0, + 766.0, + 863.0, + 739.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 838.0, + 910.0, + 838.0, + 910.0, + 863.0, + 885.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 839.0, + 1061.0, + 839.0, + 1061.0, + 867.0, + 1029.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 839.0, + 1203.0, + 839.0, + 1203.0, + 862.0, + 1175.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 839.0, + 1346.0, + 839.0, + 1346.0, + 862.0, + 1325.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 156.0, + 873.0, + 181.0, + 873.0, + 181.0, + 897.0, + 156.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 194.0, + 872.0, + 215.0, + 872.0, + 215.0, + 896.0, + 194.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 874.0, + 324.0, + 874.0, + 324.0, + 893.0, + 305.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 873.0, + 362.0, + 873.0, + 362.0, + 895.0, + 339.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 874.0, + 395.0, + 874.0, + 395.0, + 893.0, + 378.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 874.0, + 468.0, + 874.0, + 468.0, + 893.0, + 450.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 877.0, + 539.0, + 877.0, + 539.0, + 892.0, + 527.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 873.0, + 617.0, + 873.0, + 617.0, + 896.0, + 594.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 874.0, + 650.0, + 874.0, + 650.0, + 893.0, + 631.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 874.0, + 761.0, + 874.0, + 761.0, + 892.0, + 740.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 874.0, + 795.0, + 874.0, + 795.0, + 893.0, + 778.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 876.0, + 829.0, + 876.0, + 829.0, + 892.0, + 817.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 873.0, + 1056.0, + 873.0, + 1056.0, + 896.0, + 1030.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 874.0, + 1087.0, + 874.0, + 1087.0, + 893.0, + 1071.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 877.0, + 1339.0, + 877.0, + 1339.0, + 889.0, + 1329.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 877.0, + 1380.0, + 877.0, + 1380.0, + 892.0, + 1361.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 157.0, + 905.0, + 181.0, + 905.0, + 181.0, + 930.0, + 157.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 194.0, + 907.0, + 217.0, + 907.0, + 217.0, + 930.0, + 194.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 910.0, + 324.0, + 910.0, + 324.0, + 927.0, + 305.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 907.0, + 362.0, + 907.0, + 362.0, + 928.0, + 339.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 908.0, + 395.0, + 908.0, + 395.0, + 927.0, + 378.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 910.0, + 469.0, + 910.0, + 469.0, + 927.0, + 450.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 910.0, + 505.0, + 910.0, + 505.0, + 926.0, + 485.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 908.0, + 540.0, + 908.0, + 540.0, + 927.0, + 523.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 905.0, + 617.0, + 905.0, + 617.0, + 928.0, + 594.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 907.0, + 652.0, + 907.0, + 652.0, + 930.0, + 630.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 908.0, + 761.0, + 908.0, + 761.0, + 926.0, + 740.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 907.0, + 800.0, + 907.0, + 800.0, + 930.0, + 775.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 908.0, + 833.0, + 908.0, + 833.0, + 927.0, + 814.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 907.0, + 1056.0, + 907.0, + 1056.0, + 930.0, + 1032.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 907.0, + 1091.0, + 907.0, + 1091.0, + 928.0, + 1068.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 911.0, + 1341.0, + 911.0, + 1341.0, + 924.0, + 1328.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 910.0, + 1381.0, + 910.0, + 1381.0, + 927.0, + 1361.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 912.0, + 1413.0, + 912.0, + 1413.0, + 923.0, + 1400.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 941.0, + 617.0, + 941.0, + 617.0, + 964.0, + 594.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 941.0, + 652.0, + 941.0, + 652.0, + 964.0, + 629.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 943.0, + 761.0, + 943.0, + 761.0, + 961.0, + 740.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 941.0, + 800.0, + 941.0, + 800.0, + 964.0, + 775.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 941.0, + 835.0, + 941.0, + 835.0, + 964.0, + 811.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 941.0, + 1058.0, + 941.0, + 1058.0, + 964.0, + 1030.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 941.0, + 1090.0, + 941.0, + 1090.0, + 964.0, + 1065.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 943.0, + 1345.0, + 943.0, + 1345.0, + 961.0, + 1325.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 942.0, + 1383.0, + 942.0, + 1383.0, + 964.0, + 1358.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 946.0, + 1413.0, + 946.0, + 1413.0, + 957.0, + 1399.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 162.0, + 971.0, + 250.0, + 971.0, + 250.0, + 998.0, + 162.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 971.0, + 398.0, + 971.0, + 398.0, + 1002.0, + 310.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 971.0, + 543.0, + 971.0, + 543.0, + 1002.0, + 456.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 969.0, + 647.0, + 969.0, + 647.0, + 1002.0, + 603.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 969.0, + 690.0, + 969.0, + 690.0, + 1002.0, + 683.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 971.0, + 793.0, + 971.0, + 793.0, + 1002.0, + 747.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 971.0, + 835.0, + 971.0, + 835.0, + 1002.0, + 829.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 971.0, + 939.0, + 971.0, + 939.0, + 1002.0, + 892.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 971.0, + 981.0, + 971.0, + 981.0, + 1002.0, + 974.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 969.0, + 1085.0, + 969.0, + 1085.0, + 1002.0, + 1039.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 969.0, + 1126.0, + 969.0, + 1126.0, + 1002.0, + 1120.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 971.0, + 1272.0, + 971.0, + 1272.0, + 1002.0, + 1185.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 971.0, + 1419.0, + 971.0, + 1419.0, + 1002.0, + 1329.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1001.0, + 831.0, + 1001.0, + 831.0, + 1038.0, + 747.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1001.0, + 977.0, + 1001.0, + 977.0, + 1038.0, + 892.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1004.0, + 1267.0, + 1004.0, + 1267.0, + 1037.0, + 1184.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 1003.0, + 1415.0, + 1003.0, + 1415.0, + 1038.0, + 1330.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 153.0, + 839.0, + 183.0, + 839.0, + 183.0, + 866.0, + 153.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 836.0, + 331.0, + 836.0, + 331.0, + 865.0, + 299.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 840.0, + 476.0, + 840.0, + 476.0, + 863.0, + 448.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 836.0, + 623.0, + 836.0, + 623.0, + 865.0, + 590.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 839.0, + 768.0, + 839.0, + 768.0, + 865.0, + 736.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 838.0, + 910.0, + 838.0, + 910.0, + 862.0, + 885.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 839.0, + 1060.0, + 839.0, + 1060.0, + 867.0, + 1028.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 836.0, + 1205.0, + 836.0, + 1205.0, + 863.0, + 1173.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 838.0, + 1348.0, + 838.0, + 1348.0, + 862.0, + 1323.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 156.0, + 873.0, + 180.0, + 873.0, + 180.0, + 897.0, + 156.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 872.0, + 217.0, + 872.0, + 217.0, + 896.0, + 193.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 874.0, + 324.0, + 874.0, + 324.0, + 893.0, + 305.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 876.0, + 360.0, + 876.0, + 360.0, + 892.0, + 341.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 874.0, + 396.0, + 874.0, + 396.0, + 893.0, + 377.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 874.0, + 468.0, + 874.0, + 468.0, + 893.0, + 449.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 873.0, + 507.0, + 873.0, + 507.0, + 895.0, + 484.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 874.0, + 541.0, + 874.0, + 541.0, + 893.0, + 525.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 873.0, + 616.0, + 873.0, + 616.0, + 896.0, + 594.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 873.0, + 652.0, + 873.0, + 652.0, + 896.0, + 630.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 874.0, + 759.0, + 874.0, + 759.0, + 893.0, + 742.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 876.0, + 792.0, + 876.0, + 792.0, + 892.0, + 781.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 876.0, + 834.0, + 876.0, + 834.0, + 893.0, + 814.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 873.0, + 1054.0, + 873.0, + 1054.0, + 896.0, + 1031.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 873.0, + 1090.0, + 873.0, + 1090.0, + 895.0, + 1067.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 874.0, + 1344.0, + 874.0, + 1344.0, + 892.0, + 1325.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 873.0, + 1381.0, + 873.0, + 1381.0, + 896.0, + 1360.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 876.0, + 1412.0, + 876.0, + 1412.0, + 892.0, + 1400.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 157.0, + 905.0, + 180.0, + 905.0, + 180.0, + 930.0, + 157.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 907.0, + 217.0, + 907.0, + 217.0, + 930.0, + 193.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 910.0, + 325.0, + 910.0, + 325.0, + 927.0, + 306.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 908.0, + 360.0, + 908.0, + 360.0, + 926.0, + 340.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 908.0, + 395.0, + 908.0, + 395.0, + 928.0, + 377.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 907.0, + 471.0, + 907.0, + 471.0, + 930.0, + 448.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 908.0, + 504.0, + 908.0, + 504.0, + 927.0, + 486.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 908.0, + 541.0, + 908.0, + 541.0, + 927.0, + 522.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 905.0, + 616.0, + 905.0, + 616.0, + 930.0, + 594.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 905.0, + 653.0, + 905.0, + 653.0, + 931.0, + 629.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 910.0, + 759.0, + 910.0, + 759.0, + 927.0, + 742.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 908.0, + 800.0, + 908.0, + 800.0, + 928.0, + 775.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 910.0, + 834.0, + 910.0, + 834.0, + 927.0, + 814.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 908.0, + 1053.0, + 908.0, + 1053.0, + 927.0, + 1034.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 907.0, + 1090.0, + 907.0, + 1090.0, + 928.0, + 1067.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 908.0, + 1344.0, + 908.0, + 1344.0, + 926.0, + 1325.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 907.0, + 1383.0, + 907.0, + 1383.0, + 930.0, + 1360.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 911.0, + 1413.0, + 911.0, + 1413.0, + 924.0, + 1400.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 939.0, + 617.0, + 939.0, + 617.0, + 964.0, + 593.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 939.0, + 653.0, + 939.0, + 653.0, + 965.0, + 629.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 941.0, + 761.0, + 941.0, + 761.0, + 964.0, + 739.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 941.0, + 801.0, + 941.0, + 801.0, + 964.0, + 774.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 941.0, + 836.0, + 941.0, + 836.0, + 964.0, + 811.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 941.0, + 1056.0, + 941.0, + 1056.0, + 964.0, + 1031.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 941.0, + 1090.0, + 941.0, + 1090.0, + 964.0, + 1066.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 941.0, + 1345.0, + 941.0, + 1345.0, + 964.0, + 1322.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 941.0, + 1383.0, + 941.0, + 1383.0, + 965.0, + 1358.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 943.0, + 1416.0, + 943.0, + 1416.0, + 961.0, + 1399.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 162.0, + 971.0, + 250.0, + 971.0, + 250.0, + 998.0, + 162.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 971.0, + 397.0, + 971.0, + 397.0, + 1002.0, + 311.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 971.0, + 544.0, + 971.0, + 544.0, + 1002.0, + 457.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 969.0, + 647.0, + 969.0, + 647.0, + 1002.0, + 601.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 969.0, + 688.0, + 969.0, + 688.0, + 1002.0, + 683.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 971.0, + 793.0, + 971.0, + 793.0, + 1002.0, + 748.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 971.0, + 836.0, + 971.0, + 836.0, + 1002.0, + 829.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 971.0, + 939.0, + 971.0, + 939.0, + 1002.0, + 892.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 971.0, + 980.0, + 971.0, + 980.0, + 1002.0, + 974.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 969.0, + 1085.0, + 969.0, + 1085.0, + 1002.0, + 1040.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 969.0, + 1127.0, + 969.0, + 1127.0, + 1002.0, + 1120.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 971.0, + 1273.0, + 971.0, + 1273.0, + 1002.0, + 1184.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 971.0, + 1417.0, + 971.0, + 1417.0, + 1002.0, + 1331.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1001.0, + 831.0, + 1001.0, + 831.0, + 1038.0, + 747.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1001.0, + 977.0, + 1001.0, + 977.0, + 1038.0, + 893.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1004.0, + 1267.0, + 1004.0, + 1267.0, + 1037.0, + 1184.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 1001.0, + 1414.0, + 1001.0, + 1414.0, + 1038.0, + 1330.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 1848.0, + 483.0, + 1848.0, + 483.0, + 1887.0, + 220.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 1523.0, + 479.0, + 1523.0, + 479.0, + 1562.0, + 216.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 125.0, + 295.0, + 839.0, + 295.0, + 839.0, + 352.0, + 125.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 215.0, + 1223.0, + 480.0, + 1223.0, + 480.0, + 1262.0, + 215.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 1523.0, + 479.0, + 1523.0, + 479.0, + 1562.0, + 216.0, + 1562.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 223, + 963, + 1458, + 963, + 1458, + 1363, + 223, + 1363 + ], + "score": 0.908 + }, + { + "category_id": 3, + "poly": [ + 215, + 440, + 1484, + 440, + 1484, + 839, + 215, + 839 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 140, + 290, + 623, + 290, + 623, + 317, + 140, + 317 + ], + "score": 0.798 + }, + { + "category_id": 1, + "poly": [ + 139, + 359, + 397, + 359, + 397, + 386, + 139, + 386 + ], + "score": 0.722 + }, + { + "category_id": 1, + "poly": [ + 132, + 1573, + 617, + 1573, + 617, + 1602, + 132, + 1602 + ], + "score": 0.638 + }, + { + "category_id": 1, + "poly": [ + 129, + 1703, + 653, + 1703, + 653, + 1733, + 129, + 1733 + ], + "score": 0.512 + }, + { + "category_id": 1, + "poly": [ + 130, + 1638, + 609, + 1638, + 609, + 1668, + 130, + 1668 + ], + "score": 0.495 + }, + { + "category_id": 1, + "poly": [ + 135, + 899, + 390, + 899, + 390, + 926, + 135, + 926 + ], + "score": 0.481 + }, + { + "category_id": 4, + "poly": [ + 135, + 899, + 390, + 899, + 390, + 926, + 135, + 926 + ], + "score": 0.195 + }, + { + "category_id": 13, + "poly": [ + 225, + 1188, + 403, + 1188, + 403, + 1362, + 225, + 1362 + ], + "score": 0.69, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { ~ b ~ } } & { \\mathrm { ~ c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { ~ e ~ } } & { \\mathrm { ~ f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { ~ h ~ } } & { \\mathrm { ~ i ~ } } \\end{array} \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 444, + 1187, + 618, + 1187, + 618, + 1362, + 444, + 1362 + ], + "score": 0.61, + "latex": "\\begin{array} { r } { \\begin{array} { l l l l } { \\mathrm { a } } & { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } \\\\ { \\mathrm { a } } & { \\mathrm { b } } & { \\mathrm { c } } & { } \\\\ { \\mathrm { d } } & { \\mathrm { e } } & { \\mathrm { f } } & { } \\\\ { \\mathrm { g } } & { \\mathrm { h } } & { \\mathrm { i } } & { } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 662, + 977, + 837, + 977, + 837, + 1154, + 662, + 1154 + ], + "score": 0.5, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { b ~ } } & { \\mathrm { c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { e ~ } } & { \\mathrm { ~ f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { h ~ } } & { \\mathrm { i ~ } } \\end{array} \\right. } \\\\ & { \\mathrm { ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 445, + 977, + 624, + 977, + 624, + 1152, + 445, + 1152 + ], + "score": 0.47, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { b ~ } } & { \\mathrm { c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { e ~ } } & { \\mathrm { f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { h ~ } } & { \\mathrm { i ~ } } \\end{array} \\right. } \\\\ & { \\mathrm { ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 225, + 977, + 405, + 977, + 405, + 1152, + 225, + 1152 + ], + "score": 0.44, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ \\in ~ { \\mathfrak { a } } ~ } } \\\\ & { \\mathrm { ~ a ~ \\left[ \\begin{array} { l l l } { a } & { b } & { c } \\\\ { d } & { e } & { { \\mathfrak { f } } } \\end{array} \\right] ~ } } \\\\ & { \\mathrm { ~ g ~ \\left[ \\begin{array} { l l l } { \\mathfrak { g } } & { h } & { i } \\end{array} \\right] ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 666, + 662, + 845, + 662, + 845, + 837, + 666, + 837 + ], + "score": 0.42, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { ~ b ~ } } & { \\mathrm { ~ c ~ } } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { ~ e ~ } } & { \\mathrm { ~ f ~ } } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { ~ h ~ } } & { \\mathrm { ~ i ~ } } \\end{array} \\right. } \\\\ & { \\mathrm { ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 445, + 662, + 617, + 662, + 617, + 836, + 445, + 836 + ], + "score": 0.39, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { ~ b ~ } } & { \\mathrm { ~ c ~ } } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { ~ e ~ } } & { \\mathrm { ~ f ~ } } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { ~ h ~ } } & { \\mathrm { ~ i ~ } } \\end{array} \\right. } \\\\ & { \\mathrm { ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1096, + 452, + 1239, + 452, + 1239, + 626, + 1096, + 626 + ], + "score": 0.3, + "latex": "{ \\begin{array} { l } { { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } { \\Biggl [ } { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ d ~ } } \\ { \\mathrm { ~ e ~ } } \\ { \\mathrm { ~ f ~ } } } \\\\ { { \\mathrm { ~ g ~ } } { \\mathrm { ~ h ~ } } \\ { \\mathrm { ~ i ~ } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 877, + 976, + 1053, + 976, + 1053, + 1152, + 877, + 1152 + ], + "score": 0.3, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma _ \\alpha ~ a ~ \\gamma _ \\alpha ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ \\left[ \\begin{array} { l l l } { a } & { b } & { c } \\\\ { d } & { e } & { f } \\end{array} \\right] ~ } } \\\\ & { \\mathrm { ~ g ~ \\left[ \\begin{array} { l l l } { \\mathrm { ~ g ~ \\cdot ~ h ~ { ~ i ~ } ~ } } & { } \\\\ { } & { h } & { \\mathrm { ~ i ~ } } \\end{array} \\right] ~ } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1099, + 977, + 1238, + 977, + 1238, + 1151, + 1099, + 1151 + ], + "score": 0.29, + "latex": "{ \\begin{array} { l } { { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } { \\Biggl [ } { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ d ~ } } \\ { \\mathrm { ~ e ~ } } \\ { \\mathrm { ~ f ~ } } } \\\\ { { \\mathrm { ~ g ~ } } { \\Biggl [ } \\ { \\mathrm { ~ g ~ } } \\ { \\mathrm { ~ h ~ } } \\ { \\mathrm { ~ i ~ } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 1263, + 978, + 1461, + 978, + 1461, + 1152, + 1263, + 1152 + ], + "score": 0.26, + "latex": "\\begin{array} { r l } & { \\mathrm { ~ a ~ \\gamma ~ a ~ b ~ c ~ } } \\\\ & { \\mathrm { ~ a ~ } \\left| \\begin{array} { l l l l } { \\mathrm { ~ a ~ } } & { \\mathrm { b ~ } } & { \\mathrm { c ~ } } & { } \\\\ { \\mathrm { ~ d ~ } } & { \\mathrm { e ~ } } & { \\mathrm { f ~ } } & { } \\\\ { \\mathrm { ~ g ~ } } & { \\mathrm { h ~ } } & { \\mathrm { i ~ } } \\end{array} \\right. } \\\\ & { \\mathrm { ~ } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1191.0, + 686.0, + 1191.0, + 686.0, + 1214.0, + 664.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1190.0, + 722.0, + 1190.0, + 722.0, + 1214.0, + 697.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1190.0, + 757.0, + 1190.0, + 757.0, + 1213.0, + 735.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1190.0, + 793.0, + 1190.0, + 793.0, + 1214.0, + 770.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 1225.0, + 685.0, + 1225.0, + 685.0, + 1250.0, + 663.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1224.0, + 722.0, + 1224.0, + 722.0, + 1250.0, + 699.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1223.0, + 757.0, + 1223.0, + 757.0, + 1248.0, + 735.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1224.0, + 793.0, + 1224.0, + 793.0, + 1248.0, + 770.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1260.0, + 682.0, + 1260.0, + 682.0, + 1282.0, + 665.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1258.0, + 722.0, + 1258.0, + 722.0, + 1284.0, + 699.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1260.0, + 757.0, + 1260.0, + 757.0, + 1284.0, + 735.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1258.0, + 795.0, + 1258.0, + 795.0, + 1281.0, + 778.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 1297.0, + 683.0, + 1297.0, + 683.0, + 1322.0, + 663.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1297.0, + 721.0, + 1297.0, + 721.0, + 1324.0, + 699.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1294.0, + 757.0, + 1294.0, + 757.0, + 1319.0, + 735.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1299.0, + 788.0, + 1299.0, + 788.0, + 1314.0, + 777.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 455.0, + 251.0, + 455.0, + 251.0, + 478.0, + 228.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 455.0, + 289.0, + 455.0, + 289.0, + 478.0, + 263.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 452.0, + 323.0, + 452.0, + 323.0, + 476.0, + 299.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 454.0, + 359.0, + 454.0, + 359.0, + 478.0, + 338.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 451.0, + 546.0, + 451.0, + 546.0, + 479.0, + 443.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 455.0, + 576.0, + 455.0, + 576.0, + 478.0, + 551.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 451.0, + 792.0, + 451.0, + 792.0, + 479.0, + 658.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 455.0, + 904.0, + 455.0, + 904.0, + 478.0, + 879.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 455.0, + 939.0, + 455.0, + 939.0, + 478.0, + 915.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 454.0, + 975.0, + 454.0, + 975.0, + 478.0, + 951.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 454.0, + 1012.0, + 454.0, + 1012.0, + 478.0, + 989.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 451.0, + 1412.0, + 451.0, + 1412.0, + 478.0, + 1310.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 457.0, + 1443.0, + 457.0, + 1443.0, + 475.0, + 1423.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 489.0, + 251.0, + 489.0, + 251.0, + 513.0, + 228.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 489.0, + 288.0, + 489.0, + 288.0, + 513.0, + 263.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 488.0, + 322.0, + 488.0, + 322.0, + 512.0, + 300.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 489.0, + 359.0, + 489.0, + 359.0, + 513.0, + 338.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 489.0, + 467.0, + 489.0, + 467.0, + 513.0, + 445.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 489.0, + 506.0, + 489.0, + 506.0, + 513.0, + 482.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 488.0, + 539.0, + 488.0, + 539.0, + 512.0, + 517.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 489.0, + 576.0, + 489.0, + 576.0, + 513.0, + 554.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 488.0, + 686.0, + 488.0, + 686.0, + 513.0, + 663.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 489.0, + 723.0, + 489.0, + 723.0, + 513.0, + 698.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 488.0, + 757.0, + 488.0, + 757.0, + 512.0, + 734.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 489.0, + 792.0, + 489.0, + 792.0, + 513.0, + 771.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 489.0, + 902.0, + 489.0, + 902.0, + 513.0, + 879.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 489.0, + 939.0, + 489.0, + 939.0, + 513.0, + 915.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 488.0, + 974.0, + 488.0, + 974.0, + 512.0, + 952.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 489.0, + 1011.0, + 489.0, + 1011.0, + 513.0, + 989.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 489.0, + 1337.0, + 489.0, + 1337.0, + 513.0, + 1314.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 489.0, + 1373.0, + 489.0, + 1373.0, + 513.0, + 1350.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 488.0, + 1409.0, + 488.0, + 1409.0, + 512.0, + 1386.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 492.0, + 1443.0, + 492.0, + 1443.0, + 512.0, + 1424.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 522.0, + 251.0, + 522.0, + 251.0, + 547.0, + 228.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 522.0, + 288.0, + 522.0, + 288.0, + 549.0, + 265.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 525.0, + 323.0, + 525.0, + 323.0, + 547.0, + 299.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 523.0, + 360.0, + 523.0, + 360.0, + 546.0, + 343.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 522.0, + 467.0, + 522.0, + 467.0, + 547.0, + 445.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 522.0, + 504.0, + 522.0, + 504.0, + 547.0, + 482.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 528.0, + 539.0, + 528.0, + 539.0, + 546.0, + 519.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 523.0, + 577.0, + 523.0, + 577.0, + 545.0, + 560.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 523.0, + 684.0, + 523.0, + 684.0, + 547.0, + 663.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 522.0, + 721.0, + 522.0, + 721.0, + 547.0, + 700.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 528.0, + 755.0, + 528.0, + 755.0, + 545.0, + 737.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 523.0, + 794.0, + 523.0, + 794.0, + 545.0, + 777.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 522.0, + 904.0, + 522.0, + 904.0, + 549.0, + 879.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 522.0, + 938.0, + 522.0, + 938.0, + 547.0, + 917.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 526.0, + 972.0, + 526.0, + 972.0, + 546.0, + 954.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 523.0, + 1011.0, + 523.0, + 1011.0, + 543.0, + 994.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 523.0, + 1337.0, + 523.0, + 1337.0, + 549.0, + 1314.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 523.0, + 1370.0, + 523.0, + 1370.0, + 546.0, + 1353.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 525.0, + 1409.0, + 525.0, + 1409.0, + 547.0, + 1386.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1428.0, + 523.0, + 1444.0, + 523.0, + 1444.0, + 543.0, + 1428.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 562.0, + 248.0, + 562.0, + 248.0, + 586.0, + 231.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 560.0, + 288.0, + 560.0, + 288.0, + 586.0, + 265.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 557.0, + 325.0, + 557.0, + 325.0, + 581.0, + 299.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 563.0, + 352.0, + 563.0, + 352.0, + 574.0, + 342.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 560.0, + 467.0, + 560.0, + 467.0, + 587.0, + 445.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 560.0, + 504.0, + 560.0, + 504.0, + 586.0, + 482.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 559.0, + 537.0, + 559.0, + 537.0, + 580.0, + 520.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 560.0, + 684.0, + 560.0, + 684.0, + 587.0, + 663.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 560.0, + 721.0, + 560.0, + 721.0, + 586.0, + 698.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 559.0, + 757.0, + 559.0, + 757.0, + 583.0, + 735.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 563.0, + 788.0, + 563.0, + 788.0, + 577.0, + 777.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 562.0, + 899.0, + 562.0, + 899.0, + 584.0, + 882.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 560.0, + 938.0, + 560.0, + 938.0, + 587.0, + 915.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 559.0, + 975.0, + 559.0, + 975.0, + 583.0, + 952.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 560.0, + 1337.0, + 560.0, + 1337.0, + 587.0, + 1314.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 560.0, + 1371.0, + 560.0, + 1371.0, + 586.0, + 1350.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 557.0, + 1409.0, + 557.0, + 1409.0, + 581.0, + 1386.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1428.0, + 563.0, + 1438.0, + 563.0, + 1438.0, + 577.0, + 1428.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 666.0, + 251.0, + 666.0, + 251.0, + 689.0, + 228.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 666.0, + 288.0, + 666.0, + 288.0, + 689.0, + 263.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 664.0, + 322.0, + 664.0, + 322.0, + 689.0, + 299.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 666.0, + 359.0, + 666.0, + 359.0, + 689.0, + 336.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 700.0, + 251.0, + 700.0, + 251.0, + 723.0, + 228.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 700.0, + 289.0, + 700.0, + 289.0, + 723.0, + 265.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 698.0, + 323.0, + 698.0, + 323.0, + 722.0, + 300.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 699.0, + 359.0, + 699.0, + 359.0, + 723.0, + 336.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 736.0, + 248.0, + 736.0, + 248.0, + 757.0, + 231.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 733.0, + 288.0, + 733.0, + 288.0, + 759.0, + 265.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 737.0, + 320.0, + 737.0, + 320.0, + 757.0, + 303.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 734.0, + 359.0, + 734.0, + 359.0, + 754.0, + 343.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 771.0, + 248.0, + 771.0, + 248.0, + 795.0, + 231.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 770.0, + 286.0, + 770.0, + 286.0, + 795.0, + 265.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 767.0, + 323.0, + 767.0, + 323.0, + 791.0, + 300.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 773.0, + 353.0, + 773.0, + 353.0, + 786.0, + 343.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 287.0, + 625.0, + 287.0, + 625.0, + 320.0, + 137.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 134.0, + 893.0, + 392.0, + 893.0, + 392.0, + 932.0, + 134.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 136.0, + 354.0, + 399.0, + 354.0, + 399.0, + 391.0, + 136.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 130.0, + 1569.0, + 619.0, + 1569.0, + 619.0, + 1607.0, + 130.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 128.0, + 1703.0, + 654.0, + 1703.0, + 654.0, + 1734.0, + 128.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 127.0, + 1635.0, + 610.0, + 1635.0, + 610.0, + 1671.0, + 127.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 134.0, + 893.0, + 392.0, + 893.0, + 392.0, + 932.0, + 134.0, + 932.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 292, + 1575, + 1392, + 1575, + 1392, + 2053, + 292, + 2053 + ], + "score": 0.946 + }, + { + "category_id": 3, + "poly": [ + 253, + 508, + 1414, + 508, + 1414, + 672, + 253, + 672 + ], + "score": 0.937 + }, + { + "category_id": 3, + "poly": [ + 297, + 787, + 1331, + 787, + 1331, + 1099, + 297, + 1099 + ], + "score": 0.922 + }, + { + "category_id": 3, + "poly": [ + 292, + 1180, + 1392, + 1180, + 1392, + 1494, + 292, + 1494 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 804, + 220, + 1514, + 220, + 1514, + 278, + 804, + 278 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 197, + 698, + 861, + 698, + 861, + 733, + 197, + 733 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 804, + 313, + 1516, + 313, + 1516, + 354, + 804, + 354 + ], + "score": 0.884 + }, + { + "category_id": 0, + "poly": [ + 198, + 463, + 681, + 463, + 681, + 496, + 198, + 496 + ], + "score": 0.849 + }, + { + "category_id": 1, + "poly": [ + 198, + 749, + 458, + 749, + 458, + 777, + 198, + 777 + ], + "score": 0.813 + }, + { + "category_id": 3, + "poly": [ + 207, + 224, + 672, + 224, + 672, + 446, + 207, + 446 + ], + "score": 0.681 + }, + { + "category_id": 4, + "poly": [ + 188, + 1524, + 449, + 1524, + 449, + 1552, + 188, + 1552 + ], + "score": 0.488 + }, + { + "category_id": 4, + "poly": [ + 194, + 1126, + 456, + 1126, + 456, + 1154, + 194, + 1154 + ], + "score": 0.444 + }, + { + "category_id": 1, + "poly": [ + 194, + 1126, + 456, + 1126, + 456, + 1154, + 194, + 1154 + ], + "score": 0.125 + }, + { + "category_id": 1, + "poly": [ + 188, + 1524, + 449, + 1524, + 449, + 1552, + 188, + 1552 + ], + "score": 0.098 + }, + { + "category_id": 13, + "poly": [ + 1296, + 513, + 1410, + 513, + 1410, + 663, + 1296, + 663 + ], + "score": 0.7, + "latex": "\\begin{array}{c} \\begin{array} { r } { \\mathrm { ~ i } ; } \\\\ { \\mathrm { ~ \\frac { 1 } { 2 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\\\ { \\mathrm { ~ \\frac { 1 } { 3 } ~ } } \\end{array} \\textstyle \\mathrm { ~ \\frac { 1 } { 3 } ~ } \\mathrm { ~ \\frac { 1 } { 3 } ~ } \\begin{array} { r } { 1 } \\\\ { 1 } \\\\ { 1 } \\end{array} \\begin{array} { r } { 1 } \\\\ { 1 } \\\\ { 1 } \\end{array} \\begin{array} { r } { 1 } \\\\ { 1 } \\\\ { 1 } \\end{array} \\\\ { \\begin{array} { r } { 1 } \\\\ { 1 } \\end{array} \\begin{array} { r } { 1 } \\\\ { 1 } \\end{array} \\begin{array} { r } { 1 } \\\\ { 1 } \\end{array} } \\\\ { \\begin{array} { r } { 2 5 } \\\\ { 0 } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 300, + 961, + 429, + 961, + 429, + 1081, + 300, + 1081 + ], + "score": 0.43, + "latex": "{ \\begin{array} { r l } & { \\mathbf { e } \\mathbf { \\theta } \\quad \\mathbf { d } \\quad \\mathbf { d } \\quad \\mathbf { e } \\quad \\mathbf { f } } \\\\ & { \\mathbf { b } \\quad \\mathbf { a } \\quad \\mathbf { a } \\quad \\mathbf { b } \\quad \\mathbf { c } } \\\\ & { \\mathbf { b } \\quad \\mathbf { a } \\quad \\mathbf { \\tilde { a } } \\quad \\mathbf { b } \\quad \\mathbf { c } } \\\\ & { \\mathbf { e } \\quad \\mathbf { d } \\quad \\mathbf { d } \\quad \\mathbf { e } \\quad \\mathbf { f } } \\\\ & { \\mathbf { h } \\quad \\mathbf { \\theta } \\quad \\mathbf { g } \\left| \\quad \\mathbf { \\tilde { g } } \\quad \\mathbf { h } \\quad \\mathbf { i } \\quad \\right. } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 1169, + 510, + 1267, + 510, + 1267, + 662, + 1169, + 662 + ], + "score": 0.36, + "latex": "\\begin{array} { c } { { \\mathrm { ~ { \\scriptstyle { \\hat { h } } } : } } } \\\\ { { \\mathrm { ~ { \\scriptstyle { \\hat { \\phi } } } 2 ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle 2 } ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\mathrm { ~ { \\scriptstyle ~ \\ s u m } = 2 5 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 906, + 509, + 1022, + 509, + 1022, + 613, + 906, + 613 + ], + "score": 0.35, + "latex": "\\begin{array} { c c c c c c } { \\mathbf { f } . } & { } & { } & { } & { } & { } & { } \\\\ { 2 } & { 2 } & { 2 } & { 2 } & { 2 } & { 2 } \\\\ { 1 } & { 1 } & { 1 } & { 1 } & { 1 } & { 1 } \\\\ { 1 } & { 1 } & { 1 } & { 1 } & { 1 } & { 1 } \\\\ { 1 } & { 1 } & { 1 } & { 1 } & { 1 } & { 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 299, + 1353, + 428, + 1353, + 428, + 1476, + 299, + 1476 + ], + "score": 0.31, + "latex": "\\begin{array} { r l } & { \\begin{array} { r l } { \\mathrm { ~ e ~ } \\quad \\mathrm { ~ d ~ } \\mathrm { ~ d ~ } \\mathrm { ~ e ~ f ~ } } \\\\ { \\mathrm { ~ b ~ } \\quad \\mathrm { ~ a ~ } \\mathrm { ~ a ~ } \\mathrm { ~ b ~ } \\mathrm { ~ c ~ } } \\end{array} } \\\\ & { \\begin{array} { r l } { \\mathrm { ~ b ~ } } & { \\mathrm { ~ a ~ } \\left[ \\mathrm { ~ a ~ } \\mathrm { ~ b ~ } \\mathrm { ~ c ~ } \\right. } \\\\ { \\mathrm { ~ e ~ } \\quad \\mathrm { ~ d ~ } \\mathrm { ~ e ~ f ~ } } \\\\ { \\mathrm { ~ h ~ } \\quad \\mathrm { ~ 9 ~ } \\left| \\mathrm { ~ g ~ } \\mathrm { ~ h ~ } \\mathrm { ~ i ~ } \\right. } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1213, + 1190, + 1346, + 1190, + 1346, + 1302, + 1213, + 1302 + ], + "score": 0.29, + "latex": "{ \\begin{array} { r l } & { \\mathbf { e } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { e } \\quad \\mathbf { \\ \" { f } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\quad \\mathbf { { a } } \\quad \\mathbf { b } \\quad \\mathbf { \\ \" { c } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\left[ \\mathbf { \\ \" { a } } \\quad \\mathbf { { b } } \\quad \\mathbf { \\ \" { c } } \\right. } \\\\ & { \\mathbf { e } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { e } \\quad \\mathbf { \\ \" { f } } } \\\\ & { \\mathbf { n } \\quad \\mathbf { \\ \" { g } } \\quad \\mathbf { \\ \" { n } } \\quad \\mathbf { \\ \" { i } } } \\end{array} }" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1590.0, + 316.0, + 1590.0, + 316.0, + 1601.0, + 304.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1590.0, + 343.0, + 1590.0, + 343.0, + 1602.0, + 332.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1590.0, + 364.0, + 1590.0, + 364.0, + 1601.0, + 353.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1590.0, + 387.0, + 1590.0, + 387.0, + 1601.0, + 374.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1590.0, + 407.0, + 1590.0, + 407.0, + 1600.0, + 397.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1588.0, + 492.0, + 1588.0, + 492.0, + 1602.0, + 476.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1587.0, + 520.0, + 1587.0, + 520.0, + 1602.0, + 503.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1587.0, + 541.0, + 1587.0, + 541.0, + 1602.0, + 523.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1587.0, + 563.0, + 1587.0, + 563.0, + 1602.0, + 544.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1588.0, + 580.0, + 1588.0, + 580.0, + 1600.0, + 570.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1588.0, + 666.0, + 1588.0, + 666.0, + 1604.0, + 650.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1587.0, + 693.0, + 1587.0, + 693.0, + 1602.0, + 676.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1586.0, + 737.0, + 1586.0, + 737.0, + 1605.0, + 694.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1587.0, + 756.0, + 1587.0, + 756.0, + 1602.0, + 741.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1588.0, + 838.0, + 1588.0, + 838.0, + 1604.0, + 823.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1587.0, + 866.0, + 1587.0, + 866.0, + 1602.0, + 848.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1587.0, + 886.0, + 1587.0, + 886.0, + 1604.0, + 871.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1588.0, + 908.0, + 1588.0, + 908.0, + 1604.0, + 892.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1590.0, + 926.0, + 1590.0, + 926.0, + 1600.0, + 917.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1590.0, + 1037.0, + 1590.0, + 1037.0, + 1601.0, + 1025.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1587.0, + 1061.0, + 1587.0, + 1061.0, + 1602.0, + 1043.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1587.0, + 1081.0, + 1587.0, + 1081.0, + 1602.0, + 1064.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1588.0, + 1103.0, + 1588.0, + 1103.0, + 1602.0, + 1086.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1588.0, + 1121.0, + 1588.0, + 1121.0, + 1600.0, + 1109.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1588.0, + 1228.0, + 1588.0, + 1228.0, + 1604.0, + 1213.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1587.0, + 1256.0, + 1587.0, + 1256.0, + 1602.0, + 1238.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1587.0, + 1276.0, + 1587.0, + 1276.0, + 1604.0, + 1259.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1588.0, + 1298.0, + 1588.0, + 1298.0, + 1604.0, + 1279.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1588.0, + 1316.0, + 1588.0, + 1316.0, + 1600.0, + 1305.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1608.0, + 316.0, + 1608.0, + 316.0, + 1620.0, + 304.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1609.0, + 343.0, + 1609.0, + 343.0, + 1622.0, + 332.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1609.0, + 364.0, + 1609.0, + 364.0, + 1620.0, + 354.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1608.0, + 386.0, + 1608.0, + 386.0, + 1620.0, + 374.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1611.0, + 406.0, + 1611.0, + 406.0, + 1618.0, + 399.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1608.0, + 489.0, + 1608.0, + 489.0, + 1620.0, + 478.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1609.0, + 517.0, + 1609.0, + 517.0, + 1620.0, + 504.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1610.0, + 538.0, + 1610.0, + 538.0, + 1620.0, + 526.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1608.0, + 559.0, + 1608.0, + 559.0, + 1620.0, + 548.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1610.0, + 581.0, + 1610.0, + 581.0, + 1620.0, + 572.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1609.0, + 663.0, + 1609.0, + 663.0, + 1620.0, + 651.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1608.0, + 691.0, + 1608.0, + 691.0, + 1623.0, + 676.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1609.0, + 711.0, + 1609.0, + 711.0, + 1620.0, + 699.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1609.0, + 733.0, + 1609.0, + 733.0, + 1620.0, + 721.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1609.0, + 755.0, + 1609.0, + 755.0, + 1620.0, + 742.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1606.0, + 838.0, + 1606.0, + 838.0, + 1623.0, + 823.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1609.0, + 865.0, + 1609.0, + 865.0, + 1623.0, + 848.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1609.0, + 884.0, + 1609.0, + 884.0, + 1620.0, + 873.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1609.0, + 906.0, + 1609.0, + 906.0, + 1620.0, + 894.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1609.0, + 927.0, + 1609.0, + 927.0, + 1620.0, + 916.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1606.0, + 1038.0, + 1606.0, + 1038.0, + 1622.0, + 1023.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1610.0, + 1058.0, + 1610.0, + 1058.0, + 1620.0, + 1044.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1609.0, + 1080.0, + 1609.0, + 1080.0, + 1620.0, + 1068.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1609.0, + 1101.0, + 1609.0, + 1101.0, + 1620.0, + 1090.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1610.0, + 1121.0, + 1610.0, + 1121.0, + 1620.0, + 1112.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1608.0, + 1226.0, + 1608.0, + 1226.0, + 1620.0, + 1214.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1608.0, + 1254.0, + 1608.0, + 1254.0, + 1623.0, + 1238.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1610.0, + 1274.0, + 1610.0, + 1274.0, + 1620.0, + 1263.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1609.0, + 1295.0, + 1609.0, + 1295.0, + 1620.0, + 1284.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1610.0, + 1316.0, + 1610.0, + 1316.0, + 1620.0, + 1307.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1627.0, + 317.0, + 1627.0, + 317.0, + 1644.0, + 303.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1628.0, + 343.0, + 1628.0, + 343.0, + 1645.0, + 329.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1628.0, + 367.0, + 1628.0, + 367.0, + 1645.0, + 351.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1628.0, + 388.0, + 1628.0, + 388.0, + 1644.0, + 373.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1631.0, + 407.0, + 1631.0, + 407.0, + 1641.0, + 397.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1630.0, + 489.0, + 1630.0, + 489.0, + 1641.0, + 478.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1631.0, + 517.0, + 1631.0, + 517.0, + 1642.0, + 504.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1628.0, + 540.0, + 1628.0, + 540.0, + 1645.0, + 523.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1628.0, + 561.0, + 1628.0, + 561.0, + 1644.0, + 546.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1631.0, + 581.0, + 1631.0, + 581.0, + 1641.0, + 572.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1627.0, + 663.0, + 1627.0, + 663.0, + 1644.0, + 650.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1631.0, + 690.0, + 1631.0, + 690.0, + 1644.0, + 678.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1630.0, + 713.0, + 1630.0, + 713.0, + 1645.0, + 698.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1628.0, + 734.0, + 1628.0, + 734.0, + 1644.0, + 720.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1631.0, + 755.0, + 1631.0, + 755.0, + 1641.0, + 744.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1627.0, + 838.0, + 1627.0, + 838.0, + 1644.0, + 823.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1631.0, + 862.0, + 1631.0, + 862.0, + 1642.0, + 850.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1628.0, + 887.0, + 1628.0, + 887.0, + 1645.0, + 871.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1628.0, + 908.0, + 1628.0, + 908.0, + 1644.0, + 892.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1630.0, + 927.0, + 1630.0, + 927.0, + 1642.0, + 916.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1627.0, + 1038.0, + 1627.0, + 1038.0, + 1642.0, + 1023.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1631.0, + 1057.0, + 1631.0, + 1057.0, + 1642.0, + 1046.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1631.0, + 1080.0, + 1631.0, + 1080.0, + 1642.0, + 1067.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1630.0, + 1102.0, + 1630.0, + 1102.0, + 1641.0, + 1090.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1631.0, + 1122.0, + 1631.0, + 1122.0, + 1641.0, + 1111.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1627.0, + 1227.0, + 1627.0, + 1227.0, + 1644.0, + 1213.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1631.0, + 1252.0, + 1631.0, + 1252.0, + 1641.0, + 1240.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1631.0, + 1274.0, + 1631.0, + 1274.0, + 1642.0, + 1262.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1628.0, + 1298.0, + 1628.0, + 1298.0, + 1642.0, + 1282.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1631.0, + 1317.0, + 1631.0, + 1317.0, + 1641.0, + 1307.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1650.0, + 316.0, + 1650.0, + 316.0, + 1662.0, + 304.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1649.0, + 342.0, + 1649.0, + 342.0, + 1662.0, + 332.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1648.0, + 366.0, + 1648.0, + 366.0, + 1663.0, + 351.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1650.0, + 387.0, + 1650.0, + 387.0, + 1662.0, + 374.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1650.0, + 489.0, + 1650.0, + 489.0, + 1662.0, + 478.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1648.0, + 518.0, + 1648.0, + 518.0, + 1663.0, + 503.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1648.0, + 541.0, + 1648.0, + 541.0, + 1664.0, + 524.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1649.0, + 561.0, + 1649.0, + 561.0, + 1664.0, + 546.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1648.0, + 586.0, + 1648.0, + 586.0, + 1663.0, + 571.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1650.0, + 662.0, + 1650.0, + 662.0, + 1662.0, + 651.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1649.0, + 690.0, + 1649.0, + 690.0, + 1662.0, + 678.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1648.0, + 713.0, + 1648.0, + 713.0, + 1663.0, + 698.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1650.0, + 733.0, + 1650.0, + 733.0, + 1662.0, + 721.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1650.0, + 837.0, + 1650.0, + 837.0, + 1662.0, + 825.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1649.0, + 862.0, + 1649.0, + 862.0, + 1662.0, + 851.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1648.0, + 886.0, + 1648.0, + 886.0, + 1663.0, + 871.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1649.0, + 908.0, + 1649.0, + 908.0, + 1664.0, + 892.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1649.0, + 928.0, + 1649.0, + 928.0, + 1662.0, + 918.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1649.0, + 1038.0, + 1649.0, + 1038.0, + 1664.0, + 1023.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1649.0, + 1057.0, + 1649.0, + 1057.0, + 1662.0, + 1046.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1648.0, + 1081.0, + 1648.0, + 1081.0, + 1663.0, + 1066.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1649.0, + 1103.0, + 1649.0, + 1103.0, + 1664.0, + 1087.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1649.0, + 1124.0, + 1649.0, + 1124.0, + 1662.0, + 1113.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1649.0, + 1228.0, + 1649.0, + 1228.0, + 1664.0, + 1213.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1650.0, + 1252.0, + 1650.0, + 1252.0, + 1662.0, + 1241.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1648.0, + 1276.0, + 1648.0, + 1276.0, + 1663.0, + 1261.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1649.0, + 1298.0, + 1649.0, + 1298.0, + 1663.0, + 1282.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1669.0, + 316.0, + 1669.0, + 316.0, + 1682.0, + 304.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1671.0, + 342.0, + 1671.0, + 342.0, + 1682.0, + 332.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1669.0, + 386.0, + 1669.0, + 386.0, + 1681.0, + 374.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1669.0, + 489.0, + 1669.0, + 489.0, + 1682.0, + 478.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1669.0, + 518.0, + 1669.0, + 518.0, + 1686.0, + 503.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1669.0, + 540.0, + 1669.0, + 540.0, + 1686.0, + 524.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1667.0, + 561.0, + 1667.0, + 561.0, + 1684.0, + 546.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1672.0, + 578.0, + 1672.0, + 578.0, + 1679.0, + 572.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1669.0, + 662.0, + 1669.0, + 662.0, + 1682.0, + 651.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1669.0, + 691.0, + 1669.0, + 691.0, + 1686.0, + 677.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1669.0, + 713.0, + 1669.0, + 713.0, + 1686.0, + 698.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1668.0, + 736.0, + 1668.0, + 736.0, + 1684.0, + 718.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1669.0, + 836.0, + 1669.0, + 836.0, + 1682.0, + 825.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1672.0, + 862.0, + 1672.0, + 862.0, + 1684.0, + 850.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1669.0, + 886.0, + 1669.0, + 886.0, + 1685.0, + 871.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1669.0, + 906.0, + 1669.0, + 906.0, + 1681.0, + 894.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1667.0, + 1038.0, + 1667.0, + 1038.0, + 1684.0, + 1023.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1672.0, + 1057.0, + 1672.0, + 1057.0, + 1685.0, + 1046.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1669.0, + 1081.0, + 1669.0, + 1081.0, + 1686.0, + 1066.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1667.0, + 1103.0, + 1667.0, + 1103.0, + 1684.0, + 1087.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1667.0, + 1228.0, + 1667.0, + 1228.0, + 1684.0, + 1213.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1672.0, + 1252.0, + 1672.0, + 1252.0, + 1682.0, + 1241.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1669.0, + 1276.0, + 1669.0, + 1276.0, + 1686.0, + 1261.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1668.0, + 1298.0, + 1668.0, + 1298.0, + 1684.0, + 1282.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1751.0, + 318.0, + 1751.0, + 318.0, + 1766.0, + 303.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1749.0, + 345.0, + 1749.0, + 345.0, + 1766.0, + 329.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1749.0, + 366.0, + 1749.0, + 366.0, + 1766.0, + 352.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1751.0, + 388.0, + 1751.0, + 388.0, + 1766.0, + 373.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1751.0, + 492.0, + 1751.0, + 492.0, + 1766.0, + 476.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1751.0, + 518.0, + 1751.0, + 518.0, + 1766.0, + 503.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1749.0, + 541.0, + 1749.0, + 541.0, + 1765.0, + 524.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1751.0, + 562.0, + 1751.0, + 562.0, + 1766.0, + 544.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1751.0, + 666.0, + 1751.0, + 666.0, + 1766.0, + 650.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1751.0, + 694.0, + 1751.0, + 694.0, + 1766.0, + 676.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1749.0, + 739.0, + 1749.0, + 739.0, + 1768.0, + 696.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1752.0, + 753.0, + 1752.0, + 753.0, + 1764.0, + 741.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1751.0, + 840.0, + 1751.0, + 840.0, + 1766.0, + 823.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1749.0, + 866.0, + 1749.0, + 866.0, + 1766.0, + 848.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1751.0, + 886.0, + 1751.0, + 886.0, + 1766.0, + 871.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1751.0, + 908.0, + 1751.0, + 908.0, + 1766.0, + 893.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1751.0, + 1039.0, + 1751.0, + 1039.0, + 1766.0, + 1023.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1749.0, + 1061.0, + 1749.0, + 1061.0, + 1766.0, + 1043.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1749.0, + 1081.0, + 1749.0, + 1081.0, + 1766.0, + 1064.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1751.0, + 1103.0, + 1751.0, + 1103.0, + 1766.0, + 1087.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1751.0, + 1121.0, + 1751.0, + 1121.0, + 1760.0, + 1109.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1751.0, + 1228.0, + 1751.0, + 1228.0, + 1766.0, + 1212.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1749.0, + 1254.0, + 1749.0, + 1254.0, + 1766.0, + 1240.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1749.0, + 1276.0, + 1749.0, + 1276.0, + 1766.0, + 1261.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1751.0, + 1298.0, + 1751.0, + 1298.0, + 1766.0, + 1281.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1752.0, + 1316.0, + 1752.0, + 1316.0, + 1761.0, + 1306.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1769.0, + 318.0, + 1769.0, + 318.0, + 1787.0, + 303.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1770.0, + 346.0, + 1770.0, + 346.0, + 1787.0, + 329.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1773.0, + 364.0, + 1773.0, + 364.0, + 1784.0, + 353.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1771.0, + 386.0, + 1771.0, + 386.0, + 1783.0, + 374.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1773.0, + 407.0, + 1773.0, + 407.0, + 1784.0, + 397.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1770.0, + 491.0, + 1770.0, + 491.0, + 1786.0, + 476.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1771.0, + 520.0, + 1771.0, + 520.0, + 1786.0, + 503.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1773.0, + 538.0, + 1773.0, + 538.0, + 1783.0, + 526.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1771.0, + 559.0, + 1771.0, + 559.0, + 1783.0, + 547.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1770.0, + 666.0, + 1770.0, + 666.0, + 1786.0, + 650.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1771.0, + 693.0, + 1771.0, + 693.0, + 1786.0, + 676.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1773.0, + 713.0, + 1773.0, + 713.0, + 1783.0, + 697.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1770.0, + 736.0, + 1770.0, + 736.0, + 1784.0, + 718.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1773.0, + 755.0, + 1773.0, + 755.0, + 1783.0, + 741.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1770.0, + 838.0, + 1770.0, + 838.0, + 1786.0, + 823.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1771.0, + 866.0, + 1771.0, + 866.0, + 1787.0, + 848.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1773.0, + 884.0, + 1773.0, + 884.0, + 1784.0, + 872.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1771.0, + 906.0, + 1771.0, + 906.0, + 1783.0, + 894.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1770.0, + 1038.0, + 1770.0, + 1038.0, + 1786.0, + 1022.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1771.0, + 1059.0, + 1771.0, + 1059.0, + 1786.0, + 1043.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1770.0, + 1081.0, + 1770.0, + 1081.0, + 1786.0, + 1064.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1769.0, + 1103.0, + 1769.0, + 1103.0, + 1784.0, + 1088.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1773.0, + 1121.0, + 1773.0, + 1121.0, + 1783.0, + 1109.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1770.0, + 1228.0, + 1770.0, + 1228.0, + 1786.0, + 1212.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1771.0, + 1256.0, + 1771.0, + 1256.0, + 1787.0, + 1238.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1771.0, + 1276.0, + 1771.0, + 1276.0, + 1786.0, + 1259.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1770.0, + 1298.0, + 1770.0, + 1298.0, + 1786.0, + 1282.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1771.0, + 1317.0, + 1771.0, + 1317.0, + 1784.0, + 1307.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1791.0, + 318.0, + 1791.0, + 318.0, + 1808.0, + 303.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1792.0, + 345.0, + 1792.0, + 345.0, + 1808.0, + 329.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1792.0, + 367.0, + 1792.0, + 367.0, + 1808.0, + 351.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1791.0, + 388.0, + 1791.0, + 388.0, + 1808.0, + 372.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1795.0, + 408.0, + 1795.0, + 408.0, + 1805.0, + 396.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1791.0, + 491.0, + 1791.0, + 491.0, + 1808.0, + 476.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1792.0, + 518.0, + 1792.0, + 518.0, + 1808.0, + 503.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1792.0, + 540.0, + 1792.0, + 540.0, + 1808.0, + 524.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1791.0, + 561.0, + 1791.0, + 561.0, + 1806.0, + 546.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1792.0, + 666.0, + 1792.0, + 666.0, + 1808.0, + 650.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1792.0, + 692.0, + 1792.0, + 692.0, + 1808.0, + 676.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1792.0, + 715.0, + 1792.0, + 715.0, + 1808.0, + 697.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1792.0, + 736.0, + 1792.0, + 736.0, + 1808.0, + 717.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1793.0, + 755.0, + 1793.0, + 755.0, + 1804.0, + 742.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1791.0, + 838.0, + 1791.0, + 838.0, + 1808.0, + 822.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1792.0, + 863.0, + 1792.0, + 863.0, + 1808.0, + 848.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1792.0, + 887.0, + 1792.0, + 887.0, + 1808.0, + 871.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1791.0, + 908.0, + 1791.0, + 908.0, + 1806.0, + 893.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1791.0, + 1038.0, + 1791.0, + 1038.0, + 1808.0, + 1023.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1792.0, + 1058.0, + 1792.0, + 1058.0, + 1808.0, + 1043.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1792.0, + 1081.0, + 1792.0, + 1081.0, + 1808.0, + 1066.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1791.0, + 1103.0, + 1791.0, + 1103.0, + 1806.0, + 1088.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1795.0, + 1122.0, + 1795.0, + 1122.0, + 1805.0, + 1109.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1791.0, + 1228.0, + 1791.0, + 1228.0, + 1808.0, + 1213.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1792.0, + 1253.0, + 1792.0, + 1253.0, + 1808.0, + 1238.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1792.0, + 1276.0, + 1792.0, + 1276.0, + 1808.0, + 1261.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1791.0, + 1298.0, + 1791.0, + 1298.0, + 1808.0, + 1282.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1795.0, + 1318.0, + 1795.0, + 1318.0, + 1805.0, + 1307.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1811.0, + 318.0, + 1811.0, + 318.0, + 1828.0, + 303.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1810.0, + 345.0, + 1810.0, + 345.0, + 1828.0, + 329.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1811.0, + 367.0, + 1811.0, + 367.0, + 1827.0, + 351.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1813.0, + 388.0, + 1813.0, + 388.0, + 1827.0, + 372.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1813.0, + 489.0, + 1813.0, + 489.0, + 1826.0, + 478.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1811.0, + 518.0, + 1811.0, + 518.0, + 1827.0, + 503.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1813.0, + 538.0, + 1813.0, + 538.0, + 1824.0, + 527.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1813.0, + 562.0, + 1813.0, + 562.0, + 1827.0, + 546.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1813.0, + 666.0, + 1813.0, + 666.0, + 1827.0, + 650.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1811.0, + 692.0, + 1811.0, + 692.0, + 1827.0, + 676.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1811.0, + 715.0, + 1811.0, + 715.0, + 1827.0, + 696.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1813.0, + 736.0, + 1813.0, + 736.0, + 1827.0, + 716.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1814.0, + 837.0, + 1814.0, + 837.0, + 1826.0, + 825.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1813.0, + 862.0, + 1813.0, + 862.0, + 1826.0, + 851.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1811.0, + 886.0, + 1811.0, + 886.0, + 1827.0, + 871.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1814.0, + 906.0, + 1814.0, + 906.0, + 1826.0, + 894.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1811.0, + 1038.0, + 1811.0, + 1038.0, + 1828.0, + 1023.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1811.0, + 1059.0, + 1811.0, + 1059.0, + 1827.0, + 1044.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1811.0, + 1081.0, + 1811.0, + 1081.0, + 1827.0, + 1066.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1814.0, + 1102.0, + 1814.0, + 1102.0, + 1826.0, + 1090.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1811.0, + 1228.0, + 1811.0, + 1228.0, + 1827.0, + 1212.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1813.0, + 1252.0, + 1813.0, + 1252.0, + 1826.0, + 1241.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1810.0, + 1276.0, + 1810.0, + 1276.0, + 1827.0, + 1261.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1814.0, + 1296.0, + 1814.0, + 1296.0, + 1826.0, + 1284.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1831.0, + 318.0, + 1831.0, + 318.0, + 1848.0, + 303.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1832.0, + 345.0, + 1832.0, + 345.0, + 1849.0, + 329.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1832.0, + 366.0, + 1832.0, + 366.0, + 1848.0, + 351.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1833.0, + 386.0, + 1833.0, + 386.0, + 1844.0, + 374.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1833.0, + 489.0, + 1833.0, + 489.0, + 1845.0, + 478.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1835.0, + 516.0, + 1835.0, + 516.0, + 1848.0, + 504.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1835.0, + 538.0, + 1835.0, + 538.0, + 1848.0, + 527.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1833.0, + 559.0, + 1833.0, + 559.0, + 1845.0, + 547.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1832.0, + 666.0, + 1832.0, + 666.0, + 1846.0, + 650.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1833.0, + 691.0, + 1833.0, + 691.0, + 1849.0, + 676.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1833.0, + 715.0, + 1833.0, + 715.0, + 1849.0, + 697.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1832.0, + 736.0, + 1832.0, + 736.0, + 1846.0, + 720.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1832.0, + 840.0, + 1832.0, + 840.0, + 1846.0, + 822.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1833.0, + 863.0, + 1833.0, + 863.0, + 1849.0, + 848.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1833.0, + 886.0, + 1833.0, + 886.0, + 1849.0, + 871.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1833.0, + 904.0, + 1833.0, + 904.0, + 1842.0, + 894.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1831.0, + 1038.0, + 1831.0, + 1038.0, + 1848.0, + 1022.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1832.0, + 1059.0, + 1832.0, + 1059.0, + 1849.0, + 1044.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1832.0, + 1081.0, + 1832.0, + 1081.0, + 1849.0, + 1066.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1833.0, + 1102.0, + 1833.0, + 1102.0, + 1845.0, + 1090.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1831.0, + 1228.0, + 1831.0, + 1228.0, + 1846.0, + 1212.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1833.0, + 1253.0, + 1833.0, + 1253.0, + 1849.0, + 1238.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1832.0, + 1276.0, + 1832.0, + 1276.0, + 1849.0, + 1261.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1833.0, + 1295.0, + 1833.0, + 1295.0, + 1844.0, + 1284.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1913.0, + 318.0, + 1913.0, + 318.0, + 1929.0, + 303.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1913.0, + 346.0, + 1913.0, + 346.0, + 1929.0, + 329.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1913.0, + 367.0, + 1913.0, + 367.0, + 1929.0, + 351.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1913.0, + 388.0, + 1913.0, + 388.0, + 1929.0, + 373.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1915.0, + 406.0, + 1915.0, + 406.0, + 1925.0, + 397.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1913.0, + 541.0, + 1913.0, + 541.0, + 1929.0, + 524.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1913.0, + 562.0, + 1913.0, + 562.0, + 1929.0, + 546.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1913.0, + 583.0, + 1913.0, + 583.0, + 1929.0, + 567.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1913.0, + 605.0, + 1913.0, + 605.0, + 1929.0, + 589.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1915.0, + 753.0, + 1915.0, + 753.0, + 1929.0, + 736.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1913.0, + 779.0, + 1913.0, + 779.0, + 1929.0, + 761.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1913.0, + 800.0, + 1913.0, + 800.0, + 1929.0, + 781.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1915.0, + 823.0, + 1915.0, + 823.0, + 1929.0, + 801.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1916.0, + 841.0, + 1916.0, + 841.0, + 1925.0, + 827.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1933.0, + 318.0, + 1933.0, + 318.0, + 1949.0, + 303.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1934.0, + 346.0, + 1934.0, + 346.0, + 1949.0, + 329.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1934.0, + 367.0, + 1934.0, + 367.0, + 1949.0, + 351.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1933.0, + 388.0, + 1933.0, + 388.0, + 1948.0, + 373.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1935.0, + 407.0, + 1935.0, + 407.0, + 1946.0, + 397.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1933.0, + 541.0, + 1933.0, + 541.0, + 1949.0, + 524.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1934.0, + 561.0, + 1934.0, + 561.0, + 1949.0, + 546.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1934.0, + 584.0, + 1934.0, + 584.0, + 1949.0, + 567.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1933.0, + 605.0, + 1933.0, + 605.0, + 1949.0, + 589.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1934.0, + 627.0, + 1934.0, + 627.0, + 1948.0, + 611.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1934.0, + 752.0, + 1934.0, + 752.0, + 1948.0, + 736.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1934.0, + 779.0, + 1934.0, + 779.0, + 1949.0, + 762.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1934.0, + 801.0, + 1934.0, + 801.0, + 1949.0, + 782.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1933.0, + 821.0, + 1933.0, + 821.0, + 1948.0, + 804.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1934.0, + 843.0, + 1934.0, + 843.0, + 1948.0, + 827.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1955.0, + 318.0, + 1955.0, + 318.0, + 1970.0, + 303.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1955.0, + 343.0, + 1955.0, + 343.0, + 1971.0, + 329.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1955.0, + 367.0, + 1955.0, + 367.0, + 1971.0, + 351.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1953.0, + 388.0, + 1953.0, + 388.0, + 1970.0, + 372.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1956.0, + 407.0, + 1956.0, + 407.0, + 1968.0, + 397.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1953.0, + 541.0, + 1953.0, + 541.0, + 1971.0, + 524.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1955.0, + 561.0, + 1955.0, + 561.0, + 1970.0, + 546.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1955.0, + 583.0, + 1955.0, + 583.0, + 1971.0, + 568.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1953.0, + 605.0, + 1953.0, + 605.0, + 1970.0, + 589.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1956.0, + 626.0, + 1956.0, + 626.0, + 1968.0, + 613.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1955.0, + 751.0, + 1955.0, + 751.0, + 1970.0, + 736.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1955.0, + 778.0, + 1955.0, + 778.0, + 1970.0, + 762.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1955.0, + 800.0, + 1955.0, + 800.0, + 1970.0, + 783.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1953.0, + 821.0, + 1953.0, + 821.0, + 1970.0, + 804.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1957.0, + 841.0, + 1957.0, + 841.0, + 1968.0, + 831.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1975.0, + 318.0, + 1975.0, + 318.0, + 1991.0, + 303.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1973.0, + 343.0, + 1973.0, + 343.0, + 1991.0, + 329.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1974.0, + 366.0, + 1974.0, + 366.0, + 1991.0, + 351.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1975.0, + 388.0, + 1975.0, + 388.0, + 1991.0, + 372.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1975.0, + 541.0, + 1975.0, + 541.0, + 1991.0, + 524.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1974.0, + 561.0, + 1974.0, + 561.0, + 1989.0, + 546.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1974.0, + 583.0, + 1974.0, + 583.0, + 1991.0, + 568.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1977.0, + 603.0, + 1977.0, + 603.0, + 1988.0, + 591.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1977.0, + 750.0, + 1977.0, + 750.0, + 1988.0, + 737.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1974.0, + 778.0, + 1974.0, + 778.0, + 1989.0, + 763.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1974.0, + 800.0, + 1974.0, + 800.0, + 1989.0, + 784.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1977.0, + 819.0, + 1977.0, + 819.0, + 1988.0, + 807.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1993.0, + 318.0, + 1993.0, + 318.0, + 2010.0, + 303.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1995.0, + 343.0, + 1995.0, + 343.0, + 2013.0, + 329.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1995.0, + 366.0, + 1995.0, + 366.0, + 2013.0, + 351.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1993.0, + 388.0, + 1993.0, + 388.0, + 2009.0, + 373.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1993.0, + 541.0, + 1993.0, + 541.0, + 2010.0, + 524.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1996.0, + 561.0, + 1996.0, + 561.0, + 2011.0, + 546.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1996.0, + 583.0, + 1996.0, + 583.0, + 2011.0, + 568.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1996.0, + 603.0, + 1996.0, + 603.0, + 2008.0, + 591.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1996.0, + 750.0, + 1996.0, + 750.0, + 2008.0, + 737.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1997.0, + 777.0, + 1997.0, + 777.0, + 2009.0, + 765.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1996.0, + 800.0, + 1996.0, + 800.0, + 2011.0, + 784.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1996.0, + 819.0, + 1996.0, + 819.0, + 2008.0, + 807.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1670.5, + 367.0, + 1670.5, + 367.0, + 1682.5, + 348.0, + 1682.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 512.0, + 286.0, + 512.0, + 286.0, + 532.0, + 261.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 509.0, + 414.0, + 509.0, + 414.0, + 531.0, + 392.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 510.0, + 545.0, + 510.0, + 545.0, + 531.0, + 522.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 509.0, + 675.0, + 509.0, + 675.0, + 531.0, + 652.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 510.0, + 805.0, + 510.0, + 805.0, + 531.0, + 782.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 512.0, + 1065.0, + 512.0, + 1065.0, + 532.0, + 1042.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 531.0, + 280.0, + 531.0, + 280.0, + 549.0, + 265.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 531.0, + 304.0, + 531.0, + 304.0, + 549.0, + 286.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 531.0, + 323.0, + 531.0, + 323.0, + 549.0, + 308.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 531.0, + 411.0, + 531.0, + 411.0, + 547.0, + 395.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 532.0, + 430.0, + 532.0, + 430.0, + 546.0, + 418.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 531.0, + 454.0, + 531.0, + 454.0, + 549.0, + 437.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 531.0, + 474.0, + 531.0, + 474.0, + 549.0, + 458.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 531.0, + 541.0, + 531.0, + 541.0, + 549.0, + 525.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 532.0, + 563.0, + 532.0, + 563.0, + 547.0, + 546.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 534.0, + 581.0, + 534.0, + 581.0, + 545.0, + 568.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 534.0, + 604.0, + 534.0, + 604.0, + 546.0, + 590.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 532.0, + 625.0, + 532.0, + 625.0, + 546.0, + 613.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 531.0, + 672.0, + 531.0, + 672.0, + 549.0, + 655.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 531.0, + 694.0, + 531.0, + 694.0, + 549.0, + 676.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 531.0, + 713.0, + 531.0, + 713.0, + 549.0, + 698.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 531.0, + 801.0, + 531.0, + 801.0, + 547.0, + 785.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 531.0, + 822.0, + 531.0, + 822.0, + 547.0, + 806.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 534.0, + 841.0, + 534.0, + 841.0, + 546.0, + 830.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 534.0, + 862.0, + 534.0, + 862.0, + 546.0, + 852.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 531.0, + 1061.0, + 531.0, + 1061.0, + 549.0, + 1045.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 531.0, + 1082.0, + 531.0, + 1082.0, + 549.0, + 1065.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 532.0, + 1103.0, + 532.0, + 1103.0, + 549.0, + 1087.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 267.0, + 554.0, + 279.0, + 554.0, + 279.0, + 568.0, + 267.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 553.0, + 303.0, + 553.0, + 303.0, + 569.0, + 286.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 554.0, + 321.0, + 554.0, + 321.0, + 568.0, + 309.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 554.0, + 408.0, + 554.0, + 408.0, + 567.0, + 397.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 554.0, + 453.0, + 554.0, + 453.0, + 567.0, + 441.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 554.0, + 538.0, + 554.0, + 538.0, + 567.0, + 527.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 554.0, + 560.0, + 554.0, + 560.0, + 567.0, + 547.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 554.0, + 581.0, + 554.0, + 581.0, + 567.0, + 568.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 554.0, + 604.0, + 554.0, + 604.0, + 567.0, + 590.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 556.0, + 625.0, + 556.0, + 625.0, + 567.0, + 614.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 554.0, + 668.0, + 554.0, + 668.0, + 568.0, + 656.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 553.0, + 693.0, + 553.0, + 693.0, + 569.0, + 676.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 553.0, + 713.0, + 553.0, + 713.0, + 569.0, + 696.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 556.0, + 797.0, + 556.0, + 797.0, + 565.0, + 786.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 554.0, + 819.0, + 554.0, + 819.0, + 567.0, + 809.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 554.0, + 841.0, + 554.0, + 841.0, + 567.0, + 831.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 553.0, + 864.0, + 553.0, + 864.0, + 569.0, + 848.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 553.0, + 1061.0, + 553.0, + 1061.0, + 569.0, + 1045.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 553.0, + 1082.0, + 553.0, + 1082.0, + 569.0, + 1066.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 553.0, + 1103.0, + 553.0, + 1103.0, + 569.0, + 1087.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 572.0, + 280.0, + 572.0, + 280.0, + 589.0, + 263.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 572.0, + 303.0, + 572.0, + 303.0, + 589.0, + 284.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 572.0, + 323.0, + 572.0, + 323.0, + 589.0, + 307.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 572.0, + 411.0, + 572.0, + 411.0, + 587.0, + 393.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 572.0, + 432.0, + 572.0, + 432.0, + 589.0, + 416.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 574.0, + 451.0, + 574.0, + 451.0, + 586.0, + 439.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 572.0, + 474.0, + 572.0, + 474.0, + 589.0, + 458.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 574.0, + 538.0, + 574.0, + 538.0, + 587.0, + 526.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 574.0, + 560.0, + 574.0, + 560.0, + 586.0, + 547.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 574.0, + 581.0, + 574.0, + 581.0, + 586.0, + 568.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 572.0, + 605.0, + 572.0, + 605.0, + 589.0, + 588.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 572.0, + 671.0, + 572.0, + 671.0, + 589.0, + 654.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 572.0, + 693.0, + 572.0, + 693.0, + 589.0, + 676.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 572.0, + 711.0, + 572.0, + 711.0, + 589.0, + 696.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 574.0, + 797.0, + 574.0, + 797.0, + 586.0, + 786.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 574.0, + 819.0, + 574.0, + 819.0, + 587.0, + 807.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 574.0, + 840.0, + 574.0, + 840.0, + 587.0, + 828.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 572.0, + 864.0, + 572.0, + 864.0, + 589.0, + 848.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 572.0, + 1061.0, + 572.0, + 1061.0, + 589.0, + 1045.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 572.0, + 1082.0, + 572.0, + 1082.0, + 589.0, + 1065.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 572.0, + 1103.0, + 572.0, + 1103.0, + 589.0, + 1087.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 591.0, + 672.0, + 591.0, + 672.0, + 609.0, + 655.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 593.0, + 694.0, + 593.0, + 694.0, + 609.0, + 675.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 593.0, + 713.0, + 593.0, + 713.0, + 609.0, + 696.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 593.0, + 799.0, + 593.0, + 799.0, + 609.0, + 784.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 593.0, + 822.0, + 593.0, + 822.0, + 609.0, + 806.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 593.0, + 843.0, + 593.0, + 843.0, + 609.0, + 827.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 593.0, + 864.0, + 593.0, + 864.0, + 609.0, + 848.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 591.0, + 1061.0, + 591.0, + 1061.0, + 609.0, + 1044.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 591.0, + 1083.0, + 591.0, + 1083.0, + 609.0, + 1065.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 593.0, + 1103.0, + 593.0, + 1103.0, + 611.0, + 1087.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 615.0, + 1059.0, + 615.0, + 1059.0, + 627.0, + 1046.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 613.0, + 1083.0, + 613.0, + 1083.0, + 628.0, + 1065.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 615.0, + 1100.0, + 615.0, + 1100.0, + 627.0, + 1088.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 270.0, + 632.0, + 363.0, + 632.0, + 363.0, + 657.0, + 270.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 632.0, + 493.0, + 632.0, + 493.0, + 657.0, + 401.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 631.0, + 622.0, + 631.0, + 622.0, + 659.0, + 531.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 632.0, + 753.0, + 632.0, + 753.0, + 657.0, + 660.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 631.0, + 883.0, + 631.0, + 883.0, + 659.0, + 791.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 633.0, + 1012.0, + 633.0, + 1012.0, + 657.0, + 921.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 632.0, + 1143.0, + 632.0, + 1143.0, + 657.0, + 1050.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 630.0, + 1275.0, + 630.0, + 1275.0, + 661.0, + 1268.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 653.0, + 1402.0, + 653.0, + 1402.0, + 677.0, + 1329.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 572.0, + 621.0, + 572.0, + 621.0, + 588.0, + 614.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 798.0, + 317.0, + 798.0, + 317.0, + 813.0, + 304.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 797.0, + 345.0, + 797.0, + 345.0, + 813.0, + 330.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 797.0, + 366.0, + 797.0, + 366.0, + 812.0, + 351.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 798.0, + 388.0, + 798.0, + 388.0, + 813.0, + 373.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 796.0, + 409.0, + 796.0, + 409.0, + 811.0, + 395.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 798.0, + 492.0, + 798.0, + 492.0, + 812.0, + 476.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 797.0, + 518.0, + 797.0, + 518.0, + 812.0, + 503.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 797.0, + 539.0, + 797.0, + 539.0, + 811.0, + 525.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 798.0, + 561.0, + 798.0, + 561.0, + 812.0, + 546.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 798.0, + 580.0, + 798.0, + 580.0, + 810.0, + 571.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 798.0, + 665.0, + 798.0, + 665.0, + 813.0, + 650.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 798.0, + 692.0, + 798.0, + 692.0, + 812.0, + 676.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 798.0, + 714.0, + 798.0, + 714.0, + 812.0, + 697.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 798.0, + 735.0, + 798.0, + 735.0, + 813.0, + 718.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 797.0, + 754.0, + 797.0, + 754.0, + 812.0, + 741.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 798.0, + 838.0, + 798.0, + 838.0, + 814.0, + 823.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 797.0, + 865.0, + 797.0, + 865.0, + 812.0, + 849.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 797.0, + 887.0, + 797.0, + 887.0, + 812.0, + 871.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 798.0, + 907.0, + 798.0, + 907.0, + 813.0, + 892.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 798.0, + 926.0, + 798.0, + 926.0, + 810.0, + 916.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 798.0, + 1016.0, + 798.0, + 1016.0, + 813.0, + 1002.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 797.0, + 1038.0, + 797.0, + 1038.0, + 812.0, + 1022.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 797.0, + 1059.0, + 797.0, + 1059.0, + 811.0, + 1044.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 798.0, + 1081.0, + 798.0, + 1081.0, + 812.0, + 1066.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 798.0, + 1100.0, + 798.0, + 1100.0, + 810.0, + 1089.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 798.0, + 1185.0, + 798.0, + 1185.0, + 813.0, + 1171.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 798.0, + 1212.0, + 798.0, + 1212.0, + 812.0, + 1196.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 798.0, + 1235.0, + 798.0, + 1235.0, + 812.0, + 1217.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 797.0, + 1257.0, + 797.0, + 1257.0, + 814.0, + 1236.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 797.0, + 1274.0, + 797.0, + 1274.0, + 812.0, + 1261.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 817.0, + 317.0, + 817.0, + 317.0, + 833.0, + 303.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 818.0, + 345.0, + 818.0, + 345.0, + 832.0, + 330.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 818.0, + 366.0, + 818.0, + 366.0, + 832.0, + 352.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 817.0, + 388.0, + 817.0, + 388.0, + 832.0, + 373.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 818.0, + 410.0, + 818.0, + 410.0, + 832.0, + 395.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 818.0, + 490.0, + 818.0, + 490.0, + 833.0, + 476.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 818.0, + 518.0, + 818.0, + 518.0, + 832.0, + 503.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 818.0, + 539.0, + 818.0, + 539.0, + 833.0, + 526.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 817.0, + 561.0, + 817.0, + 561.0, + 832.0, + 546.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 818.0, + 584.0, + 818.0, + 584.0, + 832.0, + 568.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 817.0, + 664.0, + 817.0, + 664.0, + 833.0, + 650.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 818.0, + 691.0, + 818.0, + 691.0, + 833.0, + 676.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 818.0, + 714.0, + 818.0, + 714.0, + 833.0, + 697.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 817.0, + 735.0, + 817.0, + 735.0, + 832.0, + 719.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 818.0, + 755.0, + 818.0, + 755.0, + 833.0, + 741.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 817.0, + 838.0, + 817.0, + 838.0, + 834.0, + 823.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 818.0, + 865.0, + 818.0, + 865.0, + 832.0, + 849.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 818.0, + 886.0, + 818.0, + 886.0, + 832.0, + 872.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 817.0, + 907.0, + 817.0, + 907.0, + 832.0, + 892.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 818.0, + 929.0, + 818.0, + 929.0, + 832.0, + 914.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 817.0, + 1016.0, + 817.0, + 1016.0, + 833.0, + 1002.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 818.0, + 1037.0, + 818.0, + 1037.0, + 832.0, + 1023.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 819.0, + 1057.0, + 819.0, + 1057.0, + 831.0, + 1047.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 817.0, + 1081.0, + 817.0, + 1081.0, + 832.0, + 1067.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 818.0, + 1102.0, + 818.0, + 1102.0, + 833.0, + 1088.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 818.0, + 1184.0, + 818.0, + 1184.0, + 833.0, + 1170.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 818.0, + 1211.0, + 818.0, + 1211.0, + 833.0, + 1196.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 818.0, + 1233.0, + 818.0, + 1233.0, + 833.0, + 1217.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 817.0, + 1253.0, + 817.0, + 1253.0, + 832.0, + 1238.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 818.0, + 1275.0, + 818.0, + 1275.0, + 833.0, + 1261.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 838.0, + 317.0, + 838.0, + 317.0, + 854.0, + 304.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 838.0, + 344.0, + 838.0, + 344.0, + 853.0, + 330.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 839.0, + 366.0, + 839.0, + 366.0, + 853.0, + 352.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 838.0, + 388.0, + 838.0, + 388.0, + 853.0, + 373.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 838.0, + 410.0, + 838.0, + 410.0, + 853.0, + 395.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 838.0, + 490.0, + 838.0, + 490.0, + 853.0, + 476.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 839.0, + 518.0, + 839.0, + 518.0, + 853.0, + 503.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 839.0, + 540.0, + 839.0, + 540.0, + 853.0, + 525.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 838.0, + 561.0, + 838.0, + 561.0, + 853.0, + 546.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 839.0, + 584.0, + 839.0, + 584.0, + 854.0, + 568.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 838.0, + 664.0, + 838.0, + 664.0, + 853.0, + 650.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 840.0, + 689.0, + 840.0, + 689.0, + 852.0, + 678.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 839.0, + 712.0, + 839.0, + 712.0, + 853.0, + 698.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 838.0, + 735.0, + 838.0, + 735.0, + 853.0, + 719.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 838.0, + 755.0, + 838.0, + 755.0, + 853.0, + 742.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 838.0, + 838.0, + 838.0, + 838.0, + 853.0, + 823.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 840.0, + 861.0, + 840.0, + 861.0, + 852.0, + 852.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 839.0, + 886.0, + 839.0, + 886.0, + 853.0, + 872.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 838.0, + 907.0, + 838.0, + 907.0, + 853.0, + 893.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 839.0, + 929.0, + 839.0, + 929.0, + 854.0, + 914.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 838.0, + 1016.0, + 838.0, + 1016.0, + 853.0, + 1002.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 841.0, + 1036.0, + 841.0, + 1036.0, + 852.0, + 1025.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 840.0, + 1057.0, + 840.0, + 1057.0, + 852.0, + 1047.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 838.0, + 1081.0, + 838.0, + 1081.0, + 853.0, + 1066.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 839.0, + 1102.0, + 839.0, + 1102.0, + 854.0, + 1088.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 838.0, + 1184.0, + 838.0, + 1184.0, + 853.0, + 1170.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 841.0, + 1209.0, + 841.0, + 1209.0, + 851.0, + 1199.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 840.0, + 1231.0, + 840.0, + 1231.0, + 852.0, + 1220.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 838.0, + 1253.0, + 838.0, + 1253.0, + 853.0, + 1239.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 839.0, + 1276.0, + 839.0, + 1276.0, + 854.0, + 1262.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 859.0, + 318.0, + 859.0, + 318.0, + 874.0, + 304.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 858.0, + 343.0, + 858.0, + 343.0, + 874.0, + 330.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 858.0, + 366.0, + 858.0, + 366.0, + 874.0, + 352.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 859.0, + 388.0, + 859.0, + 388.0, + 874.0, + 373.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 858.0, + 411.0, + 858.0, + 411.0, + 872.0, + 397.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 859.0, + 490.0, + 859.0, + 490.0, + 874.0, + 476.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 858.0, + 518.0, + 858.0, + 518.0, + 874.0, + 503.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 858.0, + 539.0, + 858.0, + 539.0, + 874.0, + 525.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 859.0, + 561.0, + 859.0, + 561.0, + 874.0, + 546.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 858.0, + 585.0, + 858.0, + 585.0, + 872.0, + 571.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 860.0, + 664.0, + 860.0, + 664.0, + 874.0, + 650.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 859.0, + 690.0, + 859.0, + 690.0, + 874.0, + 677.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 858.0, + 712.0, + 858.0, + 712.0, + 874.0, + 698.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 859.0, + 734.0, + 859.0, + 734.0, + 874.0, + 719.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 860.0, + 755.0, + 860.0, + 755.0, + 870.0, + 745.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 859.0, + 838.0, + 859.0, + 838.0, + 875.0, + 823.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 860.0, + 861.0, + 860.0, + 861.0, + 872.0, + 852.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 858.0, + 886.0, + 858.0, + 886.0, + 874.0, + 872.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 859.0, + 907.0, + 859.0, + 907.0, + 874.0, + 893.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 859.0, + 927.0, + 859.0, + 927.0, + 872.0, + 918.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 860.0, + 1016.0, + 860.0, + 1016.0, + 874.0, + 1002.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 858.0, + 1037.0, + 858.0, + 1037.0, + 873.0, + 1023.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 858.0, + 1059.0, + 858.0, + 1059.0, + 873.0, + 1046.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 859.0, + 1080.0, + 859.0, + 1080.0, + 874.0, + 1066.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 859.0, + 1102.0, + 859.0, + 1102.0, + 870.0, + 1093.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 860.0, + 1184.0, + 860.0, + 1184.0, + 874.0, + 1170.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 860.0, + 1209.0, + 860.0, + 1209.0, + 872.0, + 1199.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 859.0, + 1232.0, + 859.0, + 1232.0, + 874.0, + 1219.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 859.0, + 1253.0, + 859.0, + 1253.0, + 874.0, + 1239.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 859.0, + 1275.0, + 859.0, + 1275.0, + 870.0, + 1265.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 879.0, + 317.0, + 879.0, + 317.0, + 894.0, + 303.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 880.0, + 344.0, + 880.0, + 344.0, + 895.0, + 330.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 880.0, + 366.0, + 880.0, + 366.0, + 895.0, + 351.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 879.0, + 388.0, + 879.0, + 388.0, + 894.0, + 373.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 879.0, + 490.0, + 879.0, + 490.0, + 894.0, + 476.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 880.0, + 518.0, + 880.0, + 518.0, + 896.0, + 503.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 880.0, + 539.0, + 880.0, + 539.0, + 896.0, + 525.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 879.0, + 561.0, + 879.0, + 561.0, + 895.0, + 546.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 881.0, + 580.0, + 881.0, + 580.0, + 891.0, + 571.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 879.0, + 664.0, + 879.0, + 664.0, + 894.0, + 650.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 880.0, + 690.0, + 880.0, + 690.0, + 896.0, + 677.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 880.0, + 712.0, + 880.0, + 712.0, + 897.0, + 698.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 880.0, + 734.0, + 880.0, + 734.0, + 894.0, + 718.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 881.0, + 753.0, + 881.0, + 753.0, + 891.0, + 744.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 879.0, + 836.0, + 879.0, + 836.0, + 894.0, + 823.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 880.0, + 864.0, + 880.0, + 864.0, + 895.0, + 848.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 880.0, + 886.0, + 880.0, + 886.0, + 895.0, + 870.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 879.0, + 907.0, + 879.0, + 907.0, + 894.0, + 893.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 879.0, + 1016.0, + 879.0, + 1016.0, + 895.0, + 1002.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 881.0, + 1037.0, + 881.0, + 1037.0, + 895.0, + 1023.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 881.0, + 1059.0, + 881.0, + 1059.0, + 896.0, + 1044.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 880.0, + 1081.0, + 880.0, + 1081.0, + 894.0, + 1066.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 882.0, + 1100.0, + 882.0, + 1100.0, + 891.0, + 1090.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 879.0, + 1184.0, + 879.0, + 1184.0, + 895.0, + 1170.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 882.0, + 1209.0, + 882.0, + 1209.0, + 894.0, + 1199.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 880.0, + 1232.0, + 880.0, + 1232.0, + 896.0, + 1219.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 880.0, + 1253.0, + 880.0, + 1253.0, + 894.0, + 1239.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 961.0, + 492.0, + 961.0, + 492.0, + 976.0, + 476.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 959.0, + 519.0, + 959.0, + 519.0, + 976.0, + 503.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 959.0, + 540.0, + 959.0, + 540.0, + 975.0, + 525.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 961.0, + 562.0, + 961.0, + 562.0, + 975.0, + 546.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 961.0, + 580.0, + 961.0, + 580.0, + 972.0, + 571.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 961.0, + 665.0, + 961.0, + 665.0, + 976.0, + 650.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 961.0, + 692.0, + 961.0, + 692.0, + 976.0, + 677.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 961.0, + 714.0, + 961.0, + 714.0, + 976.0, + 697.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 961.0, + 735.0, + 961.0, + 735.0, + 976.0, + 718.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 959.0, + 754.0, + 959.0, + 754.0, + 975.0, + 741.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 979.0, + 492.0, + 979.0, + 492.0, + 996.0, + 476.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 980.0, + 519.0, + 980.0, + 519.0, + 997.0, + 503.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 983.0, + 538.0, + 983.0, + 538.0, + 995.0, + 527.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 979.0, + 561.0, + 979.0, + 561.0, + 995.0, + 546.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 983.0, + 581.0, + 983.0, + 581.0, + 993.0, + 571.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 978.0, + 666.0, + 978.0, + 666.0, + 998.0, + 647.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 982.0, + 692.0, + 982.0, + 692.0, + 996.0, + 676.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 982.0, + 714.0, + 982.0, + 714.0, + 996.0, + 698.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 979.0, + 735.0, + 979.0, + 735.0, + 995.0, + 719.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 983.0, + 754.0, + 983.0, + 754.0, + 995.0, + 743.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1000.0, + 492.0, + 1000.0, + 492.0, + 1017.0, + 476.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1002.0, + 518.0, + 1002.0, + 518.0, + 1017.0, + 503.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1003.0, + 540.0, + 1003.0, + 540.0, + 1017.0, + 526.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1000.0, + 561.0, + 1000.0, + 561.0, + 1017.0, + 547.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1003.0, + 582.0, + 1003.0, + 582.0, + 1017.0, + 568.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 999.0, + 666.0, + 999.0, + 666.0, + 1018.0, + 647.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1002.0, + 691.0, + 1002.0, + 691.0, + 1018.0, + 676.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1002.0, + 712.0, + 1002.0, + 712.0, + 1017.0, + 698.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1000.0, + 735.0, + 1000.0, + 735.0, + 1017.0, + 719.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1002.0, + 755.0, + 1002.0, + 755.0, + 1017.0, + 741.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1021.0, + 492.0, + 1021.0, + 492.0, + 1038.0, + 476.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1020.0, + 518.0, + 1020.0, + 518.0, + 1037.0, + 503.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1020.0, + 540.0, + 1020.0, + 540.0, + 1035.0, + 526.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1021.0, + 561.0, + 1021.0, + 561.0, + 1037.0, + 547.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1020.0, + 584.0, + 1020.0, + 584.0, + 1035.0, + 571.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1021.0, + 665.0, + 1021.0, + 665.0, + 1038.0, + 650.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1021.0, + 691.0, + 1021.0, + 691.0, + 1037.0, + 676.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1020.0, + 712.0, + 1020.0, + 712.0, + 1035.0, + 698.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1021.0, + 735.0, + 1021.0, + 735.0, + 1037.0, + 719.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1021.0, + 755.0, + 1021.0, + 755.0, + 1034.0, + 745.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1040.0, + 492.0, + 1040.0, + 492.0, + 1057.0, + 476.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1042.0, + 518.0, + 1042.0, + 518.0, + 1059.0, + 503.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1042.0, + 539.0, + 1042.0, + 539.0, + 1059.0, + 526.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1041.0, + 561.0, + 1041.0, + 561.0, + 1057.0, + 547.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1040.0, + 664.0, + 1040.0, + 664.0, + 1057.0, + 650.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1042.0, + 691.0, + 1042.0, + 691.0, + 1059.0, + 676.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1042.0, + 712.0, + 1042.0, + 712.0, + 1059.0, + 698.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1041.0, + 735.0, + 1041.0, + 735.0, + 1057.0, + 719.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1042.0, + 753.0, + 1042.0, + 753.0, + 1054.0, + 743.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1021.5, + 405.0, + 1021.5, + 405.0, + 1032.0, + 402.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1195.0, + 316.0, + 1195.0, + 316.0, + 1206.0, + 304.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1192.0, + 345.0, + 1192.0, + 345.0, + 1209.0, + 331.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1193.0, + 364.0, + 1193.0, + 364.0, + 1206.0, + 353.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1195.0, + 387.0, + 1195.0, + 387.0, + 1206.0, + 374.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1193.0, + 408.0, + 1193.0, + 408.0, + 1205.0, + 397.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1192.0, + 492.0, + 1192.0, + 492.0, + 1208.0, + 476.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1192.0, + 520.0, + 1192.0, + 520.0, + 1208.0, + 503.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1192.0, + 541.0, + 1192.0, + 541.0, + 1208.0, + 523.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1192.0, + 563.0, + 1192.0, + 563.0, + 1208.0, + 544.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1193.0, + 580.0, + 1193.0, + 580.0, + 1205.0, + 568.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1195.0, + 663.0, + 1195.0, + 663.0, + 1206.0, + 651.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1192.0, + 693.0, + 1192.0, + 693.0, + 1208.0, + 676.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1192.0, + 715.0, + 1192.0, + 715.0, + 1208.0, + 696.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1192.0, + 736.0, + 1192.0, + 736.0, + 1208.0, + 717.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1191.0, + 756.0, + 1191.0, + 756.0, + 1206.0, + 741.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1193.0, + 838.0, + 1193.0, + 838.0, + 1209.0, + 823.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1192.0, + 866.0, + 1192.0, + 866.0, + 1208.0, + 848.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1192.0, + 886.0, + 1192.0, + 886.0, + 1208.0, + 871.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1193.0, + 908.0, + 1193.0, + 908.0, + 1208.0, + 892.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1193.0, + 927.0, + 1193.0, + 927.0, + 1205.0, + 917.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1195.0, + 1037.0, + 1195.0, + 1037.0, + 1206.0, + 1025.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1192.0, + 1061.0, + 1192.0, + 1061.0, + 1208.0, + 1044.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1192.0, + 1081.0, + 1192.0, + 1081.0, + 1208.0, + 1064.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1192.0, + 1103.0, + 1192.0, + 1103.0, + 1208.0, + 1086.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1193.0, + 1121.0, + 1193.0, + 1121.0, + 1205.0, + 1109.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1213.0, + 316.0, + 1213.0, + 316.0, + 1226.0, + 304.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1214.0, + 343.0, + 1214.0, + 343.0, + 1227.0, + 332.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1214.0, + 364.0, + 1214.0, + 364.0, + 1226.0, + 353.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1213.0, + 386.0, + 1213.0, + 386.0, + 1226.0, + 374.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1214.0, + 408.0, + 1214.0, + 408.0, + 1226.0, + 397.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1213.0, + 489.0, + 1213.0, + 489.0, + 1226.0, + 478.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1213.0, + 518.0, + 1213.0, + 518.0, + 1228.0, + 503.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1214.0, + 538.0, + 1214.0, + 538.0, + 1226.0, + 526.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1213.0, + 559.0, + 1213.0, + 559.0, + 1226.0, + 548.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1212.0, + 583.0, + 1212.0, + 583.0, + 1227.0, + 568.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1212.0, + 664.0, + 1212.0, + 664.0, + 1227.0, + 650.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1213.0, + 691.0, + 1213.0, + 691.0, + 1228.0, + 676.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1214.0, + 712.0, + 1214.0, + 712.0, + 1226.0, + 699.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1212.0, + 734.0, + 1212.0, + 734.0, + 1227.0, + 718.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1214.0, + 755.0, + 1214.0, + 755.0, + 1226.0, + 742.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1212.0, + 838.0, + 1212.0, + 838.0, + 1227.0, + 823.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1213.0, + 865.0, + 1213.0, + 865.0, + 1228.0, + 848.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1214.0, + 884.0, + 1214.0, + 884.0, + 1226.0, + 872.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1212.0, + 908.0, + 1212.0, + 908.0, + 1227.0, + 893.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1214.0, + 927.0, + 1214.0, + 927.0, + 1226.0, + 916.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1212.0, + 1038.0, + 1212.0, + 1038.0, + 1227.0, + 1023.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1213.0, + 1059.0, + 1213.0, + 1059.0, + 1227.0, + 1043.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1214.0, + 1080.0, + 1214.0, + 1080.0, + 1226.0, + 1067.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1213.0, + 1101.0, + 1213.0, + 1101.0, + 1226.0, + 1090.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1215.0, + 1122.0, + 1215.0, + 1122.0, + 1226.0, + 1112.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1232.0, + 318.0, + 1232.0, + 318.0, + 1248.0, + 303.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1232.0, + 343.0, + 1232.0, + 343.0, + 1250.0, + 329.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1232.0, + 367.0, + 1232.0, + 367.0, + 1250.0, + 351.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1235.0, + 387.0, + 1235.0, + 387.0, + 1246.0, + 374.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1235.0, + 408.0, + 1235.0, + 408.0, + 1248.0, + 397.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1235.0, + 489.0, + 1235.0, + 489.0, + 1246.0, + 478.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1234.0, + 518.0, + 1234.0, + 518.0, + 1249.0, + 503.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1232.0, + 540.0, + 1232.0, + 540.0, + 1250.0, + 523.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1232.0, + 561.0, + 1232.0, + 561.0, + 1248.0, + 546.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1235.0, + 582.0, + 1235.0, + 582.0, + 1246.0, + 571.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1235.0, + 662.0, + 1235.0, + 662.0, + 1246.0, + 651.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1235.0, + 690.0, + 1235.0, + 690.0, + 1248.0, + 678.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1234.0, + 713.0, + 1234.0, + 713.0, + 1249.0, + 698.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1234.0, + 734.0, + 1234.0, + 734.0, + 1248.0, + 718.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1236.0, + 755.0, + 1236.0, + 755.0, + 1246.0, + 743.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1232.0, + 838.0, + 1232.0, + 838.0, + 1248.0, + 823.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1236.0, + 862.0, + 1236.0, + 862.0, + 1248.0, + 850.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1234.0, + 887.0, + 1234.0, + 887.0, + 1249.0, + 871.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1232.0, + 908.0, + 1232.0, + 908.0, + 1248.0, + 893.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1235.0, + 927.0, + 1235.0, + 927.0, + 1248.0, + 916.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1232.0, + 1038.0, + 1232.0, + 1038.0, + 1248.0, + 1023.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1236.0, + 1057.0, + 1236.0, + 1057.0, + 1248.0, + 1046.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1234.0, + 1081.0, + 1234.0, + 1081.0, + 1249.0, + 1066.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1232.0, + 1103.0, + 1232.0, + 1103.0, + 1248.0, + 1087.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1235.0, + 1122.0, + 1235.0, + 1122.0, + 1246.0, + 1111.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1255.0, + 316.0, + 1255.0, + 316.0, + 1267.0, + 304.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1252.0, + 343.0, + 1252.0, + 343.0, + 1268.0, + 329.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1253.0, + 366.0, + 1253.0, + 366.0, + 1268.0, + 351.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1255.0, + 387.0, + 1255.0, + 387.0, + 1267.0, + 374.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1255.0, + 410.0, + 1255.0, + 410.0, + 1265.0, + 399.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1255.0, + 489.0, + 1255.0, + 489.0, + 1267.0, + 478.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1252.0, + 518.0, + 1252.0, + 518.0, + 1268.0, + 503.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1252.0, + 541.0, + 1252.0, + 541.0, + 1268.0, + 524.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1255.0, + 559.0, + 1255.0, + 559.0, + 1267.0, + 547.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1252.0, + 586.0, + 1252.0, + 586.0, + 1268.0, + 571.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1257.0, + 662.0, + 1257.0, + 662.0, + 1267.0, + 652.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1254.0, + 690.0, + 1254.0, + 690.0, + 1267.0, + 678.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1252.0, + 713.0, + 1252.0, + 713.0, + 1268.0, + 698.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1255.0, + 733.0, + 1255.0, + 733.0, + 1267.0, + 721.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1255.0, + 837.0, + 1255.0, + 837.0, + 1267.0, + 825.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1254.0, + 862.0, + 1254.0, + 862.0, + 1267.0, + 851.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1253.0, + 886.0, + 1253.0, + 886.0, + 1268.0, + 871.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1254.0, + 908.0, + 1254.0, + 908.0, + 1268.0, + 892.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1254.0, + 928.0, + 1254.0, + 928.0, + 1266.0, + 920.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1255.0, + 1036.0, + 1255.0, + 1036.0, + 1267.0, + 1025.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1254.0, + 1057.0, + 1254.0, + 1057.0, + 1267.0, + 1046.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1254.0, + 1080.0, + 1254.0, + 1080.0, + 1267.0, + 1068.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1254.0, + 1103.0, + 1254.0, + 1103.0, + 1268.0, + 1087.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1253.0, + 1124.0, + 1253.0, + 1124.0, + 1266.0, + 1113.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1275.0, + 316.0, + 1275.0, + 316.0, + 1288.0, + 304.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1276.0, + 342.0, + 1276.0, + 342.0, + 1289.0, + 332.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1277.0, + 364.0, + 1277.0, + 364.0, + 1289.0, + 353.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1275.0, + 386.0, + 1275.0, + 386.0, + 1288.0, + 374.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1275.0, + 489.0, + 1275.0, + 489.0, + 1288.0, + 478.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1276.0, + 518.0, + 1276.0, + 518.0, + 1292.0, + 503.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1275.0, + 540.0, + 1275.0, + 540.0, + 1292.0, + 524.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1274.0, + 561.0, + 1274.0, + 561.0, + 1290.0, + 546.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1276.0, + 662.0, + 1276.0, + 662.0, + 1288.0, + 652.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1276.0, + 691.0, + 1276.0, + 691.0, + 1292.0, + 676.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1275.0, + 713.0, + 1275.0, + 713.0, + 1292.0, + 698.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1274.0, + 734.0, + 1274.0, + 734.0, + 1290.0, + 718.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1276.0, + 836.0, + 1276.0, + 836.0, + 1288.0, + 825.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1277.0, + 862.0, + 1277.0, + 862.0, + 1289.0, + 850.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1277.0, + 884.0, + 1277.0, + 884.0, + 1289.0, + 872.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1276.0, + 906.0, + 1276.0, + 906.0, + 1288.0, + 894.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1274.0, + 1038.0, + 1274.0, + 1038.0, + 1289.0, + 1023.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1277.0, + 1057.0, + 1277.0, + 1057.0, + 1290.0, + 1046.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1276.0, + 1081.0, + 1276.0, + 1081.0, + 1292.0, + 1066.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1274.0, + 1103.0, + 1274.0, + 1103.0, + 1289.0, + 1087.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1355.0, + 492.0, + 1355.0, + 492.0, + 1372.0, + 476.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1355.0, + 518.0, + 1355.0, + 518.0, + 1371.0, + 503.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1354.0, + 541.0, + 1354.0, + 541.0, + 1371.0, + 524.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1358.0, + 559.0, + 1358.0, + 559.0, + 1369.0, + 547.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1356.0, + 580.0, + 1356.0, + 580.0, + 1367.0, + 570.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1356.0, + 666.0, + 1356.0, + 666.0, + 1372.0, + 650.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1355.0, + 693.0, + 1355.0, + 693.0, + 1371.0, + 676.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1355.0, + 739.0, + 1355.0, + 739.0, + 1373.0, + 696.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1356.0, + 753.0, + 1356.0, + 753.0, + 1368.0, + 742.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1356.0, + 840.0, + 1356.0, + 840.0, + 1372.0, + 823.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1355.0, + 865.0, + 1355.0, + 865.0, + 1371.0, + 850.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1355.0, + 886.0, + 1355.0, + 886.0, + 1371.0, + 871.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1358.0, + 907.0, + 1358.0, + 907.0, + 1369.0, + 894.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1355.0, + 927.0, + 1355.0, + 927.0, + 1368.0, + 917.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1356.0, + 1039.0, + 1356.0, + 1039.0, + 1372.0, + 1023.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1355.0, + 1061.0, + 1355.0, + 1061.0, + 1372.0, + 1043.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1355.0, + 1081.0, + 1355.0, + 1081.0, + 1371.0, + 1064.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1358.0, + 1102.0, + 1358.0, + 1102.0, + 1369.0, + 1090.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1355.0, + 1122.0, + 1355.0, + 1122.0, + 1365.0, + 1112.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1356.0, + 1228.0, + 1356.0, + 1228.0, + 1372.0, + 1212.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1355.0, + 1254.0, + 1355.0, + 1254.0, + 1372.0, + 1240.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1355.0, + 1277.0, + 1355.0, + 1277.0, + 1372.0, + 1259.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1356.0, + 1300.0, + 1356.0, + 1300.0, + 1371.0, + 1281.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1355.0, + 1317.0, + 1355.0, + 1317.0, + 1368.0, + 1306.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1375.0, + 491.0, + 1375.0, + 491.0, + 1391.0, + 476.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1375.0, + 520.0, + 1375.0, + 520.0, + 1391.0, + 503.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1375.0, + 541.0, + 1375.0, + 541.0, + 1391.0, + 523.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1375.0, + 561.0, + 1375.0, + 561.0, + 1390.0, + 546.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1376.0, + 581.0, + 1376.0, + 581.0, + 1389.0, + 570.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1375.0, + 664.0, + 1375.0, + 664.0, + 1391.0, + 650.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1376.0, + 692.0, + 1376.0, + 692.0, + 1391.0, + 676.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1376.0, + 715.0, + 1376.0, + 715.0, + 1390.0, + 696.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1375.0, + 736.0, + 1375.0, + 736.0, + 1390.0, + 720.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1377.0, + 755.0, + 1377.0, + 755.0, + 1389.0, + 742.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1375.0, + 838.0, + 1375.0, + 838.0, + 1391.0, + 823.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1376.0, + 866.0, + 1376.0, + 866.0, + 1391.0, + 848.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1375.0, + 886.0, + 1375.0, + 886.0, + 1390.0, + 870.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1377.0, + 906.0, + 1377.0, + 906.0, + 1387.0, + 894.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1376.0, + 927.0, + 1376.0, + 927.0, + 1389.0, + 916.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1375.0, + 1038.0, + 1375.0, + 1038.0, + 1391.0, + 1023.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1376.0, + 1059.0, + 1376.0, + 1059.0, + 1391.0, + 1043.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1375.0, + 1081.0, + 1375.0, + 1081.0, + 1390.0, + 1064.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1375.0, + 1103.0, + 1375.0, + 1103.0, + 1390.0, + 1087.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1375.0, + 1123.0, + 1375.0, + 1123.0, + 1390.0, + 1108.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1375.0, + 1228.0, + 1375.0, + 1228.0, + 1391.0, + 1212.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1376.0, + 1254.0, + 1376.0, + 1254.0, + 1391.0, + 1238.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1375.0, + 1276.0, + 1375.0, + 1276.0, + 1390.0, + 1259.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1375.0, + 1298.0, + 1375.0, + 1298.0, + 1390.0, + 1281.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1395.0, + 491.0, + 1395.0, + 491.0, + 1412.0, + 476.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1397.0, + 518.0, + 1397.0, + 518.0, + 1412.0, + 503.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1397.0, + 541.0, + 1397.0, + 541.0, + 1412.0, + 524.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1395.0, + 561.0, + 1395.0, + 561.0, + 1412.0, + 547.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1397.0, + 583.0, + 1397.0, + 583.0, + 1412.0, + 568.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1395.0, + 666.0, + 1395.0, + 666.0, + 1412.0, + 650.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1397.0, + 691.0, + 1397.0, + 691.0, + 1412.0, + 676.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1397.0, + 713.0, + 1397.0, + 713.0, + 1412.0, + 698.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1398.0, + 733.0, + 1398.0, + 733.0, + 1409.0, + 721.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1397.0, + 756.0, + 1397.0, + 756.0, + 1412.0, + 741.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1395.0, + 838.0, + 1395.0, + 838.0, + 1412.0, + 823.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1397.0, + 863.0, + 1397.0, + 863.0, + 1413.0, + 850.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1397.0, + 886.0, + 1397.0, + 886.0, + 1412.0, + 871.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1398.0, + 907.0, + 1398.0, + 907.0, + 1409.0, + 894.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1398.0, + 927.0, + 1398.0, + 927.0, + 1411.0, + 917.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1395.0, + 1038.0, + 1395.0, + 1038.0, + 1412.0, + 1023.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1397.0, + 1058.0, + 1397.0, + 1058.0, + 1413.0, + 1044.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1397.0, + 1081.0, + 1397.0, + 1081.0, + 1412.0, + 1066.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1395.0, + 1103.0, + 1395.0, + 1103.0, + 1412.0, + 1088.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1397.0, + 1124.0, + 1397.0, + 1124.0, + 1412.0, + 1109.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1395.0, + 1228.0, + 1395.0, + 1228.0, + 1412.0, + 1212.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1397.0, + 1253.0, + 1397.0, + 1253.0, + 1413.0, + 1238.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1397.0, + 1276.0, + 1397.0, + 1276.0, + 1413.0, + 1261.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1398.0, + 1318.0, + 1398.0, + 1318.0, + 1411.0, + 1307.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1417.0, + 491.0, + 1417.0, + 491.0, + 1433.0, + 476.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1416.0, + 518.0, + 1416.0, + 518.0, + 1433.0, + 503.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1417.0, + 538.0, + 1417.0, + 538.0, + 1430.0, + 527.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1419.0, + 559.0, + 1419.0, + 559.0, + 1430.0, + 547.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1417.0, + 666.0, + 1417.0, + 666.0, + 1433.0, + 650.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1416.0, + 691.0, + 1416.0, + 691.0, + 1431.0, + 676.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1416.0, + 713.0, + 1416.0, + 713.0, + 1431.0, + 698.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1419.0, + 733.0, + 1419.0, + 733.0, + 1430.0, + 721.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1417.0, + 838.0, + 1417.0, + 838.0, + 1433.0, + 823.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1417.0, + 862.0, + 1417.0, + 862.0, + 1430.0, + 851.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1416.0, + 886.0, + 1416.0, + 886.0, + 1431.0, + 871.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1419.0, + 907.0, + 1419.0, + 907.0, + 1430.0, + 894.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1417.0, + 1038.0, + 1417.0, + 1038.0, + 1433.0, + 1023.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1416.0, + 1058.0, + 1416.0, + 1058.0, + 1433.0, + 1044.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1416.0, + 1081.0, + 1416.0, + 1081.0, + 1433.0, + 1066.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1419.0, + 1102.0, + 1419.0, + 1102.0, + 1430.0, + 1090.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1417.0, + 1123.0, + 1417.0, + 1123.0, + 1429.0, + 1114.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1417.0, + 1228.0, + 1417.0, + 1228.0, + 1433.0, + 1213.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1416.0, + 1253.0, + 1416.0, + 1253.0, + 1433.0, + 1240.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1416.0, + 1276.0, + 1416.0, + 1276.0, + 1433.0, + 1261.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1419.0, + 1296.0, + 1419.0, + 1296.0, + 1430.0, + 1284.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1437.0, + 491.0, + 1437.0, + 491.0, + 1452.0, + 476.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1438.0, + 518.0, + 1438.0, + 518.0, + 1453.0, + 503.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1439.0, + 538.0, + 1439.0, + 538.0, + 1452.0, + 527.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1438.0, + 559.0, + 1438.0, + 559.0, + 1450.0, + 547.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1437.0, + 664.0, + 1437.0, + 664.0, + 1452.0, + 650.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1439.0, + 690.0, + 1439.0, + 690.0, + 1452.0, + 678.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1439.0, + 712.0, + 1439.0, + 712.0, + 1451.0, + 699.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1438.0, + 733.0, + 1438.0, + 733.0, + 1450.0, + 721.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1437.0, + 838.0, + 1437.0, + 838.0, + 1452.0, + 823.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1438.0, + 863.0, + 1438.0, + 863.0, + 1455.0, + 850.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1439.0, + 884.0, + 1439.0, + 884.0, + 1452.0, + 873.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1438.0, + 906.0, + 1438.0, + 906.0, + 1448.0, + 894.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1435.0, + 1038.0, + 1435.0, + 1038.0, + 1452.0, + 1023.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1437.0, + 1058.0, + 1437.0, + 1058.0, + 1455.0, + 1044.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1437.0, + 1081.0, + 1437.0, + 1081.0, + 1455.0, + 1066.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1438.0, + 1102.0, + 1438.0, + 1102.0, + 1450.0, + 1090.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1437.0, + 1228.0, + 1437.0, + 1228.0, + 1452.0, + 1213.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1439.0, + 1252.0, + 1439.0, + 1252.0, + 1452.0, + 1241.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1438.0, + 1276.0, + 1438.0, + 1276.0, + 1453.0, + 1261.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1438.0, + 1297.0, + 1438.0, + 1297.0, + 1450.0, + 1284.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1375.5, + 1322.0, + 1375.5, + 1322.0, + 1388.0, + 1302.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1397.0, + 1300.0, + 1397.0, + 1300.0, + 1410.0, + 1279.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 209.0, + 1520.0, + 209.0, + 1520.0, + 286.0, + 799.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 197.0, + 699.0, + 860.0, + 699.0, + 860.0, + 734.0, + 197.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 462.0, + 684.0, + 462.0, + 684.0, + 501.0, + 196.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 221.0, + 315.0, + 221.0, + 315.0, + 258.0, + 216.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 225.0, + 484.0, + 225.0, + 484.0, + 256.0, + 390.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 224.0, + 675.0, + 224.0, + 675.0, + 259.0, + 516.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 249.0, + 305.0, + 249.0, + 305.0, + 284.0, + 233.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 250.0, + 477.0, + 250.0, + 477.0, + 285.0, + 405.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 249.0, + 675.0, + 249.0, + 675.0, + 285.0, + 517.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 302.0, + 362.0, + 302.0, + 362.0, + 315.0, + 349.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 302.0, + 384.0, + 302.0, + 384.0, + 314.0, + 370.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 301.0, + 407.0, + 301.0, + 407.0, + 316.0, + 388.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 301.0, + 428.0, + 301.0, + 428.0, + 317.0, + 412.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 321.0, + 364.0, + 321.0, + 364.0, + 338.0, + 347.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 324.0, + 384.0, + 324.0, + 384.0, + 337.0, + 370.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 324.0, + 405.0, + 324.0, + 405.0, + 336.0, + 389.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 323.0, + 427.0, + 323.0, + 427.0, + 336.0, + 413.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 325.0, + 445.0, + 325.0, + 445.0, + 334.0, + 438.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 340.0, + 257.0, + 340.0, + 257.0, + 359.0, + 234.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 342.0, + 275.0, + 342.0, + 275.0, + 358.0, + 259.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 339.0, + 366.0, + 339.0, + 366.0, + 359.0, + 345.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 342.0, + 385.0, + 342.0, + 385.0, + 359.0, + 368.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 341.0, + 408.0, + 341.0, + 408.0, + 360.0, + 389.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 341.0, + 428.0, + 341.0, + 428.0, + 358.0, + 411.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 344.0, + 447.0, + 344.0, + 447.0, + 357.0, + 434.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 338.0, + 652.0, + 338.0, + 652.0, + 360.0, + 541.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 214.0, + 357.0, + 258.0, + 357.0, + 258.0, + 381.0, + 214.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 361.0, + 364.0, + 361.0, + 364.0, + 379.0, + 347.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 360.0, + 385.0, + 360.0, + 385.0, + 378.0, + 368.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 359.0, + 428.0, + 359.0, + 428.0, + 379.0, + 388.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 358.0, + 651.0, + 358.0, + 651.0, + 381.0, + 541.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 217.0, + 382.0, + 235.0, + 382.0, + 235.0, + 401.0, + 217.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 381.0, + 255.0, + 381.0, + 255.0, + 397.0, + 236.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 379.0, + 366.0, + 379.0, + 366.0, + 399.0, + 345.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 382.0, + 385.0, + 382.0, + 385.0, + 399.0, + 368.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 382.0, + 408.0, + 382.0, + 408.0, + 401.0, + 389.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 381.0, + 428.0, + 381.0, + 428.0, + 397.0, + 410.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 379.0, + 651.0, + 379.0, + 651.0, + 401.0, + 541.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 399.0, + 652.0, + 399.0, + 652.0, + 423.0, + 541.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 420.0, + 652.0, + 420.0, + 652.0, + 443.0, + 539.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 212.25, + 343.0, + 238.25, + 343.0, + 238.25, + 356.0, + 212.25, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 188.0, + 1518.0, + 450.0, + 1518.0, + 450.0, + 1557.0, + 188.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 1120.0, + 457.0, + 1120.0, + 457.0, + 1159.0, + 193.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 306.0, + 1515.0, + 306.0, + 1515.0, + 361.0, + 803.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 744.0, + 460.0, + 744.0, + 460.0, + 782.0, + 196.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 1120.0, + 457.0, + 1120.0, + 457.0, + 1159.0, + 193.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 188.0, + 1518.0, + 450.0, + 1518.0, + 450.0, + 1557.0, + 188.0, + 1557.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 265, + 984, + 1324, + 984, + 1324, + 1696, + 265, + 1696 + ], + "score": 0.949 + }, + { + "category_id": 3, + "poly": [ + 292, + 311, + 1323, + 311, + 1323, + 857, + 292, + 857 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 198, + 1872, + 678, + 1872, + 678, + 1902, + 198, + 1902 + ], + "score": 0.753 + }, + { + "category_id": 1, + "poly": [ + 201, + 1808, + 684, + 1808, + 684, + 1837, + 201, + 1837 + ], + "score": 0.723 + }, + { + "category_id": 2, + "poly": [ + 222, + 180, + 705, + 180, + 705, + 207, + 222, + 207 + ], + "score": 0.706 + }, + { + "category_id": 1, + "poly": [ + 194, + 1937, + 720, + 1937, + 720, + 1967, + 194, + 1967 + ], + "score": 0.654 + }, + { + "category_id": 1, + "poly": [ + 220, + 235, + 480, + 235, + 480, + 262, + 220, + 262 + ], + "score": 0.539 + }, + { + "category_id": 4, + "poly": [ + 221, + 934, + 477, + 934, + 477, + 961, + 221, + 961 + ], + "score": 0.223 + }, + { + "category_id": 1, + "poly": [ + 222, + 180, + 705, + 180, + 705, + 207, + 222, + 207 + ], + "score": 0.121 + }, + { + "category_id": 13, + "poly": [ + 475, + 1172, + 607, + 1172, + 607, + 1298, + 475, + 1298 + ], + "score": 0.36, + "latex": "{ \\begin{array} { l } { \\mathbf { e } \\mathbf { \\Lambda } { \\textbf { \\textsf { d } } } \\mathbf { e } \\mathbf { \\Lambda } { \\mathbf { \\textsf { f } } } } \\\\ { \\mathbf { b } \\mathbf { \\Lambda } { \\textbf { \\textsf { a } } } \\mathbf { \\Lambda } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } \\mathbf { b } \\mathbf { \\Lambda } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } } \\\\ { \\mathbf { b } \\mathbf { \\Sigma } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } } \\\\ { \\mathbf { e } \\mathbf { \\Lambda } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } } \\\\ { \\mathbf { h } \\mathbf { \\Sigma } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } { \\mathbf { \\Sigma } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 300, + 1172, + 435, + 1172, + 435, + 1299, + 300, + 1299 + ], + "score": 0.3, + "latex": "{ \\begin{array} { r l } & { \\mathbf { e } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { e } \\quad \\mathbf { \\ \" { f } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\quad \\mathbf { \\ \" { a } } \\quad \\mathbf { b } \\quad \\mathbf { \\ \" { c } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\left[ \\ \\mathbf { \\ \" { a } } \\quad \\mathbf { b } \\quad \\mathbf { \\ \" { c } } \\right. } \\\\ & { \\mathbf { e } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { e } \\quad \\mathbf { \\ \" { f } } } \\\\ & { \\mathbf { h } \\quad \\mathbf { \\ \" { g } } \\left| \\ \\mathbf { \\ \" { p } } \\quad \\mathbf { h } \\quad \\mathbf { \\ \" { i } } \\right. } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 474, + 504, + 603, + 504, + 603, + 631, + 474, + 631 + ], + "score": 0.3, + "latex": "{ \\begin{array} { l } { \\begin{array} { r l l l l l } { \\mathbf { e } } & { \\mathbf { \\mathbb { d } } } & { \\mathbf { \\mathbb { d } } } & { \\mathbf { e } } & { \\mathbf { \\mathbb { f } } } \\\\ { \\mathbf { b } } & { \\mathbf { \\mathbb { a } } } & { \\mathbf { \\mathbb { a } } } & { \\mathbf { \\mathbb { b } } } & { \\mathbf { \\mathbb { c } } } \\\\ { \\mathbf { b } } & { \\mathbf { \\mathbb { a } } } & { \\mathbf { \\mathbb { a } } } & { \\mathbf { \\mathbb { b } } } & { \\mathbf { \\mathbb { c } } } \\\\ { \\mathbf { e } } & { \\mathbf { \\mathbb { d } } } & { \\mathbf { \\mathbb { e } } } & { \\mathbf { \\mathbb { f } } } \\\\ { \\mathbf { h } } & { \\mathbf { \\mathbb { 9 } } } & { \\mathbf { \\mathbb { h } } } & { \\mathbf { \\mathbb { i } } } \\end{array} } } \\end{array} " + }, + { + "category_id": 13, + "poly": [ + 300, + 506, + 432, + 506, + 432, + 633, + 300, + 633 + ], + "score": 0.28, + "latex": "{ \\begin{array} { r l } & { \\mathbf { e } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { \\ \" { d } } \\quad \\mathbf { e } \\quad \\mathbf { \\ \" { f } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\quad \\mathbf { \\ \" { a } } \\quad \\mathbf { b } \\quad \\mathbf { \\ \" { c } } } \\\\ & { \\mathbf { b } \\quad \\mathbf { \\ \" { a } } \\quad { \\left[ \\begin{array} { l l l l } { \\mathbf { a } } & { \\mathbf { b } } & { \\mathbf { c } } & { \\mathbf { \\ \" { d } } } \\\\ { \\mathbf { a } } & { \\mathbf { b } } & { \\mathbf { c } } & { \\mathbf { \\ \" { d } } } \\\\ { \\mathbf { e } } & { \\mathbf { \\ \" { d } } } & { \\mathbf { e } } & { \\mathbf { \\ \" { f } } } \\\\ { \\mathbf { h } } & { \\mathbf { \\ \" { g } } } & { \\mathbf { \\ \" { h } } } & { \\mathbf { \\ \" { i } } } \\end{array} \\right] } } \\end{array} }" + }, + { + "category_id": 15, + "poly": [ + 264.0, + 993.0, + 279.0, + 993.0, + 279.0, + 1008.0, + 264.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 996.0, + 316.0, + 996.0, + 316.0, + 1007.0, + 304.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 993.0, + 344.0, + 993.0, + 344.0, + 1009.0, + 330.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 993.0, + 367.0, + 993.0, + 367.0, + 1008.0, + 351.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 994.0, + 388.0, + 994.0, + 388.0, + 1009.0, + 372.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 994.0, + 408.0, + 994.0, + 408.0, + 1005.0, + 397.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 994.0, + 491.0, + 994.0, + 491.0, + 1009.0, + 477.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 993.0, + 518.0, + 993.0, + 518.0, + 1009.0, + 504.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 993.0, + 540.0, + 993.0, + 540.0, + 1008.0, + 524.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 994.0, + 561.0, + 994.0, + 561.0, + 1009.0, + 546.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 994.0, + 580.0, + 994.0, + 580.0, + 1005.0, + 570.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 990.0, + 758.0, + 990.0, + 758.0, + 1013.0, + 648.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 994.0, + 840.0, + 994.0, + 840.0, + 1009.0, + 823.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 992.0, + 927.0, + 992.0, + 927.0, + 1009.0, + 846.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 993.0, + 1011.0, + 993.0, + 1011.0, + 1009.0, + 996.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 993.0, + 1038.0, + 993.0, + 1038.0, + 1008.0, + 1023.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 993.0, + 1059.0, + 993.0, + 1059.0, + 1009.0, + 1044.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 994.0, + 1081.0, + 994.0, + 1081.0, + 1008.0, + 1065.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 993.0, + 1184.0, + 993.0, + 1184.0, + 1010.0, + 1169.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 993.0, + 1211.0, + 993.0, + 1211.0, + 1009.0, + 1197.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 993.0, + 1233.0, + 993.0, + 1233.0, + 1009.0, + 1217.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 994.0, + 1253.0, + 994.0, + 1253.0, + 1009.0, + 1239.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1014.0, + 316.0, + 1014.0, + 316.0, + 1029.0, + 303.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1015.0, + 344.0, + 1015.0, + 344.0, + 1029.0, + 330.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1015.0, + 366.0, + 1015.0, + 366.0, + 1029.0, + 351.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1014.0, + 388.0, + 1014.0, + 388.0, + 1029.0, + 373.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1015.0, + 411.0, + 1015.0, + 411.0, + 1029.0, + 395.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1015.0, + 488.0, + 1015.0, + 488.0, + 1028.0, + 478.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1016.0, + 516.0, + 1016.0, + 516.0, + 1028.0, + 505.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1015.0, + 540.0, + 1015.0, + 540.0, + 1029.0, + 525.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1014.0, + 560.0, + 1014.0, + 560.0, + 1029.0, + 547.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1016.0, + 582.0, + 1016.0, + 582.0, + 1027.0, + 570.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1014.0, + 664.0, + 1014.0, + 664.0, + 1028.0, + 650.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1015.0, + 692.0, + 1015.0, + 692.0, + 1029.0, + 676.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1015.0, + 712.0, + 1015.0, + 712.0, + 1029.0, + 697.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1015.0, + 734.0, + 1015.0, + 734.0, + 1028.0, + 718.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1016.0, + 753.0, + 1016.0, + 753.0, + 1027.0, + 743.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1014.0, + 839.0, + 1014.0, + 839.0, + 1029.0, + 823.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1015.0, + 867.0, + 1015.0, + 867.0, + 1029.0, + 849.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1015.0, + 887.0, + 1015.0, + 887.0, + 1029.0, + 869.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1015.0, + 908.0, + 1015.0, + 908.0, + 1028.0, + 892.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1016.0, + 927.0, + 1016.0, + 927.0, + 1027.0, + 915.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1014.0, + 1011.0, + 1014.0, + 1011.0, + 1029.0, + 996.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1015.0, + 1037.0, + 1015.0, + 1037.0, + 1029.0, + 1023.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1016.0, + 1058.0, + 1016.0, + 1058.0, + 1027.0, + 1046.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1016.0, + 1079.0, + 1016.0, + 1079.0, + 1027.0, + 1067.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1016.0, + 1101.0, + 1016.0, + 1101.0, + 1027.0, + 1089.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1013.0, + 1184.0, + 1013.0, + 1184.0, + 1029.0, + 1169.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1015.0, + 1211.0, + 1015.0, + 1211.0, + 1029.0, + 1197.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1015.0, + 1233.0, + 1015.0, + 1233.0, + 1029.0, + 1218.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1015.0, + 1251.0, + 1015.0, + 1251.0, + 1027.0, + 1241.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1033.0, + 316.0, + 1033.0, + 316.0, + 1049.0, + 303.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1034.0, + 344.0, + 1034.0, + 344.0, + 1050.0, + 330.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1034.0, + 367.0, + 1034.0, + 367.0, + 1049.0, + 351.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1033.0, + 388.0, + 1033.0, + 388.0, + 1049.0, + 373.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1034.0, + 409.0, + 1034.0, + 409.0, + 1049.0, + 395.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1036.0, + 489.0, + 1036.0, + 489.0, + 1048.0, + 478.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1037.0, + 516.0, + 1037.0, + 516.0, + 1049.0, + 505.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1034.0, + 540.0, + 1034.0, + 540.0, + 1050.0, + 525.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1033.0, + 561.0, + 1033.0, + 561.0, + 1049.0, + 546.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1034.0, + 583.0, + 1034.0, + 583.0, + 1049.0, + 569.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1036.0, + 662.0, + 1036.0, + 662.0, + 1048.0, + 652.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1034.0, + 691.0, + 1034.0, + 691.0, + 1050.0, + 677.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1034.0, + 714.0, + 1034.0, + 714.0, + 1050.0, + 697.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1034.0, + 734.0, + 1034.0, + 734.0, + 1049.0, + 720.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1037.0, + 755.0, + 1037.0, + 755.0, + 1048.0, + 743.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1034.0, + 839.0, + 1034.0, + 839.0, + 1049.0, + 822.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1034.0, + 865.0, + 1034.0, + 865.0, + 1050.0, + 849.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1034.0, + 887.0, + 1034.0, + 887.0, + 1049.0, + 871.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1034.0, + 908.0, + 1034.0, + 908.0, + 1049.0, + 892.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1037.0, + 927.0, + 1037.0, + 927.0, + 1048.0, + 915.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1034.0, + 1011.0, + 1034.0, + 1011.0, + 1049.0, + 996.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1037.0, + 1036.0, + 1037.0, + 1036.0, + 1048.0, + 1024.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1034.0, + 1061.0, + 1034.0, + 1061.0, + 1050.0, + 1044.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1033.0, + 1081.0, + 1033.0, + 1081.0, + 1049.0, + 1066.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1037.0, + 1101.0, + 1037.0, + 1101.0, + 1048.0, + 1089.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1033.0, + 1184.0, + 1033.0, + 1184.0, + 1050.0, + 1170.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1037.0, + 1210.0, + 1037.0, + 1210.0, + 1048.0, + 1198.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1034.0, + 1233.0, + 1034.0, + 1233.0, + 1050.0, + 1218.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1034.0, + 1254.0, + 1034.0, + 1254.0, + 1049.0, + 1239.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1037.0, + 1274.0, + 1037.0, + 1274.0, + 1048.0, + 1263.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1056.0, + 315.0, + 1056.0, + 315.0, + 1067.0, + 304.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1055.0, + 342.0, + 1055.0, + 342.0, + 1067.0, + 331.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1054.0, + 366.0, + 1054.0, + 366.0, + 1068.0, + 351.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1055.0, + 388.0, + 1055.0, + 388.0, + 1069.0, + 373.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1057.0, + 495.0, + 1057.0, + 495.0, + 1067.0, + 483.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1055.0, + 516.0, + 1055.0, + 516.0, + 1067.0, + 505.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1054.0, + 540.0, + 1054.0, + 540.0, + 1069.0, + 525.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1055.0, + 560.0, + 1055.0, + 560.0, + 1071.0, + 547.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1057.0, + 668.0, + 1057.0, + 668.0, + 1068.0, + 657.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1054.0, + 691.0, + 1054.0, + 691.0, + 1069.0, + 677.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1054.0, + 714.0, + 1054.0, + 714.0, + 1069.0, + 698.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1055.0, + 734.0, + 1055.0, + 734.0, + 1069.0, + 718.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1057.0, + 842.0, + 1057.0, + 842.0, + 1067.0, + 831.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1055.0, + 862.0, + 1055.0, + 862.0, + 1067.0, + 851.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1054.0, + 886.0, + 1054.0, + 886.0, + 1068.0, + 871.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1056.0, + 906.0, + 1056.0, + 906.0, + 1067.0, + 893.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1057.0, + 1009.0, + 1057.0, + 1009.0, + 1068.0, + 997.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1056.0, + 1036.0, + 1056.0, + 1036.0, + 1067.0, + 1024.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1054.0, + 1060.0, + 1054.0, + 1060.0, + 1068.0, + 1044.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1056.0, + 1079.0, + 1056.0, + 1079.0, + 1067.0, + 1067.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1057.0, + 1183.0, + 1057.0, + 1183.0, + 1068.0, + 1171.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1056.0, + 1209.0, + 1056.0, + 1209.0, + 1067.0, + 1198.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1054.0, + 1233.0, + 1054.0, + 1233.0, + 1069.0, + 1218.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1055.0, + 1254.0, + 1055.0, + 1254.0, + 1069.0, + 1239.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1077.0, + 314.0, + 1077.0, + 314.0, + 1086.0, + 304.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1079.0, + 342.0, + 1079.0, + 342.0, + 1089.0, + 332.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1077.0, + 385.0, + 1077.0, + 385.0, + 1087.0, + 376.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1078.0, + 489.0, + 1078.0, + 489.0, + 1087.0, + 478.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1079.0, + 516.0, + 1079.0, + 516.0, + 1091.0, + 506.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1077.0, + 540.0, + 1077.0, + 540.0, + 1092.0, + 525.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1075.0, + 561.0, + 1075.0, + 561.0, + 1090.0, + 547.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1078.0, + 662.0, + 1078.0, + 662.0, + 1087.0, + 652.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1077.0, + 689.0, + 1077.0, + 689.0, + 1092.0, + 676.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1077.0, + 714.0, + 1077.0, + 714.0, + 1092.0, + 698.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1075.0, + 735.0, + 1075.0, + 735.0, + 1090.0, + 718.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1078.0, + 836.0, + 1078.0, + 836.0, + 1087.0, + 826.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1079.0, + 884.0, + 1079.0, + 884.0, + 1091.0, + 873.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1078.0, + 906.0, + 1078.0, + 906.0, + 1087.0, + 895.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1075.0, + 1011.0, + 1075.0, + 1011.0, + 1090.0, + 996.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 1079.0, + 1058.0, + 1079.0, + 1058.0, + 1089.0, + 1048.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1075.0, + 1184.0, + 1075.0, + 1184.0, + 1090.0, + 1170.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1079.0, + 1209.0, + 1079.0, + 1209.0, + 1090.0, + 1198.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1077.0, + 1233.0, + 1077.0, + 1233.0, + 1092.0, + 1218.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1077.0, + 1252.0, + 1077.0, + 1252.0, + 1089.0, + 1241.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1178.0, + 665.0, + 1178.0, + 665.0, + 1193.0, + 648.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1178.0, + 692.0, + 1178.0, + 692.0, + 1193.0, + 676.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1177.0, + 738.0, + 1177.0, + 738.0, + 1194.0, + 697.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1179.0, + 753.0, + 1179.0, + 753.0, + 1188.0, + 744.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1178.0, + 840.0, + 1178.0, + 840.0, + 1193.0, + 822.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1177.0, + 908.0, + 1177.0, + 908.0, + 1193.0, + 862.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1179.0, + 925.0, + 1179.0, + 925.0, + 1188.0, + 915.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1176.0, + 1104.0, + 1176.0, + 1104.0, + 1195.0, + 995.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1176.0, + 1258.0, + 1176.0, + 1258.0, + 1195.0, + 1168.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1179.0, + 1273.0, + 1179.0, + 1273.0, + 1189.0, + 1263.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1197.0, + 665.0, + 1197.0, + 665.0, + 1213.0, + 648.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1199.0, + 693.0, + 1199.0, + 693.0, + 1212.0, + 676.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1200.0, + 711.0, + 1200.0, + 711.0, + 1211.0, + 700.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1200.0, + 734.0, + 1200.0, + 734.0, + 1209.0, + 720.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1200.0, + 753.0, + 1200.0, + 753.0, + 1209.0, + 744.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1197.0, + 839.0, + 1197.0, + 839.0, + 1213.0, + 822.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1199.0, + 866.0, + 1199.0, + 866.0, + 1213.0, + 848.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1199.0, + 907.0, + 1199.0, + 907.0, + 1209.0, + 895.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1200.0, + 927.0, + 1200.0, + 927.0, + 1209.0, + 915.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1197.0, + 1013.0, + 1197.0, + 1013.0, + 1213.0, + 996.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1199.0, + 1040.0, + 1199.0, + 1040.0, + 1212.0, + 1020.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1199.0, + 1063.0, + 1199.0, + 1063.0, + 1212.0, + 1043.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1200.0, + 1081.0, + 1200.0, + 1081.0, + 1209.0, + 1065.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1200.0, + 1101.0, + 1200.0, + 1101.0, + 1209.0, + 1088.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1197.0, + 1186.0, + 1197.0, + 1186.0, + 1213.0, + 1169.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1199.0, + 1212.0, + 1199.0, + 1212.0, + 1213.0, + 1194.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1199.0, + 1235.0, + 1199.0, + 1235.0, + 1212.0, + 1217.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1197.0, + 1256.0, + 1197.0, + 1256.0, + 1211.0, + 1238.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1200.0, + 1274.0, + 1200.0, + 1274.0, + 1211.0, + 1263.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1218.0, + 664.0, + 1218.0, + 664.0, + 1232.0, + 650.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1220.0, + 689.0, + 1220.0, + 689.0, + 1231.0, + 677.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1218.0, + 715.0, + 1218.0, + 715.0, + 1232.0, + 697.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1218.0, + 735.0, + 1218.0, + 735.0, + 1231.0, + 718.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1218.0, + 756.0, + 1218.0, + 756.0, + 1232.0, + 740.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1218.0, + 840.0, + 1218.0, + 840.0, + 1232.0, + 822.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1219.0, + 865.0, + 1219.0, + 865.0, + 1232.0, + 848.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1218.0, + 889.0, + 1218.0, + 889.0, + 1232.0, + 871.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1218.0, + 909.0, + 1218.0, + 909.0, + 1231.0, + 890.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1218.0, + 930.0, + 1218.0, + 930.0, + 1231.0, + 914.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1218.0, + 1013.0, + 1218.0, + 1013.0, + 1232.0, + 996.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1218.0, + 1038.0, + 1218.0, + 1038.0, + 1232.0, + 1021.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1218.0, + 1063.0, + 1218.0, + 1063.0, + 1232.0, + 1043.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1218.0, + 1082.0, + 1218.0, + 1082.0, + 1231.0, + 1064.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1218.0, + 1104.0, + 1218.0, + 1104.0, + 1232.0, + 1088.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1218.0, + 1186.0, + 1218.0, + 1186.0, + 1232.0, + 1169.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1218.0, + 1211.0, + 1218.0, + 1211.0, + 1232.0, + 1195.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1218.0, + 1235.0, + 1218.0, + 1235.0, + 1232.0, + 1215.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1218.0, + 1276.0, + 1218.0, + 1276.0, + 1232.0, + 1262.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1241.0, + 689.0, + 1241.0, + 689.0, + 1251.0, + 679.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1238.0, + 714.0, + 1238.0, + 714.0, + 1252.0, + 698.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1240.0, + 733.0, + 1240.0, + 733.0, + 1251.0, + 721.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1240.0, + 840.0, + 1240.0, + 840.0, + 1253.0, + 822.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1238.0, + 863.0, + 1238.0, + 863.0, + 1252.0, + 848.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1238.0, + 887.0, + 1238.0, + 887.0, + 1252.0, + 872.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1241.0, + 907.0, + 1241.0, + 907.0, + 1251.0, + 893.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1240.0, + 1013.0, + 1240.0, + 1013.0, + 1253.0, + 996.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1238.0, + 1060.0, + 1238.0, + 1060.0, + 1252.0, + 1044.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1242.0, + 1079.0, + 1242.0, + 1079.0, + 1251.0, + 1067.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1240.0, + 1184.0, + 1240.0, + 1184.0, + 1254.0, + 1169.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1238.0, + 1211.0, + 1238.0, + 1211.0, + 1253.0, + 1195.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1237.0, + 1233.0, + 1237.0, + 1233.0, + 1253.0, + 1217.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1240.0, + 1254.0, + 1240.0, + 1254.0, + 1253.0, + 1239.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1259.0, + 665.0, + 1259.0, + 665.0, + 1272.0, + 650.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1260.0, + 691.0, + 1260.0, + 691.0, + 1276.0, + 676.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1263.0, + 710.0, + 1263.0, + 710.0, + 1272.0, + 698.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1261.0, + 733.0, + 1261.0, + 733.0, + 1270.0, + 722.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1258.0, + 839.0, + 1258.0, + 839.0, + 1272.0, + 822.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1260.0, + 865.0, + 1260.0, + 865.0, + 1276.0, + 850.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1261.0, + 887.0, + 1261.0, + 887.0, + 1275.0, + 871.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1261.0, + 907.0, + 1261.0, + 907.0, + 1270.0, + 895.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1261.0, + 1037.0, + 1261.0, + 1037.0, + 1276.0, + 1021.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1263.0, + 1058.0, + 1263.0, + 1058.0, + 1274.0, + 1046.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1260.0, + 1183.0, + 1260.0, + 1183.0, + 1271.0, + 1171.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1260.0, + 1211.0, + 1260.0, + 1211.0, + 1276.0, + 1195.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1260.0, + 1233.0, + 1260.0, + 1233.0, + 1275.0, + 1218.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1259.0, + 1254.0, + 1259.0, + 1254.0, + 1272.0, + 1239.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1362.0, + 318.0, + 1362.0, + 318.0, + 1376.0, + 303.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1361.0, + 345.0, + 1361.0, + 345.0, + 1376.0, + 330.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1361.0, + 366.0, + 1361.0, + 366.0, + 1375.0, + 351.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1363.0, + 386.0, + 1363.0, + 386.0, + 1374.0, + 376.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1362.0, + 516.0, + 1362.0, + 516.0, + 1376.0, + 499.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1359.0, + 586.0, + 1359.0, + 586.0, + 1377.0, + 522.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1361.0, + 601.0, + 1361.0, + 601.0, + 1371.0, + 592.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1357.0, + 800.0, + 1357.0, + 800.0, + 1377.0, + 688.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1359.0, + 996.0, + 1359.0, + 996.0, + 1379.0, + 887.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1381.0, + 318.0, + 1381.0, + 318.0, + 1397.0, + 303.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1382.0, + 345.0, + 1382.0, + 345.0, + 1397.0, + 330.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1381.0, + 367.0, + 1381.0, + 367.0, + 1396.0, + 350.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1381.0, + 388.0, + 1381.0, + 388.0, + 1396.0, + 372.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1383.0, + 407.0, + 1383.0, + 407.0, + 1394.0, + 396.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1381.0, + 514.0, + 1381.0, + 514.0, + 1397.0, + 498.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1382.0, + 541.0, + 1382.0, + 541.0, + 1397.0, + 523.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1381.0, + 561.0, + 1381.0, + 561.0, + 1396.0, + 545.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1381.0, + 584.0, + 1381.0, + 584.0, + 1396.0, + 565.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1383.0, + 603.0, + 1383.0, + 603.0, + 1394.0, + 590.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1381.0, + 710.0, + 1381.0, + 710.0, + 1397.0, + 693.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1382.0, + 758.0, + 1382.0, + 758.0, + 1397.0, + 717.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1381.0, + 779.0, + 1381.0, + 779.0, + 1394.0, + 763.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1383.0, + 797.0, + 1383.0, + 797.0, + 1393.0, + 787.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1382.0, + 906.0, + 1382.0, + 906.0, + 1397.0, + 887.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1380.0, + 996.0, + 1380.0, + 996.0, + 1399.0, + 910.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1402.0, + 318.0, + 1402.0, + 318.0, + 1416.0, + 303.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1402.0, + 344.0, + 1402.0, + 344.0, + 1417.0, + 330.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1402.0, + 367.0, + 1402.0, + 367.0, + 1417.0, + 350.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1400.0, + 388.0, + 1400.0, + 388.0, + 1416.0, + 373.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1402.0, + 409.0, + 1402.0, + 409.0, + 1416.0, + 394.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1402.0, + 513.0, + 1402.0, + 513.0, + 1416.0, + 499.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1403.0, + 540.0, + 1403.0, + 540.0, + 1416.0, + 523.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1402.0, + 561.0, + 1402.0, + 561.0, + 1416.0, + 546.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1402.0, + 583.0, + 1402.0, + 583.0, + 1416.0, + 568.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1403.0, + 603.0, + 1403.0, + 603.0, + 1415.0, + 592.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1402.0, + 710.0, + 1402.0, + 710.0, + 1416.0, + 693.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1402.0, + 734.0, + 1402.0, + 734.0, + 1416.0, + 717.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1402.0, + 757.0, + 1402.0, + 757.0, + 1417.0, + 740.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1399.0, + 799.0, + 1399.0, + 799.0, + 1417.0, + 759.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1402.0, + 904.0, + 1402.0, + 904.0, + 1416.0, + 887.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1403.0, + 930.0, + 1403.0, + 930.0, + 1416.0, + 914.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1402.0, + 954.0, + 1402.0, + 954.0, + 1416.0, + 935.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1402.0, + 977.0, + 1402.0, + 977.0, + 1416.0, + 956.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1403.0, + 992.0, + 1403.0, + 992.0, + 1414.0, + 980.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1422.0, + 318.0, + 1422.0, + 318.0, + 1437.0, + 303.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1421.0, + 343.0, + 1421.0, + 343.0, + 1437.0, + 330.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1421.0, + 367.0, + 1421.0, + 367.0, + 1437.0, + 351.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1425.0, + 386.0, + 1425.0, + 386.0, + 1434.0, + 376.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1422.0, + 513.0, + 1422.0, + 513.0, + 1437.0, + 499.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1422.0, + 540.0, + 1422.0, + 540.0, + 1435.0, + 524.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1421.0, + 561.0, + 1421.0, + 561.0, + 1435.0, + 546.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1423.0, + 582.0, + 1423.0, + 582.0, + 1434.0, + 570.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1423.0, + 601.0, + 1423.0, + 601.0, + 1433.0, + 592.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1422.0, + 709.0, + 1422.0, + 709.0, + 1437.0, + 693.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1421.0, + 734.0, + 1421.0, + 734.0, + 1435.0, + 720.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1421.0, + 757.0, + 1421.0, + 757.0, + 1437.0, + 741.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1425.0, + 776.0, + 1425.0, + 776.0, + 1434.0, + 764.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1425.0, + 902.0, + 1425.0, + 902.0, + 1434.0, + 889.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1421.0, + 929.0, + 1421.0, + 929.0, + 1435.0, + 914.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1421.0, + 950.0, + 1421.0, + 950.0, + 1435.0, + 936.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1422.0, + 974.0, + 1422.0, + 974.0, + 1435.0, + 957.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1443.0, + 318.0, + 1443.0, + 318.0, + 1457.0, + 303.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1444.0, + 343.0, + 1444.0, + 343.0, + 1460.0, + 330.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1444.0, + 366.0, + 1444.0, + 366.0, + 1460.0, + 351.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1444.0, + 385.0, + 1444.0, + 385.0, + 1454.0, + 374.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1443.0, + 513.0, + 1443.0, + 513.0, + 1457.0, + 498.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1444.0, + 539.0, + 1444.0, + 539.0, + 1460.0, + 525.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1444.0, + 561.0, + 1444.0, + 561.0, + 1460.0, + 547.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1443.0, + 583.0, + 1443.0, + 583.0, + 1457.0, + 568.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1443.0, + 709.0, + 1443.0, + 709.0, + 1457.0, + 693.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1444.0, + 734.0, + 1444.0, + 734.0, + 1460.0, + 720.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1444.0, + 756.0, + 1444.0, + 756.0, + 1460.0, + 743.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1444.0, + 776.0, + 1444.0, + 776.0, + 1455.0, + 766.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1443.0, + 903.0, + 1443.0, + 903.0, + 1457.0, + 887.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1444.0, + 929.0, + 1444.0, + 929.0, + 1460.0, + 914.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1444.0, + 951.0, + 1444.0, + 951.0, + 1460.0, + 936.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1445.0, + 971.0, + 1445.0, + 971.0, + 1455.0, + 960.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1545.0, + 339.0, + 1545.0, + 339.0, + 1560.0, + 325.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1544.0, + 367.0, + 1544.0, + 367.0, + 1560.0, + 351.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1544.0, + 388.0, + 1544.0, + 388.0, + 1560.0, + 372.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1545.0, + 409.0, + 1545.0, + 409.0, + 1560.0, + 395.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1545.0, + 557.0, + 1545.0, + 557.0, + 1560.0, + 542.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1543.0, + 586.0, + 1543.0, + 586.0, + 1561.0, + 565.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1544.0, + 605.0, + 1544.0, + 605.0, + 1560.0, + 588.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1545.0, + 627.0, + 1545.0, + 627.0, + 1560.0, + 609.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1545.0, + 645.0, + 1545.0, + 645.0, + 1555.0, + 635.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1545.0, + 774.0, + 1545.0, + 774.0, + 1560.0, + 758.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1545.0, + 802.0, + 1545.0, + 802.0, + 1560.0, + 784.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1544.0, + 863.0, + 1544.0, + 863.0, + 1561.0, + 819.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1545.0, + 991.0, + 1545.0, + 991.0, + 1560.0, + 974.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1545.0, + 1019.0, + 1545.0, + 1019.0, + 1560.0, + 1001.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1544.0, + 1079.0, + 1544.0, + 1079.0, + 1561.0, + 1020.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1564.0, + 339.0, + 1564.0, + 339.0, + 1580.0, + 325.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1565.0, + 367.0, + 1565.0, + 367.0, + 1580.0, + 351.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1565.0, + 388.0, + 1565.0, + 388.0, + 1579.0, + 372.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1564.0, + 409.0, + 1564.0, + 409.0, + 1579.0, + 395.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1566.0, + 430.0, + 1566.0, + 430.0, + 1577.0, + 418.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1564.0, + 557.0, + 1564.0, + 557.0, + 1580.0, + 541.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1564.0, + 586.0, + 1564.0, + 586.0, + 1582.0, + 565.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1565.0, + 605.0, + 1565.0, + 605.0, + 1579.0, + 588.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1564.0, + 627.0, + 1564.0, + 627.0, + 1579.0, + 611.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 1564.0, + 647.0, + 1564.0, + 647.0, + 1579.0, + 632.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1564.0, + 774.0, + 1564.0, + 774.0, + 1580.0, + 758.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1565.0, + 801.0, + 1565.0, + 801.0, + 1580.0, + 784.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 1565.0, + 821.0, + 1565.0, + 821.0, + 1579.0, + 805.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1564.0, + 843.0, + 1564.0, + 843.0, + 1578.0, + 827.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1566.0, + 862.0, + 1566.0, + 862.0, + 1577.0, + 850.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1564.0, + 990.0, + 1564.0, + 990.0, + 1580.0, + 974.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1565.0, + 1017.0, + 1565.0, + 1017.0, + 1579.0, + 1001.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1565.0, + 1038.0, + 1565.0, + 1038.0, + 1579.0, + 1020.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1564.0, + 1060.0, + 1564.0, + 1060.0, + 1579.0, + 1043.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1565.0, + 1081.0, + 1565.0, + 1081.0, + 1578.0, + 1065.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1584.0, + 339.0, + 1584.0, + 339.0, + 1600.0, + 325.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1585.0, + 366.0, + 1585.0, + 366.0, + 1600.0, + 351.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1585.0, + 388.0, + 1585.0, + 388.0, + 1600.0, + 373.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1584.0, + 409.0, + 1584.0, + 409.0, + 1600.0, + 395.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1585.0, + 431.0, + 1585.0, + 431.0, + 1600.0, + 417.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1584.0, + 557.0, + 1584.0, + 557.0, + 1600.0, + 541.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1585.0, + 583.0, + 1585.0, + 583.0, + 1600.0, + 566.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1585.0, + 605.0, + 1585.0, + 605.0, + 1601.0, + 589.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1584.0, + 627.0, + 1584.0, + 627.0, + 1600.0, + 611.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1585.0, + 647.0, + 1585.0, + 647.0, + 1600.0, + 633.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1584.0, + 773.0, + 1584.0, + 773.0, + 1600.0, + 758.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1585.0, + 799.0, + 1585.0, + 799.0, + 1600.0, + 784.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 1585.0, + 821.0, + 1585.0, + 821.0, + 1600.0, + 805.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1584.0, + 843.0, + 1584.0, + 843.0, + 1600.0, + 827.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1585.0, + 863.0, + 1585.0, + 863.0, + 1600.0, + 849.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1584.0, + 990.0, + 1584.0, + 990.0, + 1600.0, + 974.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1585.0, + 1015.0, + 1585.0, + 1015.0, + 1600.0, + 1001.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1585.0, + 1038.0, + 1585.0, + 1038.0, + 1600.0, + 1023.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1584.0, + 1060.0, + 1584.0, + 1060.0, + 1600.0, + 1043.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1585.0, + 1081.0, + 1585.0, + 1081.0, + 1600.0, + 1065.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1607.0, + 339.0, + 1607.0, + 339.0, + 1622.0, + 325.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1605.0, + 366.0, + 1605.0, + 366.0, + 1622.0, + 351.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1605.0, + 388.0, + 1605.0, + 388.0, + 1622.0, + 373.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1606.0, + 409.0, + 1606.0, + 409.0, + 1620.0, + 395.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1606.0, + 557.0, + 1606.0, + 557.0, + 1622.0, + 542.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1605.0, + 583.0, + 1605.0, + 583.0, + 1622.0, + 568.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1605.0, + 605.0, + 1605.0, + 605.0, + 1622.0, + 589.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1606.0, + 627.0, + 1606.0, + 627.0, + 1620.0, + 611.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1606.0, + 773.0, + 1606.0, + 773.0, + 1622.0, + 757.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1606.0, + 799.0, + 1606.0, + 799.0, + 1620.0, + 784.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1606.0, + 821.0, + 1606.0, + 821.0, + 1620.0, + 807.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1608.0, + 840.0, + 1608.0, + 840.0, + 1619.0, + 828.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1606.0, + 990.0, + 1606.0, + 990.0, + 1622.0, + 974.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1606.0, + 1015.0, + 1606.0, + 1015.0, + 1622.0, + 1001.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1605.0, + 1037.0, + 1605.0, + 1037.0, + 1622.0, + 1023.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1606.0, + 1059.0, + 1606.0, + 1059.0, + 1622.0, + 1044.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1607.0, + 1077.0, + 1607.0, + 1077.0, + 1616.0, + 1067.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1625.0, + 339.0, + 1625.0, + 339.0, + 1641.0, + 325.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1628.0, + 366.0, + 1628.0, + 366.0, + 1643.0, + 351.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1626.0, + 388.0, + 1626.0, + 388.0, + 1643.0, + 373.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1625.0, + 409.0, + 1625.0, + 409.0, + 1640.0, + 395.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1625.0, + 557.0, + 1625.0, + 557.0, + 1641.0, + 542.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1628.0, + 582.0, + 1628.0, + 582.0, + 1643.0, + 569.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1626.0, + 604.0, + 1626.0, + 604.0, + 1643.0, + 589.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 1625.0, + 627.0, + 1625.0, + 627.0, + 1640.0, + 611.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1625.0, + 773.0, + 1625.0, + 773.0, + 1641.0, + 758.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1628.0, + 798.0, + 1628.0, + 798.0, + 1643.0, + 784.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1626.0, + 821.0, + 1626.0, + 821.0, + 1643.0, + 807.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1628.0, + 840.0, + 1628.0, + 840.0, + 1638.0, + 828.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1625.0, + 990.0, + 1625.0, + 990.0, + 1640.0, + 974.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1626.0, + 1015.0, + 1626.0, + 1015.0, + 1643.0, + 1001.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1626.0, + 1038.0, + 1626.0, + 1038.0, + 1643.0, + 1023.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1625.0, + 1059.0, + 1625.0, + 1059.0, + 1640.0, + 1044.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1179.0, + 869.0, + 1179.0, + 869.0, + 1190.0, + 846.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.25, + 1197.5, + 890.25, + 1197.5, + 890.25, + 1211.0, + 869.25, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1216.5, + 1253.0, + 1216.5, + 1253.0, + 1231.5, + 1238.0, + 1231.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.75, + 1239.5, + 1039.75, + 1239.5, + 1039.75, + 1252.0, + 1017.75, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1259.0, + 1007.0, + 1259.0, + 1007.0, + 1272.0, + 1001.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.75, + 1545.0, + 824.75, + 1545.0, + 824.75, + 1559.0, + 800.75, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 332.0, + 314.0, + 332.0, + 314.0, + 338.0, + 308.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 328.0, + 344.0, + 328.0, + 344.0, + 343.0, + 330.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 328.0, + 366.0, + 328.0, + 366.0, + 342.0, + 352.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 330.0, + 386.0, + 330.0, + 386.0, + 342.0, + 375.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 330.0, + 407.0, + 330.0, + 407.0, + 339.0, + 398.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 330.0, + 489.0, + 330.0, + 489.0, + 341.0, + 479.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 328.0, + 518.0, + 328.0, + 518.0, + 342.0, + 504.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 328.0, + 540.0, + 328.0, + 540.0, + 342.0, + 525.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 330.0, + 559.0, + 330.0, + 559.0, + 341.0, + 548.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 329.0, + 579.0, + 329.0, + 579.0, + 339.0, + 570.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 329.0, + 665.0, + 329.0, + 665.0, + 343.0, + 650.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 328.0, + 692.0, + 328.0, + 692.0, + 342.0, + 677.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 328.0, + 713.0, + 328.0, + 713.0, + 342.0, + 697.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 329.0, + 734.0, + 329.0, + 734.0, + 342.0, + 718.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 329.0, + 753.0, + 329.0, + 753.0, + 341.0, + 743.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 329.0, + 838.0, + 329.0, + 838.0, + 343.0, + 824.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 326.0, + 910.0, + 326.0, + 910.0, + 344.0, + 848.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 328.0, + 928.0, + 328.0, + 928.0, + 342.0, + 914.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 330.0, + 1009.0, + 330.0, + 1009.0, + 342.0, + 998.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 328.0, + 1037.0, + 328.0, + 1037.0, + 342.0, + 1023.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 328.0, + 1058.0, + 328.0, + 1058.0, + 342.0, + 1044.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 330.0, + 1078.0, + 330.0, + 1078.0, + 342.0, + 1068.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 330.0, + 1182.0, + 330.0, + 1182.0, + 342.0, + 1172.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 328.0, + 1212.0, + 328.0, + 1212.0, + 342.0, + 1196.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 328.0, + 1233.0, + 328.0, + 1233.0, + 342.0, + 1217.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 330.0, + 1252.0, + 330.0, + 1252.0, + 341.0, + 1241.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 349.0, + 315.0, + 349.0, + 315.0, + 362.0, + 306.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 348.0, + 344.0, + 348.0, + 344.0, + 362.0, + 330.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 350.0, + 365.0, + 350.0, + 365.0, + 361.0, + 353.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 348.0, + 387.0, + 348.0, + 387.0, + 362.0, + 373.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 350.0, + 408.0, + 350.0, + 408.0, + 360.0, + 396.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 349.0, + 488.0, + 349.0, + 488.0, + 361.0, + 478.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 349.0, + 518.0, + 349.0, + 518.0, + 363.0, + 504.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 349.0, + 538.0, + 349.0, + 538.0, + 361.0, + 527.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 349.0, + 559.0, + 349.0, + 559.0, + 361.0, + 548.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 349.0, + 582.0, + 349.0, + 582.0, + 360.0, + 571.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 348.0, + 664.0, + 348.0, + 664.0, + 363.0, + 650.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 349.0, + 692.0, + 349.0, + 692.0, + 363.0, + 677.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 348.0, + 712.0, + 348.0, + 712.0, + 362.0, + 697.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 348.0, + 734.0, + 348.0, + 734.0, + 362.0, + 719.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 348.0, + 754.0, + 348.0, + 754.0, + 362.0, + 740.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 348.0, + 838.0, + 348.0, + 838.0, + 363.0, + 823.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 348.0, + 865.0, + 348.0, + 865.0, + 363.0, + 849.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 348.0, + 888.0, + 348.0, + 888.0, + 362.0, + 870.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 348.0, + 908.0, + 348.0, + 908.0, + 362.0, + 891.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 348.0, + 929.0, + 348.0, + 929.0, + 362.0, + 914.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 348.0, + 1010.0, + 348.0, + 1010.0, + 363.0, + 997.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 348.0, + 1037.0, + 348.0, + 1037.0, + 362.0, + 1023.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 349.0, + 1057.0, + 349.0, + 1057.0, + 360.0, + 1045.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 349.0, + 1078.0, + 349.0, + 1078.0, + 360.0, + 1068.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 348.0, + 1183.0, + 348.0, + 1183.0, + 363.0, + 1169.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 348.0, + 1210.0, + 348.0, + 1210.0, + 362.0, + 1196.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 349.0, + 1230.0, + 349.0, + 1230.0, + 361.0, + 1220.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 349.0, + 1252.0, + 349.0, + 1252.0, + 360.0, + 1241.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 351.0, + 1272.0, + 351.0, + 1272.0, + 357.0, + 1266.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 367.0, + 316.0, + 367.0, + 316.0, + 382.0, + 303.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 368.0, + 343.0, + 368.0, + 343.0, + 383.0, + 330.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 368.0, + 366.0, + 368.0, + 366.0, + 383.0, + 352.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 368.0, + 388.0, + 368.0, + 388.0, + 382.0, + 373.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 368.0, + 409.0, + 368.0, + 409.0, + 382.0, + 395.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 367.0, + 491.0, + 367.0, + 491.0, + 382.0, + 476.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 368.0, + 517.0, + 368.0, + 517.0, + 383.0, + 504.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 368.0, + 540.0, + 368.0, + 540.0, + 383.0, + 524.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 368.0, + 561.0, + 368.0, + 561.0, + 382.0, + 546.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 368.0, + 582.0, + 368.0, + 582.0, + 383.0, + 568.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 367.0, + 664.0, + 367.0, + 664.0, + 382.0, + 650.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 368.0, + 691.0, + 368.0, + 691.0, + 383.0, + 677.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 368.0, + 712.0, + 368.0, + 712.0, + 383.0, + 698.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 368.0, + 734.0, + 368.0, + 734.0, + 382.0, + 719.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 368.0, + 757.0, + 368.0, + 757.0, + 383.0, + 742.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 367.0, + 838.0, + 367.0, + 838.0, + 382.0, + 823.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 367.0, + 863.0, + 367.0, + 863.0, + 383.0, + 849.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 366.0, + 888.0, + 366.0, + 888.0, + 385.0, + 869.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 367.0, + 908.0, + 367.0, + 908.0, + 382.0, + 891.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 368.0, + 929.0, + 368.0, + 929.0, + 382.0, + 914.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 367.0, + 1010.0, + 367.0, + 1010.0, + 383.0, + 996.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 369.0, + 1035.0, + 369.0, + 1035.0, + 381.0, + 1024.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 368.0, + 1058.0, + 368.0, + 1058.0, + 383.0, + 1044.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 367.0, + 1082.0, + 367.0, + 1082.0, + 382.0, + 1066.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 368.0, + 1102.0, + 368.0, + 1102.0, + 383.0, + 1087.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 367.0, + 1183.0, + 367.0, + 1183.0, + 382.0, + 1169.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 369.0, + 1208.0, + 369.0, + 1208.0, + 381.0, + 1197.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 368.0, + 1233.0, + 368.0, + 1233.0, + 382.0, + 1219.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 368.0, + 1254.0, + 368.0, + 1254.0, + 382.0, + 1239.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 370.0, + 1274.0, + 370.0, + 1274.0, + 381.0, + 1262.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 392.0, + 315.0, + 392.0, + 315.0, + 401.0, + 306.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 389.0, + 342.0, + 389.0, + 342.0, + 401.0, + 332.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 388.0, + 366.0, + 388.0, + 366.0, + 402.0, + 352.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 392.0, + 386.0, + 392.0, + 386.0, + 401.0, + 375.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 392.0, + 407.0, + 392.0, + 407.0, + 398.0, + 401.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 392.0, + 494.0, + 392.0, + 494.0, + 401.0, + 485.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 388.0, + 517.0, + 388.0, + 517.0, + 404.0, + 504.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 388.0, + 540.0, + 388.0, + 540.0, + 404.0, + 525.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 390.0, + 560.0, + 390.0, + 560.0, + 401.0, + 548.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 389.0, + 582.0, + 389.0, + 582.0, + 401.0, + 572.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 392.0, + 667.0, + 392.0, + 667.0, + 401.0, + 658.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 388.0, + 691.0, + 388.0, + 691.0, + 402.0, + 677.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 388.0, + 712.0, + 388.0, + 712.0, + 404.0, + 698.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 389.0, + 734.0, + 389.0, + 734.0, + 402.0, + 719.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 389.0, + 756.0, + 389.0, + 756.0, + 400.0, + 745.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 388.0, + 843.0, + 388.0, + 843.0, + 402.0, + 828.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 388.0, + 864.0, + 388.0, + 864.0, + 404.0, + 850.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 387.0, + 886.0, + 387.0, + 886.0, + 404.0, + 871.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 388.0, + 908.0, + 388.0, + 908.0, + 402.0, + 892.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 388.0, + 928.0, + 388.0, + 928.0, + 401.0, + 918.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 389.0, + 1011.0, + 389.0, + 1011.0, + 405.0, + 997.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 388.0, + 1036.0, + 388.0, + 1036.0, + 402.0, + 1022.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 388.0, + 1058.0, + 388.0, + 1058.0, + 402.0, + 1044.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 390.0, + 1078.0, + 390.0, + 1078.0, + 401.0, + 1068.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 389.0, + 1102.0, + 389.0, + 1102.0, + 400.0, + 1093.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 392.0, + 1182.0, + 392.0, + 1182.0, + 402.0, + 1172.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 390.0, + 1208.0, + 390.0, + 1208.0, + 401.0, + 1197.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 388.0, + 1233.0, + 388.0, + 1233.0, + 402.0, + 1219.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 393.0, + 1249.0, + 393.0, + 1249.0, + 399.0, + 1243.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 411.0, + 315.0, + 411.0, + 315.0, + 421.0, + 304.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 412.0, + 342.0, + 412.0, + 342.0, + 423.0, + 332.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 412.0, + 365.0, + 412.0, + 365.0, + 423.0, + 353.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 408.0, + 387.0, + 408.0, + 387.0, + 423.0, + 373.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 411.0, + 489.0, + 411.0, + 489.0, + 421.0, + 479.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 409.0, + 517.0, + 409.0, + 517.0, + 426.0, + 504.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 411.0, + 540.0, + 411.0, + 540.0, + 426.0, + 525.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 408.0, + 561.0, + 408.0, + 561.0, + 423.0, + 547.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 411.0, + 663.0, + 411.0, + 663.0, + 421.0, + 652.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 409.0, + 691.0, + 409.0, + 691.0, + 426.0, + 677.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 409.0, + 712.0, + 409.0, + 712.0, + 426.0, + 698.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 408.0, + 734.0, + 408.0, + 734.0, + 424.0, + 719.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 407.0, + 838.0, + 407.0, + 838.0, + 424.0, + 823.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 409.0, + 863.0, + 409.0, + 863.0, + 426.0, + 850.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 409.0, + 886.0, + 409.0, + 886.0, + 426.0, + 871.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 407.0, + 908.0, + 407.0, + 908.0, + 424.0, + 892.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 408.0, + 1010.0, + 408.0, + 1010.0, + 424.0, + 996.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 409.0, + 1036.0, + 409.0, + 1036.0, + 425.0, + 1022.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 412.0, + 1056.0, + 412.0, + 1056.0, + 421.0, + 1047.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 411.0, + 1078.0, + 411.0, + 1078.0, + 420.0, + 1069.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 411.0, + 1182.0, + 411.0, + 1182.0, + 421.0, + 1172.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 412.0, + 1208.0, + 412.0, + 1208.0, + 424.0, + 1199.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 412.0, + 1230.0, + 412.0, + 1230.0, + 424.0, + 1220.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 411.0, + 1252.0, + 411.0, + 1252.0, + 421.0, + 1241.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 512.0, + 665.0, + 512.0, + 665.0, + 526.0, + 648.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 510.0, + 692.0, + 510.0, + 692.0, + 526.0, + 676.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 508.0, + 716.0, + 508.0, + 716.0, + 526.0, + 696.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 512.0, + 734.0, + 512.0, + 734.0, + 526.0, + 719.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 512.0, + 753.0, + 512.0, + 753.0, + 521.0, + 744.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 512.0, + 839.0, + 512.0, + 839.0, + 526.0, + 823.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 509.0, + 910.0, + 509.0, + 910.0, + 527.0, + 848.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 509.0, + 928.0, + 509.0, + 928.0, + 525.0, + 915.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 512.0, + 1011.0, + 512.0, + 1011.0, + 526.0, + 996.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 509.0, + 1060.0, + 509.0, + 1060.0, + 527.0, + 1021.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 512.0, + 1083.0, + 512.0, + 1083.0, + 526.0, + 1064.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 510.0, + 1101.0, + 510.0, + 1101.0, + 525.0, + 1084.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 512.0, + 1187.0, + 512.0, + 1187.0, + 526.0, + 1169.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 508.0, + 1258.0, + 508.0, + 1258.0, + 528.0, + 1189.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 513.0, + 1273.0, + 513.0, + 1273.0, + 523.0, + 1262.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 532.0, + 665.0, + 532.0, + 665.0, + 547.0, + 648.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 531.0, + 693.0, + 531.0, + 693.0, + 548.0, + 673.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 533.0, + 714.0, + 533.0, + 714.0, + 546.0, + 697.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 532.0, + 736.0, + 532.0, + 736.0, + 545.0, + 719.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 534.0, + 754.0, + 534.0, + 754.0, + 544.0, + 744.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 532.0, + 839.0, + 532.0, + 839.0, + 547.0, + 823.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 533.0, + 865.0, + 533.0, + 865.0, + 547.0, + 849.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 531.0, + 891.0, + 531.0, + 891.0, + 547.0, + 868.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 531.0, + 908.0, + 531.0, + 908.0, + 545.0, + 892.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 532.0, + 929.0, + 532.0, + 929.0, + 546.0, + 915.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 532.0, + 1011.0, + 532.0, + 1011.0, + 547.0, + 996.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 533.0, + 1040.0, + 533.0, + 1040.0, + 547.0, + 1022.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 532.0, + 1061.0, + 532.0, + 1061.0, + 546.0, + 1042.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 532.0, + 1083.0, + 532.0, + 1083.0, + 546.0, + 1066.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 532.0, + 1102.0, + 532.0, + 1102.0, + 546.0, + 1087.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 532.0, + 1186.0, + 532.0, + 1186.0, + 547.0, + 1169.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 532.0, + 1256.0, + 532.0, + 1256.0, + 550.0, + 1193.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 532.0, + 1275.0, + 532.0, + 1275.0, + 546.0, + 1261.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 552.0, + 665.0, + 552.0, + 665.0, + 567.0, + 650.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 553.0, + 691.0, + 553.0, + 691.0, + 567.0, + 674.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 552.0, + 714.0, + 552.0, + 714.0, + 566.0, + 697.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 550.0, + 737.0, + 550.0, + 737.0, + 567.0, + 717.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 552.0, + 756.0, + 552.0, + 756.0, + 566.0, + 740.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 550.0, + 842.0, + 550.0, + 842.0, + 569.0, + 822.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 553.0, + 865.0, + 553.0, + 865.0, + 567.0, + 849.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 552.0, + 889.0, + 552.0, + 889.0, + 566.0, + 870.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 551.0, + 908.0, + 551.0, + 908.0, + 566.0, + 892.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 552.0, + 929.0, + 552.0, + 929.0, + 566.0, + 915.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 551.0, + 1011.0, + 551.0, + 1011.0, + 567.0, + 996.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 552.0, + 1037.0, + 552.0, + 1037.0, + 567.0, + 1022.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 552.0, + 1060.0, + 552.0, + 1060.0, + 566.0, + 1044.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 551.0, + 1082.0, + 551.0, + 1082.0, + 566.0, + 1066.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 552.0, + 1103.0, + 552.0, + 1103.0, + 567.0, + 1087.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 551.0, + 1184.0, + 551.0, + 1184.0, + 567.0, + 1169.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 552.0, + 1210.0, + 552.0, + 1210.0, + 567.0, + 1195.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 552.0, + 1233.0, + 552.0, + 1233.0, + 567.0, + 1217.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 551.0, + 1254.0, + 551.0, + 1254.0, + 566.0, + 1239.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 552.0, + 1275.0, + 552.0, + 1275.0, + 566.0, + 1261.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 572.0, + 665.0, + 572.0, + 665.0, + 586.0, + 648.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 570.0, + 693.0, + 570.0, + 693.0, + 588.0, + 673.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 572.0, + 714.0, + 572.0, + 714.0, + 586.0, + 697.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 572.0, + 734.0, + 572.0, + 734.0, + 585.0, + 719.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 572.0, + 839.0, + 572.0, + 839.0, + 588.0, + 823.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 572.0, + 865.0, + 572.0, + 865.0, + 586.0, + 849.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 572.0, + 889.0, + 572.0, + 889.0, + 586.0, + 870.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 572.0, + 908.0, + 572.0, + 908.0, + 586.0, + 891.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 572.0, + 1011.0, + 572.0, + 1011.0, + 588.0, + 996.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 572.0, + 1037.0, + 572.0, + 1037.0, + 586.0, + 1022.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 571.0, + 1058.0, + 571.0, + 1058.0, + 586.0, + 1044.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 573.0, + 1078.0, + 573.0, + 1078.0, + 585.0, + 1068.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 572.0, + 1186.0, + 572.0, + 1186.0, + 588.0, + 1169.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 571.0, + 1212.0, + 571.0, + 1212.0, + 588.0, + 1196.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 571.0, + 1233.0, + 571.0, + 1233.0, + 586.0, + 1219.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 572.0, + 1254.0, + 572.0, + 1254.0, + 586.0, + 1240.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 592.0, + 665.0, + 592.0, + 665.0, + 608.0, + 648.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 595.0, + 691.0, + 595.0, + 691.0, + 610.0, + 676.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 596.0, + 711.0, + 596.0, + 711.0, + 607.0, + 699.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 595.0, + 732.0, + 595.0, + 732.0, + 605.0, + 720.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 592.0, + 838.0, + 592.0, + 838.0, + 607.0, + 823.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 595.0, + 864.0, + 595.0, + 864.0, + 610.0, + 850.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 596.0, + 884.0, + 596.0, + 884.0, + 608.0, + 873.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 594.0, + 908.0, + 594.0, + 908.0, + 607.0, + 892.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 592.0, + 1011.0, + 592.0, + 1011.0, + 607.0, + 996.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 595.0, + 1037.0, + 595.0, + 1037.0, + 610.0, + 1023.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 595.0, + 1058.0, + 595.0, + 1058.0, + 609.0, + 1044.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 592.0, + 1081.0, + 592.0, + 1081.0, + 607.0, + 1067.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 592.0, + 1186.0, + 592.0, + 1186.0, + 608.0, + 1169.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 595.0, + 1210.0, + 595.0, + 1210.0, + 610.0, + 1196.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 595.0, + 1233.0, + 595.0, + 1233.0, + 610.0, + 1219.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 592.0, + 1254.0, + 592.0, + 1254.0, + 607.0, + 1240.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 696.0, + 317.0, + 696.0, + 317.0, + 710.0, + 303.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 694.0, + 344.0, + 694.0, + 344.0, + 710.0, + 329.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 694.0, + 366.0, + 694.0, + 366.0, + 710.0, + 352.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 697.0, + 386.0, + 697.0, + 386.0, + 708.0, + 376.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 696.0, + 513.0, + 696.0, + 513.0, + 710.0, + 499.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 694.0, + 540.0, + 694.0, + 540.0, + 710.0, + 525.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 694.0, + 561.0, + 694.0, + 561.0, + 710.0, + 546.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 696.0, + 582.0, + 696.0, + 582.0, + 709.0, + 567.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 696.0, + 601.0, + 696.0, + 601.0, + 705.0, + 593.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 696.0, + 711.0, + 696.0, + 711.0, + 710.0, + 693.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 696.0, + 736.0, + 696.0, + 736.0, + 710.0, + 719.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 696.0, + 758.0, + 696.0, + 758.0, + 710.0, + 740.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 697.0, + 780.0, + 697.0, + 780.0, + 709.0, + 762.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 696.0, + 905.0, + 696.0, + 905.0, + 710.0, + 886.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 694.0, + 980.0, + 694.0, + 980.0, + 711.0, + 914.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 697.0, + 991.0, + 697.0, + 991.0, + 706.0, + 982.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 715.0, + 317.0, + 715.0, + 317.0, + 730.0, + 303.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 716.0, + 344.0, + 716.0, + 344.0, + 730.0, + 329.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 715.0, + 366.0, + 715.0, + 366.0, + 729.0, + 352.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 716.0, + 386.0, + 716.0, + 386.0, + 726.0, + 375.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 716.0, + 407.0, + 716.0, + 407.0, + 728.0, + 396.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 715.0, + 513.0, + 715.0, + 513.0, + 730.0, + 499.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 716.0, + 540.0, + 716.0, + 540.0, + 730.0, + 525.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 715.0, + 561.0, + 715.0, + 561.0, + 729.0, + 546.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 713.0, + 582.0, + 713.0, + 582.0, + 728.0, + 567.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 716.0, + 603.0, + 716.0, + 603.0, + 728.0, + 592.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 715.0, + 710.0, + 715.0, + 710.0, + 730.0, + 693.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 716.0, + 737.0, + 716.0, + 737.0, + 730.0, + 719.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 716.0, + 758.0, + 716.0, + 758.0, + 729.0, + 739.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 715.0, + 778.0, + 715.0, + 778.0, + 729.0, + 763.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 717.0, + 797.0, + 717.0, + 797.0, + 726.0, + 784.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 713.0, + 905.0, + 713.0, + 905.0, + 731.0, + 885.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 713.0, + 994.0, + 713.0, + 994.0, + 731.0, + 912.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 734.0, + 317.0, + 734.0, + 317.0, + 750.0, + 303.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 735.0, + 343.0, + 735.0, + 343.0, + 750.0, + 330.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 735.0, + 366.0, + 735.0, + 366.0, + 750.0, + 352.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 735.0, + 387.0, + 735.0, + 387.0, + 749.0, + 373.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 736.0, + 407.0, + 736.0, + 407.0, + 748.0, + 396.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 734.0, + 513.0, + 734.0, + 513.0, + 750.0, + 499.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 735.0, + 540.0, + 735.0, + 540.0, + 750.0, + 525.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 735.0, + 561.0, + 735.0, + 561.0, + 749.0, + 546.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 735.0, + 582.0, + 735.0, + 582.0, + 749.0, + 568.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 736.0, + 605.0, + 736.0, + 605.0, + 749.0, + 590.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 735.0, + 710.0, + 735.0, + 710.0, + 749.0, + 693.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 735.0, + 733.0, + 735.0, + 733.0, + 749.0, + 719.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 735.0, + 758.0, + 735.0, + 758.0, + 749.0, + 740.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 735.0, + 778.0, + 735.0, + 778.0, + 749.0, + 763.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 736.0, + 799.0, + 736.0, + 799.0, + 749.0, + 784.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 735.0, + 904.0, + 735.0, + 904.0, + 749.0, + 886.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 736.0, + 929.0, + 736.0, + 929.0, + 750.0, + 915.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 735.0, + 952.0, + 735.0, + 952.0, + 749.0, + 936.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 735.0, + 975.0, + 735.0, + 975.0, + 749.0, + 956.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 735.0, + 994.0, + 735.0, + 994.0, + 749.0, + 978.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 756.0, + 319.0, + 756.0, + 319.0, + 772.0, + 303.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 755.0, + 343.0, + 755.0, + 343.0, + 770.0, + 330.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 755.0, + 366.0, + 755.0, + 366.0, + 770.0, + 353.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 759.0, + 386.0, + 759.0, + 386.0, + 769.0, + 376.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 756.0, + 513.0, + 756.0, + 513.0, + 770.0, + 499.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 755.0, + 540.0, + 755.0, + 540.0, + 770.0, + 525.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 757.0, + 559.0, + 757.0, + 559.0, + 769.0, + 548.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 756.0, + 709.0, + 756.0, + 709.0, + 772.0, + 693.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 755.0, + 733.0, + 755.0, + 733.0, + 770.0, + 720.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 756.0, + 757.0, + 756.0, + 757.0, + 770.0, + 742.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 757.0, + 776.0, + 757.0, + 776.0, + 768.0, + 765.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 755.0, + 905.0, + 755.0, + 905.0, + 773.0, + 885.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 755.0, + 929.0, + 755.0, + 929.0, + 770.0, + 914.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 756.0, + 972.0, + 756.0, + 972.0, + 770.0, + 957.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 757.0, + 992.0, + 757.0, + 992.0, + 767.0, + 984.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 775.0, + 319.0, + 775.0, + 319.0, + 791.0, + 303.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 776.0, + 343.0, + 776.0, + 343.0, + 792.0, + 330.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 776.0, + 366.0, + 776.0, + 366.0, + 792.0, + 352.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 778.0, + 386.0, + 778.0, + 386.0, + 787.0, + 375.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 775.0, + 513.0, + 775.0, + 513.0, + 791.0, + 499.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 778.0, + 539.0, + 778.0, + 539.0, + 792.0, + 525.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 776.0, + 560.0, + 776.0, + 560.0, + 792.0, + 547.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 778.0, + 580.0, + 778.0, + 580.0, + 787.0, + 570.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 775.0, + 707.0, + 775.0, + 707.0, + 791.0, + 693.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 776.0, + 733.0, + 776.0, + 733.0, + 792.0, + 719.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 778.0, + 757.0, + 778.0, + 757.0, + 792.0, + 742.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 775.0, + 778.0, + 775.0, + 778.0, + 789.0, + 764.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 774.0, + 905.0, + 774.0, + 905.0, + 792.0, + 885.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 776.0, + 929.0, + 776.0, + 929.0, + 793.0, + 915.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 776.0, + 951.0, + 776.0, + 951.0, + 792.0, + 936.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 775.0, + 974.0, + 775.0, + 974.0, + 789.0, + 958.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.75, + 755.5, + 954.75, + 755.5, + 954.75, + 771.0, + 931.75, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 219.0, + 177.0, + 707.0, + 177.0, + 707.0, + 210.0, + 219.0, + 210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 928.0, + 479.0, + 928.0, + 479.0, + 967.0, + 220.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 1871.0, + 679.0, + 1871.0, + 679.0, + 1905.0, + 195.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 198.0, + 1803.0, + 687.0, + 1803.0, + 687.0, + 1841.0, + 198.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 1937.0, + 722.0, + 1937.0, + 722.0, + 1968.0, + 196.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 230.0, + 482.0, + 230.0, + 482.0, + 267.0, + 218.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 219.0, + 177.0, + 707.0, + 177.0, + 707.0, + 210.0, + 219.0, + 210.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 300, + 1586, + 1440, + 1586, + 1440, + 2089, + 300, + 2089 + ], + "score": 0.964 + }, + { + "category_id": 3, + "poly": [ + 299, + 1179, + 1439, + 1179, + 1439, + 1514, + 299, + 1514 + ], + "score": 0.96 + }, + { + "category_id": 3, + "poly": [ + 299, + 761, + 1367, + 761, + 1367, + 1096, + 299, + 1096 + ], + "score": 0.948 + }, + { + "category_id": 3, + "poly": [ + 258, + 493, + 1056, + 493, + 1056, + 635, + 258, + 635 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 804, + 220, + 1514, + 220, + 1514, + 278, + 804, + 278 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 804, + 312, + 1515, + 312, + 1515, + 390, + 804, + 390 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 188, + 659, + 853, + 659, + 853, + 693, + 188, + 693 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 198, + 451, + 680, + 451, + 680, + 484, + 198, + 484 + ], + "score": 0.891 + }, + { + "category_id": 3, + "poly": [ + 212, + 224, + 704, + 224, + 704, + 433, + 212, + 433 + ], + "score": 0.853 + }, + { + "category_id": 1, + "poly": [ + 217, + 716, + 479, + 716, + 479, + 745, + 217, + 745 + ], + "score": 0.481 + }, + { + "category_id": 4, + "poly": [ + 219, + 1529, + 480, + 1529, + 480, + 1557, + 219, + 1557 + ], + "score": 0.297 + }, + { + "category_id": 4, + "poly": [ + 218, + 1127, + 481, + 1127, + 481, + 1155, + 218, + 1155 + ], + "score": 0.262 + }, + { + "category_id": 1, + "poly": [ + 218, + 1127, + 481, + 1127, + 481, + 1155, + 218, + 1155 + ], + "score": 0.196 + }, + { + "category_id": 1, + "poly": [ + 219, + 1529, + 480, + 1529, + 480, + 1557, + 219, + 1557 + ], + "score": 0.123 + }, + { + "category_id": 13, + "poly": [ + 661, + 588, + 695, + 588, + 695, + 609, + 661, + 609 + ], + "score": 0.53, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 572, + 588, + 606, + 588, + 606, + 609, + 572, + 609 + ], + "score": 0.53, + "latex": "= 9" + }, + { + "category_id": 13, + "poly": [ + 482, + 588, + 516, + 588, + 516, + 609, + 482, + 609 + ], + "score": 0.52, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 392, + 588, + 426, + 588, + 426, + 609, + 392, + 609 + ], + "score": 0.52, + "latex": "= 9" + }, + { + "category_id": 13, + "poly": [ + 750, + 588, + 784, + 588, + 784, + 609, + 750, + 609 + ], + "score": 0.48, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 1022, + 314, + 1085, + 314, + 1085, + 350, + 1022, + 350 + ], + "score": 0.46, + "latex": "\\mathbf { 3 \\times 3 }" + }, + { + "category_id": 13, + "poly": [ + 930, + 588, + 963, + 588, + 963, + 609, + 930, + 609 + ], + "score": 0.44, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 303, + 588, + 337, + 588, + 337, + 609, + 303, + 609 + ], + "score": 0.43, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 840, + 588, + 874, + 588, + 874, + 609, + 840, + 609 + ], + "score": 0.41, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1604.0, + 325.0, + 1604.0, + 325.0, + 1615.0, + 312.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1600.0, + 355.0, + 1600.0, + 355.0, + 1616.0, + 339.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1596.0, + 379.0, + 1596.0, + 379.0, + 1618.0, + 358.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1601.0, + 399.0, + 1601.0, + 399.0, + 1616.0, + 383.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1596.0, + 424.0, + 1596.0, + 424.0, + 1616.0, + 404.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1601.0, + 506.0, + 1601.0, + 506.0, + 1616.0, + 490.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1601.0, + 533.0, + 1601.0, + 533.0, + 1618.0, + 517.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1600.0, + 556.0, + 1600.0, + 556.0, + 1616.0, + 538.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1601.0, + 578.0, + 1601.0, + 578.0, + 1616.0, + 561.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1601.0, + 597.0, + 1601.0, + 597.0, + 1614.0, + 586.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1601.0, + 685.0, + 1601.0, + 685.0, + 1618.0, + 670.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1601.0, + 712.0, + 1601.0, + 712.0, + 1616.0, + 696.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1600.0, + 734.0, + 1600.0, + 734.0, + 1616.0, + 718.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1601.0, + 758.0, + 1601.0, + 758.0, + 1616.0, + 740.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1601.0, + 776.0, + 1601.0, + 776.0, + 1614.0, + 764.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1604.0, + 861.0, + 1604.0, + 861.0, + 1615.0, + 849.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1599.0, + 893.0, + 1599.0, + 893.0, + 1615.0, + 875.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1600.0, + 914.0, + 1600.0, + 914.0, + 1616.0, + 897.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1601.0, + 936.0, + 1601.0, + 936.0, + 1616.0, + 919.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1601.0, + 1070.0, + 1601.0, + 1070.0, + 1618.0, + 1055.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1600.0, + 1093.0, + 1600.0, + 1093.0, + 1616.0, + 1075.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1601.0, + 1114.0, + 1601.0, + 1114.0, + 1616.0, + 1099.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1601.0, + 1137.0, + 1601.0, + 1137.0, + 1616.0, + 1121.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1601.0, + 1157.0, + 1601.0, + 1157.0, + 1614.0, + 1146.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1603.0, + 1267.0, + 1603.0, + 1267.0, + 1618.0, + 1251.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1601.0, + 1294.0, + 1601.0, + 1294.0, + 1616.0, + 1278.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1601.0, + 1316.0, + 1601.0, + 1316.0, + 1616.0, + 1300.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1601.0, + 1339.0, + 1601.0, + 1339.0, + 1616.0, + 1322.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1601.0, + 1358.0, + 1601.0, + 1358.0, + 1614.0, + 1348.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1621.0, + 326.0, + 1621.0, + 326.0, + 1638.0, + 311.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1623.0, + 355.0, + 1623.0, + 355.0, + 1639.0, + 338.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1623.0, + 378.0, + 1623.0, + 378.0, + 1638.0, + 360.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1621.0, + 399.0, + 1621.0, + 399.0, + 1638.0, + 383.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1623.0, + 421.0, + 1623.0, + 421.0, + 1638.0, + 405.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1621.0, + 506.0, + 1621.0, + 506.0, + 1638.0, + 489.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1623.0, + 533.0, + 1623.0, + 533.0, + 1639.0, + 516.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1619.0, + 559.0, + 1619.0, + 559.0, + 1638.0, + 537.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1620.0, + 578.0, + 1620.0, + 578.0, + 1638.0, + 561.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1624.0, + 599.0, + 1624.0, + 599.0, + 1635.0, + 586.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1621.0, + 685.0, + 1621.0, + 685.0, + 1638.0, + 669.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1623.0, + 712.0, + 1623.0, + 712.0, + 1638.0, + 696.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1623.0, + 734.0, + 1623.0, + 734.0, + 1638.0, + 716.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1621.0, + 758.0, + 1621.0, + 758.0, + 1637.0, + 740.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1623.0, + 778.0, + 1623.0, + 778.0, + 1638.0, + 763.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1623.0, + 864.0, + 1623.0, + 864.0, + 1638.0, + 848.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1623.0, + 892.0, + 1623.0, + 892.0, + 1638.0, + 874.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1623.0, + 914.0, + 1623.0, + 914.0, + 1638.0, + 897.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1623.0, + 936.0, + 1623.0, + 936.0, + 1638.0, + 919.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1624.0, + 955.0, + 1624.0, + 955.0, + 1635.0, + 944.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1621.0, + 1070.0, + 1621.0, + 1070.0, + 1638.0, + 1055.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1620.0, + 1095.0, + 1620.0, + 1095.0, + 1639.0, + 1073.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1619.0, + 1117.0, + 1619.0, + 1117.0, + 1639.0, + 1096.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1620.0, + 1137.0, + 1620.0, + 1137.0, + 1637.0, + 1119.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1621.0, + 1159.0, + 1621.0, + 1159.0, + 1637.0, + 1143.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1621.0, + 1267.0, + 1621.0, + 1267.0, + 1638.0, + 1251.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1623.0, + 1294.0, + 1623.0, + 1294.0, + 1638.0, + 1277.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1623.0, + 1316.0, + 1623.0, + 1316.0, + 1638.0, + 1300.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1621.0, + 1339.0, + 1621.0, + 1339.0, + 1638.0, + 1322.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1624.0, + 1358.0, + 1624.0, + 1358.0, + 1635.0, + 1347.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1643.0, + 326.0, + 1643.0, + 326.0, + 1659.0, + 311.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1643.0, + 353.0, + 1643.0, + 353.0, + 1660.0, + 338.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1642.0, + 379.0, + 1642.0, + 379.0, + 1662.0, + 358.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1643.0, + 399.0, + 1643.0, + 399.0, + 1660.0, + 382.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1642.0, + 426.0, + 1642.0, + 426.0, + 1662.0, + 404.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1643.0, + 506.0, + 1643.0, + 506.0, + 1659.0, + 490.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1644.0, + 533.0, + 1644.0, + 533.0, + 1659.0, + 517.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1643.0, + 556.0, + 1643.0, + 556.0, + 1659.0, + 538.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1643.0, + 579.0, + 1643.0, + 579.0, + 1659.0, + 561.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1643.0, + 600.0, + 1643.0, + 600.0, + 1659.0, + 582.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1643.0, + 684.0, + 1643.0, + 684.0, + 1660.0, + 669.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1643.0, + 712.0, + 1643.0, + 712.0, + 1660.0, + 696.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1642.0, + 737.0, + 1642.0, + 737.0, + 1662.0, + 715.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1640.0, + 760.0, + 1640.0, + 760.0, + 1662.0, + 738.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1642.0, + 782.0, + 1642.0, + 782.0, + 1662.0, + 761.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1643.0, + 864.0, + 1643.0, + 864.0, + 1659.0, + 848.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1643.0, + 891.0, + 1643.0, + 891.0, + 1660.0, + 874.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1643.0, + 914.0, + 1643.0, + 914.0, + 1660.0, + 897.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1642.0, + 938.0, + 1642.0, + 938.0, + 1660.0, + 917.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1643.0, + 959.0, + 1643.0, + 959.0, + 1659.0, + 942.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1643.0, + 1070.0, + 1643.0, + 1070.0, + 1659.0, + 1055.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1644.0, + 1092.0, + 1644.0, + 1092.0, + 1659.0, + 1075.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1643.0, + 1115.0, + 1643.0, + 1115.0, + 1660.0, + 1099.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1643.0, + 1137.0, + 1643.0, + 1137.0, + 1659.0, + 1119.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1643.0, + 1159.0, + 1643.0, + 1159.0, + 1659.0, + 1143.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1643.0, + 1267.0, + 1643.0, + 1267.0, + 1659.0, + 1251.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1643.0, + 1294.0, + 1643.0, + 1294.0, + 1659.0, + 1276.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1643.0, + 1317.0, + 1643.0, + 1317.0, + 1660.0, + 1300.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1643.0, + 1339.0, + 1643.0, + 1339.0, + 1659.0, + 1322.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1643.0, + 1361.0, + 1643.0, + 1361.0, + 1660.0, + 1345.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1664.0, + 326.0, + 1664.0, + 326.0, + 1680.0, + 311.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1663.0, + 353.0, + 1663.0, + 353.0, + 1680.0, + 339.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1662.0, + 379.0, + 1662.0, + 379.0, + 1682.0, + 358.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1664.0, + 399.0, + 1664.0, + 399.0, + 1680.0, + 382.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1663.0, + 423.0, + 1663.0, + 423.0, + 1679.0, + 406.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1664.0, + 506.0, + 1664.0, + 506.0, + 1680.0, + 489.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1663.0, + 533.0, + 1663.0, + 533.0, + 1680.0, + 516.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1662.0, + 560.0, + 1662.0, + 560.0, + 1682.0, + 537.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1663.0, + 578.0, + 1663.0, + 578.0, + 1680.0, + 561.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1663.0, + 603.0, + 1663.0, + 603.0, + 1679.0, + 587.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1664.0, + 685.0, + 1664.0, + 685.0, + 1680.0, + 669.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1663.0, + 711.0, + 1663.0, + 711.0, + 1680.0, + 696.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1663.0, + 734.0, + 1663.0, + 734.0, + 1680.0, + 718.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1665.0, + 758.0, + 1665.0, + 758.0, + 1680.0, + 741.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1665.0, + 864.0, + 1665.0, + 864.0, + 1680.0, + 848.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1663.0, + 891.0, + 1663.0, + 891.0, + 1680.0, + 874.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1663.0, + 914.0, + 1663.0, + 914.0, + 1680.0, + 897.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1664.0, + 936.0, + 1664.0, + 936.0, + 1680.0, + 919.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1664.0, + 1070.0, + 1664.0, + 1070.0, + 1680.0, + 1055.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1662.0, + 1095.0, + 1662.0, + 1095.0, + 1682.0, + 1073.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1663.0, + 1115.0, + 1663.0, + 1115.0, + 1680.0, + 1099.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1662.0, + 1139.0, + 1662.0, + 1139.0, + 1682.0, + 1118.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1663.0, + 1161.0, + 1663.0, + 1161.0, + 1679.0, + 1145.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1665.0, + 1267.0, + 1665.0, + 1267.0, + 1680.0, + 1251.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1663.0, + 1292.0, + 1663.0, + 1292.0, + 1679.0, + 1277.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1663.0, + 1316.0, + 1663.0, + 1316.0, + 1680.0, + 1300.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1665.0, + 1339.0, + 1665.0, + 1339.0, + 1680.0, + 1323.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1665.0, + 1358.0, + 1665.0, + 1358.0, + 1675.0, + 1349.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1683.0, + 328.0, + 1683.0, + 328.0, + 1701.0, + 311.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1685.0, + 353.0, + 1685.0, + 353.0, + 1703.0, + 338.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1682.0, + 379.0, + 1682.0, + 379.0, + 1703.0, + 357.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1683.0, + 399.0, + 1683.0, + 399.0, + 1701.0, + 382.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1683.0, + 423.0, + 1683.0, + 423.0, + 1699.0, + 405.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1684.0, + 506.0, + 1684.0, + 506.0, + 1701.0, + 489.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1687.0, + 533.0, + 1687.0, + 533.0, + 1703.0, + 517.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1685.0, + 556.0, + 1685.0, + 556.0, + 1702.0, + 538.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1684.0, + 578.0, + 1684.0, + 578.0, + 1701.0, + 561.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1683.0, + 685.0, + 1683.0, + 685.0, + 1701.0, + 669.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1687.0, + 711.0, + 1687.0, + 711.0, + 1702.0, + 696.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1683.0, + 738.0, + 1683.0, + 738.0, + 1702.0, + 715.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1682.0, + 760.0, + 1682.0, + 760.0, + 1702.0, + 740.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1683.0, + 780.0, + 1683.0, + 780.0, + 1699.0, + 763.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1684.0, + 864.0, + 1684.0, + 864.0, + 1701.0, + 848.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1682.0, + 893.0, + 1682.0, + 893.0, + 1703.0, + 870.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1682.0, + 938.0, + 1682.0, + 938.0, + 1704.0, + 896.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1684.0, + 1070.0, + 1684.0, + 1070.0, + 1701.0, + 1055.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1685.0, + 1092.0, + 1685.0, + 1092.0, + 1702.0, + 1075.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1687.0, + 1113.0, + 1687.0, + 1113.0, + 1703.0, + 1099.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1684.0, + 1137.0, + 1684.0, + 1137.0, + 1701.0, + 1119.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1688.0, + 1157.0, + 1688.0, + 1157.0, + 1697.0, + 1146.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1684.0, + 1267.0, + 1684.0, + 1267.0, + 1701.0, + 1251.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1684.0, + 1295.0, + 1684.0, + 1295.0, + 1703.0, + 1273.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1687.0, + 1316.0, + 1687.0, + 1316.0, + 1703.0, + 1300.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1684.0, + 1339.0, + 1684.0, + 1339.0, + 1699.0, + 1322.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1771.0, + 325.0, + 1771.0, + 325.0, + 1783.0, + 313.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1768.0, + 356.0, + 1768.0, + 356.0, + 1785.0, + 339.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1768.0, + 378.0, + 1768.0, + 378.0, + 1785.0, + 360.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1768.0, + 399.0, + 1768.0, + 399.0, + 1785.0, + 382.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1770.0, + 506.0, + 1770.0, + 506.0, + 1785.0, + 490.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1768.0, + 533.0, + 1768.0, + 533.0, + 1785.0, + 517.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1768.0, + 555.0, + 1768.0, + 555.0, + 1785.0, + 539.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1770.0, + 578.0, + 1770.0, + 578.0, + 1785.0, + 561.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1770.0, + 596.0, + 1770.0, + 596.0, + 1780.0, + 587.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1770.0, + 685.0, + 1770.0, + 685.0, + 1785.0, + 670.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1768.0, + 712.0, + 1768.0, + 712.0, + 1785.0, + 696.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1768.0, + 734.0, + 1768.0, + 734.0, + 1785.0, + 718.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1770.0, + 758.0, + 1770.0, + 758.0, + 1785.0, + 740.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1771.0, + 776.0, + 1771.0, + 776.0, + 1782.0, + 764.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1770.0, + 864.0, + 1770.0, + 864.0, + 1786.0, + 848.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1768.0, + 892.0, + 1768.0, + 892.0, + 1785.0, + 875.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1768.0, + 914.0, + 1768.0, + 914.0, + 1785.0, + 897.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1768.0, + 936.0, + 1768.0, + 936.0, + 1785.0, + 918.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1770.0, + 1070.0, + 1770.0, + 1070.0, + 1785.0, + 1055.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1770.0, + 1092.0, + 1770.0, + 1092.0, + 1785.0, + 1077.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1770.0, + 1114.0, + 1770.0, + 1114.0, + 1785.0, + 1097.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1770.0, + 1136.0, + 1770.0, + 1136.0, + 1785.0, + 1121.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1770.0, + 1157.0, + 1770.0, + 1157.0, + 1781.0, + 1146.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1770.0, + 1267.0, + 1770.0, + 1267.0, + 1785.0, + 1251.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1768.0, + 1294.0, + 1768.0, + 1294.0, + 1785.0, + 1278.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1768.0, + 1317.0, + 1768.0, + 1317.0, + 1785.0, + 1300.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1770.0, + 1339.0, + 1770.0, + 1339.0, + 1785.0, + 1322.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1771.0, + 1357.0, + 1771.0, + 1357.0, + 1781.0, + 1348.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1792.0, + 325.0, + 1792.0, + 325.0, + 1805.0, + 313.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1794.0, + 352.0, + 1794.0, + 352.0, + 1806.0, + 340.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1792.0, + 377.0, + 1792.0, + 377.0, + 1806.0, + 360.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1790.0, + 399.0, + 1790.0, + 399.0, + 1806.0, + 383.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1794.0, + 419.0, + 1794.0, + 419.0, + 1805.0, + 408.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1791.0, + 506.0, + 1791.0, + 506.0, + 1807.0, + 490.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1791.0, + 533.0, + 1791.0, + 533.0, + 1807.0, + 517.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1791.0, + 556.0, + 1791.0, + 556.0, + 1806.0, + 541.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1789.0, + 578.0, + 1789.0, + 578.0, + 1806.0, + 561.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1794.0, + 599.0, + 1794.0, + 599.0, + 1804.0, + 588.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1790.0, + 684.0, + 1790.0, + 684.0, + 1807.0, + 670.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1792.0, + 711.0, + 1792.0, + 711.0, + 1807.0, + 696.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1792.0, + 734.0, + 1792.0, + 734.0, + 1806.0, + 718.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1790.0, + 756.0, + 1790.0, + 756.0, + 1806.0, + 741.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1794.0, + 777.0, + 1794.0, + 777.0, + 1805.0, + 764.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1791.0, + 864.0, + 1791.0, + 864.0, + 1807.0, + 848.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1792.0, + 891.0, + 1792.0, + 891.0, + 1807.0, + 875.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1791.0, + 913.0, + 1791.0, + 913.0, + 1806.0, + 897.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1790.0, + 936.0, + 1790.0, + 936.0, + 1806.0, + 920.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1794.0, + 955.0, + 1794.0, + 955.0, + 1805.0, + 945.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1790.0, + 1070.0, + 1790.0, + 1070.0, + 1807.0, + 1055.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1792.0, + 1091.0, + 1792.0, + 1091.0, + 1807.0, + 1075.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1791.0, + 1114.0, + 1791.0, + 1114.0, + 1806.0, + 1099.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1789.0, + 1137.0, + 1789.0, + 1137.0, + 1806.0, + 1121.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1792.0, + 1158.0, + 1792.0, + 1158.0, + 1805.0, + 1146.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1792.0, + 1264.0, + 1792.0, + 1264.0, + 1805.0, + 1252.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1794.0, + 1291.0, + 1794.0, + 1291.0, + 1806.0, + 1279.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1794.0, + 1314.0, + 1794.0, + 1314.0, + 1805.0, + 1301.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1792.0, + 1336.0, + 1792.0, + 1336.0, + 1805.0, + 1325.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1794.0, + 1358.0, + 1794.0, + 1358.0, + 1804.0, + 1348.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1812.0, + 325.0, + 1812.0, + 325.0, + 1825.0, + 313.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1814.0, + 352.0, + 1814.0, + 352.0, + 1826.0, + 340.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1809.0, + 379.0, + 1809.0, + 379.0, + 1829.0, + 358.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1810.0, + 399.0, + 1810.0, + 399.0, + 1826.0, + 382.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1811.0, + 423.0, + 1811.0, + 423.0, + 1826.0, + 406.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1811.0, + 506.0, + 1811.0, + 506.0, + 1826.0, + 490.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1812.0, + 533.0, + 1812.0, + 533.0, + 1827.0, + 517.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1811.0, + 556.0, + 1811.0, + 556.0, + 1826.0, + 538.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1812.0, + 575.0, + 1812.0, + 575.0, + 1825.0, + 564.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1814.0, + 599.0, + 1814.0, + 599.0, + 1825.0, + 586.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 1812.0, + 683.0, + 1812.0, + 683.0, + 1825.0, + 671.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1811.0, + 711.0, + 1811.0, + 711.0, + 1827.0, + 696.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1811.0, + 736.0, + 1811.0, + 736.0, + 1826.0, + 718.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1810.0, + 758.0, + 1810.0, + 758.0, + 1827.0, + 741.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1811.0, + 780.0, + 1811.0, + 780.0, + 1826.0, + 763.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1811.0, + 864.0, + 1811.0, + 864.0, + 1827.0, + 848.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1814.0, + 888.0, + 1814.0, + 888.0, + 1826.0, + 876.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1810.0, + 914.0, + 1810.0, + 914.0, + 1827.0, + 897.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1809.0, + 938.0, + 1809.0, + 938.0, + 1827.0, + 917.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1811.0, + 958.0, + 1811.0, + 958.0, + 1826.0, + 942.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1810.0, + 1070.0, + 1810.0, + 1070.0, + 1827.0, + 1055.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1812.0, + 1091.0, + 1812.0, + 1091.0, + 1827.0, + 1075.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1811.0, + 1114.0, + 1811.0, + 1114.0, + 1827.0, + 1099.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1810.0, + 1137.0, + 1810.0, + 1137.0, + 1826.0, + 1121.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1814.0, + 1157.0, + 1814.0, + 1157.0, + 1825.0, + 1146.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1811.0, + 1267.0, + 1811.0, + 1267.0, + 1827.0, + 1251.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1814.0, + 1291.0, + 1814.0, + 1291.0, + 1826.0, + 1279.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1811.0, + 1316.0, + 1811.0, + 1316.0, + 1827.0, + 1300.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1810.0, + 1339.0, + 1810.0, + 1339.0, + 1826.0, + 1322.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1815.0, + 1358.0, + 1815.0, + 1358.0, + 1825.0, + 1348.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1834.0, + 326.0, + 1834.0, + 326.0, + 1849.0, + 311.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1832.0, + 355.0, + 1832.0, + 355.0, + 1848.0, + 339.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1831.0, + 377.0, + 1831.0, + 377.0, + 1848.0, + 360.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1835.0, + 397.0, + 1835.0, + 397.0, + 1846.0, + 384.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1834.0, + 421.0, + 1834.0, + 421.0, + 1845.0, + 410.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1835.0, + 503.0, + 1835.0, + 503.0, + 1846.0, + 492.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1832.0, + 533.0, + 1832.0, + 533.0, + 1848.0, + 517.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1831.0, + 557.0, + 1831.0, + 557.0, + 1846.0, + 539.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1835.0, + 575.0, + 1835.0, + 575.0, + 1846.0, + 563.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1832.0, + 684.0, + 1832.0, + 684.0, + 1849.0, + 670.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1831.0, + 711.0, + 1831.0, + 711.0, + 1848.0, + 696.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1831.0, + 734.0, + 1831.0, + 734.0, + 1846.0, + 718.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1835.0, + 755.0, + 1835.0, + 755.0, + 1846.0, + 742.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1834.0, + 777.0, + 1834.0, + 777.0, + 1844.0, + 768.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1832.0, + 864.0, + 1832.0, + 864.0, + 1849.0, + 848.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1831.0, + 889.0, + 1831.0, + 889.0, + 1849.0, + 875.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1831.0, + 913.0, + 1831.0, + 913.0, + 1849.0, + 897.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1832.0, + 936.0, + 1832.0, + 936.0, + 1848.0, + 919.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1832.0, + 1070.0, + 1832.0, + 1070.0, + 1849.0, + 1055.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1834.0, + 1090.0, + 1834.0, + 1090.0, + 1846.0, + 1078.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1831.0, + 1114.0, + 1831.0, + 1114.0, + 1848.0, + 1099.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1835.0, + 1136.0, + 1835.0, + 1136.0, + 1845.0, + 1122.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1835.0, + 1158.0, + 1835.0, + 1158.0, + 1844.0, + 1148.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1835.0, + 1264.0, + 1835.0, + 1264.0, + 1848.0, + 1252.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1831.0, + 1292.0, + 1831.0, + 1292.0, + 1848.0, + 1278.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1832.0, + 1316.0, + 1832.0, + 1316.0, + 1848.0, + 1300.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1835.0, + 1336.0, + 1835.0, + 1336.0, + 1845.0, + 1325.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1854.0, + 326.0, + 1854.0, + 326.0, + 1870.0, + 311.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1855.0, + 355.0, + 1855.0, + 355.0, + 1871.0, + 339.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1854.0, + 378.0, + 1854.0, + 378.0, + 1870.0, + 360.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1854.0, + 399.0, + 1854.0, + 399.0, + 1869.0, + 383.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1854.0, + 506.0, + 1854.0, + 506.0, + 1869.0, + 490.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1855.0, + 532.0, + 1855.0, + 532.0, + 1871.0, + 517.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1855.0, + 555.0, + 1855.0, + 555.0, + 1870.0, + 538.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1854.0, + 578.0, + 1854.0, + 578.0, + 1869.0, + 561.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1854.0, + 685.0, + 1854.0, + 685.0, + 1870.0, + 669.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1855.0, + 711.0, + 1855.0, + 711.0, + 1871.0, + 696.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1855.0, + 734.0, + 1855.0, + 734.0, + 1870.0, + 718.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1854.0, + 758.0, + 1854.0, + 758.0, + 1870.0, + 741.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1854.0, + 864.0, + 1854.0, + 864.0, + 1870.0, + 848.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1855.0, + 889.0, + 1855.0, + 889.0, + 1873.0, + 875.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1854.0, + 914.0, + 1854.0, + 914.0, + 1871.0, + 897.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1851.0, + 938.0, + 1851.0, + 938.0, + 1870.0, + 917.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1854.0, + 1070.0, + 1854.0, + 1070.0, + 1870.0, + 1055.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1855.0, + 1091.0, + 1855.0, + 1091.0, + 1871.0, + 1075.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1855.0, + 1114.0, + 1855.0, + 1114.0, + 1871.0, + 1099.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1854.0, + 1137.0, + 1854.0, + 1137.0, + 1869.0, + 1119.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1854.0, + 1267.0, + 1854.0, + 1267.0, + 1870.0, + 1251.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1855.0, + 1292.0, + 1855.0, + 1292.0, + 1871.0, + 1278.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1855.0, + 1316.0, + 1855.0, + 1316.0, + 1871.0, + 1301.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1941.0, + 325.0, + 1941.0, + 325.0, + 1952.0, + 313.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1938.0, + 355.0, + 1938.0, + 355.0, + 1954.0, + 339.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1938.0, + 377.0, + 1938.0, + 377.0, + 1953.0, + 361.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1941.0, + 397.0, + 1941.0, + 397.0, + 1952.0, + 384.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1939.0, + 556.0, + 1939.0, + 556.0, + 1954.0, + 541.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1938.0, + 578.0, + 1938.0, + 578.0, + 1954.0, + 561.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1938.0, + 600.0, + 1938.0, + 600.0, + 1954.0, + 585.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1939.0, + 622.0, + 1939.0, + 622.0, + 1954.0, + 606.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1939.0, + 641.0, + 1939.0, + 641.0, + 1951.0, + 630.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1939.0, + 773.0, + 1939.0, + 773.0, + 1954.0, + 758.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1938.0, + 802.0, + 1938.0, + 802.0, + 1954.0, + 786.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1938.0, + 822.0, + 1938.0, + 822.0, + 1954.0, + 808.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1939.0, + 845.0, + 1939.0, + 845.0, + 1954.0, + 830.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1939.0, + 865.0, + 1939.0, + 865.0, + 1951.0, + 855.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1959.0, + 326.0, + 1959.0, + 326.0, + 1974.0, + 311.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1959.0, + 353.0, + 1959.0, + 353.0, + 1976.0, + 339.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1959.0, + 377.0, + 1959.0, + 377.0, + 1974.0, + 361.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1961.0, + 397.0, + 1961.0, + 397.0, + 1972.0, + 384.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1961.0, + 419.0, + 1961.0, + 419.0, + 1972.0, + 408.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1958.0, + 556.0, + 1958.0, + 556.0, + 1976.0, + 539.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1959.0, + 578.0, + 1959.0, + 578.0, + 1974.0, + 561.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1959.0, + 600.0, + 1959.0, + 600.0, + 1974.0, + 583.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1958.0, + 622.0, + 1958.0, + 622.0, + 1974.0, + 606.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1959.0, + 647.0, + 1959.0, + 647.0, + 1974.0, + 628.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1958.0, + 773.0, + 1958.0, + 773.0, + 1974.0, + 758.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1959.0, + 802.0, + 1959.0, + 802.0, + 1974.0, + 786.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1961.0, + 821.0, + 1961.0, + 821.0, + 1973.0, + 809.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1961.0, + 844.0, + 1961.0, + 844.0, + 1972.0, + 831.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1961.0, + 866.0, + 1961.0, + 866.0, + 1972.0, + 855.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1979.0, + 326.0, + 1979.0, + 326.0, + 1996.0, + 311.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1979.0, + 353.0, + 1979.0, + 353.0, + 1997.0, + 338.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1978.0, + 379.0, + 1978.0, + 379.0, + 1998.0, + 358.0, + 1998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1979.0, + 399.0, + 1979.0, + 399.0, + 1996.0, + 382.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1981.0, + 423.0, + 1981.0, + 423.0, + 1996.0, + 406.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1979.0, + 556.0, + 1979.0, + 556.0, + 1997.0, + 539.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1979.0, + 577.0, + 1979.0, + 577.0, + 1997.0, + 561.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1979.0, + 600.0, + 1979.0, + 600.0, + 1997.0, + 583.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1979.0, + 622.0, + 1979.0, + 622.0, + 1996.0, + 605.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1979.0, + 645.0, + 1979.0, + 645.0, + 1996.0, + 628.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1979.0, + 774.0, + 1979.0, + 774.0, + 1996.0, + 758.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1979.0, + 802.0, + 1979.0, + 802.0, + 1997.0, + 785.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1979.0, + 824.0, + 1979.0, + 824.0, + 1997.0, + 807.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1979.0, + 847.0, + 1979.0, + 847.0, + 1996.0, + 830.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1983.0, + 867.0, + 1983.0, + 867.0, + 1995.0, + 857.0, + 1995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 2002.0, + 328.0, + 2002.0, + 328.0, + 2017.0, + 311.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 2000.0, + 353.0, + 2000.0, + 353.0, + 2017.0, + 338.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 2000.0, + 377.0, + 2000.0, + 377.0, + 2017.0, + 360.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 2002.0, + 399.0, + 2002.0, + 399.0, + 2017.0, + 382.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 2000.0, + 422.0, + 2000.0, + 422.0, + 2016.0, + 408.0, + 2016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 2002.0, + 556.0, + 2002.0, + 556.0, + 2017.0, + 539.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 2001.0, + 577.0, + 2001.0, + 577.0, + 2017.0, + 561.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 2000.0, + 600.0, + 2000.0, + 600.0, + 2017.0, + 585.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 2001.0, + 623.0, + 2001.0, + 623.0, + 2017.0, + 606.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 2001.0, + 644.0, + 2001.0, + 644.0, + 2015.0, + 635.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 2001.0, + 773.0, + 2001.0, + 773.0, + 2017.0, + 758.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 2001.0, + 800.0, + 2001.0, + 800.0, + 2017.0, + 786.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 2001.0, + 824.0, + 2001.0, + 824.0, + 2017.0, + 808.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2002.0, + 845.0, + 2002.0, + 845.0, + 2017.0, + 830.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 2003.0, + 867.0, + 2003.0, + 867.0, + 2013.0, + 857.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 2021.0, + 328.0, + 2021.0, + 328.0, + 2037.0, + 311.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 2023.0, + 355.0, + 2023.0, + 355.0, + 2040.0, + 338.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 2022.0, + 377.0, + 2022.0, + 377.0, + 2040.0, + 360.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 2021.0, + 399.0, + 2021.0, + 399.0, + 2037.0, + 382.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 2025.0, + 418.0, + 2025.0, + 418.0, + 2036.0, + 408.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 2022.0, + 556.0, + 2022.0, + 556.0, + 2038.0, + 541.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 2025.0, + 577.0, + 2025.0, + 577.0, + 2040.0, + 561.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 2025.0, + 600.0, + 2025.0, + 600.0, + 2041.0, + 585.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 2023.0, + 619.0, + 2023.0, + 619.0, + 2036.0, + 608.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 2022.0, + 773.0, + 2022.0, + 773.0, + 2037.0, + 758.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 2023.0, + 800.0, + 2023.0, + 800.0, + 2040.0, + 786.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 2023.0, + 822.0, + 2023.0, + 822.0, + 2040.0, + 808.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2021.0, + 845.0, + 2021.0, + 845.0, + 2037.0, + 831.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1661.5, + 957.0, + 1661.5, + 957.0, + 1678.0, + 946.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1196.0, + 327.0, + 1196.0, + 327.0, + 1212.0, + 311.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1194.0, + 355.0, + 1194.0, + 355.0, + 1212.0, + 338.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1192.0, + 378.0, + 1192.0, + 378.0, + 1213.0, + 357.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1196.0, + 399.0, + 1196.0, + 399.0, + 1212.0, + 383.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1192.0, + 425.0, + 1192.0, + 425.0, + 1212.0, + 403.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1196.0, + 506.0, + 1196.0, + 506.0, + 1212.0, + 489.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1196.0, + 533.0, + 1196.0, + 533.0, + 1212.0, + 516.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1194.0, + 555.0, + 1194.0, + 555.0, + 1212.0, + 538.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1196.0, + 578.0, + 1196.0, + 578.0, + 1212.0, + 560.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1194.0, + 599.0, + 1194.0, + 599.0, + 1211.0, + 584.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1196.0, + 686.0, + 1196.0, + 686.0, + 1212.0, + 669.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1196.0, + 713.0, + 1196.0, + 713.0, + 1212.0, + 696.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1196.0, + 735.0, + 1196.0, + 735.0, + 1212.0, + 718.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1196.0, + 758.0, + 1196.0, + 758.0, + 1212.0, + 740.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1197.0, + 776.0, + 1197.0, + 776.0, + 1209.0, + 763.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1197.0, + 861.0, + 1197.0, + 861.0, + 1211.0, + 850.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1192.0, + 895.0, + 1192.0, + 895.0, + 1213.0, + 870.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1194.0, + 914.0, + 1194.0, + 914.0, + 1212.0, + 896.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1192.0, + 939.0, + 1192.0, + 939.0, + 1213.0, + 916.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1197.0, + 954.0, + 1197.0, + 954.0, + 1209.0, + 944.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1196.0, + 1071.0, + 1196.0, + 1071.0, + 1212.0, + 1055.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1196.0, + 1094.0, + 1196.0, + 1094.0, + 1212.0, + 1073.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1196.0, + 1116.0, + 1196.0, + 1116.0, + 1212.0, + 1098.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1196.0, + 1138.0, + 1196.0, + 1138.0, + 1212.0, + 1118.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1197.0, + 1156.0, + 1197.0, + 1156.0, + 1209.0, + 1145.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1196.0, + 1267.0, + 1196.0, + 1267.0, + 1212.0, + 1251.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1196.0, + 1295.0, + 1196.0, + 1295.0, + 1212.0, + 1277.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1196.0, + 1317.0, + 1196.0, + 1317.0, + 1212.0, + 1299.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 1197.0, + 1339.0, + 1197.0, + 1339.0, + 1212.0, + 1321.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1197.0, + 1357.0, + 1197.0, + 1357.0, + 1209.0, + 1347.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1216.0, + 328.0, + 1216.0, + 328.0, + 1234.0, + 311.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1217.0, + 355.0, + 1217.0, + 355.0, + 1234.0, + 338.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1217.0, + 377.0, + 1217.0, + 377.0, + 1234.0, + 359.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1216.0, + 399.0, + 1216.0, + 399.0, + 1232.0, + 382.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1217.0, + 421.0, + 1217.0, + 421.0, + 1234.0, + 405.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1216.0, + 506.0, + 1216.0, + 506.0, + 1232.0, + 489.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1216.0, + 533.0, + 1216.0, + 533.0, + 1234.0, + 515.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1213.0, + 603.0, + 1213.0, + 603.0, + 1234.0, + 536.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1216.0, + 686.0, + 1216.0, + 686.0, + 1232.0, + 669.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1217.0, + 713.0, + 1217.0, + 713.0, + 1234.0, + 696.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1217.0, + 735.0, + 1217.0, + 735.0, + 1234.0, + 717.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1216.0, + 758.0, + 1216.0, + 758.0, + 1232.0, + 740.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1216.0, + 779.0, + 1216.0, + 779.0, + 1234.0, + 762.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1216.0, + 863.0, + 1216.0, + 863.0, + 1234.0, + 847.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1217.0, + 892.0, + 1217.0, + 892.0, + 1234.0, + 873.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1217.0, + 914.0, + 1217.0, + 914.0, + 1234.0, + 896.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1216.0, + 936.0, + 1216.0, + 936.0, + 1232.0, + 918.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1219.0, + 956.0, + 1219.0, + 956.0, + 1231.0, + 944.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1216.0, + 1071.0, + 1216.0, + 1071.0, + 1232.0, + 1054.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1213.0, + 1116.0, + 1213.0, + 1116.0, + 1234.0, + 1072.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1215.0, + 1138.0, + 1215.0, + 1138.0, + 1231.0, + 1118.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1215.0, + 1161.0, + 1215.0, + 1161.0, + 1232.0, + 1143.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1216.0, + 1266.0, + 1216.0, + 1266.0, + 1234.0, + 1250.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1217.0, + 1294.0, + 1217.0, + 1294.0, + 1234.0, + 1277.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1217.0, + 1316.0, + 1217.0, + 1316.0, + 1234.0, + 1299.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1216.0, + 1338.0, + 1216.0, + 1338.0, + 1232.0, + 1322.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1219.0, + 1359.0, + 1219.0, + 1359.0, + 1231.0, + 1347.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1239.0, + 325.0, + 1239.0, + 325.0, + 1251.0, + 314.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1238.0, + 354.0, + 1238.0, + 354.0, + 1255.0, + 338.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1234.0, + 378.0, + 1234.0, + 378.0, + 1255.0, + 357.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1235.0, + 425.0, + 1235.0, + 425.0, + 1255.0, + 379.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1236.0, + 506.0, + 1236.0, + 506.0, + 1254.0, + 489.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1240.0, + 531.0, + 1240.0, + 531.0, + 1253.0, + 519.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1239.0, + 556.0, + 1239.0, + 556.0, + 1254.0, + 538.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1238.0, + 578.0, + 1238.0, + 578.0, + 1254.0, + 562.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1240.0, + 598.0, + 1240.0, + 598.0, + 1253.0, + 585.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1235.0, + 686.0, + 1235.0, + 686.0, + 1253.0, + 669.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1239.0, + 711.0, + 1239.0, + 711.0, + 1255.0, + 695.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1235.0, + 737.0, + 1235.0, + 737.0, + 1255.0, + 715.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1234.0, + 781.0, + 1234.0, + 781.0, + 1257.0, + 739.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1236.0, + 863.0, + 1236.0, + 863.0, + 1254.0, + 847.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1235.0, + 892.0, + 1235.0, + 892.0, + 1257.0, + 870.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1236.0, + 914.0, + 1236.0, + 914.0, + 1254.0, + 896.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1234.0, + 939.0, + 1234.0, + 939.0, + 1255.0, + 916.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1238.0, + 959.0, + 1238.0, + 959.0, + 1254.0, + 941.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1236.0, + 1071.0, + 1236.0, + 1071.0, + 1254.0, + 1054.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1238.0, + 1092.0, + 1238.0, + 1092.0, + 1255.0, + 1074.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1238.0, + 1114.0, + 1238.0, + 1114.0, + 1255.0, + 1098.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1236.0, + 1138.0, + 1236.0, + 1138.0, + 1254.0, + 1120.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1240.0, + 1158.0, + 1240.0, + 1158.0, + 1253.0, + 1145.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1236.0, + 1266.0, + 1236.0, + 1266.0, + 1254.0, + 1250.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1240.0, + 1291.0, + 1240.0, + 1291.0, + 1253.0, + 1278.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1238.0, + 1317.0, + 1238.0, + 1317.0, + 1255.0, + 1299.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1236.0, + 1339.0, + 1236.0, + 1339.0, + 1254.0, + 1322.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1238.0, + 1361.0, + 1238.0, + 1361.0, + 1254.0, + 1346.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1258.0, + 327.0, + 1258.0, + 327.0, + 1274.0, + 310.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1258.0, + 354.0, + 1258.0, + 354.0, + 1276.0, + 338.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1257.0, + 377.0, + 1257.0, + 377.0, + 1274.0, + 359.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1260.0, + 399.0, + 1260.0, + 399.0, + 1276.0, + 383.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1257.0, + 423.0, + 1257.0, + 423.0, + 1274.0, + 405.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1257.0, + 506.0, + 1257.0, + 506.0, + 1274.0, + 489.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1258.0, + 532.0, + 1258.0, + 532.0, + 1274.0, + 515.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1254.0, + 559.0, + 1254.0, + 559.0, + 1276.0, + 538.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1258.0, + 578.0, + 1258.0, + 578.0, + 1274.0, + 560.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1255.0, + 603.0, + 1255.0, + 603.0, + 1274.0, + 586.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1261.0, + 683.0, + 1261.0, + 683.0, + 1273.0, + 670.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1258.0, + 711.0, + 1258.0, + 711.0, + 1274.0, + 696.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1258.0, + 733.0, + 1258.0, + 733.0, + 1274.0, + 718.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1260.0, + 757.0, + 1260.0, + 757.0, + 1276.0, + 741.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1260.0, + 779.0, + 1260.0, + 779.0, + 1273.0, + 768.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1260.0, + 863.0, + 1260.0, + 863.0, + 1276.0, + 847.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1255.0, + 891.0, + 1255.0, + 891.0, + 1274.0, + 874.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1258.0, + 914.0, + 1258.0, + 914.0, + 1276.0, + 897.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1258.0, + 936.0, + 1258.0, + 936.0, + 1274.0, + 918.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1260.0, + 957.0, + 1260.0, + 957.0, + 1273.0, + 945.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1258.0, + 1069.0, + 1258.0, + 1069.0, + 1274.0, + 1054.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1254.0, + 1095.0, + 1254.0, + 1095.0, + 1277.0, + 1072.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1257.0, + 1114.0, + 1257.0, + 1114.0, + 1276.0, + 1098.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1255.0, + 1140.0, + 1255.0, + 1140.0, + 1276.0, + 1117.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1257.0, + 1161.0, + 1257.0, + 1161.0, + 1274.0, + 1144.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1260.0, + 1267.0, + 1260.0, + 1267.0, + 1276.0, + 1250.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1257.0, + 1293.0, + 1257.0, + 1293.0, + 1274.0, + 1277.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1258.0, + 1316.0, + 1258.0, + 1316.0, + 1276.0, + 1300.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1261.0, + 1337.0, + 1261.0, + 1337.0, + 1273.0, + 1324.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1261.0, + 1360.0, + 1261.0, + 1360.0, + 1273.0, + 1349.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1276.0, + 330.0, + 1276.0, + 330.0, + 1298.0, + 308.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1279.0, + 354.0, + 1279.0, + 354.0, + 1299.0, + 338.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1276.0, + 379.0, + 1276.0, + 379.0, + 1299.0, + 356.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1279.0, + 399.0, + 1279.0, + 399.0, + 1296.0, + 382.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1273.0, + 425.0, + 1273.0, + 425.0, + 1298.0, + 403.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1279.0, + 506.0, + 1279.0, + 506.0, + 1296.0, + 489.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1283.0, + 531.0, + 1283.0, + 531.0, + 1296.0, + 519.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1280.0, + 555.0, + 1280.0, + 555.0, + 1296.0, + 538.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1279.0, + 578.0, + 1279.0, + 578.0, + 1296.0, + 562.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1280.0, + 598.0, + 1280.0, + 598.0, + 1293.0, + 587.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1276.0, + 688.0, + 1276.0, + 688.0, + 1298.0, + 668.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1281.0, + 711.0, + 1281.0, + 711.0, + 1298.0, + 696.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1277.0, + 739.0, + 1277.0, + 739.0, + 1298.0, + 717.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1277.0, + 758.0, + 1277.0, + 758.0, + 1296.0, + 741.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1276.0, + 781.0, + 1276.0, + 781.0, + 1298.0, + 759.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1279.0, + 863.0, + 1279.0, + 863.0, + 1296.0, + 847.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1276.0, + 892.0, + 1276.0, + 892.0, + 1299.0, + 869.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1280.0, + 914.0, + 1280.0, + 914.0, + 1298.0, + 897.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1276.0, + 940.0, + 1276.0, + 940.0, + 1298.0, + 916.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1281.0, + 954.0, + 1281.0, + 954.0, + 1292.0, + 944.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1279.0, + 1069.0, + 1279.0, + 1069.0, + 1296.0, + 1054.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1279.0, + 1095.0, + 1279.0, + 1095.0, + 1299.0, + 1073.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1280.0, + 1114.0, + 1280.0, + 1114.0, + 1298.0, + 1098.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1279.0, + 1138.0, + 1279.0, + 1138.0, + 1296.0, + 1120.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1281.0, + 1156.0, + 1281.0, + 1156.0, + 1293.0, + 1145.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1279.0, + 1266.0, + 1279.0, + 1266.0, + 1296.0, + 1251.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1277.0, + 1295.0, + 1277.0, + 1295.0, + 1298.0, + 1273.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1281.0, + 1316.0, + 1281.0, + 1316.0, + 1298.0, + 1300.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1280.0, + 1337.0, + 1280.0, + 1337.0, + 1293.0, + 1325.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1366.0, + 325.0, + 1366.0, + 325.0, + 1379.0, + 314.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1363.0, + 355.0, + 1363.0, + 355.0, + 1380.0, + 338.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1360.0, + 378.0, + 1360.0, + 378.0, + 1382.0, + 357.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1364.0, + 399.0, + 1364.0, + 399.0, + 1380.0, + 383.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1363.0, + 422.0, + 1363.0, + 422.0, + 1378.0, + 405.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1364.0, + 506.0, + 1364.0, + 506.0, + 1380.0, + 489.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1363.0, + 533.0, + 1363.0, + 533.0, + 1380.0, + 516.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1363.0, + 555.0, + 1363.0, + 555.0, + 1380.0, + 538.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1364.0, + 577.0, + 1364.0, + 577.0, + 1380.0, + 562.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1364.0, + 598.0, + 1364.0, + 598.0, + 1376.0, + 587.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1364.0, + 684.0, + 1364.0, + 684.0, + 1380.0, + 669.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1364.0, + 713.0, + 1364.0, + 713.0, + 1380.0, + 696.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1363.0, + 733.0, + 1363.0, + 733.0, + 1380.0, + 718.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1364.0, + 758.0, + 1364.0, + 758.0, + 1380.0, + 740.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1366.0, + 776.0, + 1366.0, + 776.0, + 1378.0, + 764.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1364.0, + 863.0, + 1364.0, + 863.0, + 1380.0, + 847.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1363.0, + 891.0, + 1363.0, + 891.0, + 1380.0, + 875.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1363.0, + 913.0, + 1363.0, + 913.0, + 1380.0, + 897.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1364.0, + 936.0, + 1364.0, + 936.0, + 1380.0, + 918.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1366.0, + 954.0, + 1366.0, + 954.0, + 1376.0, + 944.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1364.0, + 1071.0, + 1364.0, + 1071.0, + 1380.0, + 1054.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1364.0, + 1092.0, + 1364.0, + 1092.0, + 1380.0, + 1076.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1363.0, + 1114.0, + 1363.0, + 1114.0, + 1380.0, + 1098.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1364.0, + 1136.0, + 1364.0, + 1136.0, + 1380.0, + 1120.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1364.0, + 1156.0, + 1364.0, + 1156.0, + 1376.0, + 1145.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1364.0, + 1266.0, + 1364.0, + 1266.0, + 1380.0, + 1250.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1363.0, + 1294.0, + 1363.0, + 1294.0, + 1380.0, + 1278.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1363.0, + 1316.0, + 1363.0, + 1316.0, + 1380.0, + 1300.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1364.0, + 1338.0, + 1364.0, + 1338.0, + 1380.0, + 1322.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1366.0, + 1357.0, + 1366.0, + 1357.0, + 1376.0, + 1347.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1385.0, + 327.0, + 1385.0, + 327.0, + 1402.0, + 311.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1386.0, + 355.0, + 1386.0, + 355.0, + 1402.0, + 338.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1386.0, + 377.0, + 1386.0, + 377.0, + 1402.0, + 359.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1385.0, + 399.0, + 1385.0, + 399.0, + 1401.0, + 383.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1387.0, + 420.0, + 1387.0, + 420.0, + 1400.0, + 407.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1383.0, + 506.0, + 1383.0, + 506.0, + 1401.0, + 489.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1386.0, + 533.0, + 1386.0, + 533.0, + 1402.0, + 516.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1385.0, + 556.0, + 1385.0, + 556.0, + 1401.0, + 538.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1383.0, + 578.0, + 1383.0, + 578.0, + 1401.0, + 560.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1385.0, + 602.0, + 1385.0, + 602.0, + 1401.0, + 584.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1385.0, + 684.0, + 1385.0, + 684.0, + 1401.0, + 669.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1386.0, + 711.0, + 1386.0, + 711.0, + 1402.0, + 696.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1387.0, + 732.0, + 1387.0, + 732.0, + 1400.0, + 719.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1385.0, + 757.0, + 1385.0, + 757.0, + 1401.0, + 741.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1387.0, + 777.0, + 1387.0, + 777.0, + 1400.0, + 764.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1385.0, + 863.0, + 1385.0, + 863.0, + 1402.0, + 848.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1386.0, + 891.0, + 1386.0, + 891.0, + 1402.0, + 875.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1386.0, + 913.0, + 1386.0, + 913.0, + 1402.0, + 896.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1385.0, + 936.0, + 1385.0, + 936.0, + 1401.0, + 919.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1387.0, + 956.0, + 1387.0, + 956.0, + 1400.0, + 944.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1385.0, + 1071.0, + 1385.0, + 1071.0, + 1402.0, + 1054.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1386.0, + 1092.0, + 1386.0, + 1092.0, + 1402.0, + 1074.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1385.0, + 1114.0, + 1385.0, + 1114.0, + 1401.0, + 1098.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1383.0, + 1136.0, + 1383.0, + 1136.0, + 1400.0, + 1121.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1387.0, + 1158.0, + 1387.0, + 1158.0, + 1400.0, + 1145.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1385.0, + 1266.0, + 1385.0, + 1266.0, + 1401.0, + 1250.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1386.0, + 1294.0, + 1386.0, + 1294.0, + 1402.0, + 1278.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1386.0, + 1316.0, + 1386.0, + 1316.0, + 1401.0, + 1299.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1385.0, + 1338.0, + 1385.0, + 1338.0, + 1401.0, + 1322.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1387.0, + 1359.0, + 1387.0, + 1359.0, + 1400.0, + 1347.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1406.0, + 327.0, + 1406.0, + 327.0, + 1424.0, + 311.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1408.0, + 354.0, + 1408.0, + 354.0, + 1424.0, + 338.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1406.0, + 377.0, + 1406.0, + 377.0, + 1424.0, + 359.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1405.0, + 400.0, + 1405.0, + 400.0, + 1423.0, + 382.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1406.0, + 422.0, + 1406.0, + 422.0, + 1423.0, + 404.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1406.0, + 506.0, + 1406.0, + 506.0, + 1424.0, + 491.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1408.0, + 532.0, + 1408.0, + 532.0, + 1424.0, + 516.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1408.0, + 555.0, + 1408.0, + 555.0, + 1423.0, + 538.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1406.0, + 577.0, + 1406.0, + 577.0, + 1423.0, + 562.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1409.0, + 598.0, + 1409.0, + 598.0, + 1421.0, + 585.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1406.0, + 684.0, + 1406.0, + 684.0, + 1424.0, + 669.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1408.0, + 711.0, + 1408.0, + 711.0, + 1424.0, + 696.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1408.0, + 736.0, + 1408.0, + 736.0, + 1424.0, + 718.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1405.0, + 758.0, + 1405.0, + 758.0, + 1423.0, + 741.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1408.0, + 780.0, + 1408.0, + 780.0, + 1423.0, + 762.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1406.0, + 863.0, + 1406.0, + 863.0, + 1424.0, + 847.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 1409.0, + 888.0, + 1409.0, + 888.0, + 1423.0, + 877.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1408.0, + 913.0, + 1408.0, + 913.0, + 1424.0, + 897.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1405.0, + 935.0, + 1405.0, + 935.0, + 1421.0, + 918.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1406.0, + 959.0, + 1406.0, + 959.0, + 1423.0, + 943.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1406.0, + 1071.0, + 1406.0, + 1071.0, + 1424.0, + 1054.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1408.0, + 1092.0, + 1408.0, + 1092.0, + 1424.0, + 1074.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1408.0, + 1114.0, + 1408.0, + 1114.0, + 1424.0, + 1098.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1406.0, + 1136.0, + 1406.0, + 1136.0, + 1423.0, + 1121.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1409.0, + 1158.0, + 1409.0, + 1158.0, + 1421.0, + 1145.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1406.0, + 1266.0, + 1406.0, + 1266.0, + 1424.0, + 1250.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 1409.0, + 1291.0, + 1409.0, + 1291.0, + 1423.0, + 1280.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1408.0, + 1316.0, + 1408.0, + 1316.0, + 1424.0, + 1300.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1408.0, + 1337.0, + 1408.0, + 1337.0, + 1420.0, + 1324.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1409.0, + 1360.0, + 1409.0, + 1360.0, + 1421.0, + 1347.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1429.0, + 325.0, + 1429.0, + 325.0, + 1442.0, + 312.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1425.0, + 354.0, + 1425.0, + 354.0, + 1443.0, + 338.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1425.0, + 377.0, + 1425.0, + 377.0, + 1443.0, + 359.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1429.0, + 398.0, + 1429.0, + 398.0, + 1442.0, + 385.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1429.0, + 503.0, + 1429.0, + 503.0, + 1442.0, + 492.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1427.0, + 532.0, + 1427.0, + 532.0, + 1443.0, + 516.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1425.0, + 558.0, + 1425.0, + 558.0, + 1443.0, + 540.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1428.0, + 578.0, + 1428.0, + 578.0, + 1443.0, + 562.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1424.0, + 602.0, + 1424.0, + 602.0, + 1442.0, + 587.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1429.0, + 683.0, + 1429.0, + 683.0, + 1442.0, + 670.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1425.0, + 711.0, + 1425.0, + 711.0, + 1443.0, + 696.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1427.0, + 733.0, + 1427.0, + 733.0, + 1443.0, + 718.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1429.0, + 755.0, + 1429.0, + 755.0, + 1442.0, + 742.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1431.0, + 776.0, + 1431.0, + 776.0, + 1438.0, + 770.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1428.0, + 863.0, + 1428.0, + 863.0, + 1444.0, + 847.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1425.0, + 890.0, + 1425.0, + 890.0, + 1443.0, + 875.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1425.0, + 913.0, + 1425.0, + 913.0, + 1443.0, + 897.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1428.0, + 936.0, + 1428.0, + 936.0, + 1443.0, + 919.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1428.0, + 957.0, + 1428.0, + 957.0, + 1440.0, + 947.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1428.0, + 1071.0, + 1428.0, + 1071.0, + 1443.0, + 1054.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1425.0, + 1092.0, + 1425.0, + 1092.0, + 1443.0, + 1076.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1425.0, + 1114.0, + 1425.0, + 1114.0, + 1443.0, + 1099.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1427.0, + 1136.0, + 1427.0, + 1136.0, + 1443.0, + 1121.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1425.0, + 1161.0, + 1425.0, + 1161.0, + 1443.0, + 1145.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1429.0, + 1264.0, + 1429.0, + 1264.0, + 1442.0, + 1251.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1425.0, + 1293.0, + 1425.0, + 1293.0, + 1443.0, + 1278.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1425.0, + 1316.0, + 1425.0, + 1316.0, + 1443.0, + 1300.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1431.0, + 1337.0, + 1431.0, + 1337.0, + 1440.0, + 1325.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1448.0, + 327.0, + 1448.0, + 327.0, + 1465.0, + 311.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1450.0, + 354.0, + 1450.0, + 354.0, + 1467.0, + 338.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1446.0, + 381.0, + 1446.0, + 381.0, + 1467.0, + 356.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1448.0, + 399.0, + 1448.0, + 399.0, + 1465.0, + 383.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1448.0, + 506.0, + 1448.0, + 506.0, + 1465.0, + 491.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1451.0, + 532.0, + 1451.0, + 532.0, + 1467.0, + 516.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1450.0, + 555.0, + 1450.0, + 555.0, + 1466.0, + 537.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1448.0, + 577.0, + 1448.0, + 577.0, + 1465.0, + 562.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1448.0, + 684.0, + 1448.0, + 684.0, + 1465.0, + 669.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1450.0, + 711.0, + 1450.0, + 711.0, + 1467.0, + 696.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1451.0, + 736.0, + 1451.0, + 736.0, + 1466.0, + 718.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1448.0, + 758.0, + 1448.0, + 758.0, + 1465.0, + 741.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1448.0, + 864.0, + 1448.0, + 864.0, + 1465.0, + 847.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1450.0, + 890.0, + 1450.0, + 890.0, + 1467.0, + 875.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1448.0, + 914.0, + 1448.0, + 914.0, + 1467.0, + 897.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1444.0, + 939.0, + 1444.0, + 939.0, + 1466.0, + 916.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1448.0, + 1071.0, + 1448.0, + 1071.0, + 1465.0, + 1054.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1450.0, + 1091.0, + 1450.0, + 1091.0, + 1467.0, + 1076.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1451.0, + 1114.0, + 1451.0, + 1114.0, + 1467.0, + 1098.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1448.0, + 1136.0, + 1448.0, + 1136.0, + 1465.0, + 1120.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1448.0, + 1266.0, + 1448.0, + 1266.0, + 1465.0, + 1250.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1448.0, + 1293.0, + 1448.0, + 1293.0, + 1467.0, + 1278.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1450.0, + 1316.0, + 1450.0, + 1316.0, + 1467.0, + 1300.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1451.0, + 1337.0, + 1451.0, + 1337.0, + 1462.0, + 1324.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 776.0, + 326.0, + 776.0, + 326.0, + 794.0, + 311.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 776.0, + 354.0, + 776.0, + 354.0, + 793.0, + 339.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 776.0, + 378.0, + 776.0, + 378.0, + 793.0, + 362.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 778.0, + 399.0, + 778.0, + 399.0, + 793.0, + 383.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 778.0, + 418.0, + 778.0, + 418.0, + 790.0, + 407.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 778.0, + 506.0, + 778.0, + 506.0, + 794.0, + 490.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 776.0, + 534.0, + 776.0, + 534.0, + 793.0, + 518.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 776.0, + 556.0, + 776.0, + 556.0, + 793.0, + 539.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 778.0, + 579.0, + 778.0, + 579.0, + 793.0, + 560.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 778.0, + 597.0, + 778.0, + 597.0, + 790.0, + 586.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 778.0, + 685.0, + 778.0, + 685.0, + 793.0, + 669.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 778.0, + 714.0, + 778.0, + 714.0, + 793.0, + 696.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 778.0, + 734.0, + 778.0, + 734.0, + 793.0, + 715.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 778.0, + 758.0, + 778.0, + 758.0, + 793.0, + 738.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 776.0, + 777.0, + 776.0, + 777.0, + 791.0, + 761.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 778.0, + 862.0, + 778.0, + 862.0, + 794.0, + 849.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 776.0, + 890.0, + 776.0, + 890.0, + 793.0, + 874.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 776.0, + 913.0, + 776.0, + 913.0, + 793.0, + 897.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 776.0, + 936.0, + 776.0, + 936.0, + 793.0, + 921.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 778.0, + 955.0, + 778.0, + 955.0, + 788.0, + 945.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 778.0, + 1048.0, + 778.0, + 1048.0, + 793.0, + 1033.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 775.0, + 1073.0, + 775.0, + 1073.0, + 794.0, + 1052.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 778.0, + 1093.0, + 778.0, + 1093.0, + 793.0, + 1075.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 778.0, + 1114.0, + 778.0, + 1114.0, + 793.0, + 1096.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 778.0, + 1134.0, + 778.0, + 1134.0, + 790.0, + 1123.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 778.0, + 1221.0, + 778.0, + 1221.0, + 793.0, + 1207.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 776.0, + 1249.0, + 776.0, + 1249.0, + 793.0, + 1233.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 778.0, + 1271.0, + 778.0, + 1271.0, + 793.0, + 1255.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 778.0, + 1294.0, + 778.0, + 1294.0, + 793.0, + 1277.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 779.0, + 1312.0, + 779.0, + 1312.0, + 790.0, + 1303.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 798.0, + 326.0, + 798.0, + 326.0, + 814.0, + 311.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 799.0, + 354.0, + 799.0, + 354.0, + 814.0, + 339.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 800.0, + 374.0, + 800.0, + 374.0, + 812.0, + 363.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 798.0, + 397.0, + 798.0, + 397.0, + 814.0, + 383.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 800.0, + 418.0, + 800.0, + 418.0, + 812.0, + 407.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 796.0, + 506.0, + 796.0, + 506.0, + 814.0, + 490.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 798.0, + 534.0, + 798.0, + 534.0, + 814.0, + 517.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 798.0, + 554.0, + 798.0, + 554.0, + 812.0, + 539.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 796.0, + 577.0, + 796.0, + 577.0, + 814.0, + 562.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 795.0, + 603.0, + 795.0, + 603.0, + 815.0, + 584.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 798.0, + 685.0, + 798.0, + 685.0, + 814.0, + 669.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 799.0, + 711.0, + 799.0, + 711.0, + 814.0, + 696.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 799.0, + 734.0, + 799.0, + 734.0, + 814.0, + 717.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 798.0, + 756.0, + 798.0, + 756.0, + 814.0, + 741.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 798.0, + 778.0, + 798.0, + 778.0, + 814.0, + 764.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 798.0, + 863.0, + 798.0, + 863.0, + 814.0, + 849.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 799.0, + 890.0, + 799.0, + 890.0, + 814.0, + 873.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 799.0, + 912.0, + 799.0, + 912.0, + 814.0, + 897.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 798.0, + 934.0, + 798.0, + 934.0, + 814.0, + 921.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 800.0, + 955.0, + 800.0, + 955.0, + 812.0, + 944.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 798.0, + 1048.0, + 798.0, + 1048.0, + 814.0, + 1033.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 798.0, + 1070.0, + 798.0, + 1070.0, + 814.0, + 1053.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 798.0, + 1092.0, + 798.0, + 1092.0, + 814.0, + 1078.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 795.0, + 1115.0, + 795.0, + 1115.0, + 815.0, + 1096.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 798.0, + 1137.0, + 798.0, + 1137.0, + 814.0, + 1121.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 798.0, + 1221.0, + 798.0, + 1221.0, + 814.0, + 1207.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 799.0, + 1248.0, + 799.0, + 1248.0, + 814.0, + 1233.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 800.0, + 1269.0, + 800.0, + 1269.0, + 812.0, + 1258.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 798.0, + 1292.0, + 798.0, + 1292.0, + 814.0, + 1278.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 801.0, + 1312.0, + 801.0, + 1312.0, + 811.0, + 1303.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 817.0, + 326.0, + 817.0, + 326.0, + 835.0, + 310.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 821.0, + 352.0, + 821.0, + 352.0, + 836.0, + 338.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 820.0, + 374.0, + 820.0, + 374.0, + 832.0, + 363.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 819.0, + 399.0, + 819.0, + 399.0, + 835.0, + 382.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 816.0, + 423.0, + 816.0, + 423.0, + 836.0, + 404.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 819.0, + 506.0, + 819.0, + 506.0, + 835.0, + 490.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 820.0, + 532.0, + 820.0, + 532.0, + 836.0, + 518.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 819.0, + 556.0, + 819.0, + 556.0, + 835.0, + 539.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 820.0, + 577.0, + 820.0, + 577.0, + 835.0, + 562.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 820.0, + 599.0, + 820.0, + 599.0, + 835.0, + 584.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 817.0, + 685.0, + 817.0, + 685.0, + 835.0, + 669.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 820.0, + 711.0, + 820.0, + 711.0, + 836.0, + 696.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 816.0, + 736.0, + 816.0, + 736.0, + 836.0, + 716.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 816.0, + 759.0, + 816.0, + 759.0, + 836.0, + 739.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 816.0, + 782.0, + 816.0, + 782.0, + 837.0, + 760.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 819.0, + 862.0, + 819.0, + 862.0, + 835.0, + 849.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 820.0, + 889.0, + 820.0, + 889.0, + 833.0, + 873.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 819.0, + 913.0, + 819.0, + 913.0, + 835.0, + 897.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 816.0, + 939.0, + 816.0, + 939.0, + 835.0, + 917.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 819.0, + 958.0, + 819.0, + 958.0, + 835.0, + 941.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 819.0, + 1047.0, + 819.0, + 1047.0, + 835.0, + 1033.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 820.0, + 1069.0, + 820.0, + 1069.0, + 835.0, + 1053.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 820.0, + 1092.0, + 820.0, + 1092.0, + 835.0, + 1076.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 819.0, + 1114.0, + 819.0, + 1114.0, + 835.0, + 1098.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 820.0, + 1136.0, + 820.0, + 1136.0, + 835.0, + 1121.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 819.0, + 1221.0, + 819.0, + 1221.0, + 833.0, + 1207.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 822.0, + 1247.0, + 822.0, + 1247.0, + 832.0, + 1235.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 822.0, + 1270.0, + 822.0, + 1270.0, + 833.0, + 1258.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 820.0, + 1293.0, + 820.0, + 1293.0, + 833.0, + 1277.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 821.0, + 1314.0, + 821.0, + 1314.0, + 832.0, + 1303.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 842.0, + 326.0, + 842.0, + 326.0, + 857.0, + 311.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 841.0, + 354.0, + 841.0, + 354.0, + 857.0, + 339.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 841.0, + 376.0, + 841.0, + 376.0, + 857.0, + 362.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 842.0, + 399.0, + 842.0, + 399.0, + 857.0, + 383.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 841.0, + 506.0, + 841.0, + 506.0, + 857.0, + 489.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 841.0, + 532.0, + 841.0, + 532.0, + 857.0, + 517.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 841.0, + 557.0, + 841.0, + 557.0, + 856.0, + 539.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 841.0, + 579.0, + 841.0, + 579.0, + 857.0, + 562.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 837.0, + 604.0, + 837.0, + 604.0, + 857.0, + 585.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 842.0, + 685.0, + 842.0, + 685.0, + 857.0, + 669.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 842.0, + 711.0, + 842.0, + 711.0, + 857.0, + 697.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 841.0, + 734.0, + 841.0, + 734.0, + 857.0, + 717.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 842.0, + 756.0, + 842.0, + 756.0, + 857.0, + 741.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 842.0, + 777.0, + 842.0, + 777.0, + 853.0, + 767.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 842.0, + 863.0, + 842.0, + 863.0, + 857.0, + 849.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 841.0, + 890.0, + 841.0, + 890.0, + 857.0, + 873.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 841.0, + 913.0, + 841.0, + 913.0, + 857.0, + 897.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 842.0, + 934.0, + 842.0, + 934.0, + 857.0, + 919.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 841.0, + 1050.0, + 841.0, + 1050.0, + 857.0, + 1033.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 841.0, + 1070.0, + 841.0, + 1070.0, + 857.0, + 1053.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 841.0, + 1093.0, + 841.0, + 1093.0, + 857.0, + 1076.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 841.0, + 1114.0, + 841.0, + 1114.0, + 856.0, + 1098.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 842.0, + 1136.0, + 842.0, + 1136.0, + 854.0, + 1125.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 842.0, + 1221.0, + 842.0, + 1221.0, + 857.0, + 1207.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 841.0, + 1248.0, + 841.0, + 1248.0, + 856.0, + 1233.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 843.0, + 1269.0, + 843.0, + 1269.0, + 854.0, + 1258.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 844.0, + 1290.0, + 844.0, + 1290.0, + 853.0, + 1281.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 861.0, + 326.0, + 861.0, + 326.0, + 877.0, + 310.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 862.0, + 354.0, + 862.0, + 354.0, + 879.0, + 338.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 859.0, + 380.0, + 859.0, + 380.0, + 879.0, + 360.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 861.0, + 399.0, + 861.0, + 399.0, + 877.0, + 382.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 861.0, + 422.0, + 861.0, + 422.0, + 875.0, + 406.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 861.0, + 506.0, + 861.0, + 506.0, + 877.0, + 490.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 862.0, + 532.0, + 862.0, + 532.0, + 879.0, + 518.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 861.0, + 554.0, + 861.0, + 554.0, + 878.0, + 539.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 861.0, + 577.0, + 861.0, + 577.0, + 877.0, + 562.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 864.0, + 597.0, + 864.0, + 597.0, + 874.0, + 587.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 861.0, + 685.0, + 861.0, + 685.0, + 877.0, + 669.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 862.0, + 711.0, + 862.0, + 711.0, + 878.0, + 696.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 859.0, + 737.0, + 859.0, + 737.0, + 879.0, + 716.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 861.0, + 758.0, + 861.0, + 758.0, + 877.0, + 741.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 862.0, + 776.0, + 862.0, + 776.0, + 872.0, + 765.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 861.0, + 863.0, + 861.0, + 863.0, + 877.0, + 848.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 859.0, + 891.0, + 859.0, + 891.0, + 879.0, + 872.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 859.0, + 915.0, + 859.0, + 915.0, + 880.0, + 896.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 861.0, + 936.0, + 861.0, + 936.0, + 877.0, + 919.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 861.0, + 1047.0, + 861.0, + 1047.0, + 877.0, + 1033.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 862.0, + 1069.0, + 862.0, + 1069.0, + 878.0, + 1053.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 863.0, + 1091.0, + 863.0, + 1091.0, + 879.0, + 1076.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 861.0, + 1114.0, + 861.0, + 1114.0, + 877.0, + 1097.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 864.0, + 1132.0, + 864.0, + 1132.0, + 873.0, + 1123.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 861.0, + 1221.0, + 861.0, + 1221.0, + 877.0, + 1207.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 862.0, + 1248.0, + 862.0, + 1248.0, + 878.0, + 1233.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 864.0, + 1269.0, + 864.0, + 1269.0, + 877.0, + 1258.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 863.0, + 1289.0, + 863.0, + 1289.0, + 873.0, + 1281.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 946.0, + 326.0, + 946.0, + 326.0, + 961.0, + 311.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 945.0, + 355.0, + 945.0, + 355.0, + 961.0, + 338.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 945.0, + 378.0, + 945.0, + 378.0, + 961.0, + 361.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 946.0, + 399.0, + 946.0, + 399.0, + 961.0, + 383.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 946.0, + 421.0, + 946.0, + 421.0, + 956.0, + 407.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 946.0, + 506.0, + 946.0, + 506.0, + 962.0, + 490.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 945.0, + 534.0, + 945.0, + 534.0, + 961.0, + 518.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 945.0, + 554.0, + 945.0, + 554.0, + 961.0, + 539.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 946.0, + 577.0, + 946.0, + 577.0, + 961.0, + 562.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 946.0, + 597.0, + 946.0, + 597.0, + 957.0, + 587.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 946.0, + 685.0, + 946.0, + 685.0, + 962.0, + 669.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 946.0, + 713.0, + 946.0, + 713.0, + 961.0, + 697.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 945.0, + 734.0, + 945.0, + 734.0, + 961.0, + 717.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 946.0, + 756.0, + 946.0, + 756.0, + 961.0, + 739.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 945.0, + 777.0, + 945.0, + 777.0, + 959.0, + 762.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 967.0, + 326.0, + 967.0, + 326.0, + 983.0, + 311.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 968.0, + 354.0, + 968.0, + 354.0, + 983.0, + 339.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 969.0, + 374.0, + 969.0, + 374.0, + 980.0, + 362.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 966.0, + 399.0, + 966.0, + 399.0, + 982.0, + 383.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 969.0, + 418.0, + 969.0, + 418.0, + 980.0, + 407.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 966.0, + 506.0, + 966.0, + 506.0, + 983.0, + 490.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 967.0, + 534.0, + 967.0, + 534.0, + 984.0, + 518.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 967.0, + 554.0, + 967.0, + 554.0, + 982.0, + 540.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 966.0, + 579.0, + 966.0, + 579.0, + 982.0, + 562.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 966.0, + 602.0, + 966.0, + 602.0, + 982.0, + 585.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 967.0, + 685.0, + 967.0, + 685.0, + 983.0, + 669.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 968.0, + 711.0, + 968.0, + 711.0, + 984.0, + 696.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 967.0, + 734.0, + 967.0, + 734.0, + 982.0, + 717.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 966.0, + 756.0, + 966.0, + 756.0, + 982.0, + 741.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 969.0, + 777.0, + 969.0, + 777.0, + 980.0, + 765.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 987.0, + 326.0, + 987.0, + 326.0, + 1004.0, + 311.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 989.0, + 354.0, + 989.0, + 354.0, + 1005.0, + 338.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 989.0, + 378.0, + 989.0, + 378.0, + 1004.0, + 362.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 985.0, + 400.0, + 985.0, + 400.0, + 1005.0, + 380.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 987.0, + 424.0, + 987.0, + 424.0, + 1005.0, + 404.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 987.0, + 506.0, + 987.0, + 506.0, + 1004.0, + 490.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 989.0, + 534.0, + 989.0, + 534.0, + 1005.0, + 518.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 989.0, + 556.0, + 989.0, + 556.0, + 1004.0, + 539.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 988.0, + 579.0, + 988.0, + 579.0, + 1004.0, + 562.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 989.0, + 601.0, + 989.0, + 601.0, + 1004.0, + 581.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 988.0, + 685.0, + 988.0, + 685.0, + 1004.0, + 669.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 989.0, + 713.0, + 989.0, + 713.0, + 1005.0, + 696.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 989.0, + 734.0, + 989.0, + 734.0, + 1004.0, + 717.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 985.0, + 759.0, + 985.0, + 759.0, + 1005.0, + 739.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 987.0, + 782.0, + 987.0, + 782.0, + 1005.0, + 761.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1009.0, + 326.0, + 1009.0, + 326.0, + 1025.0, + 311.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1009.0, + 354.0, + 1009.0, + 354.0, + 1025.0, + 338.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1009.0, + 377.0, + 1009.0, + 377.0, + 1025.0, + 361.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1010.0, + 399.0, + 1010.0, + 399.0, + 1025.0, + 383.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1010.0, + 419.0, + 1010.0, + 419.0, + 1021.0, + 411.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1009.0, + 506.0, + 1009.0, + 506.0, + 1026.0, + 490.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1009.0, + 532.0, + 1009.0, + 532.0, + 1025.0, + 518.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1009.0, + 556.0, + 1009.0, + 556.0, + 1025.0, + 539.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1009.0, + 579.0, + 1009.0, + 579.0, + 1025.0, + 562.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1009.0, + 603.0, + 1009.0, + 603.0, + 1024.0, + 587.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1009.0, + 685.0, + 1009.0, + 685.0, + 1026.0, + 669.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1009.0, + 711.0, + 1009.0, + 711.0, + 1025.0, + 696.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1009.0, + 734.0, + 1009.0, + 734.0, + 1025.0, + 717.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1010.0, + 758.0, + 1010.0, + 758.0, + 1025.0, + 741.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1029.0, + 326.0, + 1029.0, + 326.0, + 1046.0, + 311.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1031.0, + 354.0, + 1031.0, + 354.0, + 1048.0, + 339.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1027.0, + 400.0, + 1027.0, + 400.0, + 1048.0, + 359.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1030.0, + 506.0, + 1030.0, + 506.0, + 1046.0, + 490.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1031.0, + 532.0, + 1031.0, + 532.0, + 1048.0, + 518.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1030.0, + 557.0, + 1030.0, + 557.0, + 1048.0, + 536.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1030.0, + 577.0, + 1030.0, + 577.0, + 1046.0, + 562.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1030.0, + 683.0, + 1030.0, + 683.0, + 1046.0, + 669.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1031.0, + 711.0, + 1031.0, + 711.0, + 1048.0, + 696.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1029.0, + 737.0, + 1029.0, + 737.0, + 1047.0, + 715.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1030.0, + 758.0, + 1030.0, + 758.0, + 1046.0, + 742.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 495.0, + 297.0, + 495.0, + 297.0, + 521.0, + 266.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 491.0, + 387.0, + 491.0, + 387.0, + 522.0, + 354.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 494.0, + 476.0, + 494.0, + 476.0, + 521.0, + 446.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 490.0, + 567.0, + 490.0, + 567.0, + 521.0, + 533.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 494.0, + 656.0, + 494.0, + 656.0, + 521.0, + 624.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 491.0, + 741.0, + 491.0, + 741.0, + 521.0, + 714.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 494.0, + 836.0, + 494.0, + 836.0, + 525.0, + 801.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 489.0, + 926.0, + 489.0, + 926.0, + 523.0, + 891.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 494.0, + 1008.0, + 494.0, + 1008.0, + 520.0, + 984.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 271.0, + 522.0, + 310.0, + 522.0, + 310.0, + 541.0, + 271.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 521.0, + 402.0, + 521.0, + 402.0, + 544.0, + 357.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 521.0, + 512.0, + 521.0, + 512.0, + 542.0, + 448.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 522.0, + 579.0, + 522.0, + 579.0, + 542.0, + 537.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 521.0, + 669.0, + 521.0, + 669.0, + 543.0, + 628.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 522.0, + 758.0, + 522.0, + 758.0, + 541.0, + 716.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 524.0, + 779.0, + 524.0, + 779.0, + 540.0, + 761.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 519.0, + 849.0, + 519.0, + 849.0, + 545.0, + 801.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 521.0, + 939.0, + 521.0, + 939.0, + 544.0, + 894.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 524.0, + 1002.0, + 524.0, + 1002.0, + 539.0, + 986.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 525.0, + 1024.0, + 525.0, + 1024.0, + 539.0, + 1007.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 527.0, + 1045.0, + 527.0, + 1045.0, + 538.0, + 1032.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 272.0, + 543.0, + 309.0, + 543.0, + 309.0, + 562.0, + 272.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 545.0, + 377.0, + 545.0, + 377.0, + 561.0, + 360.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 543.0, + 400.0, + 543.0, + 400.0, + 563.0, + 378.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 547.0, + 464.0, + 547.0, + 464.0, + 559.0, + 451.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 545.0, + 490.0, + 545.0, + 490.0, + 560.0, + 469.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 545.0, + 509.0, + 545.0, + 509.0, + 561.0, + 493.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 543.0, + 558.0, + 543.0, + 558.0, + 563.0, + 538.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 545.0, + 577.0, + 545.0, + 577.0, + 561.0, + 560.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 545.0, + 646.0, + 545.0, + 646.0, + 561.0, + 629.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 543.0, + 668.0, + 543.0, + 668.0, + 563.0, + 648.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 547.0, + 732.0, + 547.0, + 732.0, + 559.0, + 718.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 545.0, + 757.0, + 545.0, + 757.0, + 561.0, + 740.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 547.0, + 776.0, + 547.0, + 776.0, + 559.0, + 763.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 541.0, + 848.0, + 541.0, + 848.0, + 565.0, + 803.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 542.0, + 939.0, + 542.0, + 939.0, + 565.0, + 895.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 549.0, + 1020.0, + 549.0, + 1020.0, + 557.0, + 1012.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 551.0, + 1044.0, + 551.0, + 1044.0, + 557.0, + 1036.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 562.0, + 848.0, + 562.0, + 848.0, + 585.0, + 802.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 563.0, + 937.0, + 563.0, + 937.0, + 583.0, + 892.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 570.0, + 1000.0, + 570.0, + 1000.0, + 577.0, + 990.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 569.0, + 1045.0, + 569.0, + 1045.0, + 580.0, + 1032.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 584.0, + 302.0, + 584.0, + 302.0, + 613.0, + 258.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 584.0, + 391.0, + 584.0, + 391.0, + 613.0, + 338.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 584.0, + 481.0, + 584.0, + 481.0, + 613.0, + 427.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 584.0, + 571.0, + 584.0, + 571.0, + 613.0, + 517.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 584.0, + 660.0, + 584.0, + 660.0, + 613.0, + 607.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 584.0, + 749.0, + 584.0, + 749.0, + 613.0, + 696.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 584.0, + 839.0, + 584.0, + 839.0, + 613.0, + 785.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 584.0, + 929.0, + 584.0, + 929.0, + 613.0, + 875.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 584.0, + 970.0, + 584.0, + 970.0, + 613.0, + 964.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 585.0, + 1056.0, + 585.0, + 1056.0, + 611.0, + 972.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 605.0, + 1053.0, + 605.0, + 1053.0, + 636.0, + 975.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 565.5, + 1028.0, + 565.5, + 1028.0, + 580.5, + 1004.0, + 580.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 208.0, + 1520.0, + 208.0, + 1520.0, + 287.0, + 799.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 189.0, + 658.0, + 853.0, + 658.0, + 853.0, + 694.0, + 189.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 450.0, + 683.0, + 450.0, + 683.0, + 489.0, + 193.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 224.0, + 314.0, + 224.0, + 314.0, + 256.0, + 216.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 226.0, + 481.0, + 226.0, + 481.0, + 255.0, + 388.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 224.0, + 708.0, + 224.0, + 708.0, + 260.0, + 549.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 232.0, + 250.0, + 306.0, + 250.0, + 306.0, + 284.0, + 232.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 252.0, + 473.0, + 252.0, + 473.0, + 282.0, + 401.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 250.0, + 709.0, + 250.0, + 709.0, + 284.0, + 549.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 280.0, + 481.0, + 280.0, + 481.0, + 303.0, + 368.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 302.0, + 482.0, + 302.0, + 482.0, + 325.0, + 371.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 225.0, + 329.0, + 242.0, + 329.0, + 242.0, + 346.0, + 225.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 329.0, + 264.0, + 329.0, + 264.0, + 345.0, + 245.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 270.0, + 331.0, + 283.0, + 331.0, + 283.0, + 344.0, + 270.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 323.0, + 483.0, + 323.0, + 483.0, + 346.0, + 370.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 323.0, + 668.0, + 323.0, + 668.0, + 345.0, + 580.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 224.0, + 348.0, + 264.0, + 348.0, + 264.0, + 368.0, + 224.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 347.0, + 412.0, + 347.0, + 412.0, + 365.0, + 373.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 347.0, + 459.0, + 347.0, + 459.0, + 365.0, + 419.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 349.0, + 479.0, + 349.0, + 479.0, + 360.0, + 464.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 346.0, + 667.0, + 346.0, + 667.0, + 364.0, + 584.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 224.0, + 372.0, + 241.0, + 372.0, + 241.0, + 390.0, + 224.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 372.0, + 265.0, + 372.0, + 265.0, + 387.0, + 244.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 367.0, + 412.0, + 367.0, + 412.0, + 386.0, + 372.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 370.0, + 435.0, + 370.0, + 435.0, + 386.0, + 418.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 369.0, + 458.0, + 369.0, + 458.0, + 385.0, + 438.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 366.0, + 626.0, + 366.0, + 626.0, + 386.0, + 582.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 372.0, + 641.0, + 372.0, + 641.0, + 380.0, + 633.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 369.0, + 666.0, + 369.0, + 666.0, + 382.0, + 652.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 388.0, + 665.0, + 388.0, + 665.0, + 408.0, + 581.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1524.0, + 481.0, + 1524.0, + 481.0, + 1562.0, + 218.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1121.0, + 482.0, + 1121.0, + 482.0, + 1160.0, + 218.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 308.0, + 1021.0, + 308.0, + 1021.0, + 357.0, + 802.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 308.0, + 1516.0, + 308.0, + 1516.0, + 357.0, + 1086.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 352.0, + 1348.0, + 352.0, + 1348.0, + 387.0, + 974.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 711.0, + 480.0, + 711.0, + 480.0, + 750.0, + 216.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1121.0, + 482.0, + 1121.0, + 482.0, + 1160.0, + 218.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 1524.0, + 481.0, + 1524.0, + 481.0, + 1562.0, + 218.0, + 1562.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 25, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 297, + 987, + 1370, + 987, + 1370, + 1738, + 297, + 1738 + ], + "score": 0.961 + }, + { + "category_id": 3, + "poly": [ + 299, + 286, + 1369, + 286, + 1369, + 853, + 299, + 853 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 198, + 1872, + 678, + 1872, + 678, + 1902, + 198, + 1902 + ], + "score": 0.761 + }, + { + "category_id": 2, + "poly": [ + 222, + 179, + 706, + 179, + 706, + 207, + 222, + 207 + ], + "score": 0.749 + }, + { + "category_id": 1, + "poly": [ + 201, + 1807, + 684, + 1807, + 684, + 1837, + 201, + 1837 + ], + "score": 0.713 + }, + { + "category_id": 1, + "poly": [ + 194, + 1937, + 721, + 1937, + 721, + 1967, + 194, + 1967 + ], + "score": 0.575 + }, + { + "category_id": 1, + "poly": [ + 220, + 235, + 480, + 235, + 480, + 262, + 220, + 262 + ], + "score": 0.557 + }, + { + "category_id": 4, + "poly": [ + 221, + 933, + 477, + 933, + 477, + 962, + 221, + 962 + ], + "score": 0.33 + }, + { + "category_id": 1, + "poly": [ + 221, + 933, + 477, + 933, + 477, + 962, + 221, + 962 + ], + "score": 0.116 + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1002.0, + 328.0, + 1002.0, + 328.0, + 1018.0, + 312.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1002.0, + 353.0, + 1002.0, + 353.0, + 1018.0, + 340.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1001.0, + 377.0, + 1001.0, + 377.0, + 1017.0, + 361.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1002.0, + 399.0, + 1002.0, + 399.0, + 1018.0, + 384.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1001.0, + 421.0, + 1001.0, + 421.0, + 1016.0, + 405.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1002.0, + 505.0, + 1002.0, + 505.0, + 1018.0, + 490.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1001.0, + 533.0, + 1001.0, + 533.0, + 1018.0, + 518.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1002.0, + 554.0, + 1002.0, + 554.0, + 1017.0, + 539.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1002.0, + 578.0, + 1002.0, + 578.0, + 1018.0, + 561.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1002.0, + 596.0, + 1002.0, + 596.0, + 1014.0, + 587.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1003.0, + 684.0, + 1003.0, + 684.0, + 1018.0, + 670.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1002.0, + 714.0, + 1002.0, + 714.0, + 1018.0, + 697.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1002.0, + 735.0, + 1002.0, + 735.0, + 1017.0, + 716.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1001.0, + 759.0, + 1001.0, + 759.0, + 1019.0, + 737.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1001.0, + 776.0, + 1001.0, + 776.0, + 1017.0, + 763.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1003.0, + 863.0, + 1003.0, + 863.0, + 1018.0, + 848.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1002.0, + 892.0, + 1002.0, + 892.0, + 1017.0, + 875.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1002.0, + 914.0, + 1002.0, + 914.0, + 1017.0, + 896.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1001.0, + 940.0, + 1001.0, + 940.0, + 1019.0, + 917.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1003.0, + 954.0, + 1003.0, + 954.0, + 1014.0, + 942.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1003.0, + 1042.0, + 1003.0, + 1042.0, + 1018.0, + 1028.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1001.0, + 1071.0, + 1001.0, + 1071.0, + 1018.0, + 1055.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1002.0, + 1091.0, + 1002.0, + 1091.0, + 1017.0, + 1077.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1001.0, + 1115.0, + 1001.0, + 1115.0, + 1018.0, + 1099.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1003.0, + 1134.0, + 1003.0, + 1134.0, + 1014.0, + 1124.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1003.0, + 1221.0, + 1003.0, + 1221.0, + 1018.0, + 1206.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1001.0, + 1248.0, + 1001.0, + 1248.0, + 1017.0, + 1233.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1001.0, + 1271.0, + 1001.0, + 1271.0, + 1017.0, + 1255.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1002.0, + 1293.0, + 1002.0, + 1293.0, + 1018.0, + 1276.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1002.0, + 1312.0, + 1002.0, + 1312.0, + 1014.0, + 1303.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1022.0, + 326.0, + 1022.0, + 326.0, + 1039.0, + 312.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1024.0, + 353.0, + 1024.0, + 353.0, + 1039.0, + 339.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1024.0, + 375.0, + 1024.0, + 375.0, + 1039.0, + 361.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1024.0, + 396.0, + 1024.0, + 396.0, + 1036.0, + 385.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1025.0, + 418.0, + 1025.0, + 418.0, + 1036.0, + 406.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1022.0, + 505.0, + 1022.0, + 505.0, + 1037.0, + 490.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1023.0, + 533.0, + 1023.0, + 533.0, + 1040.0, + 517.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1022.0, + 556.0, + 1022.0, + 556.0, + 1037.0, + 539.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1019.0, + 579.0, + 1019.0, + 579.0, + 1039.0, + 560.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1020.0, + 604.0, + 1020.0, + 604.0, + 1039.0, + 583.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1023.0, + 684.0, + 1023.0, + 684.0, + 1037.0, + 669.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1024.0, + 711.0, + 1024.0, + 711.0, + 1039.0, + 697.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1024.0, + 735.0, + 1024.0, + 735.0, + 1039.0, + 717.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1023.0, + 757.0, + 1023.0, + 757.0, + 1037.0, + 741.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1023.0, + 779.0, + 1023.0, + 779.0, + 1039.0, + 763.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1023.0, + 863.0, + 1023.0, + 863.0, + 1039.0, + 847.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1024.0, + 891.0, + 1024.0, + 891.0, + 1039.0, + 875.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1024.0, + 914.0, + 1024.0, + 914.0, + 1039.0, + 897.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1023.0, + 935.0, + 1023.0, + 935.0, + 1037.0, + 919.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1025.0, + 956.0, + 1025.0, + 956.0, + 1036.0, + 943.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1023.0, + 1042.0, + 1023.0, + 1042.0, + 1039.0, + 1028.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1024.0, + 1071.0, + 1024.0, + 1071.0, + 1039.0, + 1053.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1024.0, + 1091.0, + 1024.0, + 1091.0, + 1039.0, + 1077.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1023.0, + 1113.0, + 1023.0, + 1113.0, + 1037.0, + 1099.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1025.0, + 1134.0, + 1025.0, + 1134.0, + 1036.0, + 1123.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1023.0, + 1221.0, + 1023.0, + 1221.0, + 1039.0, + 1206.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1023.0, + 1249.0, + 1023.0, + 1249.0, + 1039.0, + 1232.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1023.0, + 1271.0, + 1023.0, + 1271.0, + 1039.0, + 1255.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1022.0, + 1293.0, + 1022.0, + 1293.0, + 1037.0, + 1278.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1024.0, + 1315.0, + 1024.0, + 1315.0, + 1037.0, + 1299.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1041.0, + 328.0, + 1041.0, + 328.0, + 1062.0, + 309.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1045.0, + 353.0, + 1045.0, + 353.0, + 1060.0, + 339.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1042.0, + 379.0, + 1042.0, + 379.0, + 1062.0, + 359.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1042.0, + 401.0, + 1042.0, + 401.0, + 1062.0, + 381.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1043.0, + 423.0, + 1043.0, + 423.0, + 1062.0, + 402.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1045.0, + 505.0, + 1045.0, + 505.0, + 1059.0, + 490.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1045.0, + 533.0, + 1045.0, + 533.0, + 1060.0, + 517.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1045.0, + 555.0, + 1045.0, + 555.0, + 1060.0, + 539.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1045.0, + 578.0, + 1045.0, + 578.0, + 1059.0, + 561.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1045.0, + 600.0, + 1045.0, + 600.0, + 1060.0, + 584.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1042.0, + 687.0, + 1042.0, + 687.0, + 1062.0, + 667.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1045.0, + 711.0, + 1045.0, + 711.0, + 1060.0, + 694.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1045.0, + 735.0, + 1045.0, + 735.0, + 1060.0, + 717.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1041.0, + 759.0, + 1041.0, + 759.0, + 1062.0, + 738.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1041.0, + 781.0, + 1041.0, + 781.0, + 1062.0, + 760.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1043.0, + 863.0, + 1043.0, + 863.0, + 1059.0, + 847.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1045.0, + 891.0, + 1045.0, + 891.0, + 1060.0, + 875.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1045.0, + 914.0, + 1045.0, + 914.0, + 1060.0, + 897.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1045.0, + 935.0, + 1045.0, + 935.0, + 1059.0, + 919.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1045.0, + 958.0, + 1045.0, + 958.0, + 1060.0, + 941.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1045.0, + 1042.0, + 1045.0, + 1042.0, + 1060.0, + 1028.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1045.0, + 1071.0, + 1045.0, + 1071.0, + 1060.0, + 1055.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1045.0, + 1093.0, + 1045.0, + 1093.0, + 1060.0, + 1075.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1041.0, + 1117.0, + 1041.0, + 1117.0, + 1060.0, + 1096.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1042.0, + 1139.0, + 1042.0, + 1139.0, + 1062.0, + 1119.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1045.0, + 1221.0, + 1045.0, + 1221.0, + 1059.0, + 1206.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1045.0, + 1248.0, + 1045.0, + 1248.0, + 1060.0, + 1232.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1045.0, + 1271.0, + 1045.0, + 1271.0, + 1060.0, + 1255.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1045.0, + 1293.0, + 1045.0, + 1293.0, + 1059.0, + 1278.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1045.0, + 1315.0, + 1045.0, + 1315.0, + 1060.0, + 1300.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1065.0, + 326.0, + 1065.0, + 326.0, + 1081.0, + 312.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1065.0, + 353.0, + 1065.0, + 353.0, + 1081.0, + 339.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1064.0, + 377.0, + 1064.0, + 377.0, + 1081.0, + 361.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1066.0, + 399.0, + 1066.0, + 399.0, + 1081.0, + 383.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1065.0, + 421.0, + 1065.0, + 421.0, + 1078.0, + 411.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1063.0, + 512.0, + 1063.0, + 512.0, + 1082.0, + 491.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1064.0, + 533.0, + 1064.0, + 533.0, + 1081.0, + 517.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1063.0, + 559.0, + 1063.0, + 559.0, + 1082.0, + 538.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1064.0, + 578.0, + 1064.0, + 578.0, + 1081.0, + 562.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1062.0, + 604.0, + 1062.0, + 604.0, + 1082.0, + 585.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1066.0, + 691.0, + 1066.0, + 691.0, + 1081.0, + 673.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1065.0, + 710.0, + 1065.0, + 710.0, + 1081.0, + 697.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1063.0, + 736.0, + 1063.0, + 736.0, + 1082.0, + 716.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1066.0, + 758.0, + 1066.0, + 758.0, + 1081.0, + 742.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1066.0, + 777.0, + 1066.0, + 777.0, + 1077.0, + 768.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1063.0, + 873.0, + 1063.0, + 873.0, + 1082.0, + 849.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1064.0, + 890.0, + 1064.0, + 890.0, + 1081.0, + 874.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1063.0, + 917.0, + 1063.0, + 917.0, + 1082.0, + 897.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1065.0, + 935.0, + 1065.0, + 935.0, + 1081.0, + 919.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1064.0, + 959.0, + 1064.0, + 959.0, + 1080.0, + 943.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1066.0, + 1042.0, + 1066.0, + 1042.0, + 1081.0, + 1028.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1064.0, + 1071.0, + 1064.0, + 1071.0, + 1081.0, + 1055.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1065.0, + 1091.0, + 1065.0, + 1091.0, + 1080.0, + 1077.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1065.0, + 1115.0, + 1065.0, + 1115.0, + 1081.0, + 1097.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1066.0, + 1134.0, + 1066.0, + 1134.0, + 1077.0, + 1124.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1066.0, + 1221.0, + 1066.0, + 1221.0, + 1081.0, + 1206.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1063.0, + 1250.0, + 1063.0, + 1250.0, + 1082.0, + 1229.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1064.0, + 1271.0, + 1064.0, + 1271.0, + 1081.0, + 1255.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1065.0, + 1293.0, + 1065.0, + 1293.0, + 1080.0, + 1277.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1065.0, + 1312.0, + 1065.0, + 1312.0, + 1076.0, + 1303.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1082.0, + 328.0, + 1082.0, + 328.0, + 1103.0, + 309.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1088.0, + 353.0, + 1088.0, + 353.0, + 1104.0, + 339.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1082.0, + 379.0, + 1082.0, + 379.0, + 1104.0, + 359.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1084.0, + 399.0, + 1084.0, + 399.0, + 1101.0, + 383.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1082.0, + 423.0, + 1082.0, + 423.0, + 1101.0, + 403.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1086.0, + 505.0, + 1086.0, + 505.0, + 1101.0, + 489.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1088.0, + 533.0, + 1088.0, + 533.0, + 1105.0, + 517.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1087.0, + 555.0, + 1087.0, + 555.0, + 1103.0, + 539.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1086.0, + 577.0, + 1086.0, + 577.0, + 1101.0, + 562.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1088.0, + 596.0, + 1088.0, + 596.0, + 1099.0, + 587.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1082.0, + 687.0, + 1082.0, + 687.0, + 1103.0, + 667.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1088.0, + 711.0, + 1088.0, + 711.0, + 1104.0, + 695.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1087.0, + 736.0, + 1087.0, + 736.0, + 1101.0, + 719.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1083.0, + 759.0, + 1083.0, + 759.0, + 1103.0, + 739.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1082.0, + 780.0, + 1082.0, + 780.0, + 1101.0, + 760.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1086.0, + 863.0, + 1086.0, + 863.0, + 1101.0, + 847.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1088.0, + 890.0, + 1088.0, + 890.0, + 1104.0, + 875.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1087.0, + 913.0, + 1087.0, + 913.0, + 1103.0, + 897.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1086.0, + 935.0, + 1086.0, + 935.0, + 1101.0, + 920.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1088.0, + 954.0, + 1088.0, + 954.0, + 1099.0, + 943.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1086.0, + 1042.0, + 1086.0, + 1042.0, + 1101.0, + 1028.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1084.0, + 1071.0, + 1084.0, + 1071.0, + 1105.0, + 1051.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1088.0, + 1091.0, + 1088.0, + 1091.0, + 1104.0, + 1077.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1082.0, + 1117.0, + 1082.0, + 1117.0, + 1103.0, + 1095.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1091.0, + 1132.0, + 1091.0, + 1132.0, + 1097.0, + 1126.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1086.0, + 1221.0, + 1086.0, + 1221.0, + 1101.0, + 1206.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1087.0, + 1248.0, + 1087.0, + 1248.0, + 1104.0, + 1232.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1088.0, + 1271.0, + 1088.0, + 1271.0, + 1104.0, + 1255.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1086.0, + 1293.0, + 1086.0, + 1293.0, + 1101.0, + 1277.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1192.0, + 326.0, + 1192.0, + 326.0, + 1207.0, + 312.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1191.0, + 353.0, + 1191.0, + 353.0, + 1205.0, + 339.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1191.0, + 377.0, + 1191.0, + 377.0, + 1207.0, + 361.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1192.0, + 400.0, + 1192.0, + 400.0, + 1207.0, + 383.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1192.0, + 417.0, + 1192.0, + 417.0, + 1204.0, + 407.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1192.0, + 505.0, + 1192.0, + 505.0, + 1207.0, + 490.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1191.0, + 533.0, + 1191.0, + 533.0, + 1207.0, + 517.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1191.0, + 555.0, + 1191.0, + 555.0, + 1207.0, + 539.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1192.0, + 578.0, + 1192.0, + 578.0, + 1207.0, + 561.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1192.0, + 596.0, + 1192.0, + 596.0, + 1203.0, + 587.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1192.0, + 684.0, + 1192.0, + 684.0, + 1207.0, + 670.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1191.0, + 713.0, + 1191.0, + 713.0, + 1207.0, + 697.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1188.0, + 736.0, + 1188.0, + 736.0, + 1208.0, + 716.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1192.0, + 758.0, + 1192.0, + 758.0, + 1207.0, + 742.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1192.0, + 776.0, + 1192.0, + 776.0, + 1203.0, + 765.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1192.0, + 863.0, + 1192.0, + 863.0, + 1207.0, + 848.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1191.0, + 891.0, + 1191.0, + 891.0, + 1207.0, + 875.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1191.0, + 914.0, + 1191.0, + 914.0, + 1205.0, + 896.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1192.0, + 935.0, + 1192.0, + 935.0, + 1207.0, + 919.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1192.0, + 954.0, + 1192.0, + 954.0, + 1203.0, + 943.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1192.0, + 1042.0, + 1192.0, + 1042.0, + 1207.0, + 1028.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1191.0, + 1071.0, + 1191.0, + 1071.0, + 1207.0, + 1055.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1191.0, + 1093.0, + 1191.0, + 1093.0, + 1207.0, + 1075.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1192.0, + 1115.0, + 1192.0, + 1115.0, + 1207.0, + 1096.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1191.0, + 1135.0, + 1191.0, + 1135.0, + 1207.0, + 1121.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1192.0, + 1221.0, + 1192.0, + 1221.0, + 1207.0, + 1206.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1191.0, + 1248.0, + 1191.0, + 1248.0, + 1207.0, + 1233.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1191.0, + 1271.0, + 1191.0, + 1271.0, + 1205.0, + 1255.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1192.0, + 1294.0, + 1192.0, + 1294.0, + 1207.0, + 1277.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1214.0, + 324.0, + 1214.0, + 324.0, + 1226.0, + 313.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1215.0, + 352.0, + 1215.0, + 352.0, + 1226.0, + 340.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1213.0, + 377.0, + 1213.0, + 377.0, + 1228.0, + 361.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1211.0, + 399.0, + 1211.0, + 399.0, + 1227.0, + 383.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1213.0, + 419.0, + 1213.0, + 419.0, + 1228.0, + 405.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1213.0, + 505.0, + 1213.0, + 505.0, + 1228.0, + 490.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1214.0, + 533.0, + 1214.0, + 533.0, + 1228.0, + 517.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1214.0, + 555.0, + 1214.0, + 555.0, + 1227.0, + 539.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1213.0, + 578.0, + 1213.0, + 578.0, + 1227.0, + 562.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1213.0, + 599.0, + 1213.0, + 599.0, + 1227.0, + 584.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1211.0, + 684.0, + 1211.0, + 684.0, + 1228.0, + 669.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1214.0, + 711.0, + 1214.0, + 711.0, + 1228.0, + 697.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1213.0, + 735.0, + 1213.0, + 735.0, + 1228.0, + 717.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1211.0, + 757.0, + 1211.0, + 757.0, + 1227.0, + 741.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1214.0, + 779.0, + 1214.0, + 779.0, + 1228.0, + 763.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1211.0, + 863.0, + 1211.0, + 863.0, + 1228.0, + 847.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1213.0, + 891.0, + 1213.0, + 891.0, + 1228.0, + 875.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1209.0, + 917.0, + 1209.0, + 917.0, + 1228.0, + 896.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1208.0, + 936.0, + 1208.0, + 936.0, + 1230.0, + 918.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1210.0, + 961.0, + 1210.0, + 961.0, + 1228.0, + 940.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1213.0, + 1042.0, + 1213.0, + 1042.0, + 1228.0, + 1028.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1213.0, + 1069.0, + 1213.0, + 1069.0, + 1228.0, + 1055.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1213.0, + 1093.0, + 1213.0, + 1093.0, + 1227.0, + 1077.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1211.0, + 1115.0, + 1211.0, + 1115.0, + 1227.0, + 1097.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1213.0, + 1135.0, + 1213.0, + 1135.0, + 1227.0, + 1121.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1213.0, + 1221.0, + 1213.0, + 1221.0, + 1228.0, + 1206.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1213.0, + 1248.0, + 1213.0, + 1248.0, + 1228.0, + 1233.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1214.0, + 1271.0, + 1214.0, + 1271.0, + 1228.0, + 1255.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1213.0, + 1293.0, + 1213.0, + 1293.0, + 1227.0, + 1277.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1215.0, + 1312.0, + 1215.0, + 1312.0, + 1226.0, + 1301.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1233.0, + 325.0, + 1233.0, + 325.0, + 1249.0, + 312.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1233.0, + 353.0, + 1233.0, + 353.0, + 1249.0, + 337.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1233.0, + 377.0, + 1233.0, + 377.0, + 1250.0, + 361.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1231.0, + 402.0, + 1231.0, + 402.0, + 1250.0, + 381.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1232.0, + 423.0, + 1232.0, + 423.0, + 1251.0, + 403.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1233.0, + 505.0, + 1233.0, + 505.0, + 1249.0, + 490.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1234.0, + 533.0, + 1234.0, + 533.0, + 1250.0, + 517.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1233.0, + 555.0, + 1233.0, + 555.0, + 1249.0, + 539.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1233.0, + 578.0, + 1233.0, + 578.0, + 1249.0, + 561.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1233.0, + 599.0, + 1233.0, + 599.0, + 1249.0, + 584.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1233.0, + 684.0, + 1233.0, + 684.0, + 1249.0, + 669.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1234.0, + 711.0, + 1234.0, + 711.0, + 1250.0, + 695.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1233.0, + 735.0, + 1233.0, + 735.0, + 1249.0, + 717.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1231.0, + 759.0, + 1231.0, + 759.0, + 1250.0, + 738.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1232.0, + 782.0, + 1232.0, + 782.0, + 1251.0, + 760.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1233.0, + 863.0, + 1233.0, + 863.0, + 1249.0, + 847.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1233.0, + 914.0, + 1233.0, + 914.0, + 1249.0, + 897.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1233.0, + 935.0, + 1233.0, + 935.0, + 1249.0, + 919.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1233.0, + 958.0, + 1233.0, + 958.0, + 1249.0, + 942.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1233.0, + 1042.0, + 1233.0, + 1042.0, + 1249.0, + 1027.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1233.0, + 1069.0, + 1233.0, + 1069.0, + 1250.0, + 1055.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1232.0, + 1096.0, + 1232.0, + 1096.0, + 1250.0, + 1074.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1230.0, + 1116.0, + 1230.0, + 1116.0, + 1251.0, + 1097.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1232.0, + 1141.0, + 1232.0, + 1141.0, + 1251.0, + 1119.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1233.0, + 1221.0, + 1233.0, + 1221.0, + 1249.0, + 1206.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1233.0, + 1248.0, + 1233.0, + 1248.0, + 1249.0, + 1233.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1233.0, + 1271.0, + 1233.0, + 1271.0, + 1250.0, + 1255.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1233.0, + 1292.0, + 1233.0, + 1292.0, + 1249.0, + 1277.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1233.0, + 1315.0, + 1233.0, + 1315.0, + 1249.0, + 1300.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1256.0, + 324.0, + 1256.0, + 324.0, + 1268.0, + 313.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1254.0, + 353.0, + 1254.0, + 353.0, + 1268.0, + 339.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1254.0, + 375.0, + 1254.0, + 375.0, + 1269.0, + 361.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1255.0, + 399.0, + 1255.0, + 399.0, + 1269.0, + 384.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1255.0, + 417.0, + 1255.0, + 417.0, + 1266.0, + 408.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1255.0, + 505.0, + 1255.0, + 505.0, + 1269.0, + 490.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1252.0, + 533.0, + 1252.0, + 533.0, + 1269.0, + 517.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1254.0, + 555.0, + 1254.0, + 555.0, + 1269.0, + 540.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1254.0, + 578.0, + 1254.0, + 578.0, + 1269.0, + 561.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1255.0, + 596.0, + 1255.0, + 596.0, + 1268.0, + 587.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1255.0, + 684.0, + 1255.0, + 684.0, + 1269.0, + 669.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1255.0, + 710.0, + 1255.0, + 710.0, + 1269.0, + 695.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1254.0, + 735.0, + 1254.0, + 735.0, + 1269.0, + 717.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1255.0, + 757.0, + 1255.0, + 757.0, + 1269.0, + 741.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1255.0, + 776.0, + 1255.0, + 776.0, + 1266.0, + 764.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1255.0, + 863.0, + 1255.0, + 863.0, + 1269.0, + 847.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1254.0, + 890.0, + 1254.0, + 890.0, + 1269.0, + 875.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1251.0, + 917.0, + 1251.0, + 917.0, + 1271.0, + 897.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1255.0, + 935.0, + 1255.0, + 935.0, + 1269.0, + 919.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1252.0, + 958.0, + 1252.0, + 958.0, + 1267.0, + 942.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1255.0, + 1042.0, + 1255.0, + 1042.0, + 1269.0, + 1028.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1254.0, + 1069.0, + 1254.0, + 1069.0, + 1269.0, + 1055.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1254.0, + 1091.0, + 1254.0, + 1091.0, + 1269.0, + 1077.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1255.0, + 1115.0, + 1255.0, + 1115.0, + 1269.0, + 1099.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1255.0, + 1134.0, + 1255.0, + 1134.0, + 1266.0, + 1124.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1255.0, + 1221.0, + 1255.0, + 1221.0, + 1269.0, + 1206.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1254.0, + 1248.0, + 1254.0, + 1248.0, + 1269.0, + 1233.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1251.0, + 1273.0, + 1251.0, + 1273.0, + 1271.0, + 1254.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1255.0, + 1294.0, + 1255.0, + 1294.0, + 1269.0, + 1278.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1277.0, + 324.0, + 1277.0, + 324.0, + 1289.0, + 313.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1278.0, + 352.0, + 1278.0, + 352.0, + 1290.0, + 342.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1277.0, + 377.0, + 1277.0, + 377.0, + 1292.0, + 361.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1274.0, + 399.0, + 1274.0, + 399.0, + 1290.0, + 383.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1274.0, + 505.0, + 1274.0, + 505.0, + 1291.0, + 490.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1277.0, + 533.0, + 1277.0, + 533.0, + 1291.0, + 517.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1277.0, + 554.0, + 1277.0, + 554.0, + 1292.0, + 539.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1275.0, + 578.0, + 1275.0, + 578.0, + 1290.0, + 561.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1278.0, + 596.0, + 1278.0, + 596.0, + 1288.0, + 587.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1274.0, + 683.0, + 1274.0, + 683.0, + 1291.0, + 669.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1277.0, + 711.0, + 1277.0, + 711.0, + 1292.0, + 697.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1277.0, + 736.0, + 1277.0, + 736.0, + 1291.0, + 717.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1273.0, + 759.0, + 1273.0, + 759.0, + 1292.0, + 739.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1273.0, + 777.0, + 1273.0, + 777.0, + 1289.0, + 761.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1274.0, + 863.0, + 1274.0, + 863.0, + 1291.0, + 847.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1277.0, + 891.0, + 1277.0, + 891.0, + 1292.0, + 876.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1277.0, + 914.0, + 1277.0, + 914.0, + 1291.0, + 897.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1275.0, + 935.0, + 1275.0, + 935.0, + 1290.0, + 920.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1274.0, + 1042.0, + 1274.0, + 1042.0, + 1291.0, + 1028.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1277.0, + 1069.0, + 1277.0, + 1069.0, + 1294.0, + 1055.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1277.0, + 1091.0, + 1277.0, + 1091.0, + 1291.0, + 1075.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1274.0, + 1115.0, + 1274.0, + 1115.0, + 1290.0, + 1099.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1274.0, + 1221.0, + 1274.0, + 1221.0, + 1290.0, + 1206.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1277.0, + 1248.0, + 1277.0, + 1248.0, + 1292.0, + 1233.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1277.0, + 1270.0, + 1277.0, + 1270.0, + 1291.0, + 1255.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1277.0, + 1290.0, + 1277.0, + 1290.0, + 1289.0, + 1279.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1382.0, + 326.0, + 1382.0, + 326.0, + 1396.0, + 312.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1382.0, + 353.0, + 1382.0, + 353.0, + 1396.0, + 339.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1382.0, + 377.0, + 1382.0, + 377.0, + 1396.0, + 361.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1383.0, + 396.0, + 1383.0, + 396.0, + 1394.0, + 386.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1382.0, + 527.0, + 1382.0, + 527.0, + 1396.0, + 512.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1381.0, + 555.0, + 1381.0, + 555.0, + 1396.0, + 540.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1381.0, + 578.0, + 1381.0, + 578.0, + 1396.0, + 562.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1382.0, + 600.0, + 1382.0, + 600.0, + 1396.0, + 583.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1382.0, + 618.0, + 1382.0, + 618.0, + 1394.0, + 607.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1382.0, + 730.0, + 1382.0, + 730.0, + 1398.0, + 714.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1381.0, + 758.0, + 1381.0, + 758.0, + 1396.0, + 741.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1381.0, + 779.0, + 1381.0, + 779.0, + 1396.0, + 763.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1382.0, + 802.0, + 1382.0, + 802.0, + 1396.0, + 783.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1382.0, + 820.0, + 1382.0, + 820.0, + 1394.0, + 810.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1382.0, + 930.0, + 1382.0, + 930.0, + 1398.0, + 915.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1381.0, + 958.0, + 1381.0, + 958.0, + 1396.0, + 942.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1381.0, + 980.0, + 1381.0, + 980.0, + 1396.0, + 964.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1382.0, + 1003.0, + 1382.0, + 1003.0, + 1396.0, + 986.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1402.0, + 326.0, + 1402.0, + 326.0, + 1418.0, + 312.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1402.0, + 353.0, + 1402.0, + 353.0, + 1418.0, + 339.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1402.0, + 377.0, + 1402.0, + 377.0, + 1417.0, + 361.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1404.0, + 396.0, + 1404.0, + 396.0, + 1416.0, + 385.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1402.0, + 419.0, + 1402.0, + 419.0, + 1417.0, + 405.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1401.0, + 528.0, + 1401.0, + 528.0, + 1418.0, + 512.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1402.0, + 555.0, + 1402.0, + 555.0, + 1418.0, + 539.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1402.0, + 578.0, + 1402.0, + 578.0, + 1418.0, + 561.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1400.0, + 604.0, + 1400.0, + 604.0, + 1418.0, + 583.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1402.0, + 622.0, + 1402.0, + 622.0, + 1417.0, + 605.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1401.0, + 730.0, + 1401.0, + 730.0, + 1418.0, + 714.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1402.0, + 758.0, + 1402.0, + 758.0, + 1418.0, + 741.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1402.0, + 779.0, + 1402.0, + 779.0, + 1417.0, + 761.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1402.0, + 802.0, + 1402.0, + 802.0, + 1417.0, + 786.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1402.0, + 823.0, + 1402.0, + 823.0, + 1417.0, + 808.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1402.0, + 930.0, + 1402.0, + 930.0, + 1418.0, + 915.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1404.0, + 958.0, + 1404.0, + 958.0, + 1418.0, + 942.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1402.0, + 979.0, + 1402.0, + 979.0, + 1417.0, + 964.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1402.0, + 1002.0, + 1402.0, + 1002.0, + 1417.0, + 987.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1404.0, + 1022.0, + 1404.0, + 1022.0, + 1416.0, + 1012.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1422.0, + 326.0, + 1422.0, + 326.0, + 1439.0, + 312.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1423.0, + 353.0, + 1423.0, + 353.0, + 1439.0, + 339.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1422.0, + 377.0, + 1422.0, + 377.0, + 1439.0, + 361.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1421.0, + 403.0, + 1421.0, + 403.0, + 1440.0, + 381.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1423.0, + 422.0, + 1423.0, + 422.0, + 1439.0, + 405.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1422.0, + 528.0, + 1422.0, + 528.0, + 1439.0, + 512.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1423.0, + 555.0, + 1423.0, + 555.0, + 1440.0, + 539.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1423.0, + 578.0, + 1423.0, + 578.0, + 1439.0, + 561.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1421.0, + 603.0, + 1421.0, + 603.0, + 1440.0, + 583.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1424.0, + 622.0, + 1424.0, + 622.0, + 1439.0, + 606.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1422.0, + 730.0, + 1422.0, + 730.0, + 1439.0, + 714.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1423.0, + 757.0, + 1423.0, + 757.0, + 1440.0, + 741.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1423.0, + 779.0, + 1423.0, + 779.0, + 1440.0, + 763.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1421.0, + 804.0, + 1421.0, + 804.0, + 1440.0, + 783.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1423.0, + 825.0, + 1423.0, + 825.0, + 1439.0, + 808.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1422.0, + 930.0, + 1422.0, + 930.0, + 1439.0, + 915.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1423.0, + 958.0, + 1423.0, + 958.0, + 1439.0, + 942.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1424.0, + 980.0, + 1424.0, + 980.0, + 1439.0, + 964.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1423.0, + 1002.0, + 1423.0, + 1002.0, + 1439.0, + 987.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1425.0, + 1023.0, + 1425.0, + 1023.0, + 1437.0, + 1012.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1445.0, + 326.0, + 1445.0, + 326.0, + 1460.0, + 312.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1443.0, + 353.0, + 1443.0, + 353.0, + 1462.0, + 339.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1443.0, + 377.0, + 1443.0, + 377.0, + 1460.0, + 361.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1445.0, + 400.0, + 1445.0, + 400.0, + 1459.0, + 383.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1446.0, + 418.0, + 1446.0, + 418.0, + 1457.0, + 408.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1445.0, + 528.0, + 1445.0, + 528.0, + 1460.0, + 512.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1443.0, + 554.0, + 1443.0, + 554.0, + 1460.0, + 539.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1443.0, + 578.0, + 1443.0, + 578.0, + 1460.0, + 562.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1445.0, + 601.0, + 1445.0, + 601.0, + 1459.0, + 584.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1443.0, + 621.0, + 1443.0, + 621.0, + 1458.0, + 606.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1445.0, + 730.0, + 1445.0, + 730.0, + 1460.0, + 714.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1443.0, + 755.0, + 1443.0, + 755.0, + 1460.0, + 742.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1443.0, + 779.0, + 1443.0, + 779.0, + 1460.0, + 763.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1445.0, + 802.0, + 1445.0, + 802.0, + 1459.0, + 785.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1446.0, + 820.0, + 1446.0, + 820.0, + 1456.0, + 810.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1445.0, + 930.0, + 1445.0, + 930.0, + 1460.0, + 915.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1443.0, + 957.0, + 1443.0, + 957.0, + 1460.0, + 942.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1443.0, + 980.0, + 1443.0, + 980.0, + 1460.0, + 965.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1447.0, + 1002.0, + 1447.0, + 1002.0, + 1457.0, + 987.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1447.0, + 1022.0, + 1447.0, + 1022.0, + 1457.0, + 1012.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1465.0, + 326.0, + 1465.0, + 326.0, + 1481.0, + 312.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1466.0, + 353.0, + 1466.0, + 353.0, + 1483.0, + 339.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1466.0, + 377.0, + 1466.0, + 377.0, + 1482.0, + 362.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1463.0, + 402.0, + 1463.0, + 402.0, + 1482.0, + 381.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1465.0, + 527.0, + 1465.0, + 527.0, + 1481.0, + 512.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1466.0, + 554.0, + 1466.0, + 554.0, + 1482.0, + 539.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1466.0, + 577.0, + 1466.0, + 577.0, + 1482.0, + 562.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1465.0, + 601.0, + 1465.0, + 601.0, + 1481.0, + 584.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1466.0, + 618.0, + 1466.0, + 618.0, + 1477.0, + 609.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1465.0, + 730.0, + 1465.0, + 730.0, + 1481.0, + 714.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1466.0, + 757.0, + 1466.0, + 757.0, + 1483.0, + 742.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1466.0, + 779.0, + 1466.0, + 779.0, + 1482.0, + 764.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1463.0, + 804.0, + 1463.0, + 804.0, + 1482.0, + 785.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1465.0, + 930.0, + 1465.0, + 930.0, + 1481.0, + 915.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1466.0, + 957.0, + 1466.0, + 957.0, + 1482.0, + 943.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1466.0, + 979.0, + 1466.0, + 979.0, + 1482.0, + 965.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1465.0, + 1003.0, + 1465.0, + 1003.0, + 1481.0, + 986.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1470.0, + 1019.0, + 1470.0, + 1019.0, + 1476.0, + 1013.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1572.0, + 348.0, + 1572.0, + 348.0, + 1586.0, + 334.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1570.0, + 375.0, + 1570.0, + 375.0, + 1585.0, + 361.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1570.0, + 397.0, + 1570.0, + 397.0, + 1586.0, + 384.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1573.0, + 419.0, + 1573.0, + 419.0, + 1584.0, + 407.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1570.0, + 572.0, + 1570.0, + 572.0, + 1586.0, + 557.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1570.0, + 600.0, + 1570.0, + 600.0, + 1586.0, + 585.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1570.0, + 622.0, + 1570.0, + 622.0, + 1586.0, + 607.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1572.0, + 644.0, + 1572.0, + 644.0, + 1586.0, + 629.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1570.0, + 665.0, + 1570.0, + 665.0, + 1583.0, + 654.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1572.0, + 796.0, + 1572.0, + 796.0, + 1586.0, + 781.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1570.0, + 823.0, + 1570.0, + 823.0, + 1586.0, + 808.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1570.0, + 846.0, + 1570.0, + 846.0, + 1586.0, + 830.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1572.0, + 869.0, + 1572.0, + 869.0, + 1586.0, + 852.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 1572.0, + 887.0, + 1572.0, + 887.0, + 1583.0, + 877.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 1570.0, + 1020.0, + 1570.0, + 1020.0, + 1586.0, + 1006.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1570.0, + 1047.0, + 1570.0, + 1047.0, + 1586.0, + 1031.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1570.0, + 1069.0, + 1570.0, + 1069.0, + 1585.0, + 1053.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1572.0, + 1093.0, + 1572.0, + 1093.0, + 1586.0, + 1077.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1572.0, + 1111.0, + 1572.0, + 1111.0, + 1583.0, + 1101.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1591.0, + 348.0, + 1591.0, + 348.0, + 1607.0, + 334.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1591.0, + 377.0, + 1591.0, + 377.0, + 1608.0, + 361.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1592.0, + 399.0, + 1592.0, + 399.0, + 1607.0, + 384.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1591.0, + 421.0, + 1591.0, + 421.0, + 1607.0, + 405.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1591.0, + 443.0, + 1591.0, + 443.0, + 1607.0, + 428.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1591.0, + 573.0, + 1591.0, + 573.0, + 1608.0, + 557.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1592.0, + 600.0, + 1592.0, + 600.0, + 1607.0, + 584.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1591.0, + 622.0, + 1591.0, + 622.0, + 1607.0, + 606.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1591.0, + 645.0, + 1591.0, + 645.0, + 1607.0, + 629.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1591.0, + 667.0, + 1591.0, + 667.0, + 1607.0, + 651.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1591.0, + 797.0, + 1591.0, + 797.0, + 1607.0, + 781.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1591.0, + 823.0, + 1591.0, + 823.0, + 1607.0, + 808.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1591.0, + 847.0, + 1591.0, + 847.0, + 1607.0, + 830.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1591.0, + 868.0, + 1591.0, + 868.0, + 1607.0, + 852.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1591.0, + 890.0, + 1591.0, + 890.0, + 1607.0, + 875.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1591.0, + 1020.0, + 1591.0, + 1020.0, + 1608.0, + 1005.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1591.0, + 1047.0, + 1591.0, + 1047.0, + 1607.0, + 1031.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1591.0, + 1071.0, + 1591.0, + 1071.0, + 1607.0, + 1055.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1591.0, + 1091.0, + 1591.0, + 1091.0, + 1607.0, + 1077.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1591.0, + 1113.0, + 1591.0, + 1113.0, + 1607.0, + 1099.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1612.0, + 348.0, + 1612.0, + 348.0, + 1627.0, + 334.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1613.0, + 375.0, + 1613.0, + 375.0, + 1630.0, + 361.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1613.0, + 399.0, + 1613.0, + 399.0, + 1628.0, + 383.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1609.0, + 447.0, + 1609.0, + 447.0, + 1628.0, + 403.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1612.0, + 573.0, + 1612.0, + 573.0, + 1628.0, + 557.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1613.0, + 600.0, + 1613.0, + 600.0, + 1628.0, + 584.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1612.0, + 622.0, + 1612.0, + 622.0, + 1628.0, + 606.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1612.0, + 645.0, + 1612.0, + 645.0, + 1627.0, + 629.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1612.0, + 666.0, + 1612.0, + 666.0, + 1627.0, + 649.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1612.0, + 797.0, + 1612.0, + 797.0, + 1627.0, + 781.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1612.0, + 823.0, + 1612.0, + 823.0, + 1628.0, + 808.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1612.0, + 846.0, + 1612.0, + 846.0, + 1628.0, + 830.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1612.0, + 869.0, + 1612.0, + 869.0, + 1627.0, + 853.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1612.0, + 891.0, + 1612.0, + 891.0, + 1627.0, + 874.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1612.0, + 1020.0, + 1612.0, + 1020.0, + 1628.0, + 1005.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1613.0, + 1047.0, + 1613.0, + 1047.0, + 1628.0, + 1031.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1612.0, + 1071.0, + 1612.0, + 1071.0, + 1628.0, + 1055.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1612.0, + 1091.0, + 1612.0, + 1091.0, + 1627.0, + 1075.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1613.0, + 1113.0, + 1613.0, + 1113.0, + 1627.0, + 1099.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1634.0, + 348.0, + 1634.0, + 348.0, + 1649.0, + 334.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1632.0, + 375.0, + 1632.0, + 375.0, + 1649.0, + 362.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1632.0, + 399.0, + 1632.0, + 399.0, + 1649.0, + 383.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1634.0, + 421.0, + 1634.0, + 421.0, + 1649.0, + 405.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1633.0, + 440.0, + 1633.0, + 440.0, + 1647.0, + 430.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1633.0, + 573.0, + 1633.0, + 573.0, + 1650.0, + 557.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1631.0, + 601.0, + 1631.0, + 601.0, + 1651.0, + 583.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1632.0, + 622.0, + 1632.0, + 622.0, + 1649.0, + 607.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1633.0, + 645.0, + 1633.0, + 645.0, + 1649.0, + 629.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1634.0, + 797.0, + 1634.0, + 797.0, + 1650.0, + 781.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1632.0, + 823.0, + 1632.0, + 823.0, + 1649.0, + 808.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1632.0, + 846.0, + 1632.0, + 846.0, + 1649.0, + 831.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1634.0, + 868.0, + 1634.0, + 868.0, + 1649.0, + 852.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 1636.0, + 887.0, + 1636.0, + 887.0, + 1645.0, + 877.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1633.0, + 1020.0, + 1633.0, + 1020.0, + 1650.0, + 1005.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1632.0, + 1047.0, + 1632.0, + 1047.0, + 1649.0, + 1031.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1632.0, + 1071.0, + 1632.0, + 1071.0, + 1649.0, + 1055.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1634.0, + 1091.0, + 1634.0, + 1091.0, + 1649.0, + 1077.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1631.0, + 1115.0, + 1631.0, + 1115.0, + 1647.0, + 1099.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 1654.0, + 348.0, + 1654.0, + 348.0, + 1670.0, + 333.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1655.0, + 375.0, + 1655.0, + 375.0, + 1672.0, + 361.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1654.0, + 397.0, + 1654.0, + 397.0, + 1672.0, + 383.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1654.0, + 422.0, + 1654.0, + 422.0, + 1670.0, + 405.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1654.0, + 573.0, + 1654.0, + 573.0, + 1670.0, + 557.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1655.0, + 599.0, + 1655.0, + 599.0, + 1672.0, + 585.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1654.0, + 622.0, + 1654.0, + 622.0, + 1672.0, + 607.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1654.0, + 645.0, + 1654.0, + 645.0, + 1670.0, + 628.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1654.0, + 797.0, + 1654.0, + 797.0, + 1670.0, + 781.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1654.0, + 823.0, + 1654.0, + 823.0, + 1672.0, + 808.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1653.0, + 847.0, + 1653.0, + 847.0, + 1673.0, + 829.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1654.0, + 868.0, + 1654.0, + 868.0, + 1670.0, + 852.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1654.0, + 1022.0, + 1654.0, + 1022.0, + 1670.0, + 1005.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1655.0, + 1046.0, + 1655.0, + 1046.0, + 1672.0, + 1031.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1655.0, + 1069.0, + 1655.0, + 1069.0, + 1672.0, + 1055.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1654.0, + 1091.0, + 1654.0, + 1091.0, + 1670.0, + 1075.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.75, + 1234.0, + 893.75, + 1234.0, + 893.75, + 1249.0, + 870.75, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 302.0, + 325.0, + 302.0, + 325.0, + 312.0, + 314.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 299.0, + 354.0, + 299.0, + 354.0, + 313.0, + 338.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 300.0, + 375.0, + 300.0, + 375.0, + 312.0, + 364.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 300.0, + 399.0, + 300.0, + 399.0, + 314.0, + 383.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 297.0, + 420.0, + 297.0, + 420.0, + 313.0, + 406.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 300.0, + 505.0, + 300.0, + 505.0, + 314.0, + 490.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 299.0, + 534.0, + 299.0, + 534.0, + 314.0, + 518.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 299.0, + 556.0, + 299.0, + 556.0, + 314.0, + 539.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 300.0, + 578.0, + 300.0, + 578.0, + 314.0, + 561.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 299.0, + 599.0, + 299.0, + 599.0, + 313.0, + 584.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 300.0, + 684.0, + 300.0, + 684.0, + 314.0, + 669.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 299.0, + 712.0, + 299.0, + 712.0, + 314.0, + 696.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 300.0, + 734.0, + 300.0, + 734.0, + 313.0, + 718.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 300.0, + 757.0, + 300.0, + 757.0, + 314.0, + 740.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 299.0, + 777.0, + 299.0, + 777.0, + 313.0, + 762.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 300.0, + 863.0, + 300.0, + 863.0, + 314.0, + 849.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 299.0, + 891.0, + 299.0, + 891.0, + 314.0, + 875.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 299.0, + 914.0, + 299.0, + 914.0, + 314.0, + 897.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 300.0, + 936.0, + 300.0, + 936.0, + 314.0, + 919.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 300.0, + 953.0, + 300.0, + 953.0, + 312.0, + 944.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 300.0, + 1042.0, + 300.0, + 1042.0, + 314.0, + 1028.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 296.0, + 1070.0, + 296.0, + 1070.0, + 313.0, + 1055.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 299.0, + 1091.0, + 299.0, + 1091.0, + 314.0, + 1077.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 302.0, + 1113.0, + 302.0, + 1113.0, + 312.0, + 1100.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 302.0, + 1134.0, + 302.0, + 1134.0, + 311.0, + 1124.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 300.0, + 1222.0, + 300.0, + 1222.0, + 316.0, + 1206.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 299.0, + 1250.0, + 299.0, + 1250.0, + 314.0, + 1233.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 299.0, + 1272.0, + 299.0, + 1272.0, + 314.0, + 1254.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 300.0, + 1293.0, + 300.0, + 1293.0, + 314.0, + 1275.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 301.0, + 1312.0, + 301.0, + 1312.0, + 312.0, + 1302.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 321.0, + 326.0, + 321.0, + 326.0, + 337.0, + 310.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 322.0, + 355.0, + 322.0, + 355.0, + 338.0, + 339.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 322.0, + 376.0, + 322.0, + 376.0, + 337.0, + 360.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 321.0, + 399.0, + 321.0, + 399.0, + 337.0, + 383.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 321.0, + 420.0, + 321.0, + 420.0, + 337.0, + 406.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 321.0, + 506.0, + 321.0, + 506.0, + 337.0, + 489.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 319.0, + 535.0, + 319.0, + 535.0, + 339.0, + 516.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 321.0, + 555.0, + 321.0, + 555.0, + 337.0, + 538.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 321.0, + 578.0, + 321.0, + 578.0, + 337.0, + 562.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 319.0, + 604.0, + 319.0, + 604.0, + 338.0, + 583.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 321.0, + 684.0, + 321.0, + 684.0, + 337.0, + 669.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 322.0, + 711.0, + 322.0, + 711.0, + 337.0, + 696.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 321.0, + 734.0, + 321.0, + 734.0, + 337.0, + 718.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 321.0, + 757.0, + 321.0, + 757.0, + 337.0, + 740.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 321.0, + 779.0, + 321.0, + 779.0, + 337.0, + 762.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 321.0, + 863.0, + 321.0, + 863.0, + 337.0, + 847.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 322.0, + 891.0, + 322.0, + 891.0, + 337.0, + 874.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 322.0, + 913.0, + 322.0, + 913.0, + 337.0, + 896.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 321.0, + 935.0, + 321.0, + 935.0, + 337.0, + 919.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 321.0, + 957.0, + 321.0, + 957.0, + 335.0, + 940.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 321.0, + 1042.0, + 321.0, + 1042.0, + 338.0, + 1027.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 322.0, + 1070.0, + 322.0, + 1070.0, + 337.0, + 1053.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 322.0, + 1091.0, + 322.0, + 1091.0, + 337.0, + 1077.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 321.0, + 1114.0, + 321.0, + 1114.0, + 337.0, + 1098.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 321.0, + 1135.0, + 321.0, + 1135.0, + 337.0, + 1122.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 321.0, + 1222.0, + 321.0, + 1222.0, + 338.0, + 1206.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 319.0, + 1252.0, + 319.0, + 1252.0, + 338.0, + 1231.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 321.0, + 1272.0, + 321.0, + 1272.0, + 337.0, + 1254.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 321.0, + 1295.0, + 321.0, + 1295.0, + 337.0, + 1278.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 321.0, + 1315.0, + 321.0, + 1315.0, + 337.0, + 1298.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 340.0, + 326.0, + 340.0, + 326.0, + 357.0, + 310.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 341.0, + 354.0, + 341.0, + 354.0, + 357.0, + 338.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 341.0, + 378.0, + 341.0, + 378.0, + 357.0, + 361.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 339.0, + 400.0, + 339.0, + 400.0, + 359.0, + 381.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 339.0, + 425.0, + 339.0, + 425.0, + 359.0, + 405.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 341.0, + 506.0, + 341.0, + 506.0, + 357.0, + 489.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 343.0, + 533.0, + 343.0, + 533.0, + 359.0, + 518.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 340.0, + 557.0, + 340.0, + 557.0, + 359.0, + 537.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 341.0, + 578.0, + 341.0, + 578.0, + 357.0, + 562.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 343.0, + 599.0, + 343.0, + 599.0, + 357.0, + 583.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 339.0, + 687.0, + 339.0, + 687.0, + 359.0, + 667.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 341.0, + 711.0, + 341.0, + 711.0, + 357.0, + 695.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 339.0, + 738.0, + 339.0, + 738.0, + 359.0, + 717.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 339.0, + 783.0, + 339.0, + 783.0, + 360.0, + 739.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 341.0, + 863.0, + 341.0, + 863.0, + 357.0, + 847.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 341.0, + 890.0, + 341.0, + 890.0, + 357.0, + 875.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 343.0, + 913.0, + 343.0, + 913.0, + 357.0, + 897.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 341.0, + 935.0, + 341.0, + 935.0, + 357.0, + 919.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 343.0, + 957.0, + 343.0, + 957.0, + 357.0, + 942.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 341.0, + 1042.0, + 341.0, + 1042.0, + 357.0, + 1028.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 340.0, + 1069.0, + 340.0, + 1069.0, + 357.0, + 1053.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 341.0, + 1092.0, + 341.0, + 1092.0, + 357.0, + 1077.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 339.0, + 1117.0, + 339.0, + 1117.0, + 359.0, + 1096.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 340.0, + 1140.0, + 340.0, + 1140.0, + 359.0, + 1119.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 341.0, + 1222.0, + 341.0, + 1222.0, + 357.0, + 1206.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 341.0, + 1248.0, + 341.0, + 1248.0, + 357.0, + 1231.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 343.0, + 1270.0, + 343.0, + 1270.0, + 359.0, + 1254.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 341.0, + 1293.0, + 341.0, + 1293.0, + 357.0, + 1276.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 343.0, + 1315.0, + 343.0, + 1315.0, + 357.0, + 1301.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 362.0, + 327.0, + 362.0, + 327.0, + 378.0, + 311.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 364.0, + 351.0, + 364.0, + 351.0, + 376.0, + 342.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 362.0, + 376.0, + 362.0, + 376.0, + 377.0, + 361.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 364.0, + 399.0, + 364.0, + 399.0, + 378.0, + 384.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 361.0, + 511.0, + 361.0, + 511.0, + 377.0, + 494.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 361.0, + 533.0, + 361.0, + 533.0, + 378.0, + 518.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 359.0, + 557.0, + 359.0, + 557.0, + 378.0, + 538.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 361.0, + 578.0, + 361.0, + 578.0, + 377.0, + 562.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 357.0, + 605.0, + 357.0, + 605.0, + 378.0, + 585.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 365.0, + 688.0, + 365.0, + 688.0, + 376.0, + 676.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 362.0, + 711.0, + 362.0, + 711.0, + 377.0, + 696.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 361.0, + 734.0, + 361.0, + 734.0, + 377.0, + 718.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 364.0, + 756.0, + 364.0, + 756.0, + 378.0, + 741.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 361.0, + 780.0, + 361.0, + 780.0, + 377.0, + 766.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 361.0, + 868.0, + 361.0, + 868.0, + 377.0, + 852.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 361.0, + 890.0, + 361.0, + 890.0, + 378.0, + 874.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 359.0, + 917.0, + 359.0, + 917.0, + 379.0, + 896.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 362.0, + 935.0, + 362.0, + 935.0, + 378.0, + 919.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 359.0, + 961.0, + 359.0, + 961.0, + 379.0, + 941.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 365.0, + 1041.0, + 365.0, + 1041.0, + 376.0, + 1029.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 361.0, + 1069.0, + 361.0, + 1069.0, + 377.0, + 1055.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 362.0, + 1091.0, + 362.0, + 1091.0, + 377.0, + 1077.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 364.0, + 1114.0, + 364.0, + 1114.0, + 377.0, + 1098.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 361.0, + 1137.0, + 361.0, + 1137.0, + 377.0, + 1123.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 365.0, + 1219.0, + 365.0, + 1219.0, + 376.0, + 1207.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 359.0, + 1251.0, + 359.0, + 1251.0, + 379.0, + 1230.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 361.0, + 1272.0, + 361.0, + 1272.0, + 377.0, + 1256.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 361.0, + 1295.0, + 361.0, + 1295.0, + 377.0, + 1279.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 381.0, + 327.0, + 381.0, + 327.0, + 402.0, + 309.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 384.0, + 354.0, + 384.0, + 354.0, + 402.0, + 338.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 384.0, + 399.0, + 384.0, + 399.0, + 399.0, + 382.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 383.0, + 422.0, + 383.0, + 422.0, + 398.0, + 405.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 384.0, + 505.0, + 384.0, + 505.0, + 399.0, + 490.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 386.0, + 533.0, + 386.0, + 533.0, + 402.0, + 518.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 383.0, + 556.0, + 383.0, + 556.0, + 403.0, + 537.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 384.0, + 577.0, + 384.0, + 577.0, + 399.0, + 562.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 384.0, + 684.0, + 384.0, + 684.0, + 399.0, + 669.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 384.0, + 711.0, + 384.0, + 711.0, + 402.0, + 695.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 384.0, + 735.0, + 384.0, + 735.0, + 400.0, + 718.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 382.0, + 760.0, + 382.0, + 760.0, + 400.0, + 739.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 382.0, + 779.0, + 382.0, + 779.0, + 398.0, + 762.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 384.0, + 862.0, + 384.0, + 862.0, + 399.0, + 847.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 386.0, + 890.0, + 386.0, + 890.0, + 402.0, + 875.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 386.0, + 914.0, + 386.0, + 914.0, + 402.0, + 899.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 384.0, + 935.0, + 384.0, + 935.0, + 399.0, + 919.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 386.0, + 953.0, + 386.0, + 953.0, + 397.0, + 944.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 384.0, + 1042.0, + 384.0, + 1042.0, + 399.0, + 1028.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 383.0, + 1070.0, + 383.0, + 1070.0, + 403.0, + 1050.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 384.0, + 1091.0, + 384.0, + 1091.0, + 402.0, + 1077.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 379.0, + 1118.0, + 379.0, + 1118.0, + 403.0, + 1095.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 386.0, + 1134.0, + 386.0, + 1134.0, + 397.0, + 1124.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 384.0, + 1220.0, + 384.0, + 1220.0, + 399.0, + 1206.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 386.0, + 1247.0, + 386.0, + 1247.0, + 402.0, + 1233.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 386.0, + 1270.0, + 386.0, + 1270.0, + 402.0, + 1256.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 384.0, + 1293.0, + 384.0, + 1293.0, + 399.0, + 1278.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 490.0, + 326.0, + 490.0, + 326.0, + 505.0, + 311.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 489.0, + 354.0, + 489.0, + 354.0, + 505.0, + 338.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 490.0, + 376.0, + 490.0, + 376.0, + 505.0, + 360.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 490.0, + 399.0, + 490.0, + 399.0, + 505.0, + 383.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 490.0, + 417.0, + 490.0, + 417.0, + 501.0, + 407.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 490.0, + 506.0, + 490.0, + 506.0, + 505.0, + 490.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 489.0, + 533.0, + 489.0, + 533.0, + 505.0, + 518.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 489.0, + 555.0, + 489.0, + 555.0, + 505.0, + 539.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 490.0, + 577.0, + 490.0, + 577.0, + 505.0, + 562.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 490.0, + 596.0, + 490.0, + 596.0, + 502.0, + 587.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 490.0, + 684.0, + 490.0, + 684.0, + 505.0, + 669.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 489.0, + 711.0, + 489.0, + 711.0, + 505.0, + 696.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 489.0, + 734.0, + 489.0, + 734.0, + 502.0, + 718.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 490.0, + 757.0, + 490.0, + 757.0, + 505.0, + 743.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 490.0, + 863.0, + 490.0, + 863.0, + 505.0, + 847.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 489.0, + 891.0, + 489.0, + 891.0, + 505.0, + 875.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 487.0, + 916.0, + 487.0, + 916.0, + 506.0, + 895.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 490.0, + 935.0, + 490.0, + 935.0, + 505.0, + 919.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 490.0, + 955.0, + 490.0, + 955.0, + 501.0, + 944.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 490.0, + 1042.0, + 490.0, + 1042.0, + 505.0, + 1028.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 489.0, + 1070.0, + 489.0, + 1070.0, + 505.0, + 1055.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 489.0, + 1092.0, + 489.0, + 1092.0, + 505.0, + 1075.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 490.0, + 1114.0, + 490.0, + 1114.0, + 505.0, + 1097.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 490.0, + 1134.0, + 490.0, + 1134.0, + 502.0, + 1123.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 490.0, + 1222.0, + 490.0, + 1222.0, + 505.0, + 1206.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 489.0, + 1250.0, + 489.0, + 1250.0, + 505.0, + 1233.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 489.0, + 1272.0, + 489.0, + 1272.0, + 505.0, + 1254.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 490.0, + 1293.0, + 490.0, + 1293.0, + 505.0, + 1276.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 509.0, + 326.0, + 509.0, + 326.0, + 525.0, + 311.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 512.0, + 354.0, + 512.0, + 354.0, + 527.0, + 338.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 511.0, + 376.0, + 511.0, + 376.0, + 525.0, + 361.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 509.0, + 398.0, + 509.0, + 398.0, + 525.0, + 382.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 512.0, + 418.0, + 512.0, + 418.0, + 524.0, + 407.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 511.0, + 505.0, + 511.0, + 505.0, + 525.0, + 490.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 512.0, + 533.0, + 512.0, + 533.0, + 527.0, + 518.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 512.0, + 555.0, + 512.0, + 555.0, + 525.0, + 539.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 512.0, + 576.0, + 512.0, + 576.0, + 523.0, + 563.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 511.0, + 599.0, + 511.0, + 599.0, + 525.0, + 584.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 509.0, + 684.0, + 509.0, + 684.0, + 525.0, + 669.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 511.0, + 711.0, + 511.0, + 711.0, + 527.0, + 696.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 512.0, + 734.0, + 512.0, + 734.0, + 525.0, + 718.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 511.0, + 756.0, + 511.0, + 756.0, + 525.0, + 741.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 513.0, + 777.0, + 513.0, + 777.0, + 524.0, + 765.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 509.0, + 863.0, + 509.0, + 863.0, + 527.0, + 847.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 511.0, + 891.0, + 511.0, + 891.0, + 527.0, + 875.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 507.0, + 936.0, + 507.0, + 936.0, + 527.0, + 894.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 506.0, + 961.0, + 506.0, + 961.0, + 525.0, + 941.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 509.0, + 1042.0, + 509.0, + 1042.0, + 527.0, + 1027.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 511.0, + 1070.0, + 511.0, + 1070.0, + 527.0, + 1055.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 511.0, + 1092.0, + 511.0, + 1092.0, + 525.0, + 1075.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 509.0, + 1114.0, + 509.0, + 1114.0, + 525.0, + 1098.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 511.0, + 1136.0, + 511.0, + 1136.0, + 525.0, + 1120.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 509.0, + 1222.0, + 509.0, + 1222.0, + 527.0, + 1206.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 511.0, + 1248.0, + 511.0, + 1248.0, + 527.0, + 1233.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 511.0, + 1272.0, + 511.0, + 1272.0, + 525.0, + 1254.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 512.0, + 1291.0, + 512.0, + 1291.0, + 523.0, + 1279.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 513.0, + 1313.0, + 513.0, + 1313.0, + 523.0, + 1302.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 531.0, + 326.0, + 531.0, + 326.0, + 547.0, + 311.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 531.0, + 354.0, + 531.0, + 354.0, + 547.0, + 338.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 530.0, + 379.0, + 530.0, + 379.0, + 549.0, + 359.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 529.0, + 400.0, + 529.0, + 400.0, + 549.0, + 381.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 530.0, + 422.0, + 530.0, + 422.0, + 547.0, + 406.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 531.0, + 506.0, + 531.0, + 506.0, + 546.0, + 490.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 531.0, + 533.0, + 531.0, + 533.0, + 547.0, + 517.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 531.0, + 555.0, + 531.0, + 555.0, + 547.0, + 539.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 533.0, + 576.0, + 533.0, + 576.0, + 544.0, + 563.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 531.0, + 599.0, + 531.0, + 599.0, + 547.0, + 584.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 531.0, + 684.0, + 531.0, + 684.0, + 547.0, + 669.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 531.0, + 711.0, + 531.0, + 711.0, + 547.0, + 696.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 531.0, + 735.0, + 531.0, + 735.0, + 547.0, + 718.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 529.0, + 760.0, + 529.0, + 760.0, + 549.0, + 740.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 531.0, + 779.0, + 531.0, + 779.0, + 547.0, + 762.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 530.0, + 862.0, + 530.0, + 862.0, + 547.0, + 847.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 531.0, + 891.0, + 531.0, + 891.0, + 547.0, + 875.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 530.0, + 916.0, + 530.0, + 916.0, + 549.0, + 895.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 531.0, + 935.0, + 531.0, + 935.0, + 547.0, + 919.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 531.0, + 957.0, + 531.0, + 957.0, + 547.0, + 942.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 530.0, + 1042.0, + 530.0, + 1042.0, + 547.0, + 1027.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 531.0, + 1070.0, + 531.0, + 1070.0, + 547.0, + 1055.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 528.0, + 1117.0, + 528.0, + 1117.0, + 549.0, + 1074.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 529.0, + 1141.0, + 529.0, + 1141.0, + 549.0, + 1119.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 530.0, + 1222.0, + 530.0, + 1222.0, + 547.0, + 1206.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 531.0, + 1248.0, + 531.0, + 1248.0, + 547.0, + 1233.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 531.0, + 1272.0, + 531.0, + 1272.0, + 547.0, + 1254.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 531.0, + 1293.0, + 531.0, + 1293.0, + 547.0, + 1278.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 531.0, + 1315.0, + 531.0, + 1315.0, + 546.0, + 1301.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 554.0, + 327.0, + 554.0, + 327.0, + 568.0, + 311.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 551.0, + 354.0, + 551.0, + 354.0, + 568.0, + 338.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 551.0, + 376.0, + 551.0, + 376.0, + 568.0, + 361.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 552.0, + 399.0, + 552.0, + 399.0, + 568.0, + 383.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 552.0, + 420.0, + 552.0, + 420.0, + 566.0, + 410.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 552.0, + 506.0, + 552.0, + 506.0, + 568.0, + 490.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 550.0, + 535.0, + 550.0, + 535.0, + 570.0, + 513.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 551.0, + 555.0, + 551.0, + 555.0, + 568.0, + 539.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 554.0, + 578.0, + 554.0, + 578.0, + 567.0, + 562.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 551.0, + 601.0, + 551.0, + 601.0, + 567.0, + 587.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 552.0, + 684.0, + 552.0, + 684.0, + 568.0, + 669.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 552.0, + 711.0, + 552.0, + 711.0, + 568.0, + 696.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 551.0, + 734.0, + 551.0, + 734.0, + 567.0, + 718.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 554.0, + 757.0, + 554.0, + 757.0, + 568.0, + 741.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 554.0, + 778.0, + 554.0, + 778.0, + 565.0, + 768.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 552.0, + 863.0, + 552.0, + 863.0, + 568.0, + 847.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 551.0, + 891.0, + 551.0, + 891.0, + 568.0, + 875.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 550.0, + 917.0, + 550.0, + 917.0, + 570.0, + 896.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 552.0, + 935.0, + 552.0, + 935.0, + 568.0, + 919.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 551.0, + 960.0, + 551.0, + 960.0, + 567.0, + 944.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 552.0, + 1042.0, + 552.0, + 1042.0, + 568.0, + 1027.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 551.0, + 1070.0, + 551.0, + 1070.0, + 568.0, + 1055.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 551.0, + 1092.0, + 551.0, + 1092.0, + 567.0, + 1075.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 552.0, + 1114.0, + 552.0, + 1114.0, + 568.0, + 1098.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 551.0, + 1137.0, + 551.0, + 1137.0, + 566.0, + 1123.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 552.0, + 1222.0, + 552.0, + 1222.0, + 568.0, + 1206.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 551.0, + 1248.0, + 551.0, + 1248.0, + 568.0, + 1233.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 550.0, + 1273.0, + 550.0, + 1273.0, + 570.0, + 1253.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 552.0, + 1295.0, + 552.0, + 1295.0, + 568.0, + 1278.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 573.0, + 326.0, + 573.0, + 326.0, + 588.0, + 311.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 574.0, + 354.0, + 574.0, + 354.0, + 590.0, + 338.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 574.0, + 377.0, + 574.0, + 377.0, + 592.0, + 361.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 570.0, + 400.0, + 570.0, + 400.0, + 590.0, + 381.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 576.0, + 417.0, + 576.0, + 417.0, + 585.0, + 407.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 573.0, + 506.0, + 573.0, + 506.0, + 589.0, + 490.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 574.0, + 533.0, + 574.0, + 533.0, + 590.0, + 517.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 574.0, + 555.0, + 574.0, + 555.0, + 590.0, + 539.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 573.0, + 577.0, + 573.0, + 577.0, + 588.0, + 562.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 573.0, + 684.0, + 573.0, + 684.0, + 589.0, + 669.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 576.0, + 711.0, + 576.0, + 711.0, + 592.0, + 696.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 574.0, + 735.0, + 574.0, + 735.0, + 589.0, + 718.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 571.0, + 760.0, + 571.0, + 760.0, + 590.0, + 740.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 571.0, + 778.0, + 571.0, + 778.0, + 588.0, + 762.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 573.0, + 862.0, + 573.0, + 862.0, + 589.0, + 847.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 576.0, + 891.0, + 576.0, + 891.0, + 590.0, + 877.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 574.0, + 914.0, + 574.0, + 914.0, + 590.0, + 897.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 573.0, + 935.0, + 573.0, + 935.0, + 588.0, + 921.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 576.0, + 952.0, + 576.0, + 952.0, + 585.0, + 944.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 573.0, + 1042.0, + 573.0, + 1042.0, + 589.0, + 1028.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 574.0, + 1069.0, + 574.0, + 1069.0, + 592.0, + 1055.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 572.0, + 1095.0, + 572.0, + 1095.0, + 592.0, + 1074.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 570.0, + 1117.0, + 570.0, + 1117.0, + 590.0, + 1097.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 572.0, + 1222.0, + 572.0, + 1222.0, + 589.0, + 1206.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 576.0, + 1248.0, + 576.0, + 1248.0, + 592.0, + 1233.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 573.0, + 1273.0, + 573.0, + 1273.0, + 592.0, + 1252.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 573.0, + 1293.0, + 573.0, + 1293.0, + 588.0, + 1278.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 680.0, + 325.0, + 680.0, + 325.0, + 692.0, + 314.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 680.0, + 351.0, + 680.0, + 351.0, + 692.0, + 340.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 679.0, + 376.0, + 679.0, + 376.0, + 693.0, + 361.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 681.0, + 398.0, + 681.0, + 398.0, + 691.0, + 386.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 680.0, + 417.0, + 680.0, + 417.0, + 690.0, + 409.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 679.0, + 527.0, + 679.0, + 527.0, + 695.0, + 513.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 679.0, + 555.0, + 679.0, + 555.0, + 693.0, + 540.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 679.0, + 577.0, + 679.0, + 577.0, + 693.0, + 562.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 679.0, + 599.0, + 679.0, + 599.0, + 693.0, + 584.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 680.0, + 618.0, + 680.0, + 618.0, + 690.0, + 610.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 680.0, + 727.0, + 680.0, + 727.0, + 692.0, + 716.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 679.0, + 757.0, + 679.0, + 757.0, + 693.0, + 743.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 679.0, + 779.0, + 679.0, + 779.0, + 693.0, + 763.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 680.0, + 801.0, + 680.0, + 801.0, + 693.0, + 785.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 679.0, + 930.0, + 679.0, + 930.0, + 693.0, + 916.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 679.0, + 957.0, + 679.0, + 957.0, + 693.0, + 942.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 679.0, + 980.0, + 679.0, + 980.0, + 693.0, + 964.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 679.0, + 1002.0, + 679.0, + 1002.0, + 693.0, + 986.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 698.0, + 326.0, + 698.0, + 326.0, + 715.0, + 311.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 699.0, + 354.0, + 699.0, + 354.0, + 715.0, + 339.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 701.0, + 373.0, + 701.0, + 373.0, + 713.0, + 362.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 698.0, + 398.0, + 698.0, + 398.0, + 714.0, + 383.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 701.0, + 417.0, + 701.0, + 417.0, + 713.0, + 407.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 698.0, + 527.0, + 698.0, + 527.0, + 715.0, + 513.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 699.0, + 555.0, + 699.0, + 555.0, + 715.0, + 539.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 699.0, + 577.0, + 699.0, + 577.0, + 714.0, + 562.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 698.0, + 600.0, + 698.0, + 600.0, + 714.0, + 585.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 698.0, + 622.0, + 698.0, + 622.0, + 714.0, + 606.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 698.0, + 729.0, + 698.0, + 729.0, + 715.0, + 715.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 699.0, + 757.0, + 699.0, + 757.0, + 714.0, + 741.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 699.0, + 779.0, + 699.0, + 779.0, + 714.0, + 762.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 699.0, + 801.0, + 699.0, + 801.0, + 714.0, + 786.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 699.0, + 823.0, + 699.0, + 823.0, + 714.0, + 808.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 698.0, + 929.0, + 698.0, + 929.0, + 715.0, + 916.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 699.0, + 958.0, + 699.0, + 958.0, + 714.0, + 942.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 699.0, + 979.0, + 699.0, + 979.0, + 714.0, + 964.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 701.0, + 1001.0, + 701.0, + 1001.0, + 712.0, + 989.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 702.0, + 1022.0, + 702.0, + 1022.0, + 713.0, + 1012.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 722.0, + 326.0, + 722.0, + 326.0, + 737.0, + 311.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 722.0, + 354.0, + 722.0, + 354.0, + 739.0, + 339.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 723.0, + 377.0, + 723.0, + 377.0, + 737.0, + 361.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 719.0, + 399.0, + 719.0, + 399.0, + 735.0, + 383.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 722.0, + 421.0, + 722.0, + 421.0, + 736.0, + 406.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 722.0, + 527.0, + 722.0, + 527.0, + 737.0, + 513.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 724.0, + 552.0, + 724.0, + 552.0, + 736.0, + 541.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 723.0, + 577.0, + 723.0, + 577.0, + 737.0, + 562.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 720.0, + 599.0, + 720.0, + 599.0, + 736.0, + 584.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 723.0, + 619.0, + 723.0, + 619.0, + 735.0, + 610.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 720.0, + 729.0, + 720.0, + 729.0, + 737.0, + 715.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 722.0, + 756.0, + 722.0, + 756.0, + 737.0, + 741.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 720.0, + 779.0, + 720.0, + 779.0, + 737.0, + 763.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 719.0, + 802.0, + 719.0, + 802.0, + 735.0, + 785.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 724.0, + 822.0, + 724.0, + 822.0, + 735.0, + 811.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 720.0, + 930.0, + 720.0, + 930.0, + 737.0, + 916.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 722.0, + 957.0, + 722.0, + 957.0, + 737.0, + 944.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 723.0, + 979.0, + 723.0, + 979.0, + 737.0, + 964.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 723.0, + 1001.0, + 723.0, + 1001.0, + 734.0, + 989.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 724.0, + 1023.0, + 724.0, + 1023.0, + 735.0, + 1012.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 742.0, + 326.0, + 742.0, + 326.0, + 758.0, + 311.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 741.0, + 354.0, + 741.0, + 354.0, + 758.0, + 339.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 744.0, + 373.0, + 744.0, + 373.0, + 757.0, + 364.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 742.0, + 399.0, + 742.0, + 399.0, + 758.0, + 383.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 742.0, + 528.0, + 742.0, + 528.0, + 758.0, + 513.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 742.0, + 554.0, + 742.0, + 554.0, + 758.0, + 540.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 744.0, + 576.0, + 744.0, + 576.0, + 756.0, + 565.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 746.0, + 598.0, + 746.0, + 598.0, + 755.0, + 588.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 742.0, + 729.0, + 742.0, + 729.0, + 758.0, + 715.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 742.0, + 756.0, + 742.0, + 756.0, + 758.0, + 741.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 742.0, + 779.0, + 742.0, + 779.0, + 758.0, + 763.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 744.0, + 801.0, + 744.0, + 801.0, + 758.0, + 786.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 742.0, + 930.0, + 742.0, + 930.0, + 758.0, + 914.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 741.0, + 957.0, + 741.0, + 957.0, + 758.0, + 942.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 744.0, + 978.0, + 744.0, + 978.0, + 756.0, + 967.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 762.0, + 326.0, + 762.0, + 326.0, + 778.0, + 311.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 763.0, + 353.0, + 763.0, + 353.0, + 779.0, + 339.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 763.0, + 376.0, + 763.0, + 376.0, + 779.0, + 362.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 762.0, + 399.0, + 762.0, + 399.0, + 778.0, + 382.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 762.0, + 528.0, + 762.0, + 528.0, + 778.0, + 512.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 763.0, + 554.0, + 763.0, + 554.0, + 779.0, + 540.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 763.0, + 577.0, + 763.0, + 577.0, + 779.0, + 562.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 762.0, + 600.0, + 762.0, + 600.0, + 778.0, + 584.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 762.0, + 729.0, + 762.0, + 729.0, + 778.0, + 715.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 763.0, + 756.0, + 763.0, + 756.0, + 780.0, + 741.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 763.0, + 778.0, + 763.0, + 778.0, + 779.0, + 765.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 762.0, + 801.0, + 762.0, + 801.0, + 777.0, + 786.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 762.0, + 930.0, + 762.0, + 930.0, + 778.0, + 916.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 763.0, + 957.0, + 763.0, + 957.0, + 779.0, + 944.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 763.0, + 979.0, + 763.0, + 979.0, + 779.0, + 966.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.25, + 360.5, + 425.25, + 360.5, + 425.25, + 376.0, + 403.25, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 491.0, + 1313.0, + 491.0, + 1313.0, + 499.0, + 1301.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 761.5, + 1001.0, + 761.5, + 1001.0, + 777.0, + 989.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 219.0, + 177.0, + 707.0, + 177.0, + 707.0, + 211.0, + 219.0, + 211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 927.0, + 479.0, + 927.0, + 479.0, + 967.0, + 220.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 1871.0, + 679.0, + 1871.0, + 679.0, + 1905.0, + 195.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 199.0, + 1804.0, + 686.0, + 1804.0, + 686.0, + 1840.0, + 199.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 1937.0, + 722.0, + 1937.0, + 722.0, + 1968.0, + 196.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 230.0, + 482.0, + 230.0, + 482.0, + 267.0, + 218.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 927.0, + 479.0, + 927.0, + 479.0, + 967.0, + 220.0, + 967.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 26, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 128, + 303, + 1423, + 303, + 1423, + 383, + 128, + 383 + ], + "score": 0.958 + }, + { + "category_id": 0, + "poly": [ + 126, + 210, + 1191, + 210, + 1191, + 268, + 126, + 268 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 160, + 769, + 761, + 769, + 761, + 804, + 160, + 804 + ], + "score": 0.713 + }, + { + "category_id": 1, + "poly": [ + 164, + 1109, + 648, + 1109, + 648, + 1143, + 164, + 1143 + ], + "score": 0.696 + }, + { + "category_id": 3, + "poly": [ + 422, + 840, + 1587, + 840, + 1587, + 1070, + 422, + 1070 + ], + "score": 0.649 + }, + { + "category_id": 3, + "poly": [ + 185, + 475, + 1339, + 475, + 1339, + 740, + 185, + 740 + ], + "score": 0.601 + }, + { + "category_id": 3, + "poly": [ + 134, + 1220, + 1562, + 1220, + 1562, + 2069, + 134, + 2069 + ], + "score": 0.531 + }, + { + "category_id": 5, + "poly": [ + 134, + 1220, + 1562, + 1220, + 1562, + 2069, + 134, + 2069 + ], + "score": 0.443, + "html": "
Padded at top- left cornerPadded at bottom- left cornerPadded at top- right cornerPadded at bottom- right cornerAverage (grouped padding strategy)
Convolutions involving (a)3 22 1212 0.75
2 210.75 0.5
sum=9sum=3sum=3sum =1sum = 4
Convolutions involving (b)2 21 12 21 11.5 1.5
1 11 10.5 0.5
sum=6sum=2sum=6sum=2sum =4
Convolutions involving (c)2 12 1111.5 0.5
2 12 1111.5 0.5
sum=6sum=6sum=2sum=2sum = 4
Convolutions involving (d)1 11 11 11 11 1
1 11 11 11 11 1
sum=4sum=4sum=4sum =4sum = 4
uniformuniformuniformuniformuniform
" + }, + { + "category_id": 5, + "poly": [ + 185, + 475, + 1339, + 475, + 1339, + 740, + 185, + 740 + ], + "score": 0.294, + "html": "
Original InputPadded at top- left cornerPadded at bottom- Padded at top- left cornerright corner bPadded at bottom- right corner
a ba b a baa ba bab
C ddC dC dC d
C
" + }, + { + "category_id": 0, + "poly": [ + 164, + 1109, + 648, + 1109, + 648, + 1143, + 164, + 1143 + ], + "score": 0.152 + }, + { + "category_id": 13, + "poly": [ + 1467, + 1788, + 1497, + 1788, + 1497, + 1808, + 1467, + 1808 + ], + "score": 0.49, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 1467, + 1975, + 1498, + 1975, + 1498, + 1996, + 1467, + 1996 + ], + "score": 0.48, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 1467, + 1415, + 1497, + 1415, + 1497, + 1435, + 1467, + 1435 + ], + "score": 0.46, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 956, + 1975, + 987, + 1975, + 987, + 1996, + 956, + 1996 + ], + "score": 0.42, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 1467, + 1602, + 1497, + 1602, + 1497, + 1622, + 1467, + 1622 + ], + "score": 0.42, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 1189, + 1975, + 1219, + 1975, + 1219, + 1996, + 1189, + 1996 + ], + "score": 0.4, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 491, + 1975, + 521, + 1975, + 521, + 1996, + 491, + 1996 + ], + "score": 0.36, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 956, + 1602, + 987, + 1602, + 987, + 1623, + 956, + 1623 + ], + "score": 0.36, + "latex": "= 6" + }, + { + "category_id": 13, + "poly": [ + 723, + 1975, + 754, + 1975, + 754, + 1996, + 723, + 1996 + ], + "score": 0.34, + "latex": "= 4" + }, + { + "category_id": 13, + "poly": [ + 492, + 1788, + 522, + 1788, + 522, + 1809, + 492, + 1809 + ], + "score": 0.33, + "latex": "= 6" + }, + { + "category_id": 13, + "poly": [ + 956, + 1788, + 986, + 1788, + 986, + 1808, + 956, + 1808 + ], + "score": 0.31, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 492, + 1415, + 522, + 1415, + 522, + 1436, + 492, + 1436 + ], + "score": 0.3, + "latex": "{ } = 9" + }, + { + "category_id": 13, + "poly": [ + 1189, + 1788, + 1219, + 1788, + 1219, + 1809, + 1189, + 1809 + ], + "score": 0.29, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 491, + 1601, + 522, + 1601, + 522, + 1623, + 491, + 1623 + ], + "score": 0.28, + "latex": "= 6" + }, + { + "category_id": 13, + "poly": [ + 345, + 304, + 408, + 304, + 408, + 341, + 345, + 341 + ], + "score": 0.27, + "latex": "\\pmb { 2 \\times 2 }" + }, + { + "category_id": 13, + "poly": [ + 724, + 1788, + 754, + 1788, + 754, + 1809, + 724, + 1809 + ], + "score": 0.26, + "latex": "= 6" + }, + { + "category_id": 13, + "poly": [ + 1189, + 1602, + 1219, + 1602, + 1219, + 1623, + 1189, + 1623 + ], + "score": 0.26, + "latex": "^ { = 2 }" + }, + { + "category_id": 15, + "poly": [ + 124.0, + 202.0, + 1194.0, + 202.0, + 1194.0, + 277.0, + 124.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 845.0, + 596.0, + 845.0, + 596.0, + 883.0, + 428.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 848.0, + 853.0, + 848.0, + 853.0, + 880.0, + 645.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 845.0, + 1054.0, + 845.0, + 1054.0, + 883.0, + 887.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 845.0, + 1322.0, + 845.0, + 1322.0, + 880.0, + 1113.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 847.0, + 1589.0, + 847.0, + 1589.0, + 878.0, + 1386.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 870.0, + 575.0, + 870.0, + 575.0, + 906.0, + 448.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 874.0, + 811.0, + 874.0, + 811.0, + 903.0, + 688.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 875.0, + 1041.0, + 875.0, + 1041.0, + 905.0, + 902.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 874.0, + 1285.0, + 874.0, + 1285.0, + 903.0, + 1147.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1392.0, + 872.0, + 1590.0, + 872.0, + 1590.0, + 905.0, + 1392.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 931.0, + 473.0, + 931.0, + 473.0, + 955.0, + 451.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 932.0, + 520.0, + 932.0, + 520.0, + 954.0, + 497.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 932.0, + 566.0, + 932.0, + 566.0, + 954.0, + 546.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 929.0, + 708.0, + 929.0, + 708.0, + 956.0, + 682.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 932.0, + 753.0, + 932.0, + 753.0, + 954.0, + 730.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 932.0, + 798.0, + 932.0, + 798.0, + 954.0, + 778.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 929.0, + 943.0, + 929.0, + 943.0, + 956.0, + 915.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 932.0, + 987.0, + 932.0, + 987.0, + 954.0, + 962.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 932.0, + 1031.0, + 932.0, + 1031.0, + 954.0, + 1010.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 932.0, + 1172.0, + 932.0, + 1172.0, + 955.0, + 1149.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 931.0, + 1218.0, + 931.0, + 1218.0, + 954.0, + 1196.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 932.0, + 1263.0, + 932.0, + 1263.0, + 954.0, + 1243.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1429.0, + 932.0, + 1451.0, + 932.0, + 1451.0, + 954.0, + 1429.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1474.0, + 932.0, + 1497.0, + 932.0, + 1497.0, + 952.0, + 1474.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1524.0, + 933.0, + 1541.0, + 933.0, + 1541.0, + 951.0, + 1524.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 978.0, + 475.0, + 978.0, + 475.0, + 1001.0, + 452.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 981.0, + 518.0, + 981.0, + 518.0, + 998.0, + 500.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 982.0, + 563.0, + 982.0, + 563.0, + 998.0, + 547.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 977.0, + 707.0, + 977.0, + 707.0, + 1001.0, + 684.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 981.0, + 796.0, + 981.0, + 796.0, + 998.0, + 779.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 976.0, + 941.0, + 976.0, + 941.0, + 1003.0, + 915.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 979.0, + 987.0, + 979.0, + 987.0, + 999.0, + 962.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 978.0, + 1031.0, + 978.0, + 1031.0, + 1000.0, + 1010.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 977.0, + 1172.0, + 977.0, + 1172.0, + 1001.0, + 1149.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 979.0, + 1218.0, + 979.0, + 1218.0, + 999.0, + 1196.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 981.0, + 1261.0, + 981.0, + 1261.0, + 998.0, + 1244.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1429.0, + 978.0, + 1450.0, + 978.0, + 1450.0, + 1000.0, + 1429.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1525.0, + 981.0, + 1541.0, + 981.0, + 1541.0, + 998.0, + 1525.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1026.0, + 472.0, + 1026.0, + 472.0, + 1048.0, + 451.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1028.0, + 518.0, + 1028.0, + 518.0, + 1045.0, + 500.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1028.0, + 564.0, + 1028.0, + 564.0, + 1045.0, + 547.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1026.0, + 705.0, + 1026.0, + 705.0, + 1048.0, + 683.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1028.0, + 750.0, + 1028.0, + 750.0, + 1045.0, + 733.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1028.0, + 796.0, + 1028.0, + 796.0, + 1045.0, + 779.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1026.0, + 939.0, + 1026.0, + 939.0, + 1048.0, + 916.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1027.0, + 986.0, + 1027.0, + 986.0, + 1047.0, + 964.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1028.0, + 1030.0, + 1028.0, + 1030.0, + 1045.0, + 1012.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1026.0, + 1171.0, + 1026.0, + 1171.0, + 1048.0, + 1149.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1028.0, + 1214.0, + 1028.0, + 1214.0, + 1045.0, + 1198.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1028.0, + 1261.0, + 1028.0, + 1261.0, + 1045.0, + 1244.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1431.0, + 1028.0, + 1449.0, + 1028.0, + 1449.0, + 1045.0, + 1431.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1482.0, + 1031.0, + 1492.0, + 1031.0, + 1492.0, + 1043.0, + 1482.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1526.0, + 1031.0, + 1538.0, + 1031.0, + 1538.0, + 1043.0, + 1526.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 978.0, + 748.0, + 978.0, + 748.0, + 999.0, + 735.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 185.0, + 495.0, + 360.0, + 495.0, + 360.0, + 531.0, + 185.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 477.0, + 616.0, + 477.0, + 616.0, + 544.0, + 423.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 477.0, + 1073.0, + 477.0, + 1073.0, + 545.0, + 638.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 485.0, + 1335.0, + 485.0, + 1335.0, + 549.0, + 1103.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 559.0, + 428.0, + 559.0, + 428.0, + 582.0, + 405.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 561.0, + 476.0, + 561.0, + 476.0, + 582.0, + 451.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 558.0, + 519.0, + 558.0, + 519.0, + 581.0, + 497.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 558.0, + 944.0, + 558.0, + 944.0, + 583.0, + 915.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 558.0, + 985.0, + 558.0, + 985.0, + 582.0, + 961.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 607.0, + 241.0, + 607.0, + 241.0, + 628.0, + 220.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 267.0, + 607.0, + 284.0, + 607.0, + 284.0, + 625.0, + 267.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 607.0, + 426.0, + 607.0, + 426.0, + 628.0, + 405.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 607.0, + 475.0, + 607.0, + 475.0, + 628.0, + 451.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 605.0, + 519.0, + 605.0, + 519.0, + 628.0, + 498.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 607.0, + 659.0, + 607.0, + 659.0, + 628.0, + 638.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 609.0, + 703.0, + 609.0, + 703.0, + 627.0, + 686.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 606.0, + 752.0, + 606.0, + 752.0, + 627.0, + 731.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 606.0, + 940.0, + 606.0, + 940.0, + 628.0, + 916.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 605.0, + 984.0, + 605.0, + 984.0, + 628.0, + 963.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 607.0, + 1171.0, + 607.0, + 1171.0, + 628.0, + 1150.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 606.0, + 1215.0, + 606.0, + 1215.0, + 628.0, + 1196.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 223.0, + 654.0, + 238.0, + 654.0, + 238.0, + 672.0, + 223.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 656.0, + 424.0, + 656.0, + 424.0, + 672.0, + 408.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 656.0, + 472.0, + 656.0, + 472.0, + 671.0, + 455.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 654.0, + 516.0, + 654.0, + 516.0, + 671.0, + 502.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 656.0, + 656.0, + 656.0, + 656.0, + 672.0, + 641.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 658.0, + 702.0, + 658.0, + 702.0, + 670.0, + 690.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 654.0, + 937.0, + 654.0, + 937.0, + 672.0, + 920.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 652.0, + 984.0, + 652.0, + 984.0, + 672.0, + 963.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 654.0, + 1168.0, + 654.0, + 1168.0, + 672.0, + 1151.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 654.0, + 1214.0, + 654.0, + 1214.0, + 671.0, + 1198.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1235.0, + 578.0, + 1235.0, + 578.0, + 1277.0, + 408.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 1242.0, + 835.0, + 1242.0, + 835.0, + 1271.0, + 627.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1235.0, + 1036.0, + 1235.0, + 1036.0, + 1277.0, + 866.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1237.0, + 1302.0, + 1237.0, + 1302.0, + 1267.0, + 1093.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 1234.0, + 1570.0, + 1234.0, + 1570.0, + 1271.0, + 1360.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1263.0, + 555.0, + 1263.0, + 555.0, + 1298.0, + 430.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1267.0, + 792.0, + 1267.0, + 792.0, + 1298.0, + 670.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1265.0, + 1025.0, + 1265.0, + 1025.0, + 1301.0, + 881.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1261.0, + 1270.0, + 1261.0, + 1270.0, + 1295.0, + 1127.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 1261.0, + 1567.0, + 1261.0, + 1567.0, + 1296.0, + 1367.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 142.0, + 1313.0, + 412.0, + 1313.0, + 412.0, + 1353.0, + 142.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1320.0, + 476.0, + 1320.0, + 476.0, + 1345.0, + 450.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1321.0, + 520.0, + 1321.0, + 520.0, + 1342.0, + 499.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1321.0, + 705.0, + 1321.0, + 705.0, + 1342.0, + 684.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1321.0, + 751.0, + 1321.0, + 751.0, + 1342.0, + 732.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1321.0, + 939.0, + 1321.0, + 939.0, + 1343.0, + 918.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1321.0, + 1170.0, + 1321.0, + 1170.0, + 1343.0, + 1150.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1429.0, + 1321.0, + 1451.0, + 1321.0, + 1451.0, + 1342.0, + 1429.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1461.0, + 1318.0, + 1510.0, + 1318.0, + 1510.0, + 1346.0, + 1461.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1368.0, + 474.0, + 1368.0, + 474.0, + 1390.0, + 453.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1367.0, + 522.0, + 1367.0, + 522.0, + 1392.0, + 496.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1368.0, + 939.0, + 1368.0, + 939.0, + 1390.0, + 918.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1413.0, + 1364.0, + 1509.0, + 1364.0, + 1509.0, + 1395.0, + 1413.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1411.0, + 491.0, + 1411.0, + 491.0, + 1438.0, + 446.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1411.0, + 527.0, + 1411.0, + 527.0, + 1438.0, + 523.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1410.0, + 760.0, + 1410.0, + 760.0, + 1442.0, + 681.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1411.0, + 993.0, + 1411.0, + 993.0, + 1438.0, + 913.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1410.0, + 1225.0, + 1410.0, + 1225.0, + 1442.0, + 1146.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1412.0, + 1466.0, + 1412.0, + 1466.0, + 1438.0, + 1427.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1498.0, + 1412.0, + 1504.0, + 1412.0, + 1504.0, + 1438.0, + 1498.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 143.0, + 1495.0, + 412.0, + 1495.0, + 412.0, + 1534.0, + 143.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1509.0, + 472.0, + 1509.0, + 472.0, + 1527.0, + 453.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1509.0, + 520.0, + 1509.0, + 520.0, + 1529.0, + 498.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1509.0, + 705.0, + 1509.0, + 705.0, + 1529.0, + 684.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1509.0, + 751.0, + 1509.0, + 751.0, + 1529.0, + 730.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1509.0, + 939.0, + 1509.0, + 939.0, + 1529.0, + 918.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1509.0, + 985.0, + 1509.0, + 985.0, + 1529.0, + 963.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1509.0, + 1171.0, + 1509.0, + 1171.0, + 1529.0, + 1149.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1509.0, + 1216.0, + 1509.0, + 1216.0, + 1529.0, + 1195.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1504.0, + 1507.0, + 1504.0, + 1507.0, + 1532.0, + 1419.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1552.0, + 476.0, + 1552.0, + 476.0, + 1577.0, + 450.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1554.0, + 519.0, + 1554.0, + 519.0, + 1576.0, + 498.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1554.0, + 937.0, + 1554.0, + 937.0, + 1576.0, + 916.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1554.0, + 985.0, + 1554.0, + 985.0, + 1574.0, + 964.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1551.0, + 1461.0, + 1551.0, + 1461.0, + 1581.0, + 1419.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1464.0, + 1551.0, + 1507.0, + 1551.0, + 1507.0, + 1581.0, + 1464.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1599.0, + 490.0, + 1599.0, + 490.0, + 1626.0, + 450.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1599.0, + 526.0, + 1599.0, + 526.0, + 1626.0, + 523.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1599.0, + 759.0, + 1599.0, + 759.0, + 1626.0, + 682.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1599.0, + 955.0, + 1599.0, + 955.0, + 1626.0, + 915.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1599.0, + 993.0, + 1599.0, + 993.0, + 1626.0, + 988.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1599.0, + 1188.0, + 1599.0, + 1188.0, + 1626.0, + 1147.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1599.0, + 1224.0, + 1599.0, + 1224.0, + 1626.0, + 1220.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1599.0, + 1466.0, + 1599.0, + 1466.0, + 1626.0, + 1427.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1498.0, + 1599.0, + 1504.0, + 1599.0, + 1504.0, + 1626.0, + 1498.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 145.0, + 1682.0, + 415.0, + 1682.0, + 415.0, + 1726.0, + 145.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1696.0, + 472.0, + 1696.0, + 472.0, + 1715.0, + 453.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1694.0, + 520.0, + 1694.0, + 520.0, + 1716.0, + 499.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1694.0, + 705.0, + 1694.0, + 705.0, + 1715.0, + 684.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1696.0, + 751.0, + 1696.0, + 751.0, + 1715.0, + 732.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1698.0, + 934.0, + 1698.0, + 934.0, + 1713.0, + 920.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1696.0, + 1170.0, + 1696.0, + 1170.0, + 1715.0, + 1149.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1691.0, + 1461.0, + 1691.0, + 1461.0, + 1719.0, + 1419.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1467.0, + 1693.0, + 1505.0, + 1693.0, + 1505.0, + 1716.0, + 1467.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1740.0, + 474.0, + 1740.0, + 474.0, + 1762.0, + 453.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1740.0, + 520.0, + 1740.0, + 520.0, + 1762.0, + 499.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1738.0, + 708.0, + 1738.0, + 708.0, + 1763.0, + 682.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1741.0, + 751.0, + 1741.0, + 751.0, + 1762.0, + 732.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1740.0, + 937.0, + 1740.0, + 937.0, + 1762.0, + 918.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1740.0, + 1170.0, + 1740.0, + 1170.0, + 1762.0, + 1149.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1737.0, + 1461.0, + 1737.0, + 1461.0, + 1765.0, + 1419.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1464.0, + 1737.0, + 1507.0, + 1737.0, + 1507.0, + 1765.0, + 1464.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1784.0, + 491.0, + 1784.0, + 491.0, + 1811.0, + 446.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1784.0, + 527.0, + 1784.0, + 527.0, + 1811.0, + 523.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1785.0, + 723.0, + 1785.0, + 723.0, + 1812.0, + 682.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1785.0, + 759.0, + 1785.0, + 759.0, + 1812.0, + 755.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1785.0, + 955.0, + 1785.0, + 955.0, + 1812.0, + 916.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1785.0, + 991.0, + 1785.0, + 991.0, + 1812.0, + 987.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1785.0, + 1188.0, + 1785.0, + 1188.0, + 1812.0, + 1147.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1785.0, + 1224.0, + 1785.0, + 1224.0, + 1812.0, + 1220.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1785.0, + 1466.0, + 1785.0, + 1466.0, + 1812.0, + 1427.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1498.0, + 1785.0, + 1504.0, + 1785.0, + 1504.0, + 1812.0, + 1498.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 143.0, + 1875.0, + 410.0, + 1875.0, + 410.0, + 1915.0, + 143.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1879.0, + 476.0, + 1879.0, + 476.0, + 1904.0, + 450.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1879.0, + 522.0, + 1879.0, + 522.0, + 1904.0, + 496.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1879.0, + 708.0, + 1879.0, + 708.0, + 1904.0, + 682.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1882.0, + 751.0, + 1882.0, + 751.0, + 1902.0, + 732.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1880.0, + 939.0, + 1880.0, + 939.0, + 1902.0, + 918.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1880.0, + 985.0, + 1880.0, + 985.0, + 1902.0, + 964.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1879.0, + 1173.0, + 1879.0, + 1173.0, + 1904.0, + 1147.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1880.0, + 1216.0, + 1880.0, + 1216.0, + 1902.0, + 1195.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1879.0, + 1453.0, + 1879.0, + 1453.0, + 1904.0, + 1427.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1475.0, + 1880.0, + 1496.0, + 1880.0, + 1496.0, + 1902.0, + 1475.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1925.0, + 476.0, + 1925.0, + 476.0, + 1950.0, + 450.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1925.0, + 522.0, + 1925.0, + 522.0, + 1950.0, + 496.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1925.0, + 708.0, + 1925.0, + 708.0, + 1950.0, + 682.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1927.0, + 751.0, + 1927.0, + 751.0, + 1949.0, + 730.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1925.0, + 940.0, + 1925.0, + 940.0, + 1950.0, + 915.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1925.0, + 988.0, + 1925.0, + 988.0, + 1950.0, + 961.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1925.0, + 1173.0, + 1925.0, + 1173.0, + 1950.0, + 1147.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 1925.0, + 1219.0, + 1925.0, + 1219.0, + 1950.0, + 1193.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1925.0, + 1453.0, + 1925.0, + 1453.0, + 1950.0, + 1427.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1472.0, + 1925.0, + 1497.0, + 1925.0, + 1497.0, + 1950.0, + 1472.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1972.0, + 490.0, + 1972.0, + 490.0, + 1999.0, + 450.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1972.0, + 526.0, + 1972.0, + 526.0, + 1999.0, + 522.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1972.0, + 722.0, + 1972.0, + 722.0, + 1999.0, + 682.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1972.0, + 759.0, + 1972.0, + 759.0, + 1999.0, + 755.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1972.0, + 955.0, + 1972.0, + 955.0, + 1999.0, + 915.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1972.0, + 993.0, + 1972.0, + 993.0, + 1999.0, + 988.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1972.0, + 1188.0, + 1972.0, + 1188.0, + 1999.0, + 1147.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1972.0, + 1225.0, + 1972.0, + 1225.0, + 1999.0, + 1220.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 1972.0, + 1466.0, + 1972.0, + 1466.0, + 1999.0, + 1427.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1499.0, + 1972.0, + 1504.0, + 1972.0, + 1504.0, + 1999.0, + 1499.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 2019.0, + 523.0, + 2019.0, + 523.0, + 2046.0, + 450.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 2019.0, + 756.0, + 2019.0, + 756.0, + 2046.0, + 682.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 2019.0, + 990.0, + 2019.0, + 990.0, + 2046.0, + 915.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 2019.0, + 1220.0, + 2019.0, + 1220.0, + 2046.0, + 1147.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1427.0, + 2019.0, + 1499.0, + 2019.0, + 1499.0, + 2046.0, + 1427.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1104.0, + 650.0, + 1104.0, + 650.0, + 1147.0, + 160.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 294.0, + 344.0, + 294.0, + 344.0, + 352.0, + 123.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 294.0, + 841.0, + 294.0, + 841.0, + 352.0, + 409.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 336.0, + 1426.0, + 336.0, + 1426.0, + 388.0, + 123.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 161.0, + 770.0, + 758.0, + 770.0, + 758.0, + 807.0, + 161.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1104.0, + 650.0, + 1104.0, + 650.0, + 1147.0, + 160.0, + 1147.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 27, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 204, + 355, + 1552, + 355, + 1552, + 858, + 204, + 858 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 161, + 1848, + 648, + 1848, + 648, + 1878, + 161, + 1878 + ], + "score": 0.677 + }, + { + "category_id": 3, + "poly": [ + 206, + 1017, + 1510, + 1017, + 1510, + 1654, + 206, + 1654 + ], + "score": 0.669 + }, + { + "category_id": 1, + "poly": [ + 162, + 1779, + 646, + 1779, + 646, + 1807, + 162, + 1807 + ], + "score": 0.417 + }, + { + "category_id": 4, + "poly": [ + 148, + 239, + 814, + 239, + 814, + 273, + 148, + 273 + ], + "score": 0.384 + }, + { + "category_id": 2, + "poly": [ + 162, + 1779, + 646, + 1779, + 646, + 1807, + 162, + 1807 + ], + "score": 0.262 + }, + { + "category_id": 4, + "poly": [ + 196, + 940, + 457, + 940, + 457, + 969, + 196, + 969 + ], + "score": 0.161 + }, + { + "category_id": 1, + "poly": [ + 196, + 940, + 457, + 940, + 457, + 969, + 196, + 969 + ], + "score": 0.155 + }, + { + "category_id": 2, + "poly": [ + 161, + 1848, + 648, + 1848, + 648, + 1878, + 161, + 1878 + ], + "score": 0.108 + }, + { + "category_id": 15, + "poly": [ + 195.0, + 361.0, + 465.0, + 361.0, + 465.0, + 403.0, + 195.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 423.0, + 1560.0, + 423.0, + 1560.0, + 461.0, + 1353.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 468.0, + 447.0, + 468.0, + 447.0, + 487.0, + 427.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 465.0, + 492.0, + 465.0, + 492.0, + 490.0, + 468.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 464.0, + 537.0, + 464.0, + 537.0, + 490.0, + 513.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 467.0, + 668.0, + 467.0, + 668.0, + 489.0, + 649.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 465.0, + 715.0, + 465.0, + 715.0, + 490.0, + 691.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 464.0, + 759.0, + 464.0, + 759.0, + 490.0, + 735.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 465.0, + 895.0, + 465.0, + 895.0, + 490.0, + 869.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 465.0, + 938.0, + 465.0, + 938.0, + 490.0, + 913.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 464.0, + 982.0, + 464.0, + 982.0, + 489.0, + 958.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 467.0, + 1113.0, + 467.0, + 1113.0, + 489.0, + 1094.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 465.0, + 1162.0, + 465.0, + 1162.0, + 490.0, + 1136.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 464.0, + 1204.0, + 464.0, + 1204.0, + 489.0, + 1180.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1388.0, + 451.0, + 1528.0, + 451.0, + 1528.0, + 486.0, + 1388.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 509.0, + 448.0, + 509.0, + 448.0, + 534.0, + 424.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 508.0, + 492.0, + 508.0, + 492.0, + 534.0, + 469.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 508.0, + 537.0, + 508.0, + 537.0, + 534.0, + 513.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 511.0, + 668.0, + 511.0, + 668.0, + 533.0, + 649.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 508.0, + 715.0, + 508.0, + 715.0, + 534.0, + 691.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 506.0, + 760.0, + 506.0, + 760.0, + 534.0, + 736.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 509.0, + 893.0, + 509.0, + 893.0, + 534.0, + 869.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 509.0, + 938.0, + 509.0, + 938.0, + 534.0, + 914.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 508.0, + 982.0, + 508.0, + 982.0, + 534.0, + 958.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 512.0, + 1113.0, + 512.0, + 1113.0, + 533.0, + 1094.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 509.0, + 1160.0, + 509.0, + 1160.0, + 534.0, + 1136.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 508.0, + 1204.0, + 508.0, + 1204.0, + 534.0, + 1181.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 511.0, + 1424.0, + 511.0, + 1424.0, + 533.0, + 1404.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1450.0, + 511.0, + 1469.0, + 511.0, + 1469.0, + 531.0, + 1450.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 542.0, + 409.0, + 542.0, + 409.0, + 580.0, + 240.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 555.0, + 447.0, + 555.0, + 447.0, + 575.0, + 427.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 555.0, + 490.0, + 555.0, + 490.0, + 577.0, + 471.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 550.0, + 537.0, + 550.0, + 537.0, + 577.0, + 513.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 555.0, + 668.0, + 555.0, + 668.0, + 575.0, + 650.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 555.0, + 714.0, + 555.0, + 714.0, + 577.0, + 694.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 552.0, + 760.0, + 552.0, + 760.0, + 577.0, + 736.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 553.0, + 892.0, + 553.0, + 892.0, + 575.0, + 872.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 553.0, + 935.0, + 553.0, + 935.0, + 575.0, + 917.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 553.0, + 981.0, + 553.0, + 981.0, + 574.0, + 961.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 555.0, + 1113.0, + 555.0, + 1113.0, + 575.0, + 1095.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 553.0, + 1159.0, + 553.0, + 1159.0, + 574.0, + 1139.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 550.0, + 1204.0, + 550.0, + 1204.0, + 577.0, + 1181.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1406.0, + 555.0, + 1424.0, + 555.0, + 1424.0, + 575.0, + 1406.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1450.0, + 553.0, + 1469.0, + 553.0, + 1469.0, + 574.0, + 1450.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 262.0, + 569.0, + 387.0, + 569.0, + 387.0, + 604.0, + 262.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 685.0, + 490.0, + 685.0, + 490.0, + 705.0, + 471.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 682.0, + 537.0, + 682.0, + 537.0, + 707.0, + 513.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 685.0, + 714.0, + 685.0, + 714.0, + 705.0, + 692.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 682.0, + 760.0, + 682.0, + 760.0, + 707.0, + 735.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 710.0, + 412.0, + 710.0, + 412.0, + 751.0, + 242.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 726.0, + 493.0, + 726.0, + 493.0, + 751.0, + 469.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 724.0, + 539.0, + 724.0, + 539.0, + 751.0, + 513.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 729.0, + 714.0, + 729.0, + 714.0, + 749.0, + 692.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 727.0, + 757.0, + 727.0, + 757.0, + 748.0, + 738.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 710.0, + 1109.0, + 710.0, + 1109.0, + 745.0, + 901.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 724.0, + 1157.0, + 724.0, + 1157.0, + 745.0, + 1136.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 721.0, + 1202.0, + 721.0, + 1202.0, + 746.0, + 1177.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 720.0, + 1246.0, + 720.0, + 1246.0, + 746.0, + 1223.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 724.0, + 1377.0, + 724.0, + 1377.0, + 745.0, + 1358.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 723.0, + 1424.0, + 723.0, + 1424.0, + 743.0, + 1403.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1447.0, + 723.0, + 1466.0, + 723.0, + 1466.0, + 745.0, + 1447.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 740.0, + 400.0, + 740.0, + 400.0, + 776.0, + 257.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 737.0, + 1067.0, + 737.0, + 1067.0, + 768.0, + 944.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 770.0, + 493.0, + 770.0, + 493.0, + 795.0, + 469.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 770.0, + 537.0, + 770.0, + 537.0, + 795.0, + 513.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 771.0, + 714.0, + 771.0, + 714.0, + 793.0, + 694.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 771.0, + 757.0, + 771.0, + 757.0, + 793.0, + 738.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 768.0, + 1156.0, + 768.0, + 1156.0, + 790.0, + 1136.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 768.0, + 1201.0, + 768.0, + 1201.0, + 787.0, + 1181.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 767.0, + 1245.0, + 767.0, + 1245.0, + 789.0, + 1225.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 768.0, + 1377.0, + 768.0, + 1377.0, + 790.0, + 1359.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 768.0, + 1423.0, + 768.0, + 1423.0, + 789.0, + 1404.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1448.0, + 767.0, + 1468.0, + 767.0, + 1468.0, + 789.0, + 1448.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1043.0, + 448.0, + 1043.0, + 448.0, + 1066.0, + 425.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1043.0, + 494.0, + 1043.0, + 494.0, + 1066.0, + 469.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1042.0, + 536.0, + 1042.0, + 536.0, + 1065.0, + 513.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1043.0, + 671.0, + 1043.0, + 671.0, + 1066.0, + 647.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1043.0, + 716.0, + 1043.0, + 716.0, + 1066.0, + 691.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1043.0, + 757.0, + 1043.0, + 757.0, + 1063.0, + 738.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1043.0, + 893.0, + 1043.0, + 893.0, + 1066.0, + 870.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1043.0, + 938.0, + 1043.0, + 938.0, + 1066.0, + 914.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1042.0, + 981.0, + 1042.0, + 981.0, + 1065.0, + 957.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1043.0, + 1115.0, + 1043.0, + 1115.0, + 1066.0, + 1092.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1043.0, + 1161.0, + 1043.0, + 1161.0, + 1066.0, + 1134.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1042.0, + 1203.0, + 1042.0, + 1203.0, + 1065.0, + 1180.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1082.0, + 419.0, + 1082.0, + 419.0, + 1118.0, + 247.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1086.0, + 447.0, + 1086.0, + 447.0, + 1110.0, + 424.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1085.0, + 494.0, + 1085.0, + 494.0, + 1110.0, + 469.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1084.0, + 536.0, + 1084.0, + 536.0, + 1108.0, + 513.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1088.0, + 668.0, + 1088.0, + 668.0, + 1108.0, + 649.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1086.0, + 716.0, + 1086.0, + 716.0, + 1110.0, + 691.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1084.0, + 759.0, + 1084.0, + 759.0, + 1108.0, + 737.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1086.0, + 893.0, + 1086.0, + 893.0, + 1110.0, + 870.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1086.0, + 938.0, + 1086.0, + 938.0, + 1110.0, + 914.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1086.0, + 979.0, + 1086.0, + 979.0, + 1107.0, + 960.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1088.0, + 1112.0, + 1088.0, + 1112.0, + 1108.0, + 1093.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1086.0, + 1159.0, + 1086.0, + 1159.0, + 1110.0, + 1136.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1086.0, + 1202.0, + 1086.0, + 1202.0, + 1107.0, + 1183.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 267.0, + 1110.0, + 393.0, + 1110.0, + 393.0, + 1143.0, + 267.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1130.0, + 448.0, + 1130.0, + 448.0, + 1154.0, + 425.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1130.0, + 492.0, + 1130.0, + 492.0, + 1154.0, + 469.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1128.0, + 536.0, + 1128.0, + 536.0, + 1153.0, + 513.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1130.0, + 671.0, + 1130.0, + 671.0, + 1154.0, + 647.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1131.0, + 713.0, + 1131.0, + 713.0, + 1151.0, + 694.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1128.0, + 759.0, + 1128.0, + 759.0, + 1153.0, + 735.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1130.0, + 893.0, + 1130.0, + 893.0, + 1154.0, + 871.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1131.0, + 936.0, + 1131.0, + 936.0, + 1151.0, + 917.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1130.0, + 979.0, + 1130.0, + 979.0, + 1151.0, + 962.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1133.0, + 1114.0, + 1133.0, + 1114.0, + 1153.0, + 1095.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1133.0, + 1158.0, + 1133.0, + 1158.0, + 1151.0, + 1139.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1130.0, + 1202.0, + 1130.0, + 1202.0, + 1150.0, + 1183.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1261.0, + 494.0, + 1261.0, + 494.0, + 1282.0, + 469.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1262.0, + 535.0, + 1262.0, + 535.0, + 1279.0, + 516.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1261.0, + 716.0, + 1261.0, + 716.0, + 1284.0, + 691.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1261.0, + 757.0, + 1261.0, + 757.0, + 1279.0, + 738.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1261.0, + 938.0, + 1261.0, + 938.0, + 1284.0, + 914.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 1258.0, + 981.0, + 1258.0, + 981.0, + 1282.0, + 959.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1259.0, + 1161.0, + 1259.0, + 1161.0, + 1284.0, + 1136.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1258.0, + 1203.0, + 1258.0, + 1203.0, + 1282.0, + 1181.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1289.0, + 411.0, + 1289.0, + 411.0, + 1332.0, + 240.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1304.0, + 494.0, + 1304.0, + 494.0, + 1328.0, + 469.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1304.0, + 535.0, + 1304.0, + 535.0, + 1324.0, + 516.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1304.0, + 716.0, + 1304.0, + 716.0, + 1328.0, + 691.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1304.0, + 757.0, + 1304.0, + 757.0, + 1324.0, + 738.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1304.0, + 937.0, + 1304.0, + 937.0, + 1327.0, + 914.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1304.0, + 979.0, + 1304.0, + 979.0, + 1324.0, + 960.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1304.0, + 1161.0, + 1304.0, + 1161.0, + 1328.0, + 1136.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1304.0, + 1202.0, + 1304.0, + 1202.0, + 1324.0, + 1184.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1319.0, + 398.0, + 1319.0, + 398.0, + 1352.0, + 252.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1350.0, + 491.0, + 1350.0, + 491.0, + 1369.0, + 470.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1348.0, + 535.0, + 1348.0, + 535.0, + 1369.0, + 516.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1347.0, + 716.0, + 1347.0, + 716.0, + 1370.0, + 693.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1346.0, + 759.0, + 1346.0, + 759.0, + 1370.0, + 735.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1348.0, + 936.0, + 1348.0, + 936.0, + 1369.0, + 917.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1348.0, + 979.0, + 1348.0, + 979.0, + 1369.0, + 960.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1348.0, + 1158.0, + 1348.0, + 1158.0, + 1369.0, + 1139.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1346.0, + 1203.0, + 1346.0, + 1203.0, + 1370.0, + 1181.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1523.0, + 446.0, + 1523.0, + 446.0, + 1543.0, + 427.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1521.0, + 494.0, + 1521.0, + 494.0, + 1544.0, + 469.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1523.0, + 535.0, + 1523.0, + 535.0, + 1541.0, + 516.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1523.0, + 668.0, + 1523.0, + 668.0, + 1543.0, + 649.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1523.0, + 713.0, + 1523.0, + 713.0, + 1543.0, + 694.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1523.0, + 757.0, + 1523.0, + 757.0, + 1541.0, + 738.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1520.0, + 1119.0, + 1520.0, + 1119.0, + 1559.0, + 908.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1521.0, + 1161.0, + 1521.0, + 1161.0, + 1544.0, + 1137.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1523.0, + 1202.0, + 1523.0, + 1202.0, + 1541.0, + 1183.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 1524.0, + 1380.0, + 1524.0, + 1380.0, + 1543.0, + 1361.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1523.0, + 1424.0, + 1523.0, + 1424.0, + 1541.0, + 1405.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 210.0, + 1533.0, + 419.0, + 1533.0, + 419.0, + 1564.0, + 210.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1557.0, + 377.0, + 1557.0, + 377.0, + 1593.0, + 253.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1567.0, + 491.0, + 1567.0, + 491.0, + 1587.0, + 472.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1566.0, + 535.0, + 1566.0, + 535.0, + 1586.0, + 516.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1566.0, + 668.0, + 1566.0, + 668.0, + 1587.0, + 650.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1567.0, + 713.0, + 1567.0, + 713.0, + 1587.0, + 694.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1566.0, + 757.0, + 1566.0, + 757.0, + 1586.0, + 738.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1547.0, + 1086.0, + 1547.0, + 1086.0, + 1581.0, + 941.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1567.0, + 1158.0, + 1567.0, + 1158.0, + 1586.0, + 1139.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1566.0, + 1202.0, + 1566.0, + 1202.0, + 1586.0, + 1183.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 1567.0, + 1380.0, + 1567.0, + 1380.0, + 1587.0, + 1361.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1566.0, + 1424.0, + 1566.0, + 1424.0, + 1586.0, + 1405.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.75, + 1564.0, + 450.75, + 1564.0, + 450.75, + 1587.5, + 421.75, + 1587.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 150.0, + 238.0, + 812.0, + 238.0, + 812.0, + 275.0, + 150.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 161.0, + 1777.0, + 648.0, + 1777.0, + 648.0, + 1811.0, + 161.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 935.0, + 459.0, + 935.0, + 459.0, + 974.0, + 195.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1846.0, + 649.0, + 1846.0, + 649.0, + 1880.0, + 160.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 1846.0, + 649.0, + 1846.0, + 649.0, + 1880.0, + 160.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 161.0, + 1777.0, + 648.0, + 1777.0, + 648.0, + 1811.0, + 161.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 195.0, + 935.0, + 459.0, + 935.0, + 459.0, + 974.0, + 195.0, + 974.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 28, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 318, + 629, + 1375, + 629, + 1375, + 862, + 318, + 862 + ], + "score": 0.832, + "html": "
Padded at top-leftPadded at bottom-leftPadded at top-rightPadded at bottom-rightAverage (grouped padding strategy)
252520202015151212121515202020991212121616161616
252520202015151212121515202020991212121616161616
20201616162020161616121216161612121616161616161616
20201616162020161616121216161612121616161616161616
20201616 1620201616121216161612121616161616161616
" + }, + { + "category_id": 6, + "poly": [ + 292, + 574, + 892, + 574, + 892, + 609, + 292, + 609 + ], + "score": 0.766 + }, + { + "category_id": 3, + "poly": [ + 370, + 948, + 1297, + 948, + 1297, + 2121, + 370, + 2121 + ], + "score": 0.722 + }, + { + "category_id": 0, + "poly": [ + 131, + 189, + 1191, + 189, + 1191, + 286, + 131, + 286 + ], + "score": 0.722 + }, + { + "category_id": 5, + "poly": [ + 317, + 317, + 1386, + 317, + 1386, + 558, + 317, + 558 + ], + "score": 0.5, + "html": "
Original InputPadded at top-leftPadded at bottom-Padded at top-rightPadded at bottom left corner
edde fddefright corner
abCbabaabaC
deedb eC fbC fa da deC faab
hhgd ghie hd gd ge highde
gggh
" + }, + { + "category_id": 4, + "poly": [ + 294, + 901, + 779, + 901, + 779, + 935, + 294, + 935 + ], + "score": 0.302 + }, + { + "category_id": 6, + "poly": [ + 1196, + 326, + 1397, + 326, + 1397, + 379, + 1196, + 379 + ], + "score": 0.268 + }, + { + "category_id": 3, + "poly": [ + 318, + 629, + 1375, + 629, + 1375, + 862, + 318, + 862 + ], + "score": 0.094 + }, + { + "category_id": 13, + "poly": [ + 501, + 695, + 678, + 695, + 678, + 858, + 501, + 858 + ], + "score": 0.86, + "latex": "\\begin{array} { r } { { \\left\\{ \\begin{array} { l l l l l l l } { 1 5 } & { 1 5 } & { 1 2 } & { 1 2 } & { 1 2 } & { 1 2 } \\\\ { 1 5 } & { 1 5 } & { 1 2 } & { 1 2 } & { 1 2 } & { 1 2 } \\\\ { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\\\ { { \\left\\{ \\begin{array} { l l l l l l l } { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } & { 1 6 } \\\\ { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\\\ { { \\left\\{ \\begin{array} { l l l l l l l l } { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 311, + 692, + 485, + 692, + 485, + 858, + 311, + 858 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { { \\left( \\begin{array} { l l l l l l l } { 2 5 } & { 2 5 } & { 2 0 } & { 2 0 } & { 2 0 } & { 2 0 } \\\\ { 2 5 } & { 2 5 } & { 2 0 } & { 2 0 } & { 2 0 } & { 2 0 } \\\\ { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } & { } \\\\ { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } & { } \\\\ { 2 0 } & { 2 0 } & { 1 6 } & { 1 6 } & { 1 6 } & { } \\end{array} \\right) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 891, + 696, + 1076, + 696, + 1076, + 859, + 891, + 859 + ], + "score": 0.84, + "latex": "\\left| \\begin{array} { l l l l l l } { { 9 } } & { { 9 } } & { { 1 2 } } & { { 1 2 } } & { { 1 2 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 9 } } & { { 9 } } & { { 1 2 } } & { { 1 2 } } & { { 1 2 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 1 2 } } & { { 1 2 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 1 2 } } & { { 1 2 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\end{array} \\right." + }, + { + "category_id": 13, + "poly": [ + 1176, + 694, + 1357, + 694, + 1357, + 857, + 1176, + 857 + ], + "score": 0.74, + "latex": "\\begin{array} { c c c c c c c c } { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { \\ldots } } & { { } } & { { } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { \\ldots } } & { { } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { \\ldots } } & { { } } & { { } } \\\\ { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } \\\\ { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { 1 6 } } & { { \\ldots } } & { { } } & { { } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 697, + 694, + 870, + 694, + 870, + 858, + 697, + 858 + ], + "score": 0.72, + "latex": "\\begin{array} { r } { { \\left\\{ \\begin{array} { l l l l l l l } { 1 5 } & { 1 5 } & { 2 0 } & { 2 0 } & { 2 0 } & { 2 0 } \\\\ { 1 5 } & { 1 5 } & { 2 0 } & { 2 0 } & { 2 0 } \\\\ { 1 2 } & { 1 2 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\\\ { { \\left\\{ \\begin{array} { l l l l l l l } { 1 2 } & { 1 2 } & { 1 6 } & { 1 6 } & { 1 6 } & { 1 6 } \\\\ { 1 2 } & { 1 2 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\\\ { { \\left\\{ \\begin{array} { l l l l l l l } { 1 2 } & { 1 2 } & { 1 6 } & { 1 6 } & { 1 6 } & { 1 6 } \\\\ { 1 2 } & { 1 2 } & { 1 6 } & { 1 6 } & { 1 6 } \\end{array} \\right. } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 575.0, + 890.0, + 575.0, + 890.0, + 612.0, + 292.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 960.0, + 543.0, + 960.0, + 543.0, + 989.0, + 423.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 960.0, + 704.0, + 960.0, + 704.0, + 989.0, + 583.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 960.0, + 848.0, + 960.0, + 848.0, + 989.0, + 729.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 957.0, + 1024.0, + 957.0, + 1024.0, + 991.0, + 901.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 955.0, + 1306.0, + 955.0, + 1306.0, + 992.0, + 1123.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 986.0, + 527.0, + 986.0, + 527.0, + 1016.0, + 438.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 986.0, + 710.0, + 986.0, + 710.0, + 1015.0, + 582.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 984.0, + 844.0, + 984.0, + 844.0, + 1019.0, + 733.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 986.0, + 1037.0, + 986.0, + 1037.0, + 1016.0, + 892.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 982.0, + 1306.0, + 982.0, + 1306.0, + 1016.0, + 1128.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 1031.0, + 412.0, + 1031.0, + 412.0, + 1062.0, + 378.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1035.0, + 443.0, + 1035.0, + 443.0, + 1047.0, + 430.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1035.0, + 475.0, + 1035.0, + 475.0, + 1047.0, + 462.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1035.0, + 506.0, + 1035.0, + 506.0, + 1047.0, + 494.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1035.0, + 602.0, + 1035.0, + 602.0, + 1047.0, + 590.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1035.0, + 634.0, + 1035.0, + 634.0, + 1047.0, + 622.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1033.0, + 665.0, + 1033.0, + 665.0, + 1047.0, + 654.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1033.0, + 761.0, + 1033.0, + 761.0, + 1047.0, + 749.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1032.0, + 795.0, + 1032.0, + 795.0, + 1049.0, + 779.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1035.0, + 922.0, + 1035.0, + 922.0, + 1048.0, + 910.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1032.0, + 954.0, + 1032.0, + 954.0, + 1049.0, + 939.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1033.0, + 1177.0, + 1033.0, + 1177.0, + 1047.0, + 1165.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1032.0, + 1211.0, + 1032.0, + 1211.0, + 1049.0, + 1195.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 1032.0, + 1243.0, + 1032.0, + 1243.0, + 1049.0, + 1228.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1065.0, + 474.0, + 1065.0, + 474.0, + 1077.0, + 462.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1066.0, + 506.0, + 1066.0, + 506.0, + 1077.0, + 494.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1065.0, + 602.0, + 1065.0, + 602.0, + 1077.0, + 590.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1064.0, + 636.0, + 1064.0, + 636.0, + 1080.0, + 620.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1065.0, + 665.0, + 1065.0, + 665.0, + 1078.0, + 653.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1065.0, + 761.0, + 1065.0, + 761.0, + 1077.0, + 749.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1064.0, + 795.0, + 1064.0, + 795.0, + 1080.0, + 779.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1065.0, + 920.0, + 1065.0, + 920.0, + 1078.0, + 908.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1065.0, + 953.0, + 1065.0, + 953.0, + 1078.0, + 940.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1065.0, + 1177.0, + 1065.0, + 1177.0, + 1078.0, + 1165.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1064.0, + 1217.0, + 1064.0, + 1217.0, + 1078.0, + 1188.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1062.0, + 1252.0, + 1062.0, + 1252.0, + 1081.0, + 1219.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1097.0, + 443.0, + 1097.0, + 443.0, + 1109.0, + 430.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1094.0, + 476.0, + 1094.0, + 476.0, + 1111.0, + 459.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1097.0, + 506.0, + 1097.0, + 506.0, + 1109.0, + 494.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1094.0, + 763.0, + 1094.0, + 763.0, + 1111.0, + 747.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1094.0, + 795.0, + 1094.0, + 795.0, + 1111.0, + 779.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1097.0, + 1177.0, + 1097.0, + 1177.0, + 1110.0, + 1165.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1093.0, + 1251.0, + 1093.0, + 1251.0, + 1111.0, + 1187.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1154.0, + 513.0, + 1154.0, + 513.0, + 1175.0, + 455.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1154.0, + 672.0, + 1154.0, + 672.0, + 1175.0, + 616.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1156.0, + 831.0, + 1156.0, + 831.0, + 1173.0, + 773.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1155.0, + 987.0, + 1155.0, + 987.0, + 1174.0, + 939.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1154.0, + 1249.0, + 1154.0, + 1249.0, + 1175.0, + 1191.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1218.0, + 412.0, + 1218.0, + 412.0, + 1249.0, + 376.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1220.0, + 443.0, + 1220.0, + 443.0, + 1232.0, + 430.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1218.0, + 476.0, + 1218.0, + 476.0, + 1235.0, + 459.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 1220.0, + 506.0, + 1220.0, + 506.0, + 1232.0, + 493.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1220.0, + 537.0, + 1220.0, + 537.0, + 1232.0, + 526.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1220.0, + 602.0, + 1220.0, + 602.0, + 1232.0, + 589.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1220.0, + 633.0, + 1220.0, + 633.0, + 1232.0, + 621.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1220.0, + 665.0, + 1220.0, + 665.0, + 1232.0, + 653.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1220.0, + 697.0, + 1220.0, + 697.0, + 1232.0, + 686.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1218.0, + 763.0, + 1218.0, + 763.0, + 1233.0, + 747.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1218.0, + 795.0, + 1218.0, + 795.0, + 1235.0, + 779.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1218.0, + 827.0, + 1218.0, + 827.0, + 1233.0, + 811.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1220.0, + 920.0, + 1220.0, + 920.0, + 1233.0, + 908.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1218.0, + 954.0, + 1218.0, + 954.0, + 1235.0, + 938.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1218.0, + 986.0, + 1218.0, + 986.0, + 1235.0, + 970.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1218.0, + 1179.0, + 1218.0, + 1179.0, + 1235.0, + 1162.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1218.0, + 1211.0, + 1218.0, + 1211.0, + 1235.0, + 1195.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1220.0, + 1241.0, + 1220.0, + 1241.0, + 1233.0, + 1229.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1252.0, + 443.0, + 1252.0, + 443.0, + 1263.0, + 431.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1249.0, + 476.0, + 1249.0, + 476.0, + 1265.0, + 460.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1251.0, + 506.0, + 1251.0, + 506.0, + 1264.0, + 494.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1252.0, + 537.0, + 1252.0, + 537.0, + 1263.0, + 526.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1251.0, + 602.0, + 1251.0, + 602.0, + 1264.0, + 590.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1249.0, + 636.0, + 1249.0, + 636.0, + 1265.0, + 620.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1249.0, + 668.0, + 1249.0, + 668.0, + 1265.0, + 652.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1251.0, + 697.0, + 1251.0, + 697.0, + 1264.0, + 685.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1249.0, + 763.0, + 1249.0, + 763.0, + 1265.0, + 747.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1249.0, + 795.0, + 1249.0, + 795.0, + 1265.0, + 779.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1249.0, + 827.0, + 1249.0, + 827.0, + 1265.0, + 811.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1251.0, + 920.0, + 1251.0, + 920.0, + 1264.0, + 908.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1249.0, + 955.0, + 1249.0, + 955.0, + 1265.0, + 939.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1249.0, + 986.0, + 1249.0, + 986.0, + 1267.0, + 971.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 1248.0, + 1187.0, + 1248.0, + 1187.0, + 1267.0, + 1154.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1248.0, + 1216.0, + 1248.0, + 1216.0, + 1267.0, + 1189.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1249.0, + 1247.0, + 1249.0, + 1247.0, + 1265.0, + 1224.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1248.0, + 1285.0, + 1248.0, + 1285.0, + 1267.0, + 1252.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1284.0, + 443.0, + 1284.0, + 443.0, + 1294.0, + 430.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1281.0, + 476.0, + 1281.0, + 476.0, + 1297.0, + 459.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1281.0, + 507.0, + 1281.0, + 507.0, + 1297.0, + 491.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1283.0, + 538.0, + 1283.0, + 538.0, + 1294.0, + 526.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1281.0, + 763.0, + 1281.0, + 763.0, + 1297.0, + 747.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1281.0, + 796.0, + 1281.0, + 796.0, + 1297.0, + 779.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1281.0, + 827.0, + 1281.0, + 827.0, + 1297.0, + 811.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 1280.0, + 1187.0, + 1280.0, + 1187.0, + 1298.0, + 1154.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1280.0, + 1216.0, + 1280.0, + 1216.0, + 1298.0, + 1189.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1281.0, + 1247.0, + 1281.0, + 1247.0, + 1297.0, + 1224.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1280.0, + 1284.0, + 1280.0, + 1284.0, + 1298.0, + 1252.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1341.0, + 512.0, + 1341.0, + 512.0, + 1359.0, + 453.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1340.0, + 673.0, + 1340.0, + 673.0, + 1362.0, + 616.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1341.0, + 831.0, + 1341.0, + 831.0, + 1360.0, + 773.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1340.0, + 989.0, + 1340.0, + 989.0, + 1362.0, + 938.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1340.0, + 1249.0, + 1340.0, + 1249.0, + 1362.0, + 1191.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 1407.0, + 408.0, + 1407.0, + 408.0, + 1435.0, + 378.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1406.0, + 442.0, + 1406.0, + 442.0, + 1419.0, + 430.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1406.0, + 474.0, + 1406.0, + 474.0, + 1419.0, + 462.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1406.0, + 506.0, + 1406.0, + 506.0, + 1419.0, + 494.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1406.0, + 538.0, + 1406.0, + 538.0, + 1419.0, + 526.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1406.0, + 601.0, + 1406.0, + 601.0, + 1419.0, + 589.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1406.0, + 633.0, + 1406.0, + 633.0, + 1419.0, + 622.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1406.0, + 665.0, + 1406.0, + 665.0, + 1419.0, + 653.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1406.0, + 697.0, + 1406.0, + 697.0, + 1419.0, + 686.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1406.0, + 761.0, + 1406.0, + 761.0, + 1419.0, + 749.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1405.0, + 795.0, + 1405.0, + 795.0, + 1420.0, + 779.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1405.0, + 827.0, + 1405.0, + 827.0, + 1420.0, + 811.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1405.0, + 859.0, + 1405.0, + 859.0, + 1420.0, + 843.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1406.0, + 920.0, + 1406.0, + 920.0, + 1419.0, + 908.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1406.0, + 953.0, + 1406.0, + 953.0, + 1419.0, + 940.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1406.0, + 985.0, + 1406.0, + 985.0, + 1419.0, + 973.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1406.0, + 1017.0, + 1406.0, + 1017.0, + 1419.0, + 1005.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1405.0, + 1179.0, + 1405.0, + 1179.0, + 1420.0, + 1162.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1405.0, + 1211.0, + 1405.0, + 1211.0, + 1420.0, + 1195.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1406.0, + 1241.0, + 1406.0, + 1241.0, + 1419.0, + 1229.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1405.0, + 1276.0, + 1405.0, + 1276.0, + 1420.0, + 1260.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1439.0, + 442.0, + 1439.0, + 442.0, + 1451.0, + 430.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1436.0, + 476.0, + 1436.0, + 476.0, + 1454.0, + 459.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1436.0, + 509.0, + 1436.0, + 509.0, + 1454.0, + 491.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1436.0, + 541.0, + 1436.0, + 541.0, + 1452.0, + 523.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1439.0, + 602.0, + 1439.0, + 602.0, + 1451.0, + 589.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1436.0, + 636.0, + 1436.0, + 636.0, + 1452.0, + 620.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1439.0, + 697.0, + 1439.0, + 697.0, + 1451.0, + 685.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1438.0, + 761.0, + 1438.0, + 761.0, + 1451.0, + 749.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1436.0, + 795.0, + 1436.0, + 795.0, + 1454.0, + 779.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1436.0, + 827.0, + 1436.0, + 827.0, + 1452.0, + 811.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1436.0, + 859.0, + 1436.0, + 859.0, + 1452.0, + 843.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1439.0, + 920.0, + 1439.0, + 920.0, + 1451.0, + 910.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1438.0, + 953.0, + 1438.0, + 953.0, + 1451.0, + 940.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1438.0, + 985.0, + 1438.0, + 985.0, + 1451.0, + 973.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1439.0, + 1015.0, + 1439.0, + 1015.0, + 1451.0, + 1005.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1436.0, + 1183.0, + 1436.0, + 1183.0, + 1452.0, + 1161.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1435.0, + 1216.0, + 1435.0, + 1216.0, + 1454.0, + 1189.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1436.0, + 1247.0, + 1436.0, + 1247.0, + 1452.0, + 1224.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1435.0, + 1281.0, + 1435.0, + 1281.0, + 1454.0, + 1255.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1470.0, + 440.0, + 1470.0, + 440.0, + 1481.0, + 430.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1467.0, + 476.0, + 1467.0, + 476.0, + 1483.0, + 459.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1470.0, + 537.0, + 1470.0, + 537.0, + 1481.0, + 526.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1470.0, + 761.0, + 1470.0, + 761.0, + 1481.0, + 749.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1468.0, + 795.0, + 1468.0, + 795.0, + 1484.0, + 779.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1470.0, + 825.0, + 1470.0, + 825.0, + 1481.0, + 812.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1470.0, + 856.0, + 1470.0, + 856.0, + 1481.0, + 844.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1466.0, + 1184.0, + 1466.0, + 1184.0, + 1484.0, + 1157.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1466.0, + 1216.0, + 1466.0, + 1216.0, + 1484.0, + 1189.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1467.0, + 1247.0, + 1467.0, + 1247.0, + 1483.0, + 1224.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1466.0, + 1281.0, + 1466.0, + 1281.0, + 1484.0, + 1255.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1528.0, + 513.0, + 1528.0, + 513.0, + 1549.0, + 455.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1527.0, + 672.0, + 1527.0, + 672.0, + 1549.0, + 614.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1528.0, + 832.0, + 1528.0, + 832.0, + 1549.0, + 775.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1527.0, + 991.0, + 1527.0, + 991.0, + 1549.0, + 934.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1528.0, + 1249.0, + 1528.0, + 1249.0, + 1549.0, + 1191.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 1594.0, + 598.0, + 1594.0, + 598.0, + 1627.0, + 375.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1651.0, + 408.0, + 1651.0, + 408.0, + 1684.0, + 373.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1655.0, + 443.0, + 1655.0, + 443.0, + 1666.0, + 430.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1655.0, + 474.0, + 1655.0, + 474.0, + 1667.0, + 462.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1655.0, + 506.0, + 1655.0, + 506.0, + 1667.0, + 494.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1655.0, + 537.0, + 1655.0, + 537.0, + 1667.0, + 526.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1655.0, + 601.0, + 1655.0, + 601.0, + 1667.0, + 590.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1655.0, + 633.0, + 1655.0, + 633.0, + 1667.0, + 621.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1655.0, + 665.0, + 1655.0, + 665.0, + 1667.0, + 653.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1655.0, + 697.0, + 1655.0, + 697.0, + 1667.0, + 686.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1655.0, + 761.0, + 1655.0, + 761.0, + 1667.0, + 749.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1655.0, + 793.0, + 1655.0, + 793.0, + 1667.0, + 781.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1655.0, + 824.0, + 1655.0, + 824.0, + 1667.0, + 815.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1655.0, + 920.0, + 1655.0, + 920.0, + 1667.0, + 910.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1655.0, + 951.0, + 1655.0, + 951.0, + 1667.0, + 940.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1655.0, + 983.0, + 1655.0, + 983.0, + 1667.0, + 973.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1654.0, + 1185.0, + 1654.0, + 1185.0, + 1668.0, + 1156.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1654.0, + 1215.0, + 1654.0, + 1215.0, + 1668.0, + 1191.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1654.0, + 1247.0, + 1654.0, + 1247.0, + 1668.0, + 1225.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 1654.0, + 1283.0, + 1654.0, + 1283.0, + 1668.0, + 1253.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1686.0, + 474.0, + 1686.0, + 474.0, + 1699.0, + 462.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1687.0, + 505.0, + 1687.0, + 505.0, + 1699.0, + 494.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1687.0, + 537.0, + 1687.0, + 537.0, + 1698.0, + 526.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1687.0, + 601.0, + 1687.0, + 601.0, + 1698.0, + 590.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1687.0, + 633.0, + 1687.0, + 633.0, + 1699.0, + 621.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1687.0, + 665.0, + 1687.0, + 665.0, + 1699.0, + 653.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1687.0, + 697.0, + 1687.0, + 697.0, + 1699.0, + 685.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1687.0, + 760.0, + 1687.0, + 760.0, + 1698.0, + 751.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1684.0, + 795.0, + 1684.0, + 795.0, + 1700.0, + 779.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1687.0, + 825.0, + 1687.0, + 825.0, + 1699.0, + 813.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1686.0, + 953.0, + 1686.0, + 953.0, + 1699.0, + 940.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1686.0, + 985.0, + 1686.0, + 985.0, + 1699.0, + 973.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 1684.0, + 1183.0, + 1684.0, + 1183.0, + 1700.0, + 1160.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1686.0, + 1208.0, + 1686.0, + 1208.0, + 1699.0, + 1197.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1684.0, + 1280.0, + 1684.0, + 1280.0, + 1700.0, + 1257.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1720.0, + 439.0, + 1720.0, + 439.0, + 1727.0, + 432.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1717.0, + 474.0, + 1717.0, + 474.0, + 1729.0, + 462.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1719.0, + 505.0, + 1719.0, + 505.0, + 1729.0, + 494.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1719.0, + 537.0, + 1719.0, + 537.0, + 1728.0, + 526.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1717.0, + 601.0, + 1717.0, + 601.0, + 1729.0, + 590.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1719.0, + 633.0, + 1719.0, + 633.0, + 1729.0, + 622.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1717.0, + 665.0, + 1717.0, + 665.0, + 1731.0, + 653.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1717.0, + 697.0, + 1717.0, + 697.0, + 1729.0, + 686.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1717.0, + 761.0, + 1717.0, + 761.0, + 1729.0, + 749.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1716.0, + 795.0, + 1716.0, + 795.0, + 1732.0, + 779.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1716.0, + 827.0, + 1716.0, + 827.0, + 1732.0, + 811.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1719.0, + 919.0, + 1719.0, + 919.0, + 1729.0, + 910.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1717.0, + 953.0, + 1717.0, + 953.0, + 1729.0, + 940.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1717.0, + 985.0, + 1717.0, + 985.0, + 1731.0, + 973.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 1716.0, + 1183.0, + 1716.0, + 1183.0, + 1732.0, + 1160.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1717.0, + 1208.0, + 1717.0, + 1208.0, + 1729.0, + 1197.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1717.0, + 1241.0, + 1717.0, + 1241.0, + 1729.0, + 1229.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1716.0, + 1280.0, + 1716.0, + 1280.0, + 1732.0, + 1257.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1749.0, + 442.0, + 1749.0, + 442.0, + 1760.0, + 430.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1748.0, + 474.0, + 1748.0, + 474.0, + 1761.0, + 462.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1748.0, + 506.0, + 1748.0, + 506.0, + 1761.0, + 494.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1749.0, + 537.0, + 1749.0, + 537.0, + 1760.0, + 526.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1749.0, + 760.0, + 1749.0, + 760.0, + 1760.0, + 748.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1747.0, + 795.0, + 1747.0, + 795.0, + 1763.0, + 779.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1748.0, + 824.0, + 1748.0, + 824.0, + 1761.0, + 812.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 1745.0, + 1187.0, + 1745.0, + 1187.0, + 1764.0, + 1154.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1747.0, + 1215.0, + 1747.0, + 1215.0, + 1761.0, + 1192.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1747.0, + 1247.0, + 1747.0, + 1247.0, + 1763.0, + 1224.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 1747.0, + 1283.0, + 1747.0, + 1283.0, + 1761.0, + 1253.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1775.0, + 511.0, + 1775.0, + 511.0, + 1797.0, + 455.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1775.0, + 672.0, + 1775.0, + 672.0, + 1797.0, + 616.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1776.0, + 831.0, + 1776.0, + 831.0, + 1794.0, + 773.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1775.0, + 989.0, + 1775.0, + 989.0, + 1797.0, + 938.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1776.0, + 1248.0, + 1776.0, + 1248.0, + 1797.0, + 1191.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1838.0, + 406.0, + 1838.0, + 406.0, + 1871.0, + 373.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1841.0, + 440.0, + 1841.0, + 440.0, + 1854.0, + 430.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1841.0, + 474.0, + 1841.0, + 474.0, + 1854.0, + 462.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1841.0, + 506.0, + 1841.0, + 506.0, + 1854.0, + 494.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1841.0, + 538.0, + 1841.0, + 538.0, + 1854.0, + 526.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1841.0, + 601.0, + 1841.0, + 601.0, + 1854.0, + 590.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1841.0, + 633.0, + 1841.0, + 633.0, + 1854.0, + 621.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1841.0, + 665.0, + 1841.0, + 665.0, + 1854.0, + 653.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1841.0, + 697.0, + 1841.0, + 697.0, + 1854.0, + 685.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1841.0, + 761.0, + 1841.0, + 761.0, + 1854.0, + 749.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1841.0, + 793.0, + 1841.0, + 793.0, + 1854.0, + 781.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1839.0, + 827.0, + 1839.0, + 827.0, + 1855.0, + 811.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1839.0, + 859.0, + 1839.0, + 859.0, + 1855.0, + 843.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1842.0, + 919.0, + 1842.0, + 919.0, + 1854.0, + 908.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1841.0, + 951.0, + 1841.0, + 951.0, + 1854.0, + 940.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1841.0, + 983.0, + 1841.0, + 983.0, + 1854.0, + 973.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1841.0, + 1015.0, + 1841.0, + 1015.0, + 1854.0, + 1005.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1839.0, + 1181.0, + 1839.0, + 1181.0, + 1854.0, + 1161.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1839.0, + 1213.0, + 1839.0, + 1213.0, + 1854.0, + 1192.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1839.0, + 1247.0, + 1839.0, + 1247.0, + 1855.0, + 1225.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1839.0, + 1280.0, + 1839.0, + 1280.0, + 1855.0, + 1257.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1874.0, + 472.0, + 1874.0, + 472.0, + 1886.0, + 462.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1874.0, + 505.0, + 1874.0, + 505.0, + 1886.0, + 494.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1874.0, + 537.0, + 1874.0, + 537.0, + 1886.0, + 526.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1874.0, + 601.0, + 1874.0, + 601.0, + 1886.0, + 590.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1874.0, + 633.0, + 1874.0, + 633.0, + 1886.0, + 622.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1874.0, + 665.0, + 1874.0, + 665.0, + 1886.0, + 653.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1874.0, + 697.0, + 1874.0, + 697.0, + 1886.0, + 686.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1874.0, + 793.0, + 1874.0, + 793.0, + 1886.0, + 781.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 1877.0, + 823.0, + 1877.0, + 823.0, + 1883.0, + 816.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1875.0, + 855.0, + 1875.0, + 855.0, + 1886.0, + 846.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1875.0, + 919.0, + 1875.0, + 919.0, + 1886.0, + 910.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1874.0, + 1176.0, + 1874.0, + 1176.0, + 1886.0, + 1165.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1877.0, + 1239.0, + 1877.0, + 1239.0, + 1883.0, + 1232.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1874.0, + 1273.0, + 1874.0, + 1273.0, + 1886.0, + 1263.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1903.0, + 474.0, + 1903.0, + 474.0, + 1915.0, + 462.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1903.0, + 506.0, + 1903.0, + 506.0, + 1915.0, + 494.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1904.0, + 537.0, + 1904.0, + 537.0, + 1915.0, + 526.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1903.0, + 601.0, + 1903.0, + 601.0, + 1916.0, + 590.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1903.0, + 633.0, + 1903.0, + 633.0, + 1916.0, + 621.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1903.0, + 665.0, + 1903.0, + 665.0, + 1916.0, + 653.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1903.0, + 697.0, + 1903.0, + 697.0, + 1915.0, + 686.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1903.0, + 761.0, + 1903.0, + 761.0, + 1915.0, + 749.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1902.0, + 795.0, + 1902.0, + 795.0, + 1918.0, + 779.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1902.0, + 827.0, + 1902.0, + 827.0, + 1918.0, + 811.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1903.0, + 856.0, + 1903.0, + 856.0, + 1916.0, + 844.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1903.0, + 920.0, + 1903.0, + 920.0, + 1915.0, + 908.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1903.0, + 953.0, + 1903.0, + 953.0, + 1916.0, + 940.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1900.0, + 986.0, + 1900.0, + 986.0, + 1918.0, + 971.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1903.0, + 1015.0, + 1903.0, + 1015.0, + 1916.0, + 1005.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1903.0, + 1177.0, + 1903.0, + 1177.0, + 1916.0, + 1165.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1903.0, + 1208.0, + 1903.0, + 1208.0, + 1916.0, + 1197.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1903.0, + 1241.0, + 1903.0, + 1241.0, + 1915.0, + 1229.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1902.0, + 1276.0, + 1902.0, + 1276.0, + 1918.0, + 1260.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1936.0, + 472.0, + 1936.0, + 472.0, + 1947.0, + 462.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1935.0, + 505.0, + 1935.0, + 505.0, + 1947.0, + 494.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1936.0, + 537.0, + 1936.0, + 537.0, + 1947.0, + 526.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1935.0, + 792.0, + 1935.0, + 792.0, + 1948.0, + 780.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1935.0, + 824.0, + 1935.0, + 824.0, + 1948.0, + 812.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1935.0, + 856.0, + 1935.0, + 856.0, + 1948.0, + 846.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 1934.0, + 1183.0, + 1934.0, + 1183.0, + 1950.0, + 1160.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1934.0, + 1215.0, + 1934.0, + 1215.0, + 1950.0, + 1192.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1934.0, + 1247.0, + 1934.0, + 1247.0, + 1950.0, + 1224.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1934.0, + 1280.0, + 1934.0, + 1280.0, + 1950.0, + 1257.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1963.0, + 511.0, + 1963.0, + 511.0, + 1984.0, + 455.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1961.0, + 672.0, + 1961.0, + 672.0, + 1984.0, + 614.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1963.0, + 832.0, + 1963.0, + 832.0, + 1984.0, + 775.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1963.0, + 990.0, + 1963.0, + 990.0, + 1981.0, + 932.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1963.0, + 1248.0, + 1963.0, + 1248.0, + 1984.0, + 1191.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 2015.0, + 597.0, + 2015.0, + 597.0, + 2048.0, + 375.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 2056.0, + 593.0, + 2056.0, + 593.0, + 2085.0, + 376.0, + 2085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 2086.0, + 570.0, + 2086.0, + 570.0, + 2119.0, + 380.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.75, + 1686.5, + 921.75, + 1686.5, + 921.75, + 1697.5, + 906.75, + 1697.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1875.5, + 1017.0, + 1875.5, + 1017.0, + 1884.0, + 1003.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1875.5, + 1209.0, + 1875.5, + 1209.0, + 1884.0, + 1196.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 124.0, + 183.0, + 1192.0, + 183.0, + 1192.0, + 254.0, + 124.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 242.0, + 839.0, + 242.0, + 839.0, + 289.0, + 126.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 895.0, + 781.0, + 895.0, + 781.0, + 942.0, + 288.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 323.0, + 1402.0, + 323.0, + 1402.0, + 355.0, + 1195.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 350.0, + 1369.0, + 350.0, + 1369.0, + 381.0, + 1231.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 632.0, + 470.0, + 632.0, + 470.0, + 662.0, + 348.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 632.0, + 645.0, + 632.0, + 645.0, + 662.0, + 524.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 632.0, + 855.0, + 632.0, + 855.0, + 662.0, + 733.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 628.0, + 1038.0, + 628.0, + 1038.0, + 665.0, + 911.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 625.0, + 1374.0, + 625.0, + 1374.0, + 660.0, + 1167.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 657.0, + 455.0, + 657.0, + 455.0, + 688.0, + 359.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 658.0, + 652.0, + 658.0, + 652.0, + 688.0, + 522.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 656.0, + 849.0, + 656.0, + 849.0, + 690.0, + 738.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 656.0, + 1054.0, + 656.0, + 1054.0, + 690.0, + 904.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 652.0, + 1376.0, + 652.0, + 1376.0, + 689.0, + 1173.0, + 689.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 29, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 264, + 505, + 1407, + 505, + 1407, + 672, + 264, + 672 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 915, + 221, + 1404, + 221, + 1404, + 277, + 915, + 277 + ], + "score": 0.861 + }, + { + "category_id": 3, + "poly": [ + 307, + 1595, + 1389, + 1595, + 1389, + 2070, + 307, + 2070 + ], + "score": 0.861 + }, + { + "category_id": 0, + "poly": [ + 204, + 709, + 871, + 709, + 871, + 744, + 204, + 744 + ], + "score": 0.855 + }, + { + "category_id": 0, + "poly": [ + 218, + 469, + 703, + 469, + 703, + 502, + 218, + 502 + ], + "score": 0.757 + }, + { + "category_id": 1, + "poly": [ + 804, + 313, + 1516, + 313, + 1516, + 354, + 804, + 354 + ], + "score": 0.732 + }, + { + "category_id": 3, + "poly": [ + 308, + 1200, + 1386, + 1200, + 1386, + 1510, + 308, + 1510 + ], + "score": 0.654 + }, + { + "category_id": 3, + "poly": [ + 211, + 224, + 706, + 224, + 706, + 443, + 211, + 443 + ], + "score": 0.623 + }, + { + "category_id": 3, + "poly": [ + 308, + 806, + 1329, + 806, + 1329, + 1127, + 308, + 1127 + ], + "score": 0.607 + }, + { + "category_id": 4, + "poly": [ + 228, + 1545, + 487, + 1545, + 487, + 1572, + 228, + 1572 + ], + "score": 0.404 + }, + { + "category_id": 1, + "poly": [ + 227, + 780, + 489, + 780, + 489, + 809, + 227, + 809 + ], + "score": 0.238 + }, + { + "category_id": 1, + "poly": [ + 227, + 1155, + 488, + 1155, + 488, + 1182, + 227, + 1182 + ], + "score": 0.147 + }, + { + "category_id": 4, + "poly": [ + 227, + 1155, + 488, + 1155, + 488, + 1182, + 227, + 1182 + ], + "score": 0.132 + }, + { + "category_id": 13, + "poly": [ + 909, + 506, + 1022, + 506, + 1022, + 612, + 909, + 612 + ], + "score": 0.52, + "latex": "\\begin{array} { l } { { \\mathbf { f } . } } \\\\ { { \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } } } \\\\ { { \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } } } \\\\ { { \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } } } \\\\ { { \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } } } \\\\ { { \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } \\quad \\mathbf { \\imath } _ { 1 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1294, + 505, + 1405, + 505, + 1405, + 659, + 1294, + 659 + ], + "score": 0.45, + "latex": "\\begin{array} { r l } & { \\mathrm { i } ; } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ \\ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\vert ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ 1 ~ } ~ } } \\\\ & { \\mathrm { ~ { \\mathrm { ~ s u m } } = 2 5 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 782, + 516, + 881, + 516, + 881, + 615, + 782, + 615 + ], + "score": 0.41, + "latex": "\\begin{array} { l } { { \\bf \\in . } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ } } } \\\\ { { \\bf \\textsc { 1 ~ 1 ~ 1 ~ } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 963, + 633, + 1007, + 633, + 1007, + 656, + 963, + 656 + ], + "score": 0.38, + "latex": "= 2 0" + }, + { + "category_id": 13, + "poly": [ + 834, + 633, + 878, + 633, + 878, + 656, + 834, + 656 + ], + "score": 0.33, + "latex": "= 1 6" + }, + { + "category_id": 13, + "poly": [ + 706, + 633, + 749, + 633, + 749, + 656, + 706, + 656 + ], + "score": 0.3, + "latex": "= 2 4" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 506.0, + 298.0, + 506.0, + 298.0, + 529.0, + 274.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 504.0, + 428.0, + 504.0, + 428.0, + 530.0, + 399.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 506.0, + 553.0, + 506.0, + 553.0, + 529.0, + 531.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 505.0, + 681.0, + 505.0, + 681.0, + 529.0, + 658.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 506.0, + 811.0, + 506.0, + 811.0, + 529.0, + 786.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 508.0, + 1066.0, + 508.0, + 1066.0, + 531.0, + 1043.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 505.0, + 1195.0, + 505.0, + 1195.0, + 529.0, + 1171.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 531.0, + 292.0, + 531.0, + 292.0, + 548.0, + 276.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 531.0, + 314.0, + 531.0, + 314.0, + 548.0, + 296.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 531.0, + 333.0, + 531.0, + 333.0, + 548.0, + 316.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 534.0, + 417.0, + 534.0, + 417.0, + 547.0, + 404.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 534.0, + 439.0, + 534.0, + 439.0, + 547.0, + 426.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 533.0, + 460.0, + 533.0, + 460.0, + 547.0, + 448.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 531.0, + 482.0, + 531.0, + 482.0, + 548.0, + 466.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 531.0, + 547.0, + 531.0, + 547.0, + 548.0, + 531.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 534.0, + 567.0, + 534.0, + 567.0, + 545.0, + 553.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 534.0, + 588.0, + 534.0, + 588.0, + 547.0, + 577.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 534.0, + 610.0, + 534.0, + 610.0, + 547.0, + 599.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 534.0, + 631.0, + 534.0, + 631.0, + 545.0, + 621.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 533.0, + 674.0, + 533.0, + 674.0, + 547.0, + 662.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 536.0, + 694.0, + 536.0, + 694.0, + 545.0, + 683.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 534.0, + 715.0, + 534.0, + 715.0, + 545.0, + 705.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 534.0, + 1058.0, + 534.0, + 1058.0, + 547.0, + 1045.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 536.0, + 1079.0, + 536.0, + 1079.0, + 545.0, + 1069.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 536.0, + 1100.0, + 536.0, + 1100.0, + 547.0, + 1089.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 536.0, + 1185.0, + 536.0, + 1185.0, + 547.0, + 1175.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 536.0, + 1206.0, + 536.0, + 1206.0, + 545.0, + 1195.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 536.0, + 1228.0, + 536.0, + 1228.0, + 545.0, + 1216.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 536.0, + 1249.0, + 536.0, + 1249.0, + 545.0, + 1238.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 551.0, + 292.0, + 551.0, + 292.0, + 569.0, + 276.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 551.0, + 314.0, + 551.0, + 314.0, + 569.0, + 296.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 551.0, + 333.0, + 551.0, + 333.0, + 568.0, + 318.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 552.0, + 417.0, + 552.0, + 417.0, + 566.0, + 406.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 551.0, + 441.0, + 551.0, + 441.0, + 568.0, + 425.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 554.0, + 460.0, + 554.0, + 460.0, + 566.0, + 448.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 554.0, + 481.0, + 554.0, + 481.0, + 566.0, + 470.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 552.0, + 545.0, + 552.0, + 545.0, + 566.0, + 534.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 552.0, + 567.0, + 552.0, + 567.0, + 565.0, + 555.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 551.0, + 591.0, + 551.0, + 591.0, + 568.0, + 574.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 551.0, + 612.0, + 551.0, + 612.0, + 568.0, + 596.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 554.0, + 631.0, + 554.0, + 631.0, + 566.0, + 619.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 551.0, + 676.0, + 551.0, + 676.0, + 569.0, + 659.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 551.0, + 698.0, + 551.0, + 698.0, + 568.0, + 681.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 551.0, + 718.0, + 551.0, + 718.0, + 569.0, + 702.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 551.0, + 1061.0, + 551.0, + 1061.0, + 569.0, + 1044.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 551.0, + 1082.0, + 551.0, + 1082.0, + 568.0, + 1065.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 551.0, + 1101.0, + 551.0, + 1101.0, + 569.0, + 1087.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 552.0, + 1185.0, + 552.0, + 1185.0, + 566.0, + 1175.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 554.0, + 1207.0, + 554.0, + 1207.0, + 566.0, + 1195.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 552.0, + 1232.0, + 552.0, + 1232.0, + 568.0, + 1215.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 551.0, + 1251.0, + 551.0, + 1251.0, + 569.0, + 1236.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 572.0, + 292.0, + 572.0, + 292.0, + 590.0, + 275.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 572.0, + 314.0, + 572.0, + 314.0, + 590.0, + 296.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 573.0, + 333.0, + 573.0, + 333.0, + 590.0, + 316.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 573.0, + 419.0, + 573.0, + 419.0, + 590.0, + 403.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 575.0, + 438.0, + 575.0, + 438.0, + 587.0, + 426.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 572.0, + 461.0, + 572.0, + 461.0, + 589.0, + 446.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 572.0, + 482.0, + 572.0, + 482.0, + 590.0, + 466.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 573.0, + 547.0, + 573.0, + 547.0, + 590.0, + 531.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 573.0, + 569.0, + 573.0, + 569.0, + 590.0, + 552.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 575.0, + 588.0, + 575.0, + 588.0, + 587.0, + 577.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 575.0, + 609.0, + 575.0, + 609.0, + 587.0, + 597.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 572.0, + 632.0, + 572.0, + 632.0, + 589.0, + 617.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 575.0, + 672.0, + 575.0, + 672.0, + 589.0, + 661.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 575.0, + 696.0, + 575.0, + 696.0, + 587.0, + 683.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 573.0, + 718.0, + 573.0, + 718.0, + 590.0, + 702.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 572.0, + 1061.0, + 572.0, + 1061.0, + 590.0, + 1044.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 572.0, + 1082.0, + 572.0, + 1082.0, + 590.0, + 1065.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 573.0, + 1102.0, + 573.0, + 1102.0, + 590.0, + 1087.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 575.0, + 1186.0, + 575.0, + 1186.0, + 587.0, + 1175.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 576.0, + 1207.0, + 576.0, + 1207.0, + 586.0, + 1197.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 575.0, + 1229.0, + 575.0, + 1229.0, + 587.0, + 1216.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 573.0, + 1251.0, + 573.0, + 1251.0, + 590.0, + 1236.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 591.0, + 675.0, + 591.0, + 675.0, + 609.0, + 659.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 595.0, + 693.0, + 595.0, + 693.0, + 605.0, + 683.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 593.0, + 718.0, + 593.0, + 718.0, + 609.0, + 702.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 591.0, + 1061.0, + 591.0, + 1061.0, + 609.0, + 1044.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 593.0, + 1082.0, + 593.0, + 1082.0, + 608.0, + 1065.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 594.0, + 1100.0, + 594.0, + 1100.0, + 608.0, + 1088.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 594.0, + 1185.0, + 594.0, + 1185.0, + 607.0, + 1175.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 614.0, + 1058.0, + 614.0, + 1058.0, + 626.0, + 1045.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 614.0, + 1080.0, + 614.0, + 1080.0, + 625.0, + 1067.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 614.0, + 1100.0, + 614.0, + 1100.0, + 626.0, + 1089.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 614.0, + 1185.0, + 614.0, + 1185.0, + 626.0, + 1175.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 615.0, + 1206.0, + 615.0, + 1206.0, + 625.0, + 1195.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 614.0, + 1249.0, + 614.0, + 1249.0, + 625.0, + 1238.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 630.0, + 369.0, + 630.0, + 369.0, + 658.0, + 279.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 632.0, + 499.0, + 632.0, + 499.0, + 658.0, + 407.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 632.0, + 626.0, + 632.0, + 626.0, + 658.0, + 535.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 632.0, + 705.0, + 632.0, + 705.0, + 658.0, + 662.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 632.0, + 754.0, + 632.0, + 754.0, + 658.0, + 750.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 632.0, + 833.0, + 632.0, + 833.0, + 656.0, + 790.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 632.0, + 882.0, + 632.0, + 882.0, + 656.0, + 879.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 632.0, + 962.0, + 632.0, + 962.0, + 658.0, + 920.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 632.0, + 1138.0, + 632.0, + 1138.0, + 656.0, + 1046.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 632.0, + 1267.0, + 632.0, + 1267.0, + 658.0, + 1175.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 654.0, + 1382.0, + 654.0, + 1382.0, + 676.0, + 1306.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 550.5, + 957.0, + 550.5, + 957.0, + 566.5, + 932.0, + 566.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 593.0, + 1253.0, + 593.0, + 1253.0, + 607.5, + 1234.0, + 607.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 213.0, + 1408.0, + 213.0, + 1408.0, + 285.0, + 911.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1610.0, + 333.0, + 1610.0, + 333.0, + 1624.0, + 317.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1610.0, + 354.0, + 1610.0, + 354.0, + 1624.0, + 339.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1610.0, + 376.0, + 1610.0, + 376.0, + 1624.0, + 360.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1608.0, + 397.0, + 1608.0, + 397.0, + 1624.0, + 381.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1608.0, + 420.0, + 1608.0, + 420.0, + 1624.0, + 405.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1610.0, + 504.0, + 1610.0, + 504.0, + 1624.0, + 490.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1610.0, + 527.0, + 1610.0, + 527.0, + 1624.0, + 511.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1610.0, + 546.0, + 1610.0, + 546.0, + 1624.0, + 532.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1608.0, + 569.0, + 1608.0, + 569.0, + 1624.0, + 553.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1608.0, + 589.0, + 1608.0, + 589.0, + 1624.0, + 573.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1607.0, + 742.0, + 1607.0, + 742.0, + 1625.0, + 658.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1608.0, + 761.0, + 1608.0, + 761.0, + 1624.0, + 743.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1608.0, + 847.0, + 1608.0, + 847.0, + 1625.0, + 831.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1610.0, + 868.0, + 1610.0, + 868.0, + 1624.0, + 852.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1610.0, + 890.0, + 1610.0, + 890.0, + 1624.0, + 873.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1608.0, + 911.0, + 1608.0, + 911.0, + 1624.0, + 894.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1610.0, + 930.0, + 1610.0, + 930.0, + 1624.0, + 916.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1610.0, + 1039.0, + 1610.0, + 1039.0, + 1624.0, + 1023.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1610.0, + 1061.0, + 1610.0, + 1061.0, + 1624.0, + 1044.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1610.0, + 1081.0, + 1610.0, + 1081.0, + 1624.0, + 1066.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1608.0, + 1103.0, + 1608.0, + 1103.0, + 1624.0, + 1087.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1608.0, + 1123.0, + 1608.0, + 1123.0, + 1624.0, + 1108.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1605.0, + 1316.0, + 1605.0, + 1316.0, + 1628.0, + 1211.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1630.0, + 332.0, + 1630.0, + 332.0, + 1644.0, + 317.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1630.0, + 354.0, + 1630.0, + 354.0, + 1644.0, + 339.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1630.0, + 376.0, + 1630.0, + 376.0, + 1644.0, + 362.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1629.0, + 396.0, + 1629.0, + 396.0, + 1644.0, + 381.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1631.0, + 418.0, + 1631.0, + 418.0, + 1643.0, + 407.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1630.0, + 504.0, + 1630.0, + 504.0, + 1644.0, + 490.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1630.0, + 525.0, + 1630.0, + 525.0, + 1644.0, + 511.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1631.0, + 545.0, + 1631.0, + 545.0, + 1642.0, + 534.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1630.0, + 566.0, + 1630.0, + 566.0, + 1642.0, + 555.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1630.0, + 588.0, + 1630.0, + 588.0, + 1642.0, + 577.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1630.0, + 676.0, + 1630.0, + 676.0, + 1644.0, + 660.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1630.0, + 698.0, + 1630.0, + 698.0, + 1644.0, + 681.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1630.0, + 719.0, + 1630.0, + 719.0, + 1644.0, + 701.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1629.0, + 738.0, + 1629.0, + 738.0, + 1643.0, + 724.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1630.0, + 761.0, + 1630.0, + 761.0, + 1644.0, + 746.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1630.0, + 847.0, + 1630.0, + 847.0, + 1646.0, + 831.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1630.0, + 866.0, + 1630.0, + 866.0, + 1644.0, + 852.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1629.0, + 889.0, + 1629.0, + 889.0, + 1644.0, + 874.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1629.0, + 910.0, + 1629.0, + 910.0, + 1644.0, + 894.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1631.0, + 929.0, + 1631.0, + 929.0, + 1643.0, + 919.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1630.0, + 1038.0, + 1630.0, + 1038.0, + 1644.0, + 1023.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1631.0, + 1057.0, + 1631.0, + 1057.0, + 1643.0, + 1046.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1630.0, + 1081.0, + 1630.0, + 1081.0, + 1644.0, + 1066.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1629.0, + 1103.0, + 1629.0, + 1103.0, + 1644.0, + 1087.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1629.0, + 1124.0, + 1629.0, + 1124.0, + 1644.0, + 1109.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1630.0, + 1231.0, + 1630.0, + 1231.0, + 1644.0, + 1215.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 1630.0, + 1254.0, + 1630.0, + 1254.0, + 1644.0, + 1235.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1630.0, + 1274.0, + 1630.0, + 1274.0, + 1644.0, + 1257.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1629.0, + 1295.0, + 1629.0, + 1295.0, + 1643.0, + 1278.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1631.0, + 1313.0, + 1631.0, + 1313.0, + 1643.0, + 1302.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1650.0, + 332.0, + 1650.0, + 332.0, + 1666.0, + 317.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1649.0, + 354.0, + 1649.0, + 354.0, + 1665.0, + 339.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1649.0, + 376.0, + 1649.0, + 376.0, + 1665.0, + 362.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1648.0, + 397.0, + 1648.0, + 397.0, + 1664.0, + 381.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1649.0, + 420.0, + 1649.0, + 420.0, + 1664.0, + 405.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1650.0, + 504.0, + 1650.0, + 504.0, + 1665.0, + 490.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1652.0, + 523.0, + 1652.0, + 523.0, + 1664.0, + 512.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1649.0, + 548.0, + 1649.0, + 548.0, + 1665.0, + 532.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1649.0, + 569.0, + 1649.0, + 569.0, + 1665.0, + 554.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1649.0, + 589.0, + 1649.0, + 589.0, + 1664.0, + 575.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1650.0, + 676.0, + 1650.0, + 676.0, + 1665.0, + 660.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1649.0, + 697.0, + 1649.0, + 697.0, + 1665.0, + 682.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1649.0, + 719.0, + 1649.0, + 719.0, + 1665.0, + 703.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1649.0, + 738.0, + 1649.0, + 738.0, + 1665.0, + 724.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1652.0, + 759.0, + 1652.0, + 759.0, + 1662.0, + 748.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1649.0, + 847.0, + 1649.0, + 847.0, + 1666.0, + 831.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1649.0, + 866.0, + 1649.0, + 866.0, + 1665.0, + 853.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1649.0, + 889.0, + 1649.0, + 889.0, + 1665.0, + 874.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1648.0, + 911.0, + 1648.0, + 911.0, + 1665.0, + 895.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1649.0, + 930.0, + 1649.0, + 930.0, + 1665.0, + 917.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1652.0, + 1036.0, + 1652.0, + 1036.0, + 1664.0, + 1025.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1652.0, + 1057.0, + 1652.0, + 1057.0, + 1664.0, + 1046.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1649.0, + 1081.0, + 1649.0, + 1081.0, + 1665.0, + 1066.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1648.0, + 1103.0, + 1648.0, + 1103.0, + 1665.0, + 1087.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1649.0, + 1124.0, + 1649.0, + 1124.0, + 1665.0, + 1109.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1649.0, + 1231.0, + 1649.0, + 1231.0, + 1666.0, + 1215.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1652.0, + 1249.0, + 1652.0, + 1249.0, + 1664.0, + 1238.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1649.0, + 1273.0, + 1649.0, + 1273.0, + 1665.0, + 1258.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1649.0, + 1295.0, + 1649.0, + 1295.0, + 1664.0, + 1279.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1652.0, + 1315.0, + 1652.0, + 1315.0, + 1662.0, + 1302.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1668.0, + 332.0, + 1668.0, + 332.0, + 1685.0, + 317.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1668.0, + 354.0, + 1668.0, + 354.0, + 1685.0, + 339.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1668.0, + 376.0, + 1668.0, + 376.0, + 1685.0, + 362.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1670.0, + 397.0, + 1670.0, + 397.0, + 1685.0, + 383.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1671.0, + 418.0, + 1671.0, + 418.0, + 1680.0, + 407.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1668.0, + 504.0, + 1668.0, + 504.0, + 1685.0, + 490.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1671.0, + 523.0, + 1671.0, + 523.0, + 1684.0, + 513.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1668.0, + 546.0, + 1668.0, + 546.0, + 1685.0, + 532.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1670.0, + 569.0, + 1670.0, + 569.0, + 1685.0, + 553.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1670.0, + 589.0, + 1670.0, + 589.0, + 1682.0, + 580.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1670.0, + 674.0, + 1670.0, + 674.0, + 1685.0, + 660.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1670.0, + 697.0, + 1670.0, + 697.0, + 1685.0, + 682.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1668.0, + 719.0, + 1668.0, + 719.0, + 1685.0, + 704.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1670.0, + 740.0, + 1670.0, + 740.0, + 1685.0, + 724.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1671.0, + 761.0, + 1671.0, + 761.0, + 1682.0, + 751.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1667.0, + 848.0, + 1667.0, + 848.0, + 1688.0, + 830.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1668.0, + 866.0, + 1668.0, + 866.0, + 1685.0, + 852.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1668.0, + 889.0, + 1668.0, + 889.0, + 1685.0, + 874.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1670.0, + 910.0, + 1670.0, + 910.0, + 1685.0, + 894.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1671.0, + 930.0, + 1671.0, + 930.0, + 1682.0, + 919.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1670.0, + 1038.0, + 1670.0, + 1038.0, + 1685.0, + 1024.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1670.0, + 1059.0, + 1670.0, + 1059.0, + 1685.0, + 1045.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1668.0, + 1081.0, + 1668.0, + 1081.0, + 1684.0, + 1066.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1670.0, + 1102.0, + 1670.0, + 1102.0, + 1685.0, + 1087.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1670.0, + 1124.0, + 1670.0, + 1124.0, + 1682.0, + 1113.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1668.0, + 1231.0, + 1668.0, + 1231.0, + 1685.0, + 1216.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1671.0, + 1249.0, + 1671.0, + 1249.0, + 1683.0, + 1238.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1671.0, + 1272.0, + 1671.0, + 1272.0, + 1683.0, + 1260.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1672.0, + 1292.0, + 1672.0, + 1292.0, + 1684.0, + 1281.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1671.0, + 1315.0, + 1671.0, + 1315.0, + 1682.0, + 1306.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1692.0, + 330.0, + 1692.0, + 330.0, + 1702.0, + 319.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1690.0, + 354.0, + 1690.0, + 354.0, + 1706.0, + 339.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1692.0, + 374.0, + 1692.0, + 374.0, + 1704.0, + 363.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1691.0, + 396.0, + 1691.0, + 396.0, + 1702.0, + 385.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1691.0, + 504.0, + 1691.0, + 504.0, + 1707.0, + 490.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1690.0, + 524.0, + 1690.0, + 524.0, + 1707.0, + 511.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1690.0, + 546.0, + 1690.0, + 546.0, + 1707.0, + 532.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1689.0, + 569.0, + 1689.0, + 569.0, + 1704.0, + 553.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1691.0, + 587.0, + 1691.0, + 587.0, + 1701.0, + 578.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1690.0, + 674.0, + 1690.0, + 674.0, + 1707.0, + 660.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1691.0, + 697.0, + 1691.0, + 697.0, + 1707.0, + 682.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1690.0, + 719.0, + 1690.0, + 719.0, + 1707.0, + 703.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1689.0, + 738.0, + 1689.0, + 738.0, + 1704.0, + 724.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1690.0, + 847.0, + 1690.0, + 847.0, + 1707.0, + 831.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1690.0, + 866.0, + 1690.0, + 866.0, + 1706.0, + 850.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1690.0, + 889.0, + 1690.0, + 889.0, + 1706.0, + 874.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1691.0, + 908.0, + 1691.0, + 908.0, + 1702.0, + 896.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1690.0, + 1038.0, + 1690.0, + 1038.0, + 1707.0, + 1024.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1692.0, + 1057.0, + 1692.0, + 1057.0, + 1704.0, + 1046.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1691.0, + 1081.0, + 1691.0, + 1081.0, + 1707.0, + 1066.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1689.0, + 1103.0, + 1689.0, + 1103.0, + 1704.0, + 1087.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1692.0, + 1120.0, + 1692.0, + 1120.0, + 1702.0, + 1111.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1690.0, + 1231.0, + 1690.0, + 1231.0, + 1707.0, + 1216.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1692.0, + 1249.0, + 1692.0, + 1249.0, + 1704.0, + 1238.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1691.0, + 1273.0, + 1691.0, + 1273.0, + 1707.0, + 1258.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 1690.0, + 1295.0, + 1690.0, + 1295.0, + 1704.0, + 1280.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1770.0, + 333.0, + 1770.0, + 333.0, + 1786.0, + 317.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1770.0, + 354.0, + 1770.0, + 354.0, + 1786.0, + 339.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1769.0, + 376.0, + 1769.0, + 376.0, + 1786.0, + 362.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1769.0, + 397.0, + 1769.0, + 397.0, + 1783.0, + 381.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1771.0, + 417.0, + 1771.0, + 417.0, + 1783.0, + 406.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1770.0, + 504.0, + 1770.0, + 504.0, + 1785.0, + 490.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1770.0, + 525.0, + 1770.0, + 525.0, + 1785.0, + 511.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1771.0, + 545.0, + 1771.0, + 545.0, + 1782.0, + 533.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1770.0, + 566.0, + 1770.0, + 566.0, + 1782.0, + 555.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1770.0, + 678.0, + 1770.0, + 678.0, + 1785.0, + 661.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1770.0, + 698.0, + 1770.0, + 698.0, + 1785.0, + 679.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1770.0, + 719.0, + 1770.0, + 719.0, + 1785.0, + 703.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1768.0, + 761.0, + 1768.0, + 761.0, + 1786.0, + 721.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1770.0, + 847.0, + 1770.0, + 847.0, + 1786.0, + 831.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1770.0, + 869.0, + 1770.0, + 869.0, + 1786.0, + 852.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1770.0, + 890.0, + 1770.0, + 890.0, + 1786.0, + 873.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1767.0, + 912.0, + 1767.0, + 912.0, + 1786.0, + 892.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1769.0, + 932.0, + 1769.0, + 932.0, + 1785.0, + 916.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1770.0, + 1039.0, + 1770.0, + 1039.0, + 1786.0, + 1024.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1771.0, + 1060.0, + 1771.0, + 1060.0, + 1786.0, + 1045.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1770.0, + 1081.0, + 1770.0, + 1081.0, + 1785.0, + 1065.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1770.0, + 1100.0, + 1770.0, + 1100.0, + 1782.0, + 1089.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1773.0, + 1121.0, + 1773.0, + 1121.0, + 1782.0, + 1111.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1770.0, + 1231.0, + 1770.0, + 1231.0, + 1786.0, + 1216.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1770.0, + 1253.0, + 1770.0, + 1253.0, + 1785.0, + 1236.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1770.0, + 1274.0, + 1770.0, + 1274.0, + 1785.0, + 1257.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1769.0, + 1296.0, + 1769.0, + 1296.0, + 1785.0, + 1279.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1770.0, + 1316.0, + 1770.0, + 1316.0, + 1785.0, + 1300.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1791.0, + 333.0, + 1791.0, + 333.0, + 1806.0, + 317.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1791.0, + 354.0, + 1791.0, + 354.0, + 1805.0, + 339.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1791.0, + 376.0, + 1791.0, + 376.0, + 1805.0, + 360.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1789.0, + 397.0, + 1789.0, + 397.0, + 1804.0, + 381.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1792.0, + 417.0, + 1792.0, + 417.0, + 1803.0, + 406.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1791.0, + 504.0, + 1791.0, + 504.0, + 1805.0, + 490.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1791.0, + 525.0, + 1791.0, + 525.0, + 1805.0, + 511.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1792.0, + 545.0, + 1792.0, + 545.0, + 1803.0, + 533.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1791.0, + 567.0, + 1791.0, + 567.0, + 1801.0, + 555.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1791.0, + 678.0, + 1791.0, + 678.0, + 1805.0, + 660.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1791.0, + 698.0, + 1791.0, + 698.0, + 1805.0, + 679.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1791.0, + 719.0, + 1791.0, + 719.0, + 1804.0, + 703.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1789.0, + 741.0, + 1789.0, + 741.0, + 1804.0, + 724.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1793.0, + 759.0, + 1793.0, + 759.0, + 1801.0, + 747.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1791.0, + 847.0, + 1791.0, + 847.0, + 1806.0, + 831.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1791.0, + 866.0, + 1791.0, + 866.0, + 1805.0, + 852.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1788.0, + 889.0, + 1788.0, + 889.0, + 1804.0, + 873.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1788.0, + 911.0, + 1788.0, + 911.0, + 1804.0, + 894.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1792.0, + 929.0, + 1792.0, + 929.0, + 1803.0, + 917.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1791.0, + 1039.0, + 1791.0, + 1039.0, + 1806.0, + 1023.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1791.0, + 1060.0, + 1791.0, + 1060.0, + 1805.0, + 1044.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1791.0, + 1081.0, + 1791.0, + 1081.0, + 1805.0, + 1066.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1788.0, + 1102.0, + 1788.0, + 1102.0, + 1803.0, + 1087.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1792.0, + 1121.0, + 1792.0, + 1121.0, + 1801.0, + 1111.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1791.0, + 1231.0, + 1791.0, + 1231.0, + 1806.0, + 1215.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1791.0, + 1252.0, + 1791.0, + 1252.0, + 1805.0, + 1236.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1791.0, + 1273.0, + 1791.0, + 1273.0, + 1805.0, + 1257.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1789.0, + 1294.0, + 1789.0, + 1294.0, + 1804.0, + 1279.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1792.0, + 1315.0, + 1792.0, + 1315.0, + 1803.0, + 1302.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1810.0, + 333.0, + 1810.0, + 333.0, + 1825.0, + 319.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1810.0, + 354.0, + 1810.0, + 354.0, + 1825.0, + 339.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1810.0, + 376.0, + 1810.0, + 376.0, + 1825.0, + 360.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1809.0, + 397.0, + 1809.0, + 397.0, + 1825.0, + 381.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1810.0, + 418.0, + 1810.0, + 418.0, + 1825.0, + 404.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1811.0, + 504.0, + 1811.0, + 504.0, + 1825.0, + 490.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1812.0, + 524.0, + 1812.0, + 524.0, + 1823.0, + 512.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1812.0, + 545.0, + 1812.0, + 545.0, + 1823.0, + 534.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1812.0, + 567.0, + 1812.0, + 567.0, + 1822.0, + 555.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1811.0, + 677.0, + 1811.0, + 677.0, + 1825.0, + 660.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1811.0, + 697.0, + 1811.0, + 697.0, + 1824.0, + 679.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1810.0, + 719.0, + 1810.0, + 719.0, + 1824.0, + 703.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1810.0, + 738.0, + 1810.0, + 738.0, + 1824.0, + 724.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1810.0, + 847.0, + 1810.0, + 847.0, + 1825.0, + 831.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1810.0, + 866.0, + 1810.0, + 866.0, + 1825.0, + 852.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1810.0, + 889.0, + 1810.0, + 889.0, + 1825.0, + 873.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1809.0, + 911.0, + 1809.0, + 911.0, + 1824.0, + 894.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1810.0, + 932.0, + 1810.0, + 932.0, + 1825.0, + 916.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1810.0, + 1039.0, + 1810.0, + 1039.0, + 1825.0, + 1023.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1811.0, + 1059.0, + 1811.0, + 1059.0, + 1825.0, + 1044.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1810.0, + 1081.0, + 1810.0, + 1081.0, + 1825.0, + 1066.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1811.0, + 1100.0, + 1811.0, + 1100.0, + 1823.0, + 1089.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1811.0, + 1123.0, + 1811.0, + 1123.0, + 1823.0, + 1111.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1811.0, + 1231.0, + 1811.0, + 1231.0, + 1825.0, + 1215.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1812.0, + 1249.0, + 1812.0, + 1249.0, + 1823.0, + 1237.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1810.0, + 1273.0, + 1810.0, + 1273.0, + 1825.0, + 1258.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1809.0, + 1294.0, + 1809.0, + 1294.0, + 1824.0, + 1279.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1811.0, + 1315.0, + 1811.0, + 1315.0, + 1823.0, + 1304.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1830.0, + 333.0, + 1830.0, + 333.0, + 1846.0, + 319.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1830.0, + 354.0, + 1830.0, + 354.0, + 1846.0, + 339.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1830.0, + 376.0, + 1830.0, + 376.0, + 1846.0, + 362.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1831.0, + 397.0, + 1831.0, + 397.0, + 1846.0, + 381.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1831.0, + 417.0, + 1831.0, + 417.0, + 1843.0, + 407.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1830.0, + 504.0, + 1830.0, + 504.0, + 1846.0, + 490.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1830.0, + 524.0, + 1830.0, + 524.0, + 1846.0, + 511.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1833.0, + 545.0, + 1833.0, + 545.0, + 1843.0, + 534.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1831.0, + 569.0, + 1831.0, + 569.0, + 1846.0, + 553.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1831.0, + 676.0, + 1831.0, + 676.0, + 1846.0, + 660.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1831.0, + 697.0, + 1831.0, + 697.0, + 1846.0, + 679.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1833.0, + 717.0, + 1833.0, + 717.0, + 1843.0, + 705.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1831.0, + 740.0, + 1831.0, + 740.0, + 1846.0, + 724.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1830.0, + 847.0, + 1830.0, + 847.0, + 1846.0, + 831.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1830.0, + 866.0, + 1830.0, + 866.0, + 1846.0, + 852.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1830.0, + 890.0, + 1830.0, + 890.0, + 1846.0, + 874.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1831.0, + 911.0, + 1831.0, + 911.0, + 1846.0, + 894.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1831.0, + 930.0, + 1831.0, + 930.0, + 1842.0, + 919.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1830.0, + 1038.0, + 1830.0, + 1038.0, + 1846.0, + 1023.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1833.0, + 1057.0, + 1833.0, + 1057.0, + 1844.0, + 1046.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1830.0, + 1081.0, + 1830.0, + 1081.0, + 1846.0, + 1066.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1833.0, + 1100.0, + 1833.0, + 1100.0, + 1844.0, + 1089.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1830.0, + 1231.0, + 1830.0, + 1231.0, + 1846.0, + 1216.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1831.0, + 1251.0, + 1831.0, + 1251.0, + 1846.0, + 1236.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1830.0, + 1273.0, + 1830.0, + 1273.0, + 1846.0, + 1258.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1833.0, + 1292.0, + 1833.0, + 1292.0, + 1844.0, + 1281.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1852.0, + 332.0, + 1852.0, + 332.0, + 1868.0, + 317.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1852.0, + 354.0, + 1852.0, + 354.0, + 1867.0, + 339.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1852.0, + 376.0, + 1852.0, + 376.0, + 1866.0, + 360.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1850.0, + 397.0, + 1850.0, + 397.0, + 1865.0, + 383.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1853.0, + 504.0, + 1853.0, + 504.0, + 1867.0, + 490.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1854.0, + 523.0, + 1854.0, + 523.0, + 1866.0, + 512.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1853.0, + 546.0, + 1853.0, + 546.0, + 1867.0, + 532.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1850.0, + 569.0, + 1850.0, + 569.0, + 1865.0, + 553.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1853.0, + 676.0, + 1853.0, + 676.0, + 1867.0, + 661.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1853.0, + 697.0, + 1853.0, + 697.0, + 1867.0, + 681.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1854.0, + 716.0, + 1854.0, + 716.0, + 1866.0, + 705.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1850.0, + 741.0, + 1850.0, + 741.0, + 1865.0, + 724.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1852.0, + 847.0, + 1852.0, + 847.0, + 1868.0, + 831.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1852.0, + 866.0, + 1852.0, + 866.0, + 1868.0, + 852.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1852.0, + 890.0, + 1852.0, + 890.0, + 1868.0, + 874.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1850.0, + 910.0, + 1850.0, + 910.0, + 1865.0, + 894.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1853.0, + 1038.0, + 1853.0, + 1038.0, + 1868.0, + 1024.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1854.0, + 1057.0, + 1854.0, + 1057.0, + 1866.0, + 1046.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1853.0, + 1081.0, + 1853.0, + 1081.0, + 1868.0, + 1066.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1853.0, + 1100.0, + 1853.0, + 1100.0, + 1864.0, + 1089.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1853.0, + 1231.0, + 1853.0, + 1231.0, + 1867.0, + 1216.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1853.0, + 1251.0, + 1853.0, + 1251.0, + 1867.0, + 1236.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1853.0, + 1273.0, + 1853.0, + 1273.0, + 1867.0, + 1258.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1853.0, + 1292.0, + 1853.0, + 1292.0, + 1864.0, + 1281.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1931.0, + 333.0, + 1931.0, + 333.0, + 1946.0, + 319.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1931.0, + 354.0, + 1931.0, + 354.0, + 1946.0, + 338.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1931.0, + 376.0, + 1931.0, + 376.0, + 1946.0, + 360.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1930.0, + 397.0, + 1930.0, + 397.0, + 1946.0, + 381.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1933.0, + 417.0, + 1933.0, + 417.0, + 1944.0, + 407.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1932.0, + 548.0, + 1932.0, + 548.0, + 1946.0, + 533.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1932.0, + 569.0, + 1932.0, + 569.0, + 1945.0, + 551.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1932.0, + 591.0, + 1932.0, + 591.0, + 1946.0, + 573.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1931.0, + 612.0, + 1931.0, + 612.0, + 1945.0, + 594.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1931.0, + 633.0, + 1931.0, + 633.0, + 1945.0, + 617.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1932.0, + 762.0, + 1932.0, + 762.0, + 1946.0, + 745.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1932.0, + 784.0, + 1932.0, + 784.0, + 1946.0, + 764.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1930.0, + 847.0, + 1930.0, + 847.0, + 1947.0, + 786.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1951.0, + 333.0, + 1951.0, + 333.0, + 1967.0, + 317.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1951.0, + 354.0, + 1951.0, + 354.0, + 1965.0, + 338.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1951.0, + 376.0, + 1951.0, + 376.0, + 1965.0, + 360.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1950.0, + 397.0, + 1950.0, + 397.0, + 1965.0, + 381.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1952.0, + 417.0, + 1952.0, + 417.0, + 1964.0, + 406.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1951.0, + 549.0, + 1951.0, + 549.0, + 1965.0, + 532.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1951.0, + 569.0, + 1951.0, + 569.0, + 1965.0, + 551.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1951.0, + 589.0, + 1951.0, + 589.0, + 1965.0, + 573.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1950.0, + 610.0, + 1950.0, + 610.0, + 1964.0, + 594.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1952.0, + 631.0, + 1952.0, + 631.0, + 1963.0, + 619.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1951.0, + 762.0, + 1951.0, + 762.0, + 1967.0, + 745.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1951.0, + 782.0, + 1951.0, + 782.0, + 1965.0, + 767.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1951.0, + 805.0, + 1951.0, + 805.0, + 1965.0, + 788.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1950.0, + 825.0, + 1950.0, + 825.0, + 1964.0, + 807.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1951.0, + 846.0, + 1951.0, + 846.0, + 1965.0, + 831.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1971.0, + 333.0, + 1971.0, + 333.0, + 1987.0, + 317.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1969.0, + 354.0, + 1969.0, + 354.0, + 1987.0, + 339.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1970.0, + 376.0, + 1970.0, + 376.0, + 1987.0, + 362.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1969.0, + 397.0, + 1969.0, + 397.0, + 1986.0, + 383.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1970.0, + 420.0, + 1970.0, + 420.0, + 1985.0, + 405.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1971.0, + 548.0, + 1971.0, + 548.0, + 1987.0, + 532.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1971.0, + 569.0, + 1971.0, + 569.0, + 1987.0, + 553.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1970.0, + 589.0, + 1970.0, + 589.0, + 1987.0, + 575.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1969.0, + 610.0, + 1969.0, + 610.0, + 1986.0, + 596.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1973.0, + 631.0, + 1973.0, + 631.0, + 1983.0, + 620.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1971.0, + 761.0, + 1971.0, + 761.0, + 1987.0, + 746.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1971.0, + 780.0, + 1971.0, + 780.0, + 1987.0, + 767.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1971.0, + 804.0, + 1971.0, + 804.0, + 1987.0, + 788.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1969.0, + 825.0, + 1969.0, + 825.0, + 1986.0, + 809.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1971.0, + 844.0, + 1971.0, + 844.0, + 1983.0, + 833.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1991.0, + 333.0, + 1991.0, + 333.0, + 2006.0, + 319.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1988.0, + 354.0, + 1988.0, + 354.0, + 2006.0, + 339.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1989.0, + 376.0, + 1989.0, + 376.0, + 2007.0, + 362.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1991.0, + 397.0, + 1991.0, + 397.0, + 2006.0, + 383.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1993.0, + 417.0, + 1993.0, + 417.0, + 2003.0, + 408.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1991.0, + 548.0, + 1991.0, + 548.0, + 2006.0, + 532.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1991.0, + 569.0, + 1991.0, + 569.0, + 2006.0, + 553.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1991.0, + 589.0, + 1991.0, + 589.0, + 2006.0, + 575.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1991.0, + 610.0, + 1991.0, + 610.0, + 2006.0, + 596.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1993.0, + 631.0, + 1993.0, + 631.0, + 2003.0, + 623.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1991.0, + 761.0, + 1991.0, + 761.0, + 2006.0, + 746.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1991.0, + 780.0, + 1991.0, + 780.0, + 2006.0, + 767.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1991.0, + 804.0, + 1991.0, + 804.0, + 2006.0, + 788.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1992.0, + 825.0, + 1992.0, + 825.0, + 2006.0, + 809.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 2012.0, + 333.0, + 2012.0, + 333.0, + 2029.0, + 319.0, + 2029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 2012.0, + 353.0, + 2012.0, + 353.0, + 2029.0, + 339.0, + 2029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 2012.0, + 375.0, + 2012.0, + 375.0, + 2028.0, + 362.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 2011.0, + 397.0, + 2011.0, + 397.0, + 2027.0, + 383.0, + 2027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 2012.0, + 546.0, + 2012.0, + 546.0, + 2029.0, + 532.0, + 2029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 2013.0, + 567.0, + 2013.0, + 567.0, + 2028.0, + 553.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 2012.0, + 589.0, + 2012.0, + 589.0, + 2028.0, + 575.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 2013.0, + 609.0, + 2013.0, + 609.0, + 2024.0, + 597.0, + 2024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 2013.0, + 761.0, + 2013.0, + 761.0, + 2028.0, + 746.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 2015.0, + 779.0, + 2015.0, + 779.0, + 2027.0, + 768.0, + 2027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 2012.0, + 802.0, + 2012.0, + 802.0, + 2029.0, + 788.0, + 2029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 2011.0, + 825.0, + 2011.0, + 825.0, + 2027.0, + 809.0, + 2027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 206.0, + 709.0, + 871.0, + 709.0, + 871.0, + 745.0, + 206.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 213.0, + 463.0, + 706.0, + 463.0, + 706.0, + 508.0, + 213.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1212.0, + 333.0, + 1212.0, + 333.0, + 1227.0, + 318.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1212.0, + 354.0, + 1212.0, + 354.0, + 1228.0, + 339.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1212.0, + 376.0, + 1212.0, + 376.0, + 1227.0, + 361.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1213.0, + 395.0, + 1213.0, + 395.0, + 1224.0, + 383.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1213.0, + 418.0, + 1213.0, + 418.0, + 1226.0, + 406.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1212.0, + 505.0, + 1212.0, + 505.0, + 1226.0, + 489.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1212.0, + 527.0, + 1212.0, + 527.0, + 1226.0, + 511.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1212.0, + 548.0, + 1212.0, + 548.0, + 1226.0, + 530.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1212.0, + 568.0, + 1212.0, + 568.0, + 1226.0, + 551.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1212.0, + 589.0, + 1212.0, + 589.0, + 1226.0, + 573.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1212.0, + 678.0, + 1212.0, + 678.0, + 1227.0, + 660.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1212.0, + 697.0, + 1212.0, + 697.0, + 1227.0, + 679.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1212.0, + 719.0, + 1212.0, + 719.0, + 1227.0, + 702.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1212.0, + 740.0, + 1212.0, + 740.0, + 1227.0, + 723.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1212.0, + 760.0, + 1212.0, + 760.0, + 1228.0, + 745.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1212.0, + 846.0, + 1212.0, + 846.0, + 1228.0, + 831.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1212.0, + 868.0, + 1212.0, + 868.0, + 1227.0, + 853.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1212.0, + 889.0, + 1212.0, + 889.0, + 1227.0, + 873.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1213.0, + 908.0, + 1213.0, + 908.0, + 1224.0, + 897.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1213.0, + 929.0, + 1213.0, + 929.0, + 1226.0, + 918.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1212.0, + 1038.0, + 1212.0, + 1038.0, + 1227.0, + 1024.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1212.0, + 1061.0, + 1212.0, + 1061.0, + 1227.0, + 1043.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1212.0, + 1081.0, + 1212.0, + 1081.0, + 1226.0, + 1065.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1212.0, + 1103.0, + 1212.0, + 1103.0, + 1226.0, + 1086.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1213.0, + 1122.0, + 1213.0, + 1122.0, + 1224.0, + 1110.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1212.0, + 1231.0, + 1212.0, + 1231.0, + 1227.0, + 1216.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1212.0, + 1253.0, + 1212.0, + 1253.0, + 1227.0, + 1236.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1212.0, + 1274.0, + 1212.0, + 1274.0, + 1227.0, + 1257.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1212.0, + 1295.0, + 1212.0, + 1295.0, + 1227.0, + 1278.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1212.0, + 1315.0, + 1212.0, + 1315.0, + 1227.0, + 1300.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1231.0, + 333.0, + 1231.0, + 333.0, + 1249.0, + 318.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1233.0, + 352.0, + 1233.0, + 352.0, + 1246.0, + 341.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1233.0, + 373.0, + 1233.0, + 373.0, + 1245.0, + 364.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1231.0, + 397.0, + 1231.0, + 397.0, + 1246.0, + 382.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1233.0, + 416.0, + 1233.0, + 416.0, + 1245.0, + 406.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1232.0, + 505.0, + 1232.0, + 505.0, + 1247.0, + 490.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1233.0, + 523.0, + 1233.0, + 523.0, + 1246.0, + 512.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1233.0, + 545.0, + 1233.0, + 545.0, + 1245.0, + 534.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1232.0, + 566.0, + 1232.0, + 566.0, + 1245.0, + 555.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1233.0, + 587.0, + 1233.0, + 587.0, + 1245.0, + 577.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1233.0, + 674.0, + 1233.0, + 674.0, + 1245.0, + 662.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1233.0, + 695.0, + 1233.0, + 695.0, + 1245.0, + 683.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1233.0, + 717.0, + 1233.0, + 717.0, + 1245.0, + 705.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1231.0, + 739.0, + 1231.0, + 739.0, + 1246.0, + 724.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1233.0, + 759.0, + 1233.0, + 759.0, + 1245.0, + 748.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1231.0, + 846.0, + 1231.0, + 846.0, + 1249.0, + 831.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1231.0, + 868.0, + 1231.0, + 868.0, + 1247.0, + 852.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1233.0, + 887.0, + 1233.0, + 887.0, + 1245.0, + 875.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1229.0, + 910.0, + 1229.0, + 910.0, + 1246.0, + 895.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1232.0, + 929.0, + 1232.0, + 929.0, + 1245.0, + 918.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1232.0, + 1038.0, + 1232.0, + 1038.0, + 1247.0, + 1022.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1231.0, + 1059.0, + 1231.0, + 1059.0, + 1246.0, + 1045.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1231.0, + 1081.0, + 1231.0, + 1081.0, + 1246.0, + 1067.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1229.0, + 1101.0, + 1229.0, + 1101.0, + 1246.0, + 1086.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1233.0, + 1122.0, + 1233.0, + 1122.0, + 1245.0, + 1111.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1232.0, + 1231.0, + 1232.0, + 1231.0, + 1247.0, + 1215.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1232.0, + 1252.0, + 1232.0, + 1252.0, + 1246.0, + 1236.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1233.0, + 1272.0, + 1233.0, + 1272.0, + 1245.0, + 1259.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1232.0, + 1292.0, + 1232.0, + 1292.0, + 1245.0, + 1281.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1233.0, + 1313.0, + 1233.0, + 1313.0, + 1245.0, + 1303.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1251.0, + 333.0, + 1251.0, + 333.0, + 1268.0, + 318.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1251.0, + 354.0, + 1251.0, + 354.0, + 1268.0, + 339.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1251.0, + 376.0, + 1251.0, + 376.0, + 1267.0, + 361.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1250.0, + 398.0, + 1250.0, + 398.0, + 1267.0, + 382.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1251.0, + 419.0, + 1251.0, + 419.0, + 1267.0, + 404.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1254.0, + 503.0, + 1254.0, + 503.0, + 1265.0, + 491.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1252.0, + 523.0, + 1252.0, + 523.0, + 1265.0, + 512.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1251.0, + 546.0, + 1251.0, + 546.0, + 1267.0, + 532.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1250.0, + 568.0, + 1250.0, + 568.0, + 1267.0, + 554.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1252.0, + 588.0, + 1252.0, + 588.0, + 1265.0, + 577.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 1254.0, + 674.0, + 1254.0, + 674.0, + 1265.0, + 663.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1251.0, + 696.0, + 1251.0, + 696.0, + 1267.0, + 681.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1251.0, + 718.0, + 1251.0, + 718.0, + 1267.0, + 703.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1250.0, + 739.0, + 1250.0, + 739.0, + 1267.0, + 723.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1251.0, + 761.0, + 1251.0, + 761.0, + 1267.0, + 746.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1251.0, + 846.0, + 1251.0, + 846.0, + 1268.0, + 831.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1251.0, + 867.0, + 1251.0, + 867.0, + 1268.0, + 852.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1249.0, + 891.0, + 1249.0, + 891.0, + 1269.0, + 872.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1250.0, + 911.0, + 1250.0, + 911.0, + 1267.0, + 895.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1251.0, + 932.0, + 1251.0, + 932.0, + 1267.0, + 916.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1251.0, + 1038.0, + 1251.0, + 1038.0, + 1267.0, + 1022.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1251.0, + 1059.0, + 1251.0, + 1059.0, + 1268.0, + 1045.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1250.0, + 1081.0, + 1250.0, + 1081.0, + 1268.0, + 1067.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1250.0, + 1102.0, + 1250.0, + 1102.0, + 1267.0, + 1086.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1251.0, + 1123.0, + 1251.0, + 1123.0, + 1268.0, + 1108.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1251.0, + 1231.0, + 1251.0, + 1231.0, + 1267.0, + 1215.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1251.0, + 1251.0, + 1251.0, + 1251.0, + 1267.0, + 1236.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1251.0, + 1273.0, + 1251.0, + 1273.0, + 1267.0, + 1258.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1250.0, + 1294.0, + 1250.0, + 1294.0, + 1267.0, + 1279.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1252.0, + 1313.0, + 1252.0, + 1313.0, + 1265.0, + 1302.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1270.0, + 333.0, + 1270.0, + 333.0, + 1287.0, + 318.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1270.0, + 354.0, + 1270.0, + 354.0, + 1287.0, + 339.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1270.0, + 376.0, + 1270.0, + 376.0, + 1287.0, + 361.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1272.0, + 397.0, + 1272.0, + 397.0, + 1287.0, + 382.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1273.0, + 418.0, + 1273.0, + 418.0, + 1282.0, + 408.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1270.0, + 505.0, + 1270.0, + 505.0, + 1287.0, + 490.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1270.0, + 524.0, + 1270.0, + 524.0, + 1287.0, + 511.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1270.0, + 546.0, + 1270.0, + 546.0, + 1287.0, + 532.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1272.0, + 568.0, + 1272.0, + 568.0, + 1287.0, + 552.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1272.0, + 589.0, + 1272.0, + 589.0, + 1283.0, + 579.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1270.0, + 675.0, + 1270.0, + 675.0, + 1287.0, + 660.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1270.0, + 696.0, + 1270.0, + 696.0, + 1287.0, + 681.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1270.0, + 718.0, + 1270.0, + 718.0, + 1287.0, + 703.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1272.0, + 739.0, + 1272.0, + 739.0, + 1287.0, + 723.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1273.0, + 760.0, + 1273.0, + 760.0, + 1283.0, + 751.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1270.0, + 846.0, + 1270.0, + 846.0, + 1287.0, + 831.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1270.0, + 867.0, + 1270.0, + 867.0, + 1287.0, + 852.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1270.0, + 889.0, + 1270.0, + 889.0, + 1287.0, + 874.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1272.0, + 910.0, + 1272.0, + 910.0, + 1287.0, + 895.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1270.0, + 932.0, + 1270.0, + 932.0, + 1287.0, + 917.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1270.0, + 1037.0, + 1270.0, + 1037.0, + 1287.0, + 1024.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1270.0, + 1059.0, + 1270.0, + 1059.0, + 1287.0, + 1045.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1270.0, + 1081.0, + 1270.0, + 1081.0, + 1287.0, + 1067.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1272.0, + 1103.0, + 1272.0, + 1103.0, + 1287.0, + 1086.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1272.0, + 1123.0, + 1272.0, + 1123.0, + 1285.0, + 1112.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1270.0, + 1231.0, + 1270.0, + 1231.0, + 1287.0, + 1216.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1272.0, + 1248.0, + 1272.0, + 1248.0, + 1286.0, + 1238.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1270.0, + 1273.0, + 1270.0, + 1273.0, + 1287.0, + 1258.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1272.0, + 1294.0, + 1272.0, + 1294.0, + 1287.0, + 1279.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1272.0, + 1316.0, + 1272.0, + 1316.0, + 1283.0, + 1306.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1292.0, + 333.0, + 1292.0, + 333.0, + 1308.0, + 318.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1293.0, + 352.0, + 1293.0, + 352.0, + 1306.0, + 341.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1293.0, + 373.0, + 1293.0, + 373.0, + 1306.0, + 364.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1292.0, + 395.0, + 1292.0, + 395.0, + 1305.0, + 384.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1292.0, + 503.0, + 1292.0, + 503.0, + 1309.0, + 490.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1292.0, + 524.0, + 1292.0, + 524.0, + 1309.0, + 511.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1292.0, + 546.0, + 1292.0, + 546.0, + 1309.0, + 532.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1291.0, + 567.0, + 1291.0, + 567.0, + 1306.0, + 552.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 1295.0, + 673.0, + 1295.0, + 673.0, + 1306.0, + 663.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1292.0, + 696.0, + 1292.0, + 696.0, + 1308.0, + 681.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1292.0, + 718.0, + 1292.0, + 718.0, + 1308.0, + 703.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1291.0, + 739.0, + 1291.0, + 739.0, + 1306.0, + 724.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1292.0, + 846.0, + 1292.0, + 846.0, + 1309.0, + 831.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1293.0, + 865.0, + 1293.0, + 865.0, + 1306.0, + 854.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1292.0, + 889.0, + 1292.0, + 889.0, + 1308.0, + 873.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1291.0, + 910.0, + 1291.0, + 910.0, + 1306.0, + 895.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1295.0, + 1036.0, + 1295.0, + 1036.0, + 1308.0, + 1025.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1293.0, + 1059.0, + 1293.0, + 1059.0, + 1309.0, + 1045.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1292.0, + 1081.0, + 1292.0, + 1081.0, + 1309.0, + 1067.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1291.0, + 1103.0, + 1291.0, + 1103.0, + 1306.0, + 1086.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1292.0, + 1231.0, + 1292.0, + 1231.0, + 1309.0, + 1216.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1295.0, + 1248.0, + 1295.0, + 1248.0, + 1306.0, + 1238.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1292.0, + 1273.0, + 1292.0, + 1273.0, + 1309.0, + 1258.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1292.0, + 1292.0, + 1292.0, + 1292.0, + 1305.0, + 1281.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1372.0, + 334.0, + 1372.0, + 334.0, + 1388.0, + 318.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1372.0, + 355.0, + 1372.0, + 355.0, + 1387.0, + 339.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1372.0, + 376.0, + 1372.0, + 376.0, + 1388.0, + 361.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1370.0, + 398.0, + 1370.0, + 398.0, + 1387.0, + 382.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1372.0, + 419.0, + 1372.0, + 419.0, + 1388.0, + 404.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1372.0, + 505.0, + 1372.0, + 505.0, + 1388.0, + 490.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1372.0, + 526.0, + 1372.0, + 526.0, + 1387.0, + 511.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1372.0, + 546.0, + 1372.0, + 546.0, + 1387.0, + 532.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1372.0, + 568.0, + 1372.0, + 568.0, + 1386.0, + 552.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1373.0, + 587.0, + 1373.0, + 587.0, + 1383.0, + 577.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1372.0, + 676.0, + 1372.0, + 676.0, + 1387.0, + 660.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1373.0, + 696.0, + 1373.0, + 696.0, + 1387.0, + 680.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1372.0, + 718.0, + 1372.0, + 718.0, + 1387.0, + 703.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1372.0, + 740.0, + 1372.0, + 740.0, + 1387.0, + 723.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1372.0, + 760.0, + 1372.0, + 760.0, + 1387.0, + 745.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1372.0, + 847.0, + 1372.0, + 847.0, + 1388.0, + 831.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1372.0, + 868.0, + 1372.0, + 868.0, + 1387.0, + 851.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1372.0, + 889.0, + 1372.0, + 889.0, + 1387.0, + 873.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1370.0, + 910.0, + 1370.0, + 910.0, + 1386.0, + 895.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1372.0, + 932.0, + 1372.0, + 932.0, + 1388.0, + 917.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1372.0, + 1038.0, + 1372.0, + 1038.0, + 1388.0, + 1024.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1372.0, + 1061.0, + 1372.0, + 1061.0, + 1387.0, + 1045.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1372.0, + 1081.0, + 1372.0, + 1081.0, + 1387.0, + 1065.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1370.0, + 1102.0, + 1370.0, + 1102.0, + 1386.0, + 1088.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1373.0, + 1122.0, + 1373.0, + 1122.0, + 1384.0, + 1111.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1372.0, + 1231.0, + 1372.0, + 1231.0, + 1388.0, + 1215.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1372.0, + 1252.0, + 1372.0, + 1252.0, + 1387.0, + 1236.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1372.0, + 1274.0, + 1372.0, + 1274.0, + 1387.0, + 1258.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1370.0, + 1295.0, + 1370.0, + 1295.0, + 1387.0, + 1278.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1372.0, + 1316.0, + 1372.0, + 1316.0, + 1387.0, + 1300.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1392.0, + 333.0, + 1392.0, + 333.0, + 1409.0, + 318.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1392.0, + 355.0, + 1392.0, + 355.0, + 1409.0, + 339.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1393.0, + 373.0, + 1393.0, + 373.0, + 1405.0, + 364.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1393.0, + 395.0, + 1393.0, + 395.0, + 1405.0, + 383.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1392.0, + 419.0, + 1392.0, + 419.0, + 1408.0, + 404.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1393.0, + 505.0, + 1393.0, + 505.0, + 1409.0, + 490.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1395.0, + 524.0, + 1395.0, + 524.0, + 1406.0, + 512.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1395.0, + 545.0, + 1395.0, + 545.0, + 1405.0, + 534.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1395.0, + 566.0, + 1395.0, + 566.0, + 1404.0, + 555.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1395.0, + 587.0, + 1395.0, + 587.0, + 1404.0, + 577.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1393.0, + 676.0, + 1393.0, + 676.0, + 1409.0, + 659.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1395.0, + 695.0, + 1395.0, + 695.0, + 1406.0, + 683.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1395.0, + 717.0, + 1395.0, + 717.0, + 1405.0, + 705.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 1393.0, + 738.0, + 1393.0, + 738.0, + 1404.0, + 726.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1392.0, + 846.0, + 1392.0, + 846.0, + 1409.0, + 831.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1393.0, + 868.0, + 1393.0, + 868.0, + 1409.0, + 852.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1391.0, + 889.0, + 1391.0, + 889.0, + 1408.0, + 873.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1391.0, + 910.0, + 1391.0, + 910.0, + 1406.0, + 895.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1392.0, + 932.0, + 1392.0, + 932.0, + 1408.0, + 917.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1392.0, + 1038.0, + 1392.0, + 1038.0, + 1409.0, + 1022.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1392.0, + 1059.0, + 1392.0, + 1059.0, + 1409.0, + 1045.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1392.0, + 1081.0, + 1392.0, + 1081.0, + 1408.0, + 1065.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1393.0, + 1100.0, + 1393.0, + 1100.0, + 1405.0, + 1089.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1393.0, + 1122.0, + 1393.0, + 1122.0, + 1405.0, + 1110.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1392.0, + 1231.0, + 1392.0, + 1231.0, + 1409.0, + 1216.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1392.0, + 1252.0, + 1392.0, + 1252.0, + 1408.0, + 1236.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1392.0, + 1273.0, + 1392.0, + 1273.0, + 1408.0, + 1257.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1392.0, + 1295.0, + 1392.0, + 1295.0, + 1406.0, + 1279.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1393.0, + 1313.0, + 1393.0, + 1313.0, + 1405.0, + 1302.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1413.0, + 333.0, + 1413.0, + 333.0, + 1429.0, + 318.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1413.0, + 354.0, + 1413.0, + 354.0, + 1429.0, + 339.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1413.0, + 376.0, + 1413.0, + 376.0, + 1429.0, + 361.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1411.0, + 398.0, + 1411.0, + 398.0, + 1428.0, + 382.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1413.0, + 419.0, + 1413.0, + 419.0, + 1428.0, + 404.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1413.0, + 505.0, + 1413.0, + 505.0, + 1429.0, + 490.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1414.0, + 523.0, + 1414.0, + 523.0, + 1427.0, + 512.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1414.0, + 545.0, + 1414.0, + 545.0, + 1425.0, + 534.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1415.0, + 566.0, + 1415.0, + 566.0, + 1424.0, + 556.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1414.0, + 587.0, + 1414.0, + 587.0, + 1427.0, + 577.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1413.0, + 676.0, + 1413.0, + 676.0, + 1428.0, + 659.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1413.0, + 696.0, + 1413.0, + 696.0, + 1428.0, + 681.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1413.0, + 718.0, + 1413.0, + 718.0, + 1428.0, + 703.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1413.0, + 739.0, + 1413.0, + 739.0, + 1428.0, + 724.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1414.0, + 759.0, + 1414.0, + 759.0, + 1427.0, + 748.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1413.0, + 846.0, + 1413.0, + 846.0, + 1429.0, + 831.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1413.0, + 867.0, + 1413.0, + 867.0, + 1429.0, + 852.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1411.0, + 889.0, + 1411.0, + 889.0, + 1429.0, + 873.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1411.0, + 911.0, + 1411.0, + 911.0, + 1428.0, + 895.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1413.0, + 932.0, + 1413.0, + 932.0, + 1428.0, + 916.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1413.0, + 1038.0, + 1413.0, + 1038.0, + 1429.0, + 1022.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1413.0, + 1059.0, + 1413.0, + 1059.0, + 1429.0, + 1045.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1411.0, + 1081.0, + 1411.0, + 1081.0, + 1429.0, + 1065.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1411.0, + 1102.0, + 1411.0, + 1102.0, + 1428.0, + 1088.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1413.0, + 1123.0, + 1413.0, + 1123.0, + 1428.0, + 1108.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1413.0, + 1231.0, + 1413.0, + 1231.0, + 1429.0, + 1215.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1414.0, + 1249.0, + 1414.0, + 1249.0, + 1427.0, + 1238.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1413.0, + 1273.0, + 1413.0, + 1273.0, + 1429.0, + 1258.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1413.0, + 1292.0, + 1413.0, + 1292.0, + 1425.0, + 1281.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1413.0, + 1316.0, + 1413.0, + 1316.0, + 1428.0, + 1301.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1432.0, + 333.0, + 1432.0, + 333.0, + 1449.0, + 318.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1431.0, + 354.0, + 1431.0, + 354.0, + 1449.0, + 339.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1431.0, + 376.0, + 1431.0, + 376.0, + 1449.0, + 361.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1433.0, + 398.0, + 1433.0, + 398.0, + 1449.0, + 381.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1433.0, + 418.0, + 1433.0, + 418.0, + 1445.0, + 408.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1432.0, + 505.0, + 1432.0, + 505.0, + 1449.0, + 490.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1434.0, + 523.0, + 1434.0, + 523.0, + 1446.0, + 512.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 1434.0, + 544.0, + 1434.0, + 544.0, + 1446.0, + 534.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1436.0, + 566.0, + 1436.0, + 566.0, + 1446.0, + 555.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1432.0, + 675.0, + 1432.0, + 675.0, + 1449.0, + 660.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1432.0, + 696.0, + 1432.0, + 696.0, + 1447.0, + 681.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1433.0, + 716.0, + 1433.0, + 716.0, + 1445.0, + 706.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 1434.0, + 738.0, + 1434.0, + 738.0, + 1446.0, + 726.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1432.0, + 846.0, + 1432.0, + 846.0, + 1450.0, + 831.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1431.0, + 867.0, + 1431.0, + 867.0, + 1450.0, + 852.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1431.0, + 889.0, + 1431.0, + 889.0, + 1450.0, + 874.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1433.0, + 911.0, + 1433.0, + 911.0, + 1449.0, + 895.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1433.0, + 930.0, + 1433.0, + 930.0, + 1445.0, + 919.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1432.0, + 1038.0, + 1432.0, + 1038.0, + 1449.0, + 1024.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1431.0, + 1059.0, + 1431.0, + 1059.0, + 1449.0, + 1045.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1431.0, + 1080.0, + 1431.0, + 1080.0, + 1449.0, + 1067.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1433.0, + 1102.0, + 1433.0, + 1102.0, + 1449.0, + 1088.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1433.0, + 1122.0, + 1433.0, + 1122.0, + 1445.0, + 1113.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1432.0, + 1231.0, + 1432.0, + 1231.0, + 1449.0, + 1216.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1432.0, + 1251.0, + 1432.0, + 1251.0, + 1449.0, + 1237.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1432.0, + 1273.0, + 1432.0, + 1273.0, + 1449.0, + 1258.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1434.0, + 1292.0, + 1434.0, + 1292.0, + 1446.0, + 1281.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1433.0, + 1315.0, + 1433.0, + 1315.0, + 1443.0, + 1306.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1454.0, + 333.0, + 1454.0, + 333.0, + 1470.0, + 318.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1454.0, + 354.0, + 1454.0, + 354.0, + 1470.0, + 339.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1452.0, + 376.0, + 1452.0, + 376.0, + 1469.0, + 361.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1454.0, + 395.0, + 1454.0, + 395.0, + 1465.0, + 383.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1454.0, + 505.0, + 1454.0, + 505.0, + 1470.0, + 490.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1454.0, + 524.0, + 1454.0, + 524.0, + 1470.0, + 511.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1454.0, + 546.0, + 1454.0, + 546.0, + 1469.0, + 532.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1452.0, + 568.0, + 1452.0, + 568.0, + 1466.0, + 552.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1454.0, + 675.0, + 1454.0, + 675.0, + 1470.0, + 660.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1454.0, + 696.0, + 1454.0, + 696.0, + 1469.0, + 681.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1455.0, + 716.0, + 1455.0, + 716.0, + 1468.0, + 706.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1452.0, + 740.0, + 1452.0, + 740.0, + 1466.0, + 724.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1454.0, + 846.0, + 1454.0, + 846.0, + 1470.0, + 831.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1452.0, + 867.0, + 1452.0, + 867.0, + 1470.0, + 852.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1452.0, + 889.0, + 1452.0, + 889.0, + 1470.0, + 874.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1452.0, + 910.0, + 1452.0, + 910.0, + 1466.0, + 894.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1454.0, + 1038.0, + 1454.0, + 1038.0, + 1470.0, + 1022.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1452.0, + 1059.0, + 1452.0, + 1059.0, + 1470.0, + 1045.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1452.0, + 1081.0, + 1452.0, + 1081.0, + 1472.0, + 1067.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1451.0, + 1103.0, + 1451.0, + 1103.0, + 1466.0, + 1088.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1454.0, + 1231.0, + 1454.0, + 1231.0, + 1470.0, + 1216.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1454.0, + 1251.0, + 1454.0, + 1251.0, + 1470.0, + 1236.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1454.0, + 1273.0, + 1454.0, + 1273.0, + 1470.0, + 1258.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1454.0, + 1294.0, + 1454.0, + 1294.0, + 1465.0, + 1281.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 222.0, + 316.0, + 222.0, + 316.0, + 258.0, + 216.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 226.0, + 480.0, + 226.0, + 480.0, + 256.0, + 389.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 225.0, + 707.0, + 225.0, + 707.0, + 259.0, + 550.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 250.0, + 305.0, + 250.0, + 305.0, + 284.0, + 231.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 252.0, + 474.0, + 252.0, + 474.0, + 286.0, + 403.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 253.0, + 706.0, + 253.0, + 706.0, + 282.0, + 552.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 298.0, + 457.0, + 298.0, + 457.0, + 318.0, + 357.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 321.0, + 372.0, + 321.0, + 372.0, + 336.0, + 357.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 323.0, + 391.0, + 323.0, + 391.0, + 334.0, + 377.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 320.0, + 437.0, + 320.0, + 437.0, + 334.0, + 400.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 323.0, + 454.0, + 323.0, + 454.0, + 332.0, + 445.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 221.0, + 338.0, + 260.0, + 338.0, + 260.0, + 356.0, + 221.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 341.0, + 276.0, + 341.0, + 276.0, + 354.0, + 263.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 339.0, + 392.0, + 339.0, + 392.0, + 357.0, + 355.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 339.0, + 436.0, + 339.0, + 436.0, + 356.0, + 400.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 340.0, + 457.0, + 340.0, + 457.0, + 355.0, + 439.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 334.0, + 681.0, + 334.0, + 681.0, + 357.0, + 570.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 356.0, + 259.0, + 356.0, + 259.0, + 377.0, + 220.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 358.0, + 374.0, + 358.0, + 374.0, + 377.0, + 356.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 358.0, + 392.0, + 358.0, + 392.0, + 376.0, + 375.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 358.0, + 435.0, + 358.0, + 435.0, + 376.0, + 400.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 354.0, + 681.0, + 354.0, + 681.0, + 377.0, + 570.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 380.0, + 236.0, + 380.0, + 236.0, + 397.0, + 220.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 379.0, + 257.0, + 379.0, + 257.0, + 395.0, + 239.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 380.0, + 374.0, + 380.0, + 374.0, + 399.0, + 356.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 380.0, + 392.0, + 380.0, + 392.0, + 397.0, + 376.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 379.0, + 436.0, + 379.0, + 436.0, + 397.0, + 400.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 375.0, + 681.0, + 375.0, + 681.0, + 398.0, + 570.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 395.0, + 681.0, + 395.0, + 681.0, + 418.0, + 570.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 416.0, + 681.0, + 416.0, + 681.0, + 438.0, + 568.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 831.0, + 330.0, + 831.0, + 330.0, + 843.0, + 321.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 830.0, + 354.0, + 830.0, + 354.0, + 845.0, + 339.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 830.0, + 376.0, + 830.0, + 376.0, + 845.0, + 362.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 829.0, + 397.0, + 829.0, + 397.0, + 845.0, + 383.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 830.0, + 419.0, + 830.0, + 419.0, + 845.0, + 405.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 829.0, + 504.0, + 829.0, + 504.0, + 845.0, + 489.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 830.0, + 525.0, + 830.0, + 525.0, + 845.0, + 510.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 830.0, + 546.0, + 830.0, + 546.0, + 845.0, + 531.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 829.0, + 567.0, + 829.0, + 567.0, + 843.0, + 553.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 830.0, + 588.0, + 830.0, + 588.0, + 845.0, + 574.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 830.0, + 676.0, + 830.0, + 676.0, + 845.0, + 660.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 830.0, + 698.0, + 830.0, + 698.0, + 845.0, + 680.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 830.0, + 719.0, + 830.0, + 719.0, + 845.0, + 702.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 828.0, + 741.0, + 828.0, + 741.0, + 846.0, + 722.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 830.0, + 760.0, + 830.0, + 760.0, + 845.0, + 744.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 829.0, + 845.0, + 829.0, + 845.0, + 846.0, + 831.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 830.0, + 868.0, + 830.0, + 868.0, + 845.0, + 852.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 830.0, + 887.0, + 830.0, + 887.0, + 845.0, + 873.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 829.0, + 910.0, + 829.0, + 910.0, + 845.0, + 894.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 830.0, + 931.0, + 830.0, + 931.0, + 845.0, + 917.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 830.0, + 1017.0, + 830.0, + 1017.0, + 846.0, + 1002.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 830.0, + 1039.0, + 830.0, + 1039.0, + 845.0, + 1023.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 830.0, + 1060.0, + 830.0, + 1060.0, + 845.0, + 1044.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 829.0, + 1080.0, + 829.0, + 1080.0, + 845.0, + 1065.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 829.0, + 1102.0, + 829.0, + 1102.0, + 845.0, + 1087.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 830.0, + 1189.0, + 830.0, + 1189.0, + 845.0, + 1173.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 830.0, + 1211.0, + 830.0, + 1211.0, + 845.0, + 1191.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 830.0, + 1232.0, + 830.0, + 1232.0, + 845.0, + 1213.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 828.0, + 1254.0, + 828.0, + 1254.0, + 846.0, + 1234.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 830.0, + 1273.0, + 830.0, + 1273.0, + 845.0, + 1255.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 852.0, + 332.0, + 852.0, + 332.0, + 864.0, + 321.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 850.0, + 354.0, + 850.0, + 354.0, + 865.0, + 340.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 850.0, + 376.0, + 850.0, + 376.0, + 865.0, + 362.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 849.0, + 397.0, + 849.0, + 397.0, + 864.0, + 383.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 849.0, + 419.0, + 849.0, + 419.0, + 864.0, + 405.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 849.0, + 504.0, + 849.0, + 504.0, + 865.0, + 489.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 850.0, + 525.0, + 850.0, + 525.0, + 865.0, + 511.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 849.0, + 546.0, + 849.0, + 546.0, + 865.0, + 532.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 848.0, + 567.0, + 848.0, + 567.0, + 864.0, + 553.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 849.0, + 590.0, + 849.0, + 590.0, + 864.0, + 574.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 849.0, + 675.0, + 849.0, + 675.0, + 865.0, + 660.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 849.0, + 696.0, + 849.0, + 696.0, + 865.0, + 681.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 849.0, + 718.0, + 849.0, + 718.0, + 865.0, + 703.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 848.0, + 739.0, + 848.0, + 739.0, + 864.0, + 724.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 849.0, + 760.0, + 849.0, + 760.0, + 865.0, + 745.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 850.0, + 845.0, + 850.0, + 845.0, + 865.0, + 831.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 849.0, + 868.0, + 849.0, + 868.0, + 865.0, + 852.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 849.0, + 889.0, + 849.0, + 889.0, + 865.0, + 875.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 848.0, + 910.0, + 848.0, + 910.0, + 864.0, + 896.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 849.0, + 929.0, + 849.0, + 929.0, + 865.0, + 917.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 849.0, + 1017.0, + 849.0, + 1017.0, + 865.0, + 1002.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 850.0, + 1037.0, + 850.0, + 1037.0, + 865.0, + 1023.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 849.0, + 1059.0, + 849.0, + 1059.0, + 865.0, + 1045.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 848.0, + 1080.0, + 848.0, + 1080.0, + 864.0, + 1066.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 849.0, + 1101.0, + 849.0, + 1101.0, + 865.0, + 1087.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 850.0, + 1189.0, + 850.0, + 1189.0, + 865.0, + 1172.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 849.0, + 1210.0, + 849.0, + 1210.0, + 865.0, + 1193.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 849.0, + 1231.0, + 849.0, + 1231.0, + 865.0, + 1214.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 848.0, + 1252.0, + 848.0, + 1252.0, + 864.0, + 1237.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 850.0, + 1273.0, + 850.0, + 1273.0, + 865.0, + 1259.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 872.0, + 332.0, + 872.0, + 332.0, + 883.0, + 321.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 871.0, + 351.0, + 871.0, + 351.0, + 883.0, + 341.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 870.0, + 376.0, + 870.0, + 376.0, + 884.0, + 362.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 868.0, + 397.0, + 868.0, + 397.0, + 884.0, + 383.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 870.0, + 419.0, + 870.0, + 419.0, + 884.0, + 405.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 870.0, + 504.0, + 870.0, + 504.0, + 884.0, + 489.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 870.0, + 525.0, + 870.0, + 525.0, + 884.0, + 511.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 868.0, + 546.0, + 868.0, + 546.0, + 884.0, + 531.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 868.0, + 567.0, + 868.0, + 567.0, + 884.0, + 553.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 870.0, + 590.0, + 870.0, + 590.0, + 885.0, + 576.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 870.0, + 675.0, + 870.0, + 675.0, + 884.0, + 661.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 870.0, + 696.0, + 870.0, + 696.0, + 884.0, + 682.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 870.0, + 718.0, + 870.0, + 718.0, + 884.0, + 703.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 868.0, + 739.0, + 868.0, + 739.0, + 885.0, + 724.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 868.0, + 760.0, + 868.0, + 760.0, + 885.0, + 746.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 870.0, + 845.0, + 870.0, + 845.0, + 886.0, + 831.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 870.0, + 866.0, + 870.0, + 866.0, + 884.0, + 852.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 870.0, + 889.0, + 870.0, + 889.0, + 884.0, + 873.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 868.0, + 910.0, + 868.0, + 910.0, + 885.0, + 894.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 870.0, + 931.0, + 870.0, + 931.0, + 885.0, + 915.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 870.0, + 1017.0, + 870.0, + 1017.0, + 885.0, + 1002.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 872.0, + 1036.0, + 872.0, + 1036.0, + 883.0, + 1025.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 870.0, + 1059.0, + 870.0, + 1059.0, + 884.0, + 1045.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 868.0, + 1081.0, + 868.0, + 1081.0, + 885.0, + 1066.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 870.0, + 1102.0, + 870.0, + 1102.0, + 886.0, + 1087.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 870.0, + 1187.0, + 870.0, + 1187.0, + 885.0, + 1172.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 871.0, + 1206.0, + 871.0, + 1206.0, + 883.0, + 1196.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 870.0, + 1231.0, + 870.0, + 1231.0, + 884.0, + 1217.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 868.0, + 1252.0, + 868.0, + 1252.0, + 884.0, + 1237.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 870.0, + 1273.0, + 870.0, + 1273.0, + 884.0, + 1259.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 889.0, + 333.0, + 889.0, + 333.0, + 906.0, + 319.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 889.0, + 353.0, + 889.0, + 353.0, + 904.0, + 340.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 889.0, + 376.0, + 889.0, + 376.0, + 904.0, + 362.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 890.0, + 397.0, + 890.0, + 397.0, + 906.0, + 383.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 889.0, + 420.0, + 889.0, + 420.0, + 903.0, + 406.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 889.0, + 504.0, + 889.0, + 504.0, + 906.0, + 489.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 886.0, + 528.0, + 886.0, + 528.0, + 907.0, + 509.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 886.0, + 548.0, + 886.0, + 548.0, + 907.0, + 530.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 889.0, + 567.0, + 889.0, + 567.0, + 906.0, + 553.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 889.0, + 591.0, + 889.0, + 591.0, + 903.0, + 577.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 889.0, + 675.0, + 889.0, + 675.0, + 906.0, + 661.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 889.0, + 696.0, + 889.0, + 696.0, + 906.0, + 682.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 888.0, + 718.0, + 888.0, + 718.0, + 904.0, + 703.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 890.0, + 739.0, + 890.0, + 739.0, + 906.0, + 724.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 890.0, + 760.0, + 890.0, + 760.0, + 902.0, + 751.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 889.0, + 845.0, + 889.0, + 845.0, + 907.0, + 831.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 889.0, + 866.0, + 889.0, + 866.0, + 904.0, + 852.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 889.0, + 889.0, + 889.0, + 889.0, + 904.0, + 873.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 890.0, + 910.0, + 890.0, + 910.0, + 906.0, + 894.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 888.0, + 932.0, + 888.0, + 932.0, + 903.0, + 918.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 889.0, + 1017.0, + 889.0, + 1017.0, + 904.0, + 1002.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 889.0, + 1037.0, + 889.0, + 1037.0, + 904.0, + 1024.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 886.0, + 1061.0, + 886.0, + 1061.0, + 907.0, + 1044.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 889.0, + 1081.0, + 889.0, + 1081.0, + 906.0, + 1066.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 888.0, + 1102.0, + 888.0, + 1102.0, + 904.0, + 1089.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 889.0, + 1187.0, + 889.0, + 1187.0, + 906.0, + 1173.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 888.0, + 1207.0, + 888.0, + 1207.0, + 906.0, + 1195.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 889.0, + 1231.0, + 889.0, + 1231.0, + 904.0, + 1217.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 890.0, + 1252.0, + 890.0, + 1252.0, + 906.0, + 1237.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 888.0, + 1273.0, + 888.0, + 1273.0, + 904.0, + 1260.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 911.0, + 330.0, + 911.0, + 330.0, + 923.0, + 320.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 910.0, + 353.0, + 910.0, + 353.0, + 926.0, + 340.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 911.0, + 374.0, + 911.0, + 374.0, + 923.0, + 363.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 910.0, + 395.0, + 910.0, + 395.0, + 922.0, + 384.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 910.0, + 503.0, + 910.0, + 503.0, + 927.0, + 489.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 910.0, + 525.0, + 910.0, + 525.0, + 927.0, + 510.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 910.0, + 546.0, + 910.0, + 546.0, + 927.0, + 531.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 908.0, + 567.0, + 908.0, + 567.0, + 925.0, + 552.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 910.0, + 585.0, + 910.0, + 585.0, + 922.0, + 577.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 910.0, + 675.0, + 910.0, + 675.0, + 927.0, + 660.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 910.0, + 695.0, + 910.0, + 695.0, + 927.0, + 681.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 910.0, + 718.0, + 910.0, + 718.0, + 927.0, + 703.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 908.0, + 739.0, + 908.0, + 739.0, + 925.0, + 724.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 911.0, + 757.0, + 911.0, + 757.0, + 921.0, + 748.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 910.0, + 845.0, + 910.0, + 845.0, + 926.0, + 831.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 910.0, + 866.0, + 910.0, + 866.0, + 926.0, + 852.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 910.0, + 887.0, + 910.0, + 887.0, + 926.0, + 873.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 911.0, + 907.0, + 911.0, + 907.0, + 922.0, + 898.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 910.0, + 1016.0, + 910.0, + 1016.0, + 927.0, + 1003.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 910.0, + 1037.0, + 910.0, + 1037.0, + 927.0, + 1023.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 910.0, + 1059.0, + 910.0, + 1059.0, + 927.0, + 1045.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 907.0, + 1082.0, + 907.0, + 1082.0, + 926.0, + 1064.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 909.0, + 1099.0, + 909.0, + 1099.0, + 923.0, + 1089.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 910.0, + 1186.0, + 910.0, + 1186.0, + 927.0, + 1173.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 910.0, + 1207.0, + 910.0, + 1207.0, + 927.0, + 1195.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 910.0, + 1231.0, + 910.0, + 1231.0, + 927.0, + 1217.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 908.0, + 1252.0, + 908.0, + 1252.0, + 925.0, + 1237.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 911.0, + 1269.0, + 911.0, + 1269.0, + 921.0, + 1260.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 990.0, + 334.0, + 990.0, + 334.0, + 1006.0, + 318.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 990.0, + 355.0, + 990.0, + 355.0, + 1006.0, + 339.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 989.0, + 376.0, + 989.0, + 376.0, + 1004.0, + 362.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 987.0, + 399.0, + 987.0, + 399.0, + 1006.0, + 381.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 989.0, + 418.0, + 989.0, + 418.0, + 1006.0, + 404.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 990.0, + 506.0, + 990.0, + 506.0, + 1006.0, + 489.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 990.0, + 525.0, + 990.0, + 525.0, + 1005.0, + 509.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 990.0, + 548.0, + 990.0, + 548.0, + 1004.0, + 531.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 989.0, + 567.0, + 989.0, + 567.0, + 1004.0, + 552.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 992.0, + 587.0, + 992.0, + 587.0, + 1002.0, + 576.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 990.0, + 676.0, + 990.0, + 676.0, + 1006.0, + 660.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 990.0, + 697.0, + 990.0, + 697.0, + 1005.0, + 681.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 990.0, + 719.0, + 990.0, + 719.0, + 1005.0, + 702.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 989.0, + 740.0, + 989.0, + 740.0, + 1005.0, + 724.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 990.0, + 760.0, + 990.0, + 760.0, + 1006.0, + 744.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1008.0, + 335.0, + 1008.0, + 335.0, + 1027.0, + 316.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1010.0, + 355.0, + 1010.0, + 355.0, + 1025.0, + 339.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1010.0, + 376.0, + 1010.0, + 376.0, + 1025.0, + 362.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1008.0, + 397.0, + 1008.0, + 397.0, + 1024.0, + 383.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1010.0, + 418.0, + 1010.0, + 418.0, + 1025.0, + 404.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1010.0, + 504.0, + 1010.0, + 504.0, + 1026.0, + 489.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1010.0, + 525.0, + 1010.0, + 525.0, + 1025.0, + 510.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1011.0, + 546.0, + 1011.0, + 546.0, + 1024.0, + 531.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1008.0, + 567.0, + 1008.0, + 567.0, + 1024.0, + 553.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1011.0, + 590.0, + 1011.0, + 590.0, + 1025.0, + 574.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1008.0, + 677.0, + 1008.0, + 677.0, + 1027.0, + 659.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1011.0, + 696.0, + 1011.0, + 696.0, + 1025.0, + 681.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1010.0, + 718.0, + 1010.0, + 718.0, + 1025.0, + 703.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1008.0, + 740.0, + 1008.0, + 740.0, + 1024.0, + 724.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1010.0, + 760.0, + 1010.0, + 760.0, + 1025.0, + 745.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1031.0, + 334.0, + 1031.0, + 334.0, + 1047.0, + 318.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1031.0, + 354.0, + 1031.0, + 354.0, + 1047.0, + 339.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1030.0, + 377.0, + 1030.0, + 377.0, + 1045.0, + 362.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1030.0, + 397.0, + 1030.0, + 397.0, + 1045.0, + 382.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1030.0, + 418.0, + 1030.0, + 418.0, + 1045.0, + 404.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1030.0, + 504.0, + 1030.0, + 504.0, + 1047.0, + 489.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1031.0, + 525.0, + 1031.0, + 525.0, + 1047.0, + 509.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1030.0, + 546.0, + 1030.0, + 546.0, + 1045.0, + 531.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1030.0, + 569.0, + 1030.0, + 569.0, + 1045.0, + 553.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1031.0, + 590.0, + 1031.0, + 590.0, + 1045.0, + 576.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1031.0, + 676.0, + 1031.0, + 676.0, + 1047.0, + 660.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1031.0, + 696.0, + 1031.0, + 696.0, + 1047.0, + 681.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1030.0, + 718.0, + 1030.0, + 718.0, + 1045.0, + 703.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1030.0, + 740.0, + 1030.0, + 740.0, + 1045.0, + 724.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1030.0, + 760.0, + 1030.0, + 760.0, + 1047.0, + 745.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1050.0, + 334.0, + 1050.0, + 334.0, + 1066.0, + 319.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1049.0, + 353.0, + 1049.0, + 353.0, + 1066.0, + 339.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1050.0, + 376.0, + 1050.0, + 376.0, + 1066.0, + 362.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1050.0, + 397.0, + 1050.0, + 397.0, + 1066.0, + 382.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1049.0, + 419.0, + 1049.0, + 419.0, + 1065.0, + 406.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1050.0, + 504.0, + 1050.0, + 504.0, + 1066.0, + 489.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1049.0, + 525.0, + 1049.0, + 525.0, + 1066.0, + 510.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1050.0, + 546.0, + 1050.0, + 546.0, + 1066.0, + 532.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1051.0, + 567.0, + 1051.0, + 567.0, + 1066.0, + 553.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1050.0, + 591.0, + 1050.0, + 591.0, + 1065.0, + 577.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1050.0, + 676.0, + 1050.0, + 676.0, + 1066.0, + 660.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1050.0, + 696.0, + 1050.0, + 696.0, + 1066.0, + 681.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1049.0, + 718.0, + 1049.0, + 718.0, + 1066.0, + 703.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1051.0, + 739.0, + 1051.0, + 739.0, + 1066.0, + 724.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1049.0, + 761.0, + 1049.0, + 761.0, + 1065.0, + 748.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1072.0, + 333.0, + 1072.0, + 333.0, + 1087.0, + 319.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1072.0, + 353.0, + 1072.0, + 353.0, + 1087.0, + 340.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1072.0, + 376.0, + 1072.0, + 376.0, + 1086.0, + 362.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1071.0, + 397.0, + 1071.0, + 397.0, + 1085.0, + 383.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1072.0, + 504.0, + 1072.0, + 504.0, + 1087.0, + 490.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1072.0, + 525.0, + 1072.0, + 525.0, + 1087.0, + 510.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1072.0, + 546.0, + 1072.0, + 546.0, + 1087.0, + 532.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1069.0, + 567.0, + 1069.0, + 567.0, + 1085.0, + 553.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1072.0, + 586.0, + 1072.0, + 586.0, + 1084.0, + 577.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1072.0, + 675.0, + 1072.0, + 675.0, + 1088.0, + 660.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1072.0, + 696.0, + 1072.0, + 696.0, + 1088.0, + 681.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1072.0, + 717.0, + 1072.0, + 717.0, + 1087.0, + 703.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1069.0, + 740.0, + 1069.0, + 740.0, + 1085.0, + 724.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1072.0, + 758.0, + 1072.0, + 758.0, + 1084.0, + 747.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 226.0, + 1539.0, + 488.0, + 1539.0, + 488.0, + 1578.0, + 226.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 226.0, + 1149.0, + 490.0, + 1149.0, + 490.0, + 1187.0, + 226.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 306.0, + 1515.0, + 306.0, + 1515.0, + 361.0, + 803.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 226.0, + 775.0, + 491.0, + 775.0, + 491.0, + 814.0, + 226.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 226.0, + 1149.0, + 490.0, + 1149.0, + 490.0, + 1187.0, + 226.0, + 1187.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 30, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 308, + 1028, + 1319, + 1028, + 1319, + 1736, + 308, + 1736 + ], + "score": 0.956 + }, + { + "category_id": 3, + "poly": [ + 306, + 339, + 1321, + 339, + 1321, + 874, + 306, + 874 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 198, + 1872, + 678, + 1872, + 678, + 1901, + 198, + 1901 + ], + "score": 0.815 + }, + { + "category_id": 2, + "poly": [ + 221, + 181, + 705, + 181, + 705, + 207, + 221, + 207 + ], + "score": 0.796 + }, + { + "category_id": 1, + "poly": [ + 199, + 1808, + 683, + 1808, + 683, + 1836, + 199, + 1836 + ], + "score": 0.772 + }, + { + "category_id": 1, + "poly": [ + 194, + 1937, + 721, + 1937, + 721, + 1967, + 194, + 1967 + ], + "score": 0.728 + }, + { + "category_id": 4, + "poly": [ + 217, + 969, + 472, + 969, + 472, + 996, + 217, + 996 + ], + "score": 0.351 + }, + { + "category_id": 4, + "poly": [ + 220, + 282, + 479, + 282, + 479, + 309, + 220, + 309 + ], + "score": 0.311 + }, + { + "category_id": 13, + "poly": [ + 316, + 539, + 431, + 539, + 431, + 654, + 316, + 654 + ], + "score": 0.44, + "latex": "{ \\begin{array} { l } { { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ { ~ \\mathrm { ~ o ~ } } ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } \\ { \\mathrm { ~ a ~ } } { \\mathrm { ~ } } { \\mathrm { ~ { ~ \\mathrm { ~ a ~ } } ~ } } { \\mathrm { ~ b ~ } } \\ { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ d ~ } } \\ { \\mathrm { ~ d ~ } } { \\mathrm { ~ e ~ } } \\ { \\mathrm { ~ f ~ } } } \\\\ { { \\mathrm { ~ g ~ } } \\ { \\mathrm { ~ 9 ~ } } { \\mathrm { ~ \\ e ~ h ~ } } \\ { \\mathrm { ~ i ~ } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 487, + 1222, + 600, + 1222, + 600, + 1342, + 487, + 1342 + ], + "score": 0.4, + "latex": "\\begin{array} { r l } & { \\mathrm { { ~ a ~ } } \\mathrm { { ~ a ~ } } \\mathrm { { ~ b ~ } } \\mathrm { { ~ c ~ } } } \\\\ & { \\mathrm { { ~ a ~ } } \\mathrm { { ~ a ~ } } \\mathrm { { ~ b ~ } } \\mathrm { { ~ c ~ } } } \\\\ & { \\mathrm { { ~ a ~ } } \\mathrm { { ~ a ~ } } \\left[ \\mathrm { { ~ a ~ } } \\mathrm { { ~ b ~ } } \\mathrm { { ~ c ~ } } \\right. } \\\\ & { \\mathrm { { ~ d ~ } } \\mathrm { { ~ d ~ } } \\mathrm { { ~ e ~ } } ^ { \\mathrm { { ~ \\scriptsize { ~ f ~ } ~ } } } } \\\\ & { \\mathrm { { ~ g ~ } } \\mathrm { { ~ g ~ } } \\left| \\mathrm { { ~ g ~ } } \\mathrm { { ~ h ~ } } \\mathrm { { ~ i ~ } } \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 487, + 536, + 601, + 536, + 601, + 653, + 487, + 653 + ], + "score": 0.36, + "latex": "{ \\begin{array} { l } { { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ b ~ } } { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ b ~ } } { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ a ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ } } { \\mathrm { ~ a ~ } } { \\mathrm { ~ b ~ } } { \\mathrm { ~ c ~ } } } \\\\ { { \\mathrm { ~ d ~ } } { \\mathrm { ~ d ~ } } { \\mathrm { ~ e ~ } } { \\mathrm { ~ f ~ } } } \\\\ { { \\mathrm { ~ g ~ } } { \\mathrm { ~ g ~ } } { \\mathrm { ~ } } { \\mathrm { ~ \\left| ~ g ~ h ~ \\right. ~ i ~ } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 659, + 534, + 772, + 534, + 772, + 652, + 659, + 652 + ], + "score": 0.32, + "latex": "{ \\begin{array} { r l } { \\mathbf { a } } & { \\mathbf { a } \\quad \\mathbf { a } \\quad \\mathbf { b } \\quad \\mathbf { c } } \\\\ { \\mathbf { a } } & { \\mathbf { a } \\quad \\mathbf { a } \\quad \\mathbf { b } \\quad \\mathbf { c } } \\\\ { \\mathbf { a } } & { \\mathbf { a } \\quad \\mathbf { \\left[ \\begin{array} { l l l } { \\mathbf { a } } & { \\mathbf { b } } & { \\mathbf { c } } \\\\ { \\mathbf { d } } & { \\mathbf { b } } & { \\mathbf { c } } \\end{array} \\right] } } \\\\ { \\mathbf { d } } & { \\mathbf { d } \\quad \\mathbf { d } \\quad \\mathbf { d } \\quad \\mathbf { e } \\quad \\mathbf { f } } \\\\ { \\mathbf { g } } & { \\mathbf { 9 } \\quad \\mathbf { \\left[ \\begin{array} { l l l } { \\mathbf { g } } & { \\mathbf { h } } & { \\mathbf { i } } \\end{array} \\right] } } \\end{array} }" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1043.0, + 333.0, + 1043.0, + 333.0, + 1058.0, + 319.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1043.0, + 354.0, + 1043.0, + 354.0, + 1058.0, + 340.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1043.0, + 376.0, + 1043.0, + 376.0, + 1057.0, + 361.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1042.0, + 397.0, + 1042.0, + 397.0, + 1057.0, + 382.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1043.0, + 419.0, + 1043.0, + 419.0, + 1058.0, + 405.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1045.0, + 502.0, + 1045.0, + 502.0, + 1056.0, + 492.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1043.0, + 525.0, + 1043.0, + 525.0, + 1057.0, + 511.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1043.0, + 546.0, + 1043.0, + 546.0, + 1057.0, + 532.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1045.0, + 566.0, + 1045.0, + 566.0, + 1055.0, + 554.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1043.0, + 589.0, + 1043.0, + 589.0, + 1057.0, + 574.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1043.0, + 677.0, + 1043.0, + 677.0, + 1057.0, + 660.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1043.0, + 699.0, + 1043.0, + 699.0, + 1057.0, + 680.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1043.0, + 720.0, + 1043.0, + 720.0, + 1057.0, + 700.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1043.0, + 741.0, + 1043.0, + 741.0, + 1057.0, + 723.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1043.0, + 760.0, + 1043.0, + 760.0, + 1057.0, + 742.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1043.0, + 848.0, + 1043.0, + 848.0, + 1057.0, + 831.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1042.0, + 891.0, + 1042.0, + 891.0, + 1058.0, + 851.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1043.0, + 912.0, + 1043.0, + 912.0, + 1057.0, + 893.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1043.0, + 930.0, + 1043.0, + 930.0, + 1057.0, + 913.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1043.0, + 1016.0, + 1043.0, + 1016.0, + 1058.0, + 1002.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1043.0, + 1038.0, + 1043.0, + 1038.0, + 1058.0, + 1023.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1043.0, + 1060.0, + 1043.0, + 1060.0, + 1057.0, + 1044.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1042.0, + 1081.0, + 1042.0, + 1081.0, + 1057.0, + 1065.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1043.0, + 1102.0, + 1043.0, + 1102.0, + 1057.0, + 1088.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1043.0, + 1187.0, + 1043.0, + 1187.0, + 1058.0, + 1173.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1043.0, + 1208.0, + 1043.0, + 1208.0, + 1057.0, + 1194.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1043.0, + 1229.0, + 1043.0, + 1229.0, + 1057.0, + 1215.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1045.0, + 1250.0, + 1045.0, + 1250.0, + 1055.0, + 1239.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1045.0, + 1271.0, + 1045.0, + 1271.0, + 1055.0, + 1260.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1063.0, + 333.0, + 1063.0, + 333.0, + 1078.0, + 318.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1063.0, + 354.0, + 1063.0, + 354.0, + 1078.0, + 339.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1063.0, + 376.0, + 1063.0, + 376.0, + 1077.0, + 361.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1064.0, + 395.0, + 1064.0, + 395.0, + 1076.0, + 384.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1064.0, + 418.0, + 1064.0, + 418.0, + 1074.0, + 406.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1063.0, + 503.0, + 1063.0, + 503.0, + 1078.0, + 491.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1065.0, + 524.0, + 1065.0, + 524.0, + 1076.0, + 513.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1063.0, + 546.0, + 1063.0, + 546.0, + 1077.0, + 532.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1064.0, + 566.0, + 1064.0, + 566.0, + 1074.0, + 555.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1063.0, + 590.0, + 1063.0, + 590.0, + 1077.0, + 575.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1063.0, + 675.0, + 1063.0, + 675.0, + 1078.0, + 661.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1063.0, + 697.0, + 1063.0, + 697.0, + 1078.0, + 680.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1063.0, + 717.0, + 1063.0, + 717.0, + 1077.0, + 702.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1063.0, + 738.0, + 1063.0, + 738.0, + 1077.0, + 724.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1063.0, + 759.0, + 1063.0, + 759.0, + 1077.0, + 745.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1063.0, + 847.0, + 1063.0, + 847.0, + 1078.0, + 831.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1063.0, + 868.0, + 1063.0, + 868.0, + 1078.0, + 852.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1063.0, + 890.0, + 1063.0, + 890.0, + 1077.0, + 873.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1063.0, + 910.0, + 1063.0, + 910.0, + 1077.0, + 895.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1063.0, + 930.0, + 1063.0, + 930.0, + 1077.0, + 916.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1063.0, + 1016.0, + 1063.0, + 1016.0, + 1078.0, + 1002.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1063.0, + 1038.0, + 1063.0, + 1038.0, + 1078.0, + 1023.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1063.0, + 1059.0, + 1063.0, + 1059.0, + 1077.0, + 1045.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1063.0, + 1081.0, + 1063.0, + 1081.0, + 1076.0, + 1066.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1064.0, + 1099.0, + 1064.0, + 1099.0, + 1074.0, + 1089.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1063.0, + 1187.0, + 1063.0, + 1187.0, + 1078.0, + 1173.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1063.0, + 1208.0, + 1063.0, + 1208.0, + 1077.0, + 1194.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1065.0, + 1228.0, + 1065.0, + 1228.0, + 1076.0, + 1219.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1064.0, + 1250.0, + 1064.0, + 1250.0, + 1073.0, + 1239.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1084.0, + 333.0, + 1084.0, + 333.0, + 1099.0, + 319.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1082.0, + 353.0, + 1082.0, + 353.0, + 1097.0, + 340.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1082.0, + 376.0, + 1082.0, + 376.0, + 1097.0, + 361.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1081.0, + 397.0, + 1081.0, + 397.0, + 1097.0, + 382.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1082.0, + 419.0, + 1082.0, + 419.0, + 1097.0, + 405.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1084.0, + 505.0, + 1084.0, + 505.0, + 1097.0, + 491.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1085.0, + 523.0, + 1085.0, + 523.0, + 1096.0, + 513.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1082.0, + 547.0, + 1082.0, + 547.0, + 1097.0, + 532.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1082.0, + 568.0, + 1082.0, + 568.0, + 1097.0, + 553.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1082.0, + 589.0, + 1082.0, + 589.0, + 1097.0, + 575.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1084.0, + 675.0, + 1084.0, + 675.0, + 1097.0, + 661.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1082.0, + 695.0, + 1082.0, + 695.0, + 1097.0, + 682.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1082.0, + 719.0, + 1082.0, + 719.0, + 1097.0, + 704.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1082.0, + 739.0, + 1082.0, + 739.0, + 1097.0, + 724.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1082.0, + 760.0, + 1082.0, + 760.0, + 1097.0, + 746.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1081.0, + 841.0, + 1081.0, + 841.0, + 1096.0, + 826.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1082.0, + 867.0, + 1082.0, + 867.0, + 1097.0, + 853.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1082.0, + 889.0, + 1082.0, + 889.0, + 1097.0, + 874.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1081.0, + 910.0, + 1081.0, + 910.0, + 1096.0, + 895.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1085.0, + 929.0, + 1085.0, + 929.0, + 1095.0, + 918.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1084.0, + 1017.0, + 1084.0, + 1017.0, + 1099.0, + 1002.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1082.0, + 1037.0, + 1082.0, + 1037.0, + 1097.0, + 1023.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1082.0, + 1060.0, + 1082.0, + 1060.0, + 1097.0, + 1045.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1081.0, + 1081.0, + 1081.0, + 1081.0, + 1096.0, + 1066.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1085.0, + 1101.0, + 1085.0, + 1101.0, + 1095.0, + 1089.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1084.0, + 1187.0, + 1084.0, + 1187.0, + 1097.0, + 1173.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1085.0, + 1207.0, + 1085.0, + 1207.0, + 1095.0, + 1195.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1084.0, + 1231.0, + 1084.0, + 1231.0, + 1097.0, + 1216.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1082.0, + 1252.0, + 1082.0, + 1252.0, + 1096.0, + 1237.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1085.0, + 1272.0, + 1085.0, + 1272.0, + 1095.0, + 1260.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1103.0, + 333.0, + 1103.0, + 333.0, + 1118.0, + 318.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1104.0, + 352.0, + 1104.0, + 352.0, + 1116.0, + 341.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1102.0, + 376.0, + 1102.0, + 376.0, + 1117.0, + 361.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1103.0, + 397.0, + 1103.0, + 397.0, + 1118.0, + 382.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1104.0, + 418.0, + 1104.0, + 418.0, + 1113.0, + 408.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1103.0, + 505.0, + 1103.0, + 505.0, + 1117.0, + 491.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1103.0, + 525.0, + 1103.0, + 525.0, + 1117.0, + 511.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1102.0, + 546.0, + 1102.0, + 546.0, + 1118.0, + 532.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1103.0, + 568.0, + 1103.0, + 568.0, + 1118.0, + 553.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1103.0, + 675.0, + 1103.0, + 675.0, + 1117.0, + 661.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1103.0, + 695.0, + 1103.0, + 695.0, + 1117.0, + 682.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1103.0, + 717.0, + 1103.0, + 717.0, + 1117.0, + 704.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1103.0, + 739.0, + 1103.0, + 739.0, + 1118.0, + 724.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1103.0, + 846.0, + 1103.0, + 846.0, + 1117.0, + 831.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1102.0, + 867.0, + 1102.0, + 867.0, + 1118.0, + 853.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1102.0, + 889.0, + 1102.0, + 889.0, + 1118.0, + 875.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1103.0, + 910.0, + 1103.0, + 910.0, + 1118.0, + 895.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1103.0, + 1016.0, + 1103.0, + 1016.0, + 1118.0, + 1002.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1103.0, + 1037.0, + 1103.0, + 1037.0, + 1117.0, + 1023.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1103.0, + 1060.0, + 1103.0, + 1060.0, + 1117.0, + 1045.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1103.0, + 1081.0, + 1103.0, + 1081.0, + 1118.0, + 1065.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1104.0, + 1099.0, + 1104.0, + 1099.0, + 1115.0, + 1089.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1103.0, + 1187.0, + 1103.0, + 1187.0, + 1117.0, + 1173.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1103.0, + 1208.0, + 1103.0, + 1208.0, + 1117.0, + 1194.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1103.0, + 1231.0, + 1103.0, + 1231.0, + 1117.0, + 1216.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1103.0, + 1252.0, + 1103.0, + 1252.0, + 1117.0, + 1237.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1125.0, + 331.0, + 1125.0, + 331.0, + 1136.0, + 320.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1126.0, + 352.0, + 1126.0, + 352.0, + 1136.0, + 341.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1126.0, + 374.0, + 1126.0, + 374.0, + 1136.0, + 362.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1124.0, + 395.0, + 1124.0, + 395.0, + 1135.0, + 385.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1124.0, + 503.0, + 1124.0, + 503.0, + 1139.0, + 491.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1124.0, + 524.0, + 1124.0, + 524.0, + 1140.0, + 511.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1124.0, + 546.0, + 1124.0, + 546.0, + 1140.0, + 532.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1123.0, + 567.0, + 1123.0, + 567.0, + 1138.0, + 552.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1124.0, + 673.0, + 1124.0, + 673.0, + 1140.0, + 661.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1124.0, + 694.0, + 1124.0, + 694.0, + 1140.0, + 682.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1124.0, + 717.0, + 1124.0, + 717.0, + 1140.0, + 704.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1123.0, + 738.0, + 1123.0, + 738.0, + 1138.0, + 724.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1123.0, + 840.0, + 1123.0, + 840.0, + 1136.0, + 826.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1124.0, + 867.0, + 1124.0, + 867.0, + 1140.0, + 853.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1124.0, + 889.0, + 1124.0, + 889.0, + 1140.0, + 874.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1123.0, + 910.0, + 1123.0, + 910.0, + 1138.0, + 895.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1124.0, + 1016.0, + 1124.0, + 1016.0, + 1140.0, + 1002.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1125.0, + 1038.0, + 1125.0, + 1038.0, + 1138.0, + 1023.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1126.0, + 1058.0, + 1126.0, + 1058.0, + 1136.0, + 1046.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1124.0, + 1080.0, + 1124.0, + 1080.0, + 1134.0, + 1068.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1126.0, + 1185.0, + 1126.0, + 1185.0, + 1138.0, + 1175.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1126.0, + 1207.0, + 1126.0, + 1207.0, + 1136.0, + 1197.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1125.0, + 1230.0, + 1125.0, + 1230.0, + 1139.0, + 1216.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1123.0, + 1252.0, + 1123.0, + 1252.0, + 1136.0, + 1237.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1224.0, + 333.0, + 1224.0, + 333.0, + 1237.0, + 319.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1225.0, + 356.0, + 1225.0, + 356.0, + 1237.0, + 339.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1225.0, + 377.0, + 1225.0, + 377.0, + 1237.0, + 358.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1222.0, + 419.0, + 1222.0, + 419.0, + 1240.0, + 381.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1222.0, + 678.0, + 1222.0, + 678.0, + 1240.0, + 658.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1224.0, + 697.0, + 1224.0, + 697.0, + 1239.0, + 680.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1224.0, + 719.0, + 1224.0, + 719.0, + 1237.0, + 704.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1221.0, + 743.0, + 1221.0, + 743.0, + 1239.0, + 722.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1224.0, + 760.0, + 1224.0, + 760.0, + 1237.0, + 745.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1221.0, + 913.0, + 1221.0, + 913.0, + 1240.0, + 829.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1224.0, + 930.0, + 1224.0, + 930.0, + 1237.0, + 914.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1220.0, + 1102.0, + 1220.0, + 1102.0, + 1242.0, + 1001.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1220.0, + 1276.0, + 1220.0, + 1276.0, + 1242.0, + 1169.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1245.0, + 333.0, + 1245.0, + 333.0, + 1259.0, + 318.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1245.0, + 355.0, + 1245.0, + 355.0, + 1258.0, + 338.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1245.0, + 376.0, + 1245.0, + 376.0, + 1258.0, + 360.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1244.0, + 398.0, + 1244.0, + 398.0, + 1258.0, + 381.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1244.0, + 418.0, + 1244.0, + 418.0, + 1258.0, + 404.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1243.0, + 678.0, + 1243.0, + 678.0, + 1260.0, + 658.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1244.0, + 697.0, + 1244.0, + 697.0, + 1259.0, + 680.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1244.0, + 719.0, + 1244.0, + 719.0, + 1258.0, + 704.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1244.0, + 741.0, + 1244.0, + 741.0, + 1258.0, + 724.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1247.0, + 758.0, + 1247.0, + 758.0, + 1256.0, + 746.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1242.0, + 844.0, + 1242.0, + 844.0, + 1260.0, + 825.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1244.0, + 869.0, + 1244.0, + 869.0, + 1259.0, + 853.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1244.0, + 890.0, + 1244.0, + 890.0, + 1258.0, + 873.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1245.0, + 908.0, + 1245.0, + 908.0, + 1256.0, + 897.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1245.0, + 929.0, + 1245.0, + 929.0, + 1256.0, + 919.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1244.0, + 1018.0, + 1244.0, + 1018.0, + 1259.0, + 1002.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1245.0, + 1040.0, + 1245.0, + 1040.0, + 1259.0, + 1022.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1244.0, + 1061.0, + 1244.0, + 1061.0, + 1258.0, + 1044.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1244.0, + 1081.0, + 1244.0, + 1081.0, + 1258.0, + 1066.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1247.0, + 1101.0, + 1247.0, + 1101.0, + 1256.0, + 1089.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1243.0, + 1273.0, + 1243.0, + 1273.0, + 1259.0, + 1169.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1266.0, + 352.0, + 1266.0, + 352.0, + 1276.0, + 341.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1265.0, + 376.0, + 1265.0, + 376.0, + 1279.0, + 361.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1264.0, + 398.0, + 1264.0, + 398.0, + 1279.0, + 382.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1264.0, + 419.0, + 1264.0, + 419.0, + 1279.0, + 403.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1265.0, + 677.0, + 1265.0, + 677.0, + 1279.0, + 660.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1264.0, + 719.0, + 1264.0, + 719.0, + 1279.0, + 704.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1264.0, + 741.0, + 1264.0, + 741.0, + 1278.0, + 724.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1265.0, + 760.0, + 1265.0, + 760.0, + 1278.0, + 744.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1264.0, + 842.0, + 1264.0, + 842.0, + 1279.0, + 826.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1263.0, + 869.0, + 1263.0, + 869.0, + 1280.0, + 849.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1265.0, + 890.0, + 1265.0, + 890.0, + 1279.0, + 873.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1264.0, + 912.0, + 1264.0, + 912.0, + 1278.0, + 895.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1266.0, + 929.0, + 1266.0, + 929.0, + 1276.0, + 918.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1263.0, + 1019.0, + 1263.0, + 1019.0, + 1280.0, + 1000.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1265.0, + 1038.0, + 1265.0, + 1038.0, + 1279.0, + 1022.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1265.0, + 1061.0, + 1265.0, + 1061.0, + 1279.0, + 1044.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1264.0, + 1081.0, + 1264.0, + 1081.0, + 1278.0, + 1066.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1266.0, + 1101.0, + 1266.0, + 1101.0, + 1276.0, + 1089.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1264.0, + 1209.0, + 1264.0, + 1209.0, + 1281.0, + 1171.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1265.0, + 1232.0, + 1265.0, + 1232.0, + 1279.0, + 1214.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1264.0, + 1252.0, + 1264.0, + 1252.0, + 1278.0, + 1237.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1264.0, + 1273.0, + 1264.0, + 1273.0, + 1278.0, + 1257.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1286.0, + 351.0, + 1286.0, + 351.0, + 1296.0, + 342.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1283.0, + 375.0, + 1283.0, + 375.0, + 1298.0, + 361.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1284.0, + 398.0, + 1284.0, + 398.0, + 1298.0, + 382.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1284.0, + 676.0, + 1284.0, + 676.0, + 1298.0, + 661.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1284.0, + 695.0, + 1284.0, + 695.0, + 1298.0, + 680.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1284.0, + 719.0, + 1284.0, + 719.0, + 1298.0, + 704.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1284.0, + 741.0, + 1284.0, + 741.0, + 1298.0, + 724.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1284.0, + 842.0, + 1284.0, + 842.0, + 1298.0, + 826.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1283.0, + 867.0, + 1283.0, + 867.0, + 1298.0, + 853.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1284.0, + 890.0, + 1284.0, + 890.0, + 1298.0, + 875.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1284.0, + 910.0, + 1284.0, + 910.0, + 1298.0, + 895.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1283.0, + 1018.0, + 1283.0, + 1018.0, + 1298.0, + 1002.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1283.0, + 1038.0, + 1283.0, + 1038.0, + 1298.0, + 1022.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1284.0, + 1060.0, + 1284.0, + 1060.0, + 1298.0, + 1045.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1284.0, + 1081.0, + 1284.0, + 1081.0, + 1298.0, + 1066.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1282.0, + 1210.0, + 1282.0, + 1210.0, + 1299.0, + 1171.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1283.0, + 1231.0, + 1283.0, + 1231.0, + 1298.0, + 1215.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1284.0, + 1251.0, + 1284.0, + 1251.0, + 1298.0, + 1237.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1306.0, + 351.0, + 1306.0, + 351.0, + 1319.0, + 341.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1305.0, + 375.0, + 1305.0, + 375.0, + 1320.0, + 362.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1304.0, + 398.0, + 1304.0, + 398.0, + 1318.0, + 382.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1306.0, + 675.0, + 1306.0, + 675.0, + 1320.0, + 661.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1305.0, + 695.0, + 1305.0, + 695.0, + 1321.0, + 682.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1307.0, + 715.0, + 1307.0, + 715.0, + 1315.0, + 705.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1305.0, + 737.0, + 1305.0, + 737.0, + 1315.0, + 727.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1304.0, + 841.0, + 1304.0, + 841.0, + 1317.0, + 826.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1307.0, + 866.0, + 1307.0, + 866.0, + 1319.0, + 855.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1306.0, + 889.0, + 1306.0, + 889.0, + 1320.0, + 875.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1304.0, + 911.0, + 1304.0, + 911.0, + 1318.0, + 896.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1305.0, + 1016.0, + 1305.0, + 1016.0, + 1320.0, + 1002.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1306.0, + 1037.0, + 1306.0, + 1037.0, + 1320.0, + 1023.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 1307.0, + 1058.0, + 1307.0, + 1058.0, + 1319.0, + 1047.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1304.0, + 1081.0, + 1304.0, + 1081.0, + 1318.0, + 1067.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1305.0, + 1187.0, + 1305.0, + 1187.0, + 1321.0, + 1173.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 1305.0, + 1208.0, + 1305.0, + 1208.0, + 1321.0, + 1193.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1307.0, + 1228.0, + 1307.0, + 1228.0, + 1319.0, + 1217.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1305.0, + 1250.0, + 1305.0, + 1250.0, + 1315.0, + 1238.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1405.0, + 354.0, + 1405.0, + 354.0, + 1421.0, + 339.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1406.0, + 376.0, + 1406.0, + 376.0, + 1420.0, + 361.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 1404.0, + 397.0, + 1404.0, + 397.0, + 1419.0, + 384.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1405.0, + 418.0, + 1405.0, + 418.0, + 1420.0, + 404.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 1404.0, + 548.0, + 1404.0, + 548.0, + 1421.0, + 529.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1406.0, + 569.0, + 1406.0, + 569.0, + 1420.0, + 553.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1405.0, + 590.0, + 1405.0, + 590.0, + 1419.0, + 574.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1405.0, + 610.0, + 1405.0, + 610.0, + 1419.0, + 595.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1406.0, + 723.0, + 1406.0, + 723.0, + 1420.0, + 704.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1403.0, + 805.0, + 1403.0, + 805.0, + 1422.0, + 724.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1402.0, + 999.0, + 1402.0, + 999.0, + 1423.0, + 892.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1426.0, + 354.0, + 1426.0, + 354.0, + 1441.0, + 339.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1424.0, + 376.0, + 1424.0, + 376.0, + 1439.0, + 361.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1424.0, + 397.0, + 1424.0, + 397.0, + 1438.0, + 383.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1426.0, + 526.0, + 1426.0, + 526.0, + 1439.0, + 510.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1426.0, + 547.0, + 1426.0, + 547.0, + 1439.0, + 530.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1426.0, + 568.0, + 1426.0, + 568.0, + 1439.0, + 552.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1424.0, + 590.0, + 1424.0, + 590.0, + 1438.0, + 574.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1426.0, + 610.0, + 1426.0, + 610.0, + 1439.0, + 596.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1424.0, + 722.0, + 1424.0, + 722.0, + 1442.0, + 702.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1426.0, + 742.0, + 1426.0, + 742.0, + 1439.0, + 724.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1426.0, + 761.0, + 1426.0, + 761.0, + 1438.0, + 743.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1424.0, + 783.0, + 1424.0, + 783.0, + 1438.0, + 765.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1426.0, + 803.0, + 1426.0, + 803.0, + 1438.0, + 787.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1423.0, + 997.0, + 1423.0, + 997.0, + 1439.0, + 892.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1445.0, + 353.0, + 1445.0, + 353.0, + 1460.0, + 340.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1445.0, + 376.0, + 1445.0, + 376.0, + 1460.0, + 361.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1444.0, + 397.0, + 1444.0, + 397.0, + 1459.0, + 383.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1445.0, + 419.0, + 1445.0, + 419.0, + 1459.0, + 405.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1445.0, + 526.0, + 1445.0, + 526.0, + 1460.0, + 510.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1445.0, + 546.0, + 1445.0, + 546.0, + 1460.0, + 531.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1445.0, + 568.0, + 1445.0, + 568.0, + 1460.0, + 553.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1444.0, + 589.0, + 1444.0, + 589.0, + 1459.0, + 575.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1445.0, + 611.0, + 1445.0, + 611.0, + 1459.0, + 595.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1445.0, + 719.0, + 1445.0, + 719.0, + 1460.0, + 702.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1445.0, + 738.0, + 1445.0, + 738.0, + 1460.0, + 724.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1445.0, + 760.0, + 1445.0, + 760.0, + 1459.0, + 745.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1445.0, + 782.0, + 1445.0, + 782.0, + 1459.0, + 766.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1447.0, + 802.0, + 1447.0, + 802.0, + 1458.0, + 790.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1446.0, + 913.0, + 1446.0, + 913.0, + 1460.0, + 895.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1446.0, + 932.0, + 1446.0, + 932.0, + 1459.0, + 914.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 1445.0, + 955.0, + 1445.0, + 955.0, + 1459.0, + 937.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1445.0, + 995.0, + 1445.0, + 995.0, + 1459.0, + 978.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1463.0, + 353.0, + 1463.0, + 353.0, + 1481.0, + 340.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1463.0, + 376.0, + 1463.0, + 376.0, + 1480.0, + 362.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1466.0, + 397.0, + 1466.0, + 397.0, + 1480.0, + 383.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1467.0, + 415.0, + 1467.0, + 415.0, + 1476.0, + 406.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1465.0, + 525.0, + 1465.0, + 525.0, + 1480.0, + 511.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1465.0, + 546.0, + 1465.0, + 546.0, + 1480.0, + 531.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1465.0, + 568.0, + 1465.0, + 568.0, + 1480.0, + 553.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1467.0, + 588.0, + 1467.0, + 588.0, + 1477.0, + 577.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1465.0, + 719.0, + 1465.0, + 719.0, + 1480.0, + 704.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1463.0, + 738.0, + 1463.0, + 738.0, + 1480.0, + 724.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1465.0, + 760.0, + 1465.0, + 760.0, + 1480.0, + 746.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1466.0, + 782.0, + 1466.0, + 782.0, + 1480.0, + 767.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1465.0, + 912.0, + 1465.0, + 912.0, + 1480.0, + 896.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1463.0, + 933.0, + 1463.0, + 933.0, + 1481.0, + 913.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1465.0, + 954.0, + 1465.0, + 954.0, + 1480.0, + 938.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1466.0, + 973.0, + 1466.0, + 973.0, + 1478.0, + 958.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1485.0, + 353.0, + 1485.0, + 353.0, + 1503.0, + 340.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1485.0, + 375.0, + 1485.0, + 375.0, + 1501.0, + 361.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1484.0, + 396.0, + 1484.0, + 396.0, + 1499.0, + 383.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1486.0, + 525.0, + 1486.0, + 525.0, + 1501.0, + 511.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1486.0, + 546.0, + 1486.0, + 546.0, + 1501.0, + 531.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1485.0, + 567.0, + 1485.0, + 567.0, + 1501.0, + 553.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1486.0, + 588.0, + 1486.0, + 588.0, + 1497.0, + 576.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1486.0, + 719.0, + 1486.0, + 719.0, + 1501.0, + 704.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1486.0, + 738.0, + 1486.0, + 738.0, + 1501.0, + 724.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1486.0, + 760.0, + 1486.0, + 760.0, + 1501.0, + 745.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1486.0, + 780.0, + 1486.0, + 780.0, + 1497.0, + 768.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1486.0, + 912.0, + 1486.0, + 912.0, + 1501.0, + 896.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1486.0, + 930.0, + 1486.0, + 930.0, + 1500.0, + 914.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1486.0, + 952.0, + 1486.0, + 952.0, + 1500.0, + 938.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1486.0, + 972.0, + 1486.0, + 972.0, + 1497.0, + 960.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1585.0, + 354.0, + 1585.0, + 354.0, + 1601.0, + 339.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1585.0, + 376.0, + 1585.0, + 376.0, + 1600.0, + 360.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1584.0, + 399.0, + 1584.0, + 399.0, + 1601.0, + 380.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1585.0, + 419.0, + 1585.0, + 419.0, + 1600.0, + 403.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1586.0, + 440.0, + 1586.0, + 440.0, + 1600.0, + 426.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1586.0, + 570.0, + 1586.0, + 570.0, + 1600.0, + 553.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1586.0, + 591.0, + 1586.0, + 591.0, + 1600.0, + 572.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1584.0, + 613.0, + 1584.0, + 613.0, + 1601.0, + 594.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1585.0, + 633.0, + 1585.0, + 633.0, + 1600.0, + 617.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1585.0, + 653.0, + 1585.0, + 653.0, + 1600.0, + 639.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1586.0, + 783.0, + 1586.0, + 783.0, + 1600.0, + 766.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1586.0, + 804.0, + 1586.0, + 804.0, + 1600.0, + 785.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1586.0, + 825.0, + 1586.0, + 825.0, + 1600.0, + 807.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1585.0, + 847.0, + 1585.0, + 847.0, + 1599.0, + 830.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1585.0, + 867.0, + 1585.0, + 867.0, + 1600.0, + 851.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1581.0, + 1082.0, + 1581.0, + 1082.0, + 1604.0, + 980.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1606.0, + 354.0, + 1606.0, + 354.0, + 1622.0, + 339.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1606.0, + 376.0, + 1606.0, + 376.0, + 1622.0, + 361.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1606.0, + 398.0, + 1606.0, + 398.0, + 1621.0, + 383.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1605.0, + 419.0, + 1605.0, + 419.0, + 1620.0, + 404.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1608.0, + 437.0, + 1608.0, + 437.0, + 1618.0, + 427.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1605.0, + 570.0, + 1605.0, + 570.0, + 1623.0, + 552.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1605.0, + 591.0, + 1605.0, + 591.0, + 1623.0, + 572.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1606.0, + 611.0, + 1606.0, + 611.0, + 1620.0, + 596.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1605.0, + 632.0, + 1605.0, + 632.0, + 1620.0, + 617.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1606.0, + 653.0, + 1606.0, + 653.0, + 1620.0, + 639.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1607.0, + 782.0, + 1607.0, + 782.0, + 1622.0, + 766.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1607.0, + 803.0, + 1607.0, + 803.0, + 1621.0, + 788.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1606.0, + 824.0, + 1606.0, + 824.0, + 1620.0, + 810.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1606.0, + 846.0, + 1606.0, + 846.0, + 1620.0, + 831.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1607.0, + 864.0, + 1607.0, + 864.0, + 1618.0, + 854.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1605.0, + 998.0, + 1605.0, + 998.0, + 1623.0, + 979.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1607.0, + 1016.0, + 1607.0, + 1016.0, + 1621.0, + 1001.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1606.0, + 1039.0, + 1606.0, + 1039.0, + 1620.0, + 1023.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1606.0, + 1061.0, + 1606.0, + 1061.0, + 1620.0, + 1044.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1606.0, + 1081.0, + 1606.0, + 1081.0, + 1620.0, + 1065.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1626.0, + 354.0, + 1626.0, + 354.0, + 1641.0, + 339.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1626.0, + 376.0, + 1626.0, + 376.0, + 1641.0, + 361.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1626.0, + 398.0, + 1626.0, + 398.0, + 1641.0, + 383.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1624.0, + 418.0, + 1624.0, + 418.0, + 1641.0, + 404.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1626.0, + 440.0, + 1626.0, + 440.0, + 1640.0, + 426.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1626.0, + 569.0, + 1626.0, + 569.0, + 1641.0, + 553.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1625.0, + 591.0, + 1625.0, + 591.0, + 1643.0, + 572.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1626.0, + 611.0, + 1626.0, + 611.0, + 1641.0, + 596.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1623.0, + 633.0, + 1623.0, + 633.0, + 1643.0, + 616.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1628.0, + 651.0, + 1628.0, + 651.0, + 1639.0, + 641.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1625.0, + 785.0, + 1625.0, + 785.0, + 1643.0, + 765.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1626.0, + 802.0, + 1626.0, + 802.0, + 1641.0, + 788.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1626.0, + 824.0, + 1626.0, + 824.0, + 1641.0, + 810.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1625.0, + 846.0, + 1625.0, + 846.0, + 1640.0, + 831.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1626.0, + 867.0, + 1626.0, + 867.0, + 1640.0, + 853.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1626.0, + 996.0, + 1626.0, + 996.0, + 1641.0, + 980.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1626.0, + 1016.0, + 1626.0, + 1016.0, + 1641.0, + 1001.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1626.0, + 1038.0, + 1626.0, + 1038.0, + 1641.0, + 1023.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1625.0, + 1060.0, + 1625.0, + 1060.0, + 1640.0, + 1044.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1626.0, + 1081.0, + 1626.0, + 1081.0, + 1640.0, + 1065.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1645.0, + 354.0, + 1645.0, + 354.0, + 1661.0, + 339.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1645.0, + 375.0, + 1645.0, + 375.0, + 1661.0, + 361.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1645.0, + 398.0, + 1645.0, + 398.0, + 1661.0, + 383.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1646.0, + 418.0, + 1646.0, + 418.0, + 1661.0, + 404.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1645.0, + 568.0, + 1645.0, + 568.0, + 1661.0, + 553.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1645.0, + 589.0, + 1645.0, + 589.0, + 1661.0, + 575.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1645.0, + 611.0, + 1645.0, + 611.0, + 1661.0, + 596.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1646.0, + 632.0, + 1646.0, + 632.0, + 1661.0, + 618.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1646.0, + 782.0, + 1646.0, + 782.0, + 1661.0, + 767.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1646.0, + 802.0, + 1646.0, + 802.0, + 1661.0, + 788.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1645.0, + 824.0, + 1645.0, + 824.0, + 1661.0, + 810.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1646.0, + 846.0, + 1646.0, + 846.0, + 1661.0, + 831.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1644.0, + 998.0, + 1644.0, + 998.0, + 1662.0, + 979.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1646.0, + 1016.0, + 1646.0, + 1016.0, + 1661.0, + 1001.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1645.0, + 1038.0, + 1645.0, + 1038.0, + 1661.0, + 1023.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1646.0, + 1060.0, + 1646.0, + 1060.0, + 1661.0, + 1044.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1646.0, + 1077.0, + 1646.0, + 1077.0, + 1656.0, + 1069.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1667.0, + 353.0, + 1667.0, + 353.0, + 1683.0, + 339.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1667.0, + 375.0, + 1667.0, + 375.0, + 1683.0, + 361.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1667.0, + 397.0, + 1667.0, + 397.0, + 1683.0, + 383.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1665.0, + 418.0, + 1665.0, + 418.0, + 1679.0, + 404.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1667.0, + 567.0, + 1667.0, + 567.0, + 1683.0, + 553.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1667.0, + 589.0, + 1667.0, + 589.0, + 1683.0, + 575.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1668.0, + 611.0, + 1668.0, + 611.0, + 1683.0, + 596.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1667.0, + 629.0, + 1667.0, + 629.0, + 1678.0, + 619.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1668.0, + 782.0, + 1668.0, + 782.0, + 1683.0, + 767.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1668.0, + 802.0, + 1668.0, + 802.0, + 1683.0, + 788.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1667.0, + 824.0, + 1667.0, + 824.0, + 1683.0, + 809.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1665.0, + 846.0, + 1665.0, + 846.0, + 1680.0, + 831.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1667.0, + 995.0, + 1667.0, + 995.0, + 1683.0, + 980.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1668.0, + 1016.0, + 1668.0, + 1016.0, + 1683.0, + 1002.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1668.0, + 1037.0, + 1668.0, + 1037.0, + 1683.0, + 1023.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1665.0, + 1060.0, + 1665.0, + 1060.0, + 1680.0, + 1044.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1263.5, + 700.0, + 1263.5, + 700.0, + 1279.0, + 676.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.25, + 1404.0, + 525.25, + 1404.0, + 525.25, + 1422.0, + 511.25, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1443.0, + 977.0, + 1443.0, + 977.0, + 1460.0, + 957.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 353.0, + 333.0, + 353.0, + 333.0, + 367.0, + 319.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 354.0, + 355.0, + 354.0, + 355.0, + 367.0, + 340.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 353.0, + 376.0, + 353.0, + 376.0, + 367.0, + 361.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 352.0, + 397.0, + 352.0, + 397.0, + 367.0, + 383.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 353.0, + 419.0, + 353.0, + 419.0, + 367.0, + 405.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 353.0, + 503.0, + 353.0, + 503.0, + 367.0, + 489.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 353.0, + 525.0, + 353.0, + 525.0, + 367.0, + 512.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 353.0, + 546.0, + 353.0, + 546.0, + 367.0, + 531.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 352.0, + 568.0, + 352.0, + 568.0, + 367.0, + 553.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 353.0, + 589.0, + 353.0, + 589.0, + 367.0, + 574.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 353.0, + 676.0, + 353.0, + 676.0, + 367.0, + 660.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 353.0, + 699.0, + 353.0, + 699.0, + 367.0, + 681.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 353.0, + 719.0, + 353.0, + 719.0, + 367.0, + 701.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 352.0, + 761.0, + 352.0, + 761.0, + 370.0, + 723.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 348.0, + 934.0, + 348.0, + 934.0, + 371.0, + 831.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 353.0, + 1017.0, + 353.0, + 1017.0, + 368.0, + 1002.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 353.0, + 1038.0, + 353.0, + 1038.0, + 367.0, + 1024.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 353.0, + 1060.0, + 353.0, + 1060.0, + 367.0, + 1045.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 352.0, + 1081.0, + 352.0, + 1081.0, + 367.0, + 1066.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 353.0, + 1102.0, + 353.0, + 1102.0, + 367.0, + 1086.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 353.0, + 1187.0, + 353.0, + 1187.0, + 368.0, + 1174.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 353.0, + 1208.0, + 353.0, + 1208.0, + 367.0, + 1193.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 353.0, + 1229.0, + 353.0, + 1229.0, + 367.0, + 1215.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 353.0, + 1249.0, + 353.0, + 1249.0, + 365.0, + 1239.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 354.0, + 1271.0, + 354.0, + 1271.0, + 366.0, + 1261.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 374.0, + 333.0, + 374.0, + 333.0, + 388.0, + 318.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 374.0, + 354.0, + 374.0, + 354.0, + 388.0, + 340.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 375.0, + 374.0, + 375.0, + 374.0, + 386.0, + 363.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 372.0, + 397.0, + 372.0, + 397.0, + 386.0, + 383.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 374.0, + 417.0, + 374.0, + 417.0, + 386.0, + 407.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 374.0, + 503.0, + 374.0, + 503.0, + 388.0, + 489.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 375.0, + 523.0, + 375.0, + 523.0, + 387.0, + 513.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 375.0, + 545.0, + 375.0, + 545.0, + 386.0, + 534.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 373.0, + 565.0, + 373.0, + 565.0, + 385.0, + 556.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 374.0, + 587.0, + 374.0, + 587.0, + 386.0, + 578.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 373.0, + 675.0, + 373.0, + 675.0, + 388.0, + 660.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 373.0, + 697.0, + 373.0, + 697.0, + 388.0, + 681.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 373.0, + 718.0, + 373.0, + 718.0, + 387.0, + 702.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 373.0, + 740.0, + 373.0, + 740.0, + 387.0, + 724.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 373.0, + 760.0, + 373.0, + 760.0, + 387.0, + 745.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 373.0, + 847.0, + 373.0, + 847.0, + 388.0, + 831.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 372.0, + 869.0, + 372.0, + 869.0, + 389.0, + 850.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 373.0, + 890.0, + 373.0, + 890.0, + 387.0, + 873.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 371.0, + 912.0, + 371.0, + 912.0, + 388.0, + 892.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 373.0, + 931.0, + 373.0, + 931.0, + 387.0, + 916.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 373.0, + 1017.0, + 373.0, + 1017.0, + 389.0, + 1002.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 373.0, + 1038.0, + 373.0, + 1038.0, + 388.0, + 1024.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 375.0, + 1059.0, + 375.0, + 1059.0, + 386.0, + 1047.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 373.0, + 1081.0, + 373.0, + 1081.0, + 387.0, + 1066.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 374.0, + 1099.0, + 374.0, + 1099.0, + 386.0, + 1090.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 373.0, + 1187.0, + 373.0, + 1187.0, + 388.0, + 1172.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 375.0, + 1206.0, + 375.0, + 1206.0, + 386.0, + 1197.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 375.0, + 1228.0, + 375.0, + 1228.0, + 386.0, + 1218.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 373.0, + 1248.0, + 373.0, + 1248.0, + 385.0, + 1239.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 375.0, + 1271.0, + 375.0, + 1271.0, + 385.0, + 1262.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 394.0, + 333.0, + 394.0, + 333.0, + 408.0, + 319.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 393.0, + 354.0, + 393.0, + 354.0, + 408.0, + 340.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 393.0, + 376.0, + 393.0, + 376.0, + 408.0, + 362.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 392.0, + 397.0, + 392.0, + 397.0, + 407.0, + 383.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 393.0, + 420.0, + 393.0, + 420.0, + 408.0, + 404.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 395.0, + 502.0, + 395.0, + 502.0, + 406.0, + 491.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 395.0, + 523.0, + 395.0, + 523.0, + 406.0, + 513.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 393.0, + 546.0, + 393.0, + 546.0, + 408.0, + 531.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 393.0, + 568.0, + 393.0, + 568.0, + 408.0, + 552.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 393.0, + 589.0, + 393.0, + 589.0, + 408.0, + 575.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 393.0, + 675.0, + 393.0, + 675.0, + 408.0, + 660.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 393.0, + 695.0, + 393.0, + 695.0, + 408.0, + 681.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 393.0, + 718.0, + 393.0, + 718.0, + 408.0, + 703.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 393.0, + 740.0, + 393.0, + 740.0, + 408.0, + 724.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 393.0, + 761.0, + 393.0, + 761.0, + 408.0, + 746.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 393.0, + 846.0, + 393.0, + 846.0, + 408.0, + 831.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 393.0, + 867.0, + 393.0, + 867.0, + 408.0, + 852.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 392.0, + 891.0, + 392.0, + 891.0, + 409.0, + 872.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 393.0, + 911.0, + 393.0, + 911.0, + 408.0, + 895.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 393.0, + 931.0, + 393.0, + 931.0, + 408.0, + 915.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 393.0, + 1017.0, + 393.0, + 1017.0, + 409.0, + 1002.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 393.0, + 1038.0, + 393.0, + 1038.0, + 408.0, + 1024.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 393.0, + 1060.0, + 393.0, + 1060.0, + 408.0, + 1045.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 393.0, + 1081.0, + 393.0, + 1081.0, + 407.0, + 1066.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 393.0, + 1102.0, + 393.0, + 1102.0, + 408.0, + 1086.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 393.0, + 1187.0, + 393.0, + 1187.0, + 409.0, + 1172.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 393.0, + 1207.0, + 393.0, + 1207.0, + 408.0, + 1194.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 393.0, + 1230.0, + 393.0, + 1230.0, + 408.0, + 1215.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 393.0, + 1251.0, + 393.0, + 1251.0, + 407.0, + 1237.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 393.0, + 1272.0, + 393.0, + 1272.0, + 408.0, + 1258.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 413.0, + 331.0, + 413.0, + 331.0, + 428.0, + 319.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 414.0, + 352.0, + 414.0, + 352.0, + 426.0, + 342.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 412.0, + 376.0, + 412.0, + 376.0, + 427.0, + 362.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 413.0, + 397.0, + 413.0, + 397.0, + 427.0, + 383.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 414.0, + 419.0, + 414.0, + 419.0, + 423.0, + 409.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 414.0, + 502.0, + 414.0, + 502.0, + 426.0, + 492.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 413.0, + 524.0, + 413.0, + 524.0, + 428.0, + 512.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 413.0, + 546.0, + 413.0, + 546.0, + 428.0, + 532.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 414.0, + 567.0, + 414.0, + 567.0, + 428.0, + 553.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 414.0, + 588.0, + 414.0, + 588.0, + 423.0, + 580.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 413.0, + 675.0, + 413.0, + 675.0, + 427.0, + 661.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 414.0, + 694.0, + 414.0, + 694.0, + 426.0, + 683.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 413.0, + 717.0, + 413.0, + 717.0, + 427.0, + 703.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 413.0, + 739.0, + 413.0, + 739.0, + 428.0, + 725.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 414.0, + 760.0, + 414.0, + 760.0, + 423.0, + 751.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 410.0, + 847.0, + 410.0, + 847.0, + 429.0, + 829.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 412.0, + 867.0, + 412.0, + 867.0, + 428.0, + 853.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 412.0, + 889.0, + 412.0, + 889.0, + 428.0, + 875.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 413.0, + 910.0, + 413.0, + 910.0, + 428.0, + 895.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 412.0, + 932.0, + 412.0, + 932.0, + 427.0, + 919.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 412.0, + 1017.0, + 412.0, + 1017.0, + 428.0, + 1002.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 412.0, + 1038.0, + 412.0, + 1038.0, + 428.0, + 1024.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 413.0, + 1060.0, + 413.0, + 1060.0, + 427.0, + 1045.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 413.0, + 1081.0, + 413.0, + 1081.0, + 428.0, + 1066.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 413.0, + 1102.0, + 413.0, + 1102.0, + 426.0, + 1092.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 413.0, + 1187.0, + 413.0, + 1187.0, + 428.0, + 1174.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 414.0, + 1206.0, + 414.0, + 1206.0, + 426.0, + 1197.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 413.0, + 1230.0, + 413.0, + 1230.0, + 428.0, + 1217.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 415.0, + 1249.0, + 415.0, + 1249.0, + 426.0, + 1239.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 436.0, + 329.0, + 436.0, + 329.0, + 445.0, + 321.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 435.0, + 352.0, + 435.0, + 352.0, + 445.0, + 342.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 436.0, + 373.0, + 436.0, + 373.0, + 447.0, + 363.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 433.0, + 397.0, + 433.0, + 397.0, + 447.0, + 383.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 436.0, + 502.0, + 436.0, + 502.0, + 447.0, + 492.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 436.0, + 523.0, + 436.0, + 523.0, + 448.0, + 513.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 434.0, + 546.0, + 434.0, + 546.0, + 449.0, + 532.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 433.0, + 568.0, + 433.0, + 568.0, + 447.0, + 552.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 435.0, + 674.0, + 435.0, + 674.0, + 450.0, + 661.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 434.0, + 695.0, + 434.0, + 695.0, + 449.0, + 682.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 434.0, + 717.0, + 434.0, + 717.0, + 449.0, + 703.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 433.0, + 739.0, + 433.0, + 739.0, + 447.0, + 724.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 434.0, + 758.0, + 434.0, + 758.0, + 445.0, + 748.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 434.0, + 846.0, + 434.0, + 846.0, + 450.0, + 831.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 434.0, + 867.0, + 434.0, + 867.0, + 450.0, + 853.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 434.0, + 889.0, + 434.0, + 889.0, + 450.0, + 874.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 431.0, + 910.0, + 431.0, + 910.0, + 447.0, + 895.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 435.0, + 927.0, + 435.0, + 927.0, + 444.0, + 918.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 434.0, + 1017.0, + 434.0, + 1017.0, + 450.0, + 1003.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 433.0, + 1038.0, + 433.0, + 1038.0, + 448.0, + 1023.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 436.0, + 1057.0, + 436.0, + 1057.0, + 447.0, + 1046.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 433.0, + 1079.0, + 433.0, + 1079.0, + 447.0, + 1066.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 435.0, + 1186.0, + 435.0, + 1186.0, + 450.0, + 1174.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 436.0, + 1206.0, + 436.0, + 1206.0, + 447.0, + 1197.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 436.0, + 1228.0, + 436.0, + 1228.0, + 448.0, + 1218.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 433.0, + 1251.0, + 433.0, + 1251.0, + 447.0, + 1237.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 535.0, + 334.0, + 535.0, + 334.0, + 549.0, + 319.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 535.0, + 356.0, + 535.0, + 356.0, + 549.0, + 337.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 533.0, + 399.0, + 533.0, + 399.0, + 550.0, + 358.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 534.0, + 419.0, + 534.0, + 419.0, + 549.0, + 402.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 535.0, + 848.0, + 535.0, + 848.0, + 549.0, + 831.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 535.0, + 869.0, + 535.0, + 869.0, + 549.0, + 850.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 534.0, + 890.0, + 534.0, + 890.0, + 548.0, + 872.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 533.0, + 910.0, + 533.0, + 910.0, + 548.0, + 895.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 534.0, + 931.0, + 534.0, + 931.0, + 548.0, + 915.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 533.0, + 1041.0, + 533.0, + 1041.0, + 550.0, + 1000.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 532.0, + 1104.0, + 532.0, + 1104.0, + 552.0, + 1042.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 533.0, + 1233.0, + 533.0, + 1233.0, + 550.0, + 1172.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 534.0, + 1253.0, + 534.0, + 1253.0, + 548.0, + 1234.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 534.0, + 1272.0, + 534.0, + 1272.0, + 549.0, + 1257.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 553.0, + 869.0, + 553.0, + 869.0, + 570.0, + 830.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 554.0, + 890.0, + 554.0, + 890.0, + 568.0, + 873.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 553.0, + 910.0, + 553.0, + 910.0, + 567.0, + 895.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 555.0, + 928.0, + 555.0, + 928.0, + 566.0, + 918.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 553.0, + 1019.0, + 553.0, + 1019.0, + 571.0, + 1000.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 553.0, + 1040.0, + 553.0, + 1040.0, + 570.0, + 1020.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 554.0, + 1060.0, + 554.0, + 1060.0, + 568.0, + 1045.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 553.0, + 1082.0, + 553.0, + 1082.0, + 567.0, + 1066.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 554.0, + 1102.0, + 554.0, + 1102.0, + 568.0, + 1086.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 553.0, + 1190.0, + 553.0, + 1190.0, + 571.0, + 1171.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 555.0, + 1211.0, + 555.0, + 1211.0, + 569.0, + 1192.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 554.0, + 1232.0, + 554.0, + 1232.0, + 568.0, + 1214.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 553.0, + 1253.0, + 553.0, + 1253.0, + 567.0, + 1235.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 555.0, + 1271.0, + 555.0, + 1271.0, + 566.0, + 1259.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 573.0, + 867.0, + 573.0, + 867.0, + 590.0, + 830.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 574.0, + 889.0, + 574.0, + 889.0, + 589.0, + 874.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 574.0, + 910.0, + 574.0, + 910.0, + 588.0, + 895.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 574.0, + 931.0, + 574.0, + 931.0, + 588.0, + 917.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 573.0, + 1019.0, + 573.0, + 1019.0, + 591.0, + 1000.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 574.0, + 1038.0, + 574.0, + 1038.0, + 589.0, + 1021.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 574.0, + 1061.0, + 574.0, + 1061.0, + 589.0, + 1045.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 573.0, + 1081.0, + 573.0, + 1081.0, + 588.0, + 1066.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 574.0, + 1102.0, + 574.0, + 1102.0, + 589.0, + 1086.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 574.0, + 1187.0, + 574.0, + 1187.0, + 590.0, + 1172.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 574.0, + 1208.0, + 574.0, + 1208.0, + 589.0, + 1193.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 574.0, + 1230.0, + 574.0, + 1230.0, + 589.0, + 1214.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 573.0, + 1251.0, + 573.0, + 1251.0, + 588.0, + 1236.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 574.0, + 1272.0, + 574.0, + 1272.0, + 589.0, + 1258.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 592.0, + 869.0, + 592.0, + 869.0, + 610.0, + 830.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 594.0, + 889.0, + 594.0, + 889.0, + 608.0, + 874.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 595.0, + 910.0, + 595.0, + 910.0, + 608.0, + 895.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 594.0, + 1018.0, + 594.0, + 1018.0, + 609.0, + 1002.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 594.0, + 1038.0, + 594.0, + 1038.0, + 609.0, + 1021.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 594.0, + 1060.0, + 594.0, + 1060.0, + 608.0, + 1046.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 595.0, + 1081.0, + 595.0, + 1081.0, + 608.0, + 1066.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 594.0, + 1189.0, + 594.0, + 1189.0, + 609.0, + 1172.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 594.0, + 1208.0, + 594.0, + 1208.0, + 609.0, + 1193.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 594.0, + 1230.0, + 594.0, + 1230.0, + 609.0, + 1215.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 595.0, + 1251.0, + 595.0, + 1251.0, + 609.0, + 1237.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 616.0, + 847.0, + 616.0, + 847.0, + 630.0, + 832.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 616.0, + 867.0, + 616.0, + 867.0, + 630.0, + 852.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 616.0, + 889.0, + 616.0, + 889.0, + 630.0, + 875.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 615.0, + 910.0, + 615.0, + 910.0, + 628.0, + 895.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 615.0, + 1017.0, + 615.0, + 1017.0, + 631.0, + 1002.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 616.0, + 1038.0, + 616.0, + 1038.0, + 630.0, + 1021.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 616.0, + 1060.0, + 616.0, + 1060.0, + 630.0, + 1046.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 615.0, + 1081.0, + 615.0, + 1081.0, + 628.0, + 1066.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 615.0, + 1187.0, + 615.0, + 1187.0, + 631.0, + 1172.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 616.0, + 1208.0, + 616.0, + 1208.0, + 631.0, + 1193.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 617.0, + 1228.0, + 617.0, + 1228.0, + 629.0, + 1218.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 615.0, + 1249.0, + 615.0, + 1249.0, + 626.0, + 1239.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 716.0, + 334.0, + 716.0, + 334.0, + 730.0, + 319.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 716.0, + 355.0, + 716.0, + 355.0, + 730.0, + 338.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 716.0, + 373.0, + 716.0, + 373.0, + 728.0, + 363.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 716.0, + 395.0, + 716.0, + 395.0, + 727.0, + 384.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 715.0, + 419.0, + 715.0, + 419.0, + 729.0, + 405.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 715.0, + 527.0, + 715.0, + 527.0, + 730.0, + 510.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 715.0, + 548.0, + 715.0, + 548.0, + 730.0, + 531.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 716.0, + 568.0, + 716.0, + 568.0, + 730.0, + 552.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 714.0, + 589.0, + 714.0, + 589.0, + 729.0, + 575.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 716.0, + 609.0, + 716.0, + 609.0, + 727.0, + 597.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 716.0, + 719.0, + 716.0, + 719.0, + 730.0, + 703.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 716.0, + 741.0, + 716.0, + 741.0, + 730.0, + 724.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 716.0, + 761.0, + 716.0, + 761.0, + 730.0, + 743.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 715.0, + 783.0, + 715.0, + 783.0, + 729.0, + 766.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 716.0, + 803.0, + 716.0, + 803.0, + 730.0, + 788.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 711.0, + 999.0, + 711.0, + 999.0, + 733.0, + 894.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 735.0, + 333.0, + 735.0, + 333.0, + 750.0, + 318.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 736.0, + 354.0, + 736.0, + 354.0, + 750.0, + 340.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 735.0, + 376.0, + 735.0, + 376.0, + 749.0, + 362.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 736.0, + 394.0, + 736.0, + 394.0, + 745.0, + 384.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 737.0, + 416.0, + 737.0, + 416.0, + 748.0, + 407.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 735.0, + 527.0, + 735.0, + 527.0, + 750.0, + 510.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 735.0, + 546.0, + 735.0, + 546.0, + 750.0, + 531.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 735.0, + 568.0, + 735.0, + 568.0, + 749.0, + 552.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 735.0, + 610.0, + 735.0, + 610.0, + 749.0, + 596.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 735.0, + 719.0, + 735.0, + 719.0, + 750.0, + 703.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 734.0, + 762.0, + 734.0, + 762.0, + 751.0, + 723.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 735.0, + 782.0, + 735.0, + 782.0, + 749.0, + 766.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 735.0, + 803.0, + 735.0, + 803.0, + 749.0, + 788.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 736.0, + 933.0, + 736.0, + 933.0, + 750.0, + 894.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 734.0, + 996.0, + 734.0, + 996.0, + 750.0, + 937.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 756.0, + 333.0, + 756.0, + 333.0, + 770.0, + 319.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 755.0, + 354.0, + 755.0, + 354.0, + 770.0, + 340.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 755.0, + 376.0, + 755.0, + 376.0, + 769.0, + 362.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 755.0, + 397.0, + 755.0, + 397.0, + 769.0, + 383.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 757.0, + 416.0, + 757.0, + 416.0, + 768.0, + 406.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 756.0, + 527.0, + 756.0, + 527.0, + 770.0, + 510.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 757.0, + 544.0, + 757.0, + 544.0, + 769.0, + 534.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 755.0, + 568.0, + 755.0, + 568.0, + 770.0, + 552.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 754.0, + 589.0, + 754.0, + 589.0, + 769.0, + 574.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 755.0, + 610.0, + 755.0, + 610.0, + 769.0, + 596.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 755.0, + 718.0, + 755.0, + 718.0, + 771.0, + 703.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 755.0, + 739.0, + 755.0, + 739.0, + 770.0, + 724.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 755.0, + 760.0, + 755.0, + 760.0, + 770.0, + 745.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 754.0, + 782.0, + 754.0, + 782.0, + 769.0, + 766.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 755.0, + 803.0, + 755.0, + 803.0, + 769.0, + 789.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 756.0, + 912.0, + 756.0, + 912.0, + 770.0, + 895.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 755.0, + 931.0, + 755.0, + 931.0, + 769.0, + 915.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 756.0, + 954.0, + 756.0, + 954.0, + 770.0, + 938.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 755.0, + 975.0, + 755.0, + 975.0, + 769.0, + 958.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 755.0, + 996.0, + 755.0, + 996.0, + 769.0, + 978.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 775.0, + 333.0, + 775.0, + 333.0, + 790.0, + 319.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 775.0, + 354.0, + 775.0, + 354.0, + 790.0, + 340.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 775.0, + 374.0, + 775.0, + 374.0, + 790.0, + 362.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 778.0, + 395.0, + 778.0, + 395.0, + 786.0, + 386.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 775.0, + 527.0, + 775.0, + 527.0, + 790.0, + 512.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 775.0, + 545.0, + 775.0, + 545.0, + 790.0, + 532.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 775.0, + 568.0, + 775.0, + 568.0, + 790.0, + 553.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 777.0, + 587.0, + 777.0, + 587.0, + 786.0, + 578.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 776.0, + 609.0, + 776.0, + 609.0, + 786.0, + 601.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 772.0, + 719.0, + 772.0, + 719.0, + 792.0, + 702.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 775.0, + 739.0, + 775.0, + 739.0, + 790.0, + 724.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 775.0, + 760.0, + 775.0, + 760.0, + 790.0, + 746.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 777.0, + 780.0, + 777.0, + 780.0, + 787.0, + 769.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 777.0, + 801.0, + 777.0, + 801.0, + 786.0, + 793.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 775.0, + 911.0, + 775.0, + 911.0, + 790.0, + 895.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 772.0, + 932.0, + 772.0, + 932.0, + 791.0, + 913.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 775.0, + 953.0, + 775.0, + 953.0, + 790.0, + 938.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 777.0, + 971.0, + 777.0, + 971.0, + 787.0, + 960.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 776.0, + 995.0, + 776.0, + 995.0, + 786.0, + 985.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 796.0, + 333.0, + 796.0, + 333.0, + 811.0, + 319.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 796.0, + 354.0, + 796.0, + 354.0, + 811.0, + 340.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 796.0, + 374.0, + 796.0, + 374.0, + 811.0, + 362.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 797.0, + 394.0, + 797.0, + 394.0, + 805.0, + 385.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 797.0, + 525.0, + 797.0, + 525.0, + 811.0, + 512.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 796.0, + 545.0, + 796.0, + 545.0, + 811.0, + 532.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 797.0, + 568.0, + 797.0, + 568.0, + 811.0, + 553.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 796.0, + 718.0, + 796.0, + 718.0, + 812.0, + 703.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 796.0, + 738.0, + 796.0, + 738.0, + 811.0, + 724.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 796.0, + 760.0, + 796.0, + 760.0, + 811.0, + 746.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 794.0, + 782.0, + 794.0, + 782.0, + 808.0, + 767.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 796.0, + 911.0, + 796.0, + 911.0, + 811.0, + 896.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 797.0, + 931.0, + 797.0, + 931.0, + 811.0, + 915.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 796.0, + 952.0, + 796.0, + 952.0, + 811.0, + 938.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 794.0, + 974.0, + 794.0, + 974.0, + 810.0, + 960.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 220.0, + 178.0, + 707.0, + 178.0, + 707.0, + 212.0, + 220.0, + 212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 963.0, + 474.0, + 963.0, + 474.0, + 1002.0, + 216.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 218.0, + 277.0, + 482.0, + 277.0, + 482.0, + 314.0, + 218.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 194.0, + 1868.0, + 679.0, + 1868.0, + 679.0, + 1906.0, + 194.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 197.0, + 1804.0, + 684.0, + 1804.0, + 684.0, + 1840.0, + 197.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 1937.0, + 722.0, + 1937.0, + 722.0, + 1968.0, + 196.0, + 1968.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 31, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 128, + 210, + 768, + 210, + 768, + 268, + 128, + 268 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 128, + 302, + 839, + 302, + 839, + 344, + 128, + 344 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 128, + 1092, + 794, + 1092, + 794, + 1127, + 128, + 1127 + ], + "score": 0.839 + }, + { + "category_id": 0, + "poly": [ + 131, + 778, + 614, + 778, + 614, + 811, + 131, + 811 + ], + "score": 0.824 + }, + { + "category_id": 1, + "poly": [ + 127, + 1188, + 388, + 1188, + 388, + 1217, + 127, + 1217 + ], + "score": 0.802 + }, + { + "category_id": 1, + "poly": [ + 122, + 1488, + 382, + 1488, + 382, + 1517, + 122, + 1517 + ], + "score": 0.701 + }, + { + "category_id": 3, + "poly": [ + 124, + 825, + 1461, + 825, + 1461, + 999, + 124, + 999 + ], + "score": 0.681 + }, + { + "category_id": 1, + "poly": [ + 123, + 1789, + 382, + 1789, + 382, + 1816, + 123, + 1816 + ], + "score": 0.658 + }, + { + "category_id": 3, + "poly": [ + 207, + 1832, + 1567, + 1832, + 1567, + 2009, + 207, + 2009 + ], + "score": 0.657 + }, + { + "category_id": 3, + "poly": [ + 195, + 1533, + 1535, + 1533, + 1535, + 1710, + 195, + 1710 + ], + "score": 0.621 + }, + { + "category_id": 3, + "poly": [ + 198, + 1235, + 1078, + 1235, + 1078, + 1411, + 198, + 1411 + ], + "score": 0.603 + }, + { + "category_id": 5, + "poly": [ + 123, + 584, + 314, + 584, + 314, + 758, + 123, + 758 + ], + "score": 0.508, + "html": "
abC
def
gh
• ·
" + }, + { + "category_id": 5, + "poly": [ + 701, + 582, + 896, + 582, + 896, + 757, + 701, + 757 + ], + "score": 0.481, + "html": "
48666
816121212
612999
612999
612999
" + }, + { + "category_id": 5, + "poly": [ + 198, + 1235, + 1078, + 1235, + 1078, + 1411, + 198, + 1411 + ], + "score": 0.406, + "html": "
edeedefedefeedef
ba bCba bCba H Cba b
eefed efed e fed e
hg hihg hhg hhg h
" + }, + { + "category_id": 5, + "poly": [ + 207, + 1832, + 1567, + 1832, + 1567, + 2009, + 207, + 2009 + ], + "score": 0.397, + "html": "
edefede fedefedefedefedef
babCbab Cbab Cbab Cba bCbab C
edeed eede fed efed efed ef
hghh ghhghhgh :hghhh
g
" + }, + { + "category_id": 5, + "poly": [ + 384, + 560, + 609, + 560, + 609, + 757, + 384, + 757 + ], + "score": 0.396, + "html": "
edef. • . •
babC= =
edef
hgh
= =
" + }, + { + "category_id": 1, + "poly": [ + 676, + 495, + 934, + 495, + 934, + 559, + 676, + 559 + ], + "score": 0.395 + }, + { + "category_id": 5, + "poly": [ + 195, + 1533, + 1535, + 1533, + 1535, + 1710, + 195, + 1710 + ], + "score": 0.376, + "html": "
ede fede fedefede fedefe def
bab CbabCbabCb abCbab CbabC
ede fedefedefed efedefedef
hghhghhghhg hhghhh
g
" + }, + { + "category_id": 6, + "poly": [ + 676, + 495, + 934, + 495, + 934, + 559, + 676, + 559 + ], + "score": 0.321 + }, + { + "category_id": 5, + "poly": [ + 124, + 825, + 1461, + 825, + 1461, + 999, + 124, + 999 + ], + "score": 0.312, + "html": "
a:b:C:d:e:f:g:h:
1 1 1 12 1112 2421 1i: 1
2 112 12 11 1211
111 111
1 112 1
sum=4sum =8sum=6sum =8sum =12sum = 16sum=6sum = 12sum=9
" + }, + { + "category_id": 6, + "poly": [ + 127, + 511, + 314, + 511, + 314, + 545, + 127, + 545 + ], + "score": 0.302 + }, + { + "category_id": 0, + "poly": [ + 127, + 511, + 314, + 511, + 314, + 545, + 127, + 545 + ], + "score": 0.262 + }, + { + "category_id": 3, + "poly": [ + 701, + 582, + 896, + 582, + 896, + 757, + 701, + 757 + ], + "score": 0.206 + }, + { + "category_id": 13, + "poly": [ + 696, + 580, + 897, + 580, + 897, + 758, + 696, + 758 + ], + "score": 0.89, + "latex": "{ \\left[ \\begin{array} { l l l l l l } { 4 } & { 8 } & { 6 } & { 6 } & { 6 } \\\\ { 8 } & { 1 6 } & { 1 2 } & { 1 2 } & { 1 2 } \\\\ { 6 } & { 1 2 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 1 2 } & { 9 } & { 9 } & { 9 } \\\\ { 6 } & { 1 2 } & { 9 } & { 9 } & { 9 } \\end{array} \\right] }" + }, + { + "category_id": 13, + "poly": [ + 891, + 1840, + 1075, + 1840, + 1075, + 2008, + 891, + 2008 + ], + "score": 0.64, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 891, + 1540, + 1074, + 1540, + 1074, + 1709, + 891, + 1709 + ], + "score": 0.61, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1117, + 1539, + 1276, + 1539, + 1276, + 1709, + 1117, + 1709 + ], + "score": 0.6, + "latex": "\\begin{array} { l } \\begin{array} { l l l l } { { \\Theta } } & { { \\bf { d } } } & { { \\Theta } } & { { \\bf { f } } } \\\\ { { \\mathrm { ~ b ~ } } } & { { \\left[ \\begin{array} { l l l l } { { { \\bf { a } } } } & { { { \\bf { b } } } } & { { { \\bf { c } } } } & { { \\nonumber } } \\\\ { { { \\bf { d } } } } & { { { \\bf { e } } } } & { { \\bf { f } } } & { { \\nonumber } } \\\\ { { { \\bf { h } } } } & { { { \\bf { h } } } } & { { \\bf { i } } } & { { \\nonumber } } \\end{array} } } \\en\\right.d{array} \\end{array} \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 381, + 560, + 603, + 560, + 603, + 759, + 381, + 759 + ], + "score": 0.57, + "latex": "{ \\begin{array} { l } { { \\begin{array} { l l l l l l l l } { { \\mathrm { e } } } & { { \\mathrm { ~ d } } } & { { \\mathrm { e } } } & { { \\mathrm { ~ f } } } & { \\dots } & { \\dots } \\\\ { { \\mathrm { b } } } & { { \\mathrm { ~ b } } } & { { \\mathrm { ~ c } } } & { \\dots } & { \\dots } \\\\ { { \\mathrm { e } } } & { { \\mathrm { e } } } & { { \\mathrm { ~ f } } } & { \\dots } & { \\dots } \\\\ { { \\mathrm { h } } } & { { \\mathrm { h } } } & { { \\mathrm { ~ i } } } & { \\dots } & { \\dots } \\\\ { \\dots } & { \\dots } & { \\dots } & { \\dots } & { \\dots } \\\\ { \\dots } & { \\dots } & { \\dots } & { \\dots } & { \\dots } \\end{array} } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 1343, + 1540, + 1538, + 1540, + 1538, + 1710, + 1343, + 1710 + ], + "score": 0.55, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1116, + 1840, + 1282, + 1840, + 1282, + 2006, + 1116, + 2006 + ], + "score": 0.47, + "latex": "\\begin{array} { l } { { \\Theta \\quad \\mathrm { { ~ d ~ } \\quad \\Theta ~ \\in ~ \\Omega ~ } \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathsf { b } \\quad \\left[ \\begin{array} { l l l l } { { \\mathsf { a } } } & { { \\mathsf { b } } } & { { \\mathsf { c } } } & { { \\mathsf { \\pi } } } \\\\ { { \\mathsf { d } } } & { { \\mathsf { e } } } & { { \\mathsf { f } } } & { { \\mathsf { \\pi } } } \\\\ { { \\mathsf { h } } } & { { \\mathsf { h } } } & { { \\mathsf { i } } } & { { \\mathsf { \\pi } } } \\end{array} \\right] } } \\\\ { { \\mathsf { h } \\quad \\left[ \\begin{array} { l l l l } { { \\mathsf { g } } } & { { \\mathsf { h } } } & { { \\mathsf { i } } } & { { \\mathsf { f } } } \\\\ { { \\mathsf { d } } } & { { \\mathsf { h } } } & { { \\mathsf { i } } } & { { \\mathsf { f } } } \\end{array} \\right] } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 643, + 960, + 677, + 960, + 677, + 983, + 643, + 983 + ], + "score": 0.46, + "latex": "= 8" + }, + { + "category_id": 13, + "poly": [ + 889, + 1242, + 1080, + 1242, + 1080, + 1410, + 889, + 1410 + ], + "score": 0.45, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } & { { { \\bf b } } } & { { { \\bf c } } } & { { } } \\\\ { { { \\bf e } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf h } } } & { { { \\bf h } } } & { { { \\bf i } } } \\end{array} } } \\\\ { { \\begin{array} { l } { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 661, + 1540, + 837, + 1540, + 837, + 1709, + 661, + 1709 + ], + "score": 0.45, + "latex": "\\begin{array} { l } { { \\begin{array} { l l l l } { { \\Theta } } & { { \\mathbb { d } } } & { { \\mathbb { e } } } & { { \\mathbb { f } } } \\\\ { { \\mathbf { b } } } \\\\ { { \\Theta } } \\end{array} } } \\\\ { { \\begin{array} { l } { { \\mathsf { e } } } \\\\ { { \\mathsf { h } } } \\\\ { { \\mathsf { h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1336, + 857, + 1455, + 857, + 1455, + 953, + 1336, + 953 + ], + "score": 0.43, + "latex": "\\begin{array} { c c c } { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 435, + 1839, + 620, + 1839, + 620, + 2008, + 435, + 2008 + ], + "score": 0.43, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 662, + 1839, + 849, + 1839, + 849, + 2007, + 662, + 2007 + ], + "score": 0.41, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 865, + 831, + 1000, + 831, + 1000, + 957, + 865, + 957 + ], + "score": 0.4, + "latex": "\\begin{array} { c c c c } { \\mathbf { f } . } & & & & \\\\ { 4 } & { 2 } & { 2 } & { } \\\\ { 2 } & { 1 } & { 1 } & { } \\\\ { 2 } & { 1 } & { 1 } & { } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1186, + 855, + 1305, + 855, + 1305, + 956, + 1186, + 956 + ], + "score": 0.4, + "latex": "\\begin{array} { c c c } { { 2 } } & { { 1 } } & { { 1 } } \\\\ { { 2 } } & { { 1 } } & { { 1 } } \\\\ { { 2 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 282, + 856, + 393, + 856, + 393, + 922, + 282, + 922 + ], + "score": 0.31, + "latex": "\\begin{array} { c c c } { { 2 } } & { { 1 } } & { { 1 } } \\\\ { { 2 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1401, + 961, + 1434, + 961, + 1434, + 983, + 1401, + 983 + ], + "score": 0.3, + "latex": "{ \\ o } = 9" + }, + { + "category_id": 13, + "poly": [ + 199, + 1242, + 386, + 1242, + 386, + 1410, + 199, + 1410 + ], + "score": 0.28, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } & { { { \\bf b } } } & { { { \\bf c } } } & { { } } \\\\ { { { \\bf e } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf h } } } & { { { \\bf h } } } & { { { \\bf i } } } \\end{array} } } \\\\ { { \\begin{array} { r } { { { \\bf h } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 789, + 960, + 833, + 960, + 833, + 983, + 789, + 983 + ], + "score": 0.28, + "latex": "= 1 2" + }, + { + "category_id": 13, + "poly": [ + 733, + 852, + 851, + 852, + 851, + 966, + 733, + 966 + ], + "score": 0.27, + "latex": "\\begin{array} { c c c } { { 2 } } & { { 2 } } & { { 2 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\\\ { { 1 } } & { { 1 } } & { { 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 207, + 1838, + 398, + 1838, + 398, + 2008, + 207, + 2008 + ], + "score": 0.27, + "latex": "\\begin{array} { l } \\begin{array} { l l l l } { { \\Theta } } & { { \\mathbb { d } } } & { { \\Theta } } & { { \\mathbb { f } } } \\\\ { { \\mathsf { b } } } & { { \\left[ \\begin{array} { l l l l } { { a } } & { { \\mathsf { b } } } & { { \\mathbb { c } } } & { { } } \\\\ { { \\mathsf { d } } } & { { \\mathsf { e } } } & { { \\mathsf { f } } } & { { \\mathsf { \\Omega } } } \\\\ { { \\mathsf { h } } } & { { \\mathsf { h } } } & { { \\mathsf { i } } } & { { \\mathsf { \\Omega } } } \\end{array} } } \\\\ \\right.{ { \\begin{array} { l } { { \\mathsf { h } } } \\end{array} } { \\left\\{ \\begin{array} { l l } { { \\mathbb { d } } } & { { \\mathsf { h } } } \\end{array} \\right. } } \\end{array} \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 120.0, + 197.0, + 773.0, + 197.0, + 773.0, + 280.0, + 120.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 129.0, + 1093.0, + 793.0, + 1093.0, + 793.0, + 1128.0, + 129.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 127.0, + 773.0, + 617.0, + 773.0, + 617.0, + 817.0, + 127.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 132.0, + 829.0, + 161.0, + 829.0, + 161.0, + 852.0, + 132.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 825.0, + 315.0, + 825.0, + 315.0, + 853.0, + 282.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 828.0, + 465.0, + 828.0, + 465.0, + 852.0, + 438.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 825.0, + 617.0, + 825.0, + 617.0, + 852.0, + 584.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 828.0, + 767.0, + 828.0, + 767.0, + 852.0, + 739.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 825.0, + 915.0, + 825.0, + 915.0, + 851.0, + 891.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 829.0, + 1069.0, + 829.0, + 1069.0, + 853.0, + 1041.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 825.0, + 1224.0, + 825.0, + 1224.0, + 852.0, + 1189.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 826.0, + 1370.0, + 826.0, + 1370.0, + 851.0, + 1346.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 862.0, + 155.0, + 862.0, + 155.0, + 881.0, + 137.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 174.0, + 862.0, + 192.0, + 862.0, + 192.0, + 882.0, + 174.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 863.0, + 454.0, + 863.0, + 454.0, + 879.0, + 442.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 862.0, + 496.0, + 862.0, + 496.0, + 881.0, + 477.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 862.0, + 610.0, + 862.0, + 610.0, + 879.0, + 590.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 862.0, + 647.0, + 862.0, + 647.0, + 881.0, + 629.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 863.0, + 1060.0, + 863.0, + 1060.0, + 876.0, + 1048.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 865.0, + 1099.0, + 865.0, + 1099.0, + 875.0, + 1086.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 862.0, + 1138.0, + 862.0, + 1138.0, + 881.0, + 1122.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 135.0, + 892.0, + 158.0, + 892.0, + 158.0, + 916.0, + 135.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 173.0, + 892.0, + 195.0, + 892.0, + 195.0, + 918.0, + 173.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 895.0, + 459.0, + 895.0, + 459.0, + 915.0, + 439.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 895.0, + 496.0, + 895.0, + 496.0, + 915.0, + 477.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 895.0, + 534.0, + 895.0, + 534.0, + 913.0, + 516.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 893.0, + 613.0, + 893.0, + 613.0, + 916.0, + 589.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 896.0, + 646.0, + 896.0, + 646.0, + 913.0, + 628.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 896.0, + 1063.0, + 896.0, + 1063.0, + 913.0, + 1045.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 898.0, + 1099.0, + 898.0, + 1099.0, + 911.0, + 1084.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 895.0, + 1138.0, + 895.0, + 1138.0, + 915.0, + 1120.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 928.0, + 613.0, + 928.0, + 613.0, + 951.0, + 589.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 928.0, + 647.0, + 928.0, + 647.0, + 948.0, + 628.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 142.0, + 958.0, + 227.0, + 958.0, + 227.0, + 983.0, + 142.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 955.0, + 381.0, + 955.0, + 381.0, + 988.0, + 294.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 953.0, + 534.0, + 953.0, + 534.0, + 986.0, + 445.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 953.0, + 642.0, + 953.0, + 642.0, + 986.0, + 595.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 953.0, + 683.0, + 953.0, + 683.0, + 986.0, + 678.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 958.0, + 788.0, + 958.0, + 788.0, + 983.0, + 743.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 958.0, + 838.0, + 958.0, + 838.0, + 983.0, + 834.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 958.0, + 990.0, + 958.0, + 990.0, + 986.0, + 896.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 953.0, + 1137.0, + 953.0, + 1137.0, + 986.0, + 1050.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 958.0, + 1292.0, + 958.0, + 1292.0, + 983.0, + 1198.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 955.0, + 1400.0, + 955.0, + 1400.0, + 989.0, + 1354.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1435.0, + 955.0, + 1441.0, + 955.0, + 1441.0, + 989.0, + 1435.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1844.0, + 1369.0, + 1844.0, + 1369.0, + 1869.0, + 1345.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 1843.0, + 1409.0, + 1843.0, + 1409.0, + 1867.0, + 1383.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1419.0, + 1844.0, + 1447.0, + 1844.0, + 1447.0, + 1867.0, + 1419.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1462.0, + 1843.0, + 1479.0, + 1843.0, + 1479.0, + 1861.0, + 1462.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1874.0, + 1369.0, + 1874.0, + 1369.0, + 1903.0, + 1345.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 1877.0, + 1409.0, + 1877.0, + 1409.0, + 1903.0, + 1383.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1422.0, + 1876.0, + 1445.0, + 1876.0, + 1445.0, + 1902.0, + 1422.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1462.0, + 1880.0, + 1482.0, + 1880.0, + 1482.0, + 1900.0, + 1462.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 1910.0, + 1369.0, + 1910.0, + 1369.0, + 1936.0, + 1346.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 1909.0, + 1409.0, + 1909.0, + 1409.0, + 1936.0, + 1383.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1421.0, + 1910.0, + 1445.0, + 1910.0, + 1445.0, + 1936.0, + 1421.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1466.0, + 1910.0, + 1483.0, + 1910.0, + 1483.0, + 1931.0, + 1466.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1942.0, + 1371.0, + 1942.0, + 1371.0, + 1968.0, + 1345.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 1944.0, + 1407.0, + 1944.0, + 1407.0, + 1971.0, + 1383.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1421.0, + 1942.0, + 1445.0, + 1942.0, + 1445.0, + 1967.0, + 1421.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1465.0, + 1946.0, + 1477.0, + 1946.0, + 1477.0, + 1962.0, + 1465.0, + 1962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 211.0, + 1545.0, + 234.0, + 1545.0, + 234.0, + 1570.0, + 211.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1542.0, + 271.0, + 1542.0, + 271.0, + 1568.0, + 248.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1545.0, + 310.0, + 1545.0, + 310.0, + 1570.0, + 284.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1544.0, + 345.0, + 1544.0, + 345.0, + 1562.0, + 326.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1545.0, + 461.0, + 1545.0, + 461.0, + 1568.0, + 437.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1544.0, + 499.0, + 1544.0, + 499.0, + 1568.0, + 475.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1545.0, + 536.0, + 1545.0, + 536.0, + 1568.0, + 511.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1544.0, + 571.0, + 1544.0, + 571.0, + 1564.0, + 555.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 211.0, + 1577.0, + 234.0, + 1577.0, + 234.0, + 1601.0, + 211.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1578.0, + 272.0, + 1578.0, + 272.0, + 1603.0, + 248.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1577.0, + 310.0, + 1577.0, + 310.0, + 1601.0, + 286.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1578.0, + 347.0, + 1578.0, + 347.0, + 1604.0, + 325.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1577.0, + 460.0, + 1577.0, + 460.0, + 1603.0, + 436.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1578.0, + 499.0, + 1578.0, + 499.0, + 1603.0, + 475.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1575.0, + 536.0, + 1575.0, + 536.0, + 1601.0, + 513.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1578.0, + 574.0, + 1578.0, + 574.0, + 1604.0, + 550.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 209.0, + 1611.0, + 234.0, + 1611.0, + 234.0, + 1637.0, + 209.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1610.0, + 272.0, + 1610.0, + 272.0, + 1636.0, + 248.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1611.0, + 310.0, + 1611.0, + 310.0, + 1637.0, + 286.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1611.0, + 347.0, + 1611.0, + 347.0, + 1633.0, + 329.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1611.0, + 460.0, + 1611.0, + 460.0, + 1637.0, + 436.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1610.0, + 499.0, + 1610.0, + 499.0, + 1636.0, + 477.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1613.0, + 536.0, + 1613.0, + 536.0, + 1636.0, + 513.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1611.0, + 574.0, + 1611.0, + 574.0, + 1633.0, + 556.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 209.0, + 1643.0, + 234.0, + 1643.0, + 234.0, + 1669.0, + 209.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1646.0, + 271.0, + 1646.0, + 271.0, + 1672.0, + 248.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1643.0, + 312.0, + 1643.0, + 312.0, + 1671.0, + 283.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1647.0, + 340.0, + 1647.0, + 340.0, + 1663.0, + 328.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1643.0, + 460.0, + 1643.0, + 460.0, + 1669.0, + 436.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1646.0, + 499.0, + 1646.0, + 499.0, + 1672.0, + 477.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1645.0, + 536.0, + 1645.0, + 536.0, + 1668.0, + 513.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1647.0, + 568.0, + 1647.0, + 568.0, + 1663.0, + 556.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1243.0, + 540.0, + 1243.0, + 540.0, + 1271.0, + 433.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1244.0, + 572.0, + 1244.0, + 572.0, + 1269.0, + 551.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1245.0, + 688.0, + 1245.0, + 688.0, + 1273.0, + 665.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1243.0, + 727.0, + 1243.0, + 727.0, + 1271.0, + 702.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1244.0, + 766.0, + 1244.0, + 766.0, + 1270.0, + 739.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1245.0, + 798.0, + 1245.0, + 798.0, + 1266.0, + 781.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1279.0, + 458.0, + 1279.0, + 458.0, + 1304.0, + 438.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1281.0, + 498.0, + 1281.0, + 498.0, + 1303.0, + 476.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1279.0, + 535.0, + 1279.0, + 535.0, + 1303.0, + 514.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1280.0, + 573.0, + 1280.0, + 573.0, + 1306.0, + 552.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1279.0, + 686.0, + 1279.0, + 686.0, + 1304.0, + 666.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1280.0, + 725.0, + 1280.0, + 725.0, + 1304.0, + 702.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1281.0, + 760.0, + 1281.0, + 760.0, + 1300.0, + 743.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1280.0, + 800.0, + 1280.0, + 800.0, + 1304.0, + 780.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1310.0, + 461.0, + 1310.0, + 461.0, + 1342.0, + 440.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1312.0, + 497.0, + 1312.0, + 497.0, + 1336.0, + 477.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1313.0, + 535.0, + 1313.0, + 535.0, + 1336.0, + 514.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1310.0, + 574.0, + 1310.0, + 574.0, + 1336.0, + 555.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1313.0, + 687.0, + 1313.0, + 687.0, + 1337.0, + 666.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1310.0, + 726.0, + 1310.0, + 726.0, + 1338.0, + 702.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1315.0, + 760.0, + 1315.0, + 760.0, + 1336.0, + 743.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1312.0, + 801.0, + 1312.0, + 801.0, + 1332.0, + 785.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1345.0, + 459.0, + 1345.0, + 459.0, + 1369.0, + 438.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1347.0, + 496.0, + 1347.0, + 496.0, + 1372.0, + 476.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1345.0, + 536.0, + 1345.0, + 536.0, + 1369.0, + 514.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1345.0, + 570.0, + 1345.0, + 570.0, + 1367.0, + 555.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1345.0, + 686.0, + 1345.0, + 686.0, + 1370.0, + 666.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1346.0, + 726.0, + 1346.0, + 726.0, + 1375.0, + 702.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1346.0, + 762.0, + 1346.0, + 762.0, + 1368.0, + 741.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 492.0, + 935.0, + 492.0, + 935.0, + 530.0, + 674.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 525.0, + 935.0, + 525.0, + 935.0, + 560.0, + 676.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 509.0, + 317.0, + 509.0, + 317.0, + 549.0, + 126.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 509.0, + 317.0, + 509.0, + 317.0, + 549.0, + 126.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 125.0, + 295.0, + 839.0, + 295.0, + 839.0, + 352.0, + 125.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 126.0, + 1183.0, + 390.0, + 1183.0, + 390.0, + 1222.0, + 126.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 120.0, + 1483.0, + 384.0, + 1483.0, + 384.0, + 1522.0, + 120.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 121.0, + 1783.0, + 383.0, + 1783.0, + 383.0, + 1822.0, + 121.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 492.0, + 935.0, + 492.0, + 935.0, + 530.0, + 674.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 525.0, + 935.0, + 525.0, + 935.0, + 560.0, + 676.0, + 560.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 32, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 220, + 1112, + 1494, + 1112, + 1494, + 1498, + 220, + 1498 + ], + "score": 0.915 + }, + { + "category_id": 3, + "poly": [ + 213, + 598, + 1510, + 598, + 1510, + 986, + 213, + 986 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 144, + 405, + 628, + 405, + 628, + 431, + 144, + 431 + ], + "score": 0.843 + }, + { + "category_id": 1, + "poly": [ + 134, + 1573, + 617, + 1573, + 617, + 1602, + 134, + 1602 + ], + "score": 0.816 + }, + { + "category_id": 1, + "poly": [ + 130, + 1638, + 609, + 1638, + 609, + 1668, + 130, + 1668 + ], + "score": 0.749 + }, + { + "category_id": 1, + "poly": [ + 129, + 1703, + 653, + 1703, + 653, + 1733, + 129, + 1733 + ], + "score": 0.715 + }, + { + "category_id": 1, + "poly": [ + 143, + 524, + 402, + 524, + 402, + 551, + 143, + 551 + ], + "score": 0.549 + }, + { + "category_id": 1, + "poly": [ + 148, + 1058, + 403, + 1058, + 403, + 1086, + 148, + 1086 + ], + "score": 0.43 + }, + { + "category_id": 4, + "poly": [ + 148, + 1058, + 403, + 1058, + 403, + 1086, + 148, + 1086 + ], + "score": 0.222 + }, + { + "category_id": 4, + "poly": [ + 143, + 524, + 402, + 524, + 402, + 551, + 143, + 551 + ], + "score": 0.132 + }, + { + "category_id": 13, + "poly": [ + 447, + 812, + 627, + 812, + 627, + 984, + 447, + 984 + ], + "score": 0.71, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 671, + 812, + 859, + 812, + 859, + 984, + 671, + 984 + ], + "score": 0.69, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1118, + 1121, + 1269, + 1121, + 1269, + 1292, + 1118, + 1292 + ], + "score": 0.65, + "latex": "{ \\begin{array} { r l } { \\mathbf { e } } & { { \\mathrm { ~ d ~ } } \\mathbf { e } \\quad \\mathbf { f } } \\\\ { \\mathbf { b } } & { { \\mathrm { ~ { \\left[ ~ a ~ b ~ \\right] } ~ } } \\mathbf { e } } \\\\ { \\mathbf { e } } & { { \\mathrm { ~ { \\left[ ~ d ~ \\right] } ~ } } \\mathbf { e } \\quad \\mathbf { f } } \\\\ { \\mathbf { h } } & { { \\mathrm { ~ { \\left[ ~ { \\begin{array} { l l l l } { \\mathbf { g } } & { { \\mathrm { ~ h ~ } } } & { { \\mathrm { ~ i ~ } } } \\end{array} } \\right] } } } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 669, + 1121, + 843, + 1121, + 843, + 1294, + 669, + 1294 + ], + "score": 0.63, + "latex": "\\begin{array} { r l } & \\begin{array} { l l l l } { \\Theta } & { \\mathrm { ~ d ~ } } & { \\Theta } & { \\mathrm { ~ f ~ } } \\\\ { \\mathsf { b } } & { \\left[ \\begin{array} { l l l l } { \\mathsf { a } } & { \\mathsf { b } } & { \\mathsf { c } } & { } \\\\ { \\mathsf { d } } & { \\mathsf { e } } & { \\mathrm { ~ f ~ } } \\\\ { \\mathsf { h } } & { \\mathsf { h } } & { \\mathrm { i ~ } } \\end{array} } \\end{\\right.array} \\end{array} \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 670, + 606, + 825, + 606, + 825, + 781, + 670, + 781 + ], + "score": 0.57, + "latex": "\\begin{array} { r l } & { \\texttt { e } \\texttt { d } \\texttt { e } \\texttt { f } } \\\\ & { \\texttt { b } \\left[ \\texttt { a } \\texttt { b } \\texttt { c } \\right. } \\\\ & { \\texttt { e } } \\\\ & { \\texttt { h } \\left| \\begin{array} { l l l l } { \\texttt { d } } & { \\texttt { e } } & { \\texttt { f } } \\\\ { \\texttt { g } } & { \\texttt { h } } & { \\texttt { i } } \\end{array} \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1344, + 608, + 1532, + 608, + 1532, + 780, + 1344, + 780 + ], + "score": 0.54, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf { d } } } } & { { { \\bf { e } } } } & { { { \\bf { f } } } } \\\\ { { { \\bf { b } } } } \\\\ { { { \\bf { e } } } } \\\\ { { { \\bf { h } } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1118, + 608, + 1269, + 608, + 1269, + 778, + 1118, + 778 + ], + "score": 0.53, + "latex": "{ \\begin{array} { r l } { \\mathbf { e } } & { { \\mathrm { ~ d ~ } } \\mathbf { e } \\quad \\mathbf { f } } \\\\ { \\mathbf { b } } & { { \\mathrm { ~ } } { \\mathrm { ~ } } { \\mathrm { ~ } } } \\\\ { \\mathbf { e } } & { { \\mathrm { ~ } } { \\mathrm { ~ } } } \\\\ { \\mathbf { h } } & { { \\mathrm { ~ } } { \\mathrm { ~ } } { \\mathrm { ~ } } { \\mathrm { ~ } { ~ } } { \\mathrm { ~ } } { \\mathrm { ~ } { ~ } } } \\\\ { \\mathbf { b } } & { { \\mathrm { ~ } } { \\mathrm { ~ } } \\mathbf { h } \\quad \\mathbf { i } } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 444, + 1121, + 623, + 1121, + 623, + 1293, + 444, + 1293 + ], + "score": 0.52, + "latex": "\\begin{array} { r l } & { \\begin{array} { l l l l } { \\Theta } & { \\mathrm { ~ d ~ } } & { \\Theta } & { \\mathrm { ~ f ~ } } \\\\ { \\mathsf { b } } & { \\left[ \\begin{array} { l l l l } { \\mathrm { ~ a } } & { \\mathsf { b } } & { \\mathrm { ~ c } } & { } \\\\ { \\mathrm { ~ d } } & { \\mathsf { e } } & { \\mathrm { ~ f ~ } } \\\\ { \\mathsf { \\Theta } } & { \\mathsf { h } } & { \\mathrm { ~ i ~ } } \\end{array} \\right] } \\end{array} } \\\\ & { \\begin{array} { r l } & { \\mathsf { h } } \\\\ & { \\mathsf { h } } \\end{array} \\left[ \\begin{array} { l l l l } { \\mathsf { g } } & { \\mathsf { h } } & { \\mathrm { ~ i ~ } } \\\\ { \\mathsf { e } } & { \\mathsf { h } } & { \\mathrm { ~ c } } \\end{array} \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 221, + 812, + 406, + 812, + 406, + 985, + 221, + 985 + ], + "score": 0.52, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 220, + 607, + 408, + 607, + 408, + 779, + 220, + 779 + ], + "score": 0.44, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\mathbf { e } } } & { { \\mathbf { d } } } & { { \\mathbf { e } } } & { { \\mathbf { f } } } \\\\ { { \\mathbf { b } } } \\\\ { { \\mathbf { e } } } \\\\ { { \\mathbf { h } } } \\\\ { { \\mathbf { n } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 445, + 607, + 612, + 607, + 612, + 779, + 445, + 779 + ], + "score": 0.43, + "latex": "\\begin{array} { l } \\begin{array} { l l l l } { { \\Theta } } & { { \\mathrm { { ~ d ~ } } } } & { { \\Theta } } & { { \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathrm { { ~ b ~ } } } } & { { { \\left[ \\begin{array} { l l l l } { { a } } & { { \\mathrm { { ~ b ~ } } } } & { { \\mathrm { { ~ c ~ } } } } & { { \\mathrm { { ~ \\pi ~ } } } } \\\\ { { \\mathrm { { ~ d ~ } } } } & { { \\mathrm { { ~ e ~ } } } } & { { \\mathrm { { ~ f ~ } } } } & { { \\mathrm { { ~ \\pi ~ } } } } \\end{array} \\right] } } \\\\ { { \\mathrm { { ~ h ~ } } } } & { { { \\mathrm { ~ h ~ } } } } & { { { \\mathrm { ~ i ~ } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 220, + 1121, + 404, + 1121, + 404, + 1293, + 220, + 1293 + ], + "score": 0.43, + "latex": "\\begin{array} { l } { { \\begin{array} { l l l l } { { \\Theta } } & { { \\mathrm { { ~ d ~ } } } } & { { \\mathrm { { ~ e ~ } } } } & { { \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathrm { { ~ b ~ } } } } & { { \\mathrm { { ~ a ~ b ~ } } } } & { { \\mathrm { { ~ c ~ } } } } \\\\ { { \\Theta } } & { { \\mathrm { { ~ e ~ } } } } & { { \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathrm { { ~ h ~ } } } } & { { \\mathrm { { ~ h ~ } } } } & { { \\mathrm { { ~ i ~ } } } } \\end{array} } } \\\\ { { \\begin{array} { l } { { \\eta } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 221, + 1326, + 402, + 1326, + 402, + 1497, + 221, + 1497 + ], + "score": 0.4, + "latex": "\\begin{array} { l } { { \\begin{array} { l l l l } { { \\Theta } } & { { \\mathrm { { ~ d ~ } } } } & { { \\mathrm { { ~ e ~ } } } } & { { \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathrm { { ~ b ~ } } } } & { { \\mathrm { { ~ a ~ b ~ } } } } & { { \\mathrm { { ~ c ~ } } } } \\\\ { { \\Theta } } & { { \\mathrm { { ~ e ~ } } } } & { { \\mathrm { { ~ f ~ } } } } \\\\ { { \\mathrm { { ~ h ~ } } } } & { { \\mathrm { { ~ h ~ } } } } & { { \\mathrm { { ~ i ~ } } } } \\end{array} } } \\\\ { { \\begin{array} { l } { { \\eta } } \\end{array} } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 894, + 607, + 1068, + 607, + 1068, + 778, + 894, + 778 + ], + "score": 0.38, + "latex": "\\begin{array} { r l } & { \\textsf { e ~ d ~ } \\textsf { e ~ f } } \\\\ & { \\textsf { b } \\lceil \\begin{array} { l l l } { \\ a } & { \\ b } & { \\ c } \\\\ { \\ a } & { \\ e } & { \\ \\textsf { f } } \\\\ { \\ P } & { \\ h } & { \\ i } \\end{array} } \\\\ & \\textsf { h } \\lceil \\begin{array} { l } { \\ q } \\\\ { \\ q } \\end{array} \\rceil \\ q \\vdash \\{ \\begin{array} { l } { \\sum _ { i } \\sum _ { j = 1 } ^ { m _ { i } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { j = 1 } ^ { m _ { i } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { i } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { i } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { i } } \\sum _ { j = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } \\sum _ { i = 1 } ^ { m _ { j } } } \\end{array} \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1298, + 1120, + 1491, + 1120, + 1491, + 1293, + 1298, + 1293 + ], + "score": 0.31, + "latex": "\\begin{array} { l } { { \\begin{array} { c c c c } { { \\Theta } } & { { { \\bf d } } } & { { { \\bf e } } } & { { { \\bf f } } } \\\\ { { { \\bf b } } } \\\\ { { { \\bf e } } } \\\\ { { { \\bf h } } } \\end{array} } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1126.0, + 920.0, + 1126.0, + 920.0, + 1149.0, + 897.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1124.0, + 957.0, + 1124.0, + 957.0, + 1148.0, + 934.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1126.0, + 994.0, + 1126.0, + 994.0, + 1149.0, + 971.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1124.0, + 1033.0, + 1124.0, + 1033.0, + 1143.0, + 1017.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1157.0, + 919.0, + 1157.0, + 919.0, + 1184.0, + 897.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1160.0, + 959.0, + 1160.0, + 959.0, + 1184.0, + 934.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1157.0, + 994.0, + 1157.0, + 994.0, + 1182.0, + 973.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1160.0, + 1032.0, + 1160.0, + 1032.0, + 1184.0, + 1010.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1195.0, + 920.0, + 1195.0, + 920.0, + 1217.0, + 897.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1192.0, + 957.0, + 1192.0, + 957.0, + 1217.0, + 934.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1196.0, + 993.0, + 1196.0, + 993.0, + 1215.0, + 974.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 1193.0, + 1033.0, + 1193.0, + 1033.0, + 1212.0, + 1016.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1225.0, + 920.0, + 1225.0, + 920.0, + 1250.0, + 899.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1229.0, + 957.0, + 1229.0, + 957.0, + 1254.0, + 934.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1226.0, + 994.0, + 1226.0, + 994.0, + 1250.0, + 971.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1230.0, + 1024.0, + 1230.0, + 1024.0, + 1243.0, + 1014.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1331.0, + 472.0, + 1331.0, + 472.0, + 1355.0, + 449.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1330.0, + 511.0, + 1330.0, + 511.0, + 1353.0, + 486.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1331.0, + 548.0, + 1331.0, + 548.0, + 1353.0, + 522.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1330.0, + 581.0, + 1330.0, + 581.0, + 1348.0, + 562.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1331.0, + 697.0, + 1331.0, + 697.0, + 1353.0, + 674.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1330.0, + 735.0, + 1330.0, + 735.0, + 1353.0, + 710.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1331.0, + 773.0, + 1331.0, + 773.0, + 1353.0, + 747.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1330.0, + 804.0, + 1330.0, + 804.0, + 1349.0, + 785.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1362.0, + 471.0, + 1362.0, + 471.0, + 1389.0, + 449.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1364.0, + 509.0, + 1364.0, + 509.0, + 1391.0, + 486.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1363.0, + 546.0, + 1363.0, + 546.0, + 1388.0, + 524.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1367.0, + 582.0, + 1367.0, + 582.0, + 1386.0, + 564.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1363.0, + 695.0, + 1363.0, + 695.0, + 1389.0, + 674.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1366.0, + 735.0, + 1366.0, + 735.0, + 1389.0, + 711.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1363.0, + 771.0, + 1363.0, + 771.0, + 1388.0, + 748.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1364.0, + 808.0, + 1364.0, + 808.0, + 1388.0, + 784.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1399.0, + 472.0, + 1399.0, + 472.0, + 1421.0, + 449.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1396.0, + 511.0, + 1396.0, + 511.0, + 1422.0, + 486.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1399.0, + 546.0, + 1399.0, + 546.0, + 1421.0, + 524.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1402.0, + 582.0, + 1402.0, + 582.0, + 1413.0, + 569.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1399.0, + 695.0, + 1399.0, + 695.0, + 1422.0, + 674.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1396.0, + 735.0, + 1396.0, + 735.0, + 1421.0, + 711.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1400.0, + 768.0, + 1400.0, + 768.0, + 1420.0, + 750.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1398.0, + 808.0, + 1398.0, + 808.0, + 1418.0, + 793.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1432.0, + 471.0, + 1432.0, + 471.0, + 1455.0, + 449.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1435.0, + 508.0, + 1435.0, + 508.0, + 1460.0, + 486.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1432.0, + 546.0, + 1432.0, + 546.0, + 1455.0, + 524.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1432.0, + 695.0, + 1432.0, + 695.0, + 1457.0, + 672.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1435.0, + 734.0, + 1435.0, + 734.0, + 1460.0, + 711.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1432.0, + 770.0, + 1432.0, + 770.0, + 1457.0, + 748.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 141.0, + 400.0, + 631.0, + 400.0, + 631.0, + 436.0, + 141.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 146.0, + 1053.0, + 405.0, + 1053.0, + 405.0, + 1092.0, + 146.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 141.0, + 519.0, + 404.0, + 519.0, + 404.0, + 556.0, + 141.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 130.0, + 1569.0, + 619.0, + 1569.0, + 619.0, + 1607.0, + 130.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 127.0, + 1635.0, + 610.0, + 1635.0, + 610.0, + 1671.0, + 127.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 128.0, + 1703.0, + 653.0, + 1703.0, + 653.0, + 1734.0, + 128.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 141.0, + 519.0, + 404.0, + 519.0, + 404.0, + 556.0, + 141.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 146.0, + 1053.0, + 405.0, + 1053.0, + 405.0, + 1092.0, + 146.0, + 1092.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 33, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 165, + 656, + 1416, + 656, + 1416, + 853, + 165, + 853 + ], + "score": 0.845 + }, + { + "category_id": 0, + "poly": [ + 130, + 212, + 1046, + 212, + 1046, + 267, + 130, + 267 + ], + "score": 0.822 + }, + { + "category_id": 3, + "poly": [ + 188, + 1730, + 1525, + 1730, + 1525, + 1956, + 188, + 1956 + ], + "score": 0.798 + }, + { + "category_id": 3, + "poly": [ + 168, + 1366, + 1490, + 1366, + 1490, + 1625, + 168, + 1625 + ], + "score": 0.626 + }, + { + "category_id": 3, + "poly": [ + 172, + 969, + 1066, + 969, + 1066, + 1268, + 172, + 1268 + ], + "score": 0.604 + }, + { + "category_id": 4, + "poly": [ + 135, + 907, + 802, + 907, + 802, + 942, + 135, + 942 + ], + "score": 0.586 + }, + { + "category_id": 0, + "poly": [ + 139, + 609, + 623, + 609, + 623, + 643, + 139, + 643 + ], + "score": 0.449 + }, + { + "category_id": 1, + "poly": [ + 171, + 958, + 433, + 958, + 433, + 988, + 171, + 988 + ], + "score": 0.335 + }, + { + "category_id": 0, + "poly": [ + 135, + 907, + 802, + 907, + 802, + 942, + 135, + 942 + ], + "score": 0.298 + }, + { + "category_id": 0, + "poly": [ + 409, + 308, + 733, + 308, + 733, + 374, + 409, + 374 + ], + "score": 0.297 + }, + { + "category_id": 4, + "poly": [ + 167, + 1318, + 429, + 1318, + 429, + 1347, + 167, + 1347 + ], + "score": 0.229 + }, + { + "category_id": 4, + "poly": [ + 168, + 1681, + 428, + 1681, + 428, + 1711, + 168, + 1711 + ], + "score": 0.221 + }, + { + "category_id": 0, + "poly": [ + 229, + 320, + 305, + 320, + 305, + 353, + 229, + 353 + ], + "score": 0.188 + }, + { + "category_id": 5, + "poly": [ + 469, + 395, + 691, + 395, + 691, + 586, + 469, + 586 + ], + "score": 0.142, + "html": "
6.258.757.57.57.5
8.75 12.2510.5 10.510.5
7.510.5999
7.510.5999
7.510.5999
" + }, + { + "category_id": 1, + "poly": [ + 409, + 308, + 733, + 308, + 733, + 374, + 409, + 374 + ], + "score": 0.117 + }, + { + "category_id": 5, + "poly": [ + 174, + 396, + 358, + 396, + 358, + 587, + 174, + 587 + ], + "score": 0.101, + "html": "
abC
def
gh
" + }, + { + "category_id": 4, + "poly": [ + 139, + 609, + 623, + 609, + 623, + 643, + 139, + 643 + ], + "score": 0.097 + }, + { + "category_id": 13, + "poly": [ + 461, + 393, + 693, + 393, + 693, + 584, + 461, + 584 + ], + "score": 0.42, + "latex": "\\begin{array} { c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c } { { 6 . 2 5 } } & { { 8 . 7 5 } } & { { 7 . 5 } } & { { 7 . 5 } } & { { 7 . 5 } } & { { 7 . 5 } } & { { 7 . 5 } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { { } } & { } { } & { } { } & { } & { { } } & { { } } & { } & { { } } & { { } } & { } & { { } } & { } & { { } } & { } & { } { } & { } & \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1926, + 354, + 1926, + 354, + 1948, + 298, + 1948 + ], + "score": 0.34, + "latex": "9 / 6 =" + }, + { + "category_id": 13, + "poly": [ + 309, + 1566, + 354, + 1566, + 354, + 1587, + 309, + 1587 + ], + "score": 0.3, + "latex": "1 / 4 =" + }, + { + "category_id": 13, + "poly": [ + 928, + 808, + 941, + 808, + 941, + 822, + 928, + 822 + ], + "score": 0.26, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 185.0, + 662.0, + 209.0, + 662.0, + 209.0, + 682.0, + 185.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 662.0, + 350.0, + 662.0, + 350.0, + 682.0, + 326.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 662.0, + 491.0, + 662.0, + 491.0, + 683.0, + 467.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 660.0, + 631.0, + 660.0, + 631.0, + 682.0, + 606.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 663.0, + 769.0, + 663.0, + 769.0, + 680.0, + 750.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 660.0, + 912.0, + 660.0, + 912.0, + 682.0, + 889.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 663.0, + 1052.0, + 663.0, + 1052.0, + 684.0, + 1030.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 660.0, + 1193.0, + 660.0, + 1193.0, + 682.0, + 1169.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 662.0, + 1331.0, + 662.0, + 1331.0, + 680.0, + 1313.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 188.0, + 700.0, + 205.0, + 700.0, + 205.0, + 716.0, + 188.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 216.0, + 699.0, + 246.0, + 699.0, + 246.0, + 717.0, + 216.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 700.0, + 346.0, + 700.0, + 346.0, + 716.0, + 329.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 700.0, + 381.0, + 700.0, + 381.0, + 716.0, + 364.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 699.0, + 423.0, + 699.0, + 423.0, + 717.0, + 394.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 700.0, + 487.0, + 700.0, + 487.0, + 716.0, + 470.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 699.0, + 522.0, + 699.0, + 522.0, + 716.0, + 505.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 699.0, + 557.0, + 699.0, + 557.0, + 716.0, + 540.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 700.0, + 627.0, + 700.0, + 627.0, + 716.0, + 610.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 699.0, + 668.0, + 699.0, + 668.0, + 717.0, + 639.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 701.0, + 765.0, + 701.0, + 765.0, + 715.0, + 753.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 701.0, + 800.0, + 701.0, + 800.0, + 715.0, + 788.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 699.0, + 844.0, + 699.0, + 844.0, + 717.0, + 816.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 700.0, + 907.0, + 700.0, + 907.0, + 716.0, + 890.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 701.0, + 941.0, + 701.0, + 941.0, + 715.0, + 928.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 700.0, + 978.0, + 700.0, + 978.0, + 716.0, + 961.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 700.0, + 1048.0, + 700.0, + 1048.0, + 716.0, + 1031.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 699.0, + 1089.0, + 699.0, + 1089.0, + 717.0, + 1061.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 701.0, + 1186.0, + 701.0, + 1186.0, + 715.0, + 1173.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 701.0, + 1221.0, + 701.0, + 1221.0, + 715.0, + 1208.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 699.0, + 1266.0, + 699.0, + 1266.0, + 717.0, + 1235.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 700.0, + 1329.0, + 700.0, + 1329.0, + 716.0, + 1313.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 701.0, + 1362.0, + 701.0, + 1362.0, + 715.0, + 1349.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 701.0, + 1397.0, + 701.0, + 1397.0, + 715.0, + 1384.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 181.0, + 733.0, + 250.0, + 733.0, + 250.0, + 754.0, + 181.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 735.0, + 354.0, + 735.0, + 354.0, + 753.0, + 323.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 733.0, + 427.0, + 733.0, + 427.0, + 754.0, + 356.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 735.0, + 495.0, + 735.0, + 495.0, + 753.0, + 464.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 735.0, + 529.0, + 735.0, + 529.0, + 753.0, + 498.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 735.0, + 563.0, + 735.0, + 563.0, + 754.0, + 534.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 736.0, + 627.0, + 736.0, + 627.0, + 753.0, + 610.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 735.0, + 668.0, + 735.0, + 668.0, + 754.0, + 639.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 736.0, + 768.0, + 736.0, + 768.0, + 753.0, + 750.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 736.0, + 803.0, + 736.0, + 803.0, + 753.0, + 786.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 731.0, + 845.0, + 731.0, + 845.0, + 756.0, + 815.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 736.0, + 907.0, + 736.0, + 907.0, + 753.0, + 890.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 736.0, + 942.0, + 736.0, + 942.0, + 753.0, + 927.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 736.0, + 978.0, + 736.0, + 978.0, + 753.0, + 961.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 736.0, + 1049.0, + 736.0, + 1049.0, + 753.0, + 1031.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 733.0, + 1090.0, + 733.0, + 1090.0, + 753.0, + 1061.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 736.0, + 1189.0, + 736.0, + 1189.0, + 753.0, + 1172.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 736.0, + 1224.0, + 736.0, + 1224.0, + 753.0, + 1207.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 735.0, + 1266.0, + 735.0, + 1266.0, + 754.0, + 1235.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 736.0, + 1329.0, + 736.0, + 1329.0, + 753.0, + 1313.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 736.0, + 1365.0, + 736.0, + 1365.0, + 753.0, + 1346.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 735.0, + 1400.0, + 735.0, + 1400.0, + 753.0, + 1383.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 770.0, + 634.0, + 770.0, + 634.0, + 789.0, + 605.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 770.0, + 671.0, + 770.0, + 671.0, + 789.0, + 636.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 770.0, + 775.0, + 770.0, + 775.0, + 789.0, + 744.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 770.0, + 810.0, + 770.0, + 810.0, + 789.0, + 779.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 770.0, + 848.0, + 770.0, + 848.0, + 789.0, + 812.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 770.0, + 914.0, + 770.0, + 914.0, + 789.0, + 885.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 770.0, + 951.0, + 770.0, + 951.0, + 789.0, + 920.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 770.0, + 985.0, + 770.0, + 985.0, + 789.0, + 955.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 772.0, + 1048.0, + 772.0, + 1048.0, + 788.0, + 1031.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 770.0, + 1090.0, + 770.0, + 1090.0, + 789.0, + 1061.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 772.0, + 1189.0, + 772.0, + 1189.0, + 788.0, + 1172.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 772.0, + 1224.0, + 772.0, + 1224.0, + 789.0, + 1206.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 770.0, + 1266.0, + 770.0, + 1266.0, + 789.0, + 1237.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 772.0, + 1329.0, + 772.0, + 1329.0, + 788.0, + 1313.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 772.0, + 1365.0, + 772.0, + 1365.0, + 789.0, + 1346.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 772.0, + 1400.0, + 772.0, + 1400.0, + 789.0, + 1383.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 194.0, + 805.0, + 271.0, + 805.0, + 271.0, + 826.0, + 194.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 805.0, + 412.0, + 805.0, + 412.0, + 827.0, + 336.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 805.0, + 550.0, + 805.0, + 550.0, + 827.0, + 479.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 805.0, + 693.0, + 805.0, + 693.0, + 827.0, + 616.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 805.0, + 824.0, + 805.0, + 824.0, + 827.0, + 767.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 805.0, + 927.0, + 805.0, + 927.0, + 827.0, + 897.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 805.0, + 975.0, + 805.0, + 975.0, + 827.0, + 942.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 805.0, + 1111.0, + 805.0, + 1111.0, + 827.0, + 1042.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 807.0, + 1254.0, + 807.0, + 1254.0, + 824.0, + 1177.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 805.0, + 1386.0, + 805.0, + 1386.0, + 827.0, + 1328.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 845.0, + 1373.0, + 845.0, + 1373.0, + 854.0, + 1338.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 123.0, + 206.0, + 1054.0, + 206.0, + 1054.0, + 273.0, + 123.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1786.0, + 309.0, + 1786.0, + 309.0, + 1799.0, + 295.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1782.0, + 346.0, + 1782.0, + 346.0, + 1801.0, + 328.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1788.0, + 379.0, + 1788.0, + 379.0, + 1798.0, + 367.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1786.0, + 521.0, + 1786.0, + 521.0, + 1799.0, + 508.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1782.0, + 557.0, + 1782.0, + 557.0, + 1801.0, + 541.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1786.0, + 590.0, + 1786.0, + 590.0, + 1799.0, + 578.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1786.0, + 731.0, + 1786.0, + 731.0, + 1799.0, + 717.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1782.0, + 767.0, + 1782.0, + 767.0, + 1801.0, + 750.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1788.0, + 801.0, + 1788.0, + 801.0, + 1798.0, + 788.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1785.0, + 942.0, + 1785.0, + 942.0, + 1799.0, + 928.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1782.0, + 978.0, + 1782.0, + 978.0, + 1801.0, + 961.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1786.0, + 1012.0, + 1786.0, + 1012.0, + 1799.0, + 999.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1789.0, + 1045.0, + 1789.0, + 1045.0, + 1799.0, + 1033.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1783.0, + 1154.0, + 1783.0, + 1154.0, + 1801.0, + 1136.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1782.0, + 1189.0, + 1782.0, + 1189.0, + 1801.0, + 1172.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1786.0, + 1222.0, + 1786.0, + 1222.0, + 1799.0, + 1211.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1786.0, + 1364.0, + 1786.0, + 1364.0, + 1799.0, + 1349.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 1782.0, + 1400.0, + 1782.0, + 1400.0, + 1801.0, + 1382.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1421.0, + 1786.0, + 1433.0, + 1786.0, + 1433.0, + 1799.0, + 1421.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1490.0, + 1789.0, + 1503.0, + 1789.0, + 1503.0, + 1801.0, + 1490.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 312.0, + 1820.0, + 312.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1820.0, + 346.0, + 1820.0, + 346.0, + 1839.0, + 328.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1823.0, + 380.0, + 1823.0, + 380.0, + 1834.0, + 370.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1823.0, + 521.0, + 1823.0, + 521.0, + 1836.0, + 506.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1823.0, + 556.0, + 1823.0, + 556.0, + 1836.0, + 541.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1821.0, + 592.0, + 1821.0, + 592.0, + 1834.0, + 580.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1823.0, + 731.0, + 1823.0, + 731.0, + 1834.0, + 716.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1823.0, + 767.0, + 1823.0, + 767.0, + 1836.0, + 752.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1821.0, + 803.0, + 1821.0, + 803.0, + 1834.0, + 791.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1820.0, + 945.0, + 1820.0, + 945.0, + 1839.0, + 927.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1823.0, + 976.0, + 1823.0, + 976.0, + 1837.0, + 963.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1823.0, + 1012.0, + 1823.0, + 1012.0, + 1834.0, + 1002.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1820.0, + 1154.0, + 1820.0, + 1154.0, + 1839.0, + 1136.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1820.0, + 1189.0, + 1820.0, + 1189.0, + 1839.0, + 1171.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1823.0, + 1223.0, + 1823.0, + 1223.0, + 1834.0, + 1213.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1820.0, + 1365.0, + 1820.0, + 1365.0, + 1839.0, + 1347.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1383.0, + 1823.0, + 1398.0, + 1823.0, + 1398.0, + 1836.0, + 1383.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1422.0, + 1823.0, + 1434.0, + 1823.0, + 1434.0, + 1833.0, + 1422.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1858.0, + 310.0, + 1858.0, + 310.0, + 1876.0, + 292.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1856.0, + 346.0, + 1856.0, + 346.0, + 1874.0, + 328.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1858.0, + 523.0, + 1858.0, + 523.0, + 1875.0, + 505.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1856.0, + 557.0, + 1856.0, + 557.0, + 1874.0, + 539.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1858.0, + 732.0, + 1858.0, + 732.0, + 1875.0, + 714.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1856.0, + 768.0, + 1856.0, + 768.0, + 1874.0, + 750.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1859.0, + 800.0, + 1859.0, + 800.0, + 1871.0, + 788.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1862.0, + 837.0, + 1862.0, + 837.0, + 1874.0, + 822.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1856.0, + 943.0, + 1856.0, + 943.0, + 1876.0, + 927.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1858.0, + 976.0, + 1858.0, + 976.0, + 1872.0, + 963.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1863.0, + 1047.0, + 1863.0, + 1047.0, + 1874.0, + 1033.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1856.0, + 1154.0, + 1856.0, + 1154.0, + 1876.0, + 1136.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1856.0, + 1189.0, + 1856.0, + 1189.0, + 1874.0, + 1171.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1858.0, + 1364.0, + 1858.0, + 1364.0, + 1876.0, + 1347.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1382.0, + 1856.0, + 1400.0, + 1856.0, + 1400.0, + 1874.0, + 1382.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1898.0, + 310.0, + 1898.0, + 310.0, + 1910.0, + 297.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1898.0, + 523.0, + 1898.0, + 523.0, + 1910.0, + 508.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1898.0, + 592.0, + 1898.0, + 592.0, + 1910.0, + 578.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1898.0, + 732.0, + 1898.0, + 732.0, + 1908.0, + 719.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1898.0, + 803.0, + 1898.0, + 803.0, + 1910.0, + 788.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1898.0, + 837.0, + 1898.0, + 837.0, + 1910.0, + 822.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1898.0, + 942.0, + 1898.0, + 942.0, + 1908.0, + 930.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1898.0, + 976.0, + 1898.0, + 976.0, + 1908.0, + 963.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1898.0, + 1048.0, + 1898.0, + 1048.0, + 1910.0, + 1033.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1898.0, + 1258.0, + 1898.0, + 1258.0, + 1908.0, + 1246.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1898.0, + 1364.0, + 1898.0, + 1364.0, + 1910.0, + 1350.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 186.0, + 1924.0, + 256.0, + 1924.0, + 256.0, + 1954.0, + 186.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1926.0, + 366.0, + 1926.0, + 366.0, + 1949.0, + 355.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1924.0, + 412.0, + 1924.0, + 412.0, + 1951.0, + 373.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1926.0, + 568.0, + 1926.0, + 568.0, + 1951.0, + 506.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1927.0, + 611.0, + 1927.0, + 611.0, + 1948.0, + 592.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1926.0, + 779.0, + 1926.0, + 779.0, + 1951.0, + 717.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1924.0, + 834.0, + 1924.0, + 834.0, + 1951.0, + 795.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1926.0, + 990.0, + 1926.0, + 990.0, + 1951.0, + 927.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1927.0, + 1033.0, + 1927.0, + 1033.0, + 1948.0, + 1014.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1924.0, + 1201.0, + 1924.0, + 1201.0, + 1949.0, + 1139.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1924.0, + 1255.0, + 1924.0, + 1255.0, + 1951.0, + 1217.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1923.0, + 1413.0, + 1923.0, + 1413.0, + 1952.0, + 1347.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1436.0, + 1927.0, + 1454.0, + 1927.0, + 1454.0, + 1948.0, + 1436.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1425.0, + 309.0, + 1425.0, + 309.0, + 1438.0, + 296.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1424.0, + 345.0, + 1424.0, + 345.0, + 1437.0, + 330.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1425.0, + 380.0, + 1425.0, + 380.0, + 1438.0, + 367.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1425.0, + 521.0, + 1425.0, + 521.0, + 1437.0, + 508.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1424.0, + 555.0, + 1424.0, + 555.0, + 1437.0, + 542.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1425.0, + 589.0, + 1425.0, + 589.0, + 1437.0, + 577.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1425.0, + 731.0, + 1425.0, + 731.0, + 1438.0, + 718.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1423.0, + 767.0, + 1423.0, + 767.0, + 1437.0, + 753.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1425.0, + 802.0, + 1425.0, + 802.0, + 1438.0, + 788.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1424.0, + 976.0, + 1424.0, + 976.0, + 1435.0, + 962.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1425.0, + 1011.0, + 1425.0, + 1011.0, + 1437.0, + 999.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1424.0, + 1155.0, + 1424.0, + 1155.0, + 1441.0, + 1137.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1421.0, + 1189.0, + 1421.0, + 1189.0, + 1438.0, + 1171.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1425.0, + 1223.0, + 1425.0, + 1223.0, + 1438.0, + 1208.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1425.0, + 1364.0, + 1425.0, + 1364.0, + 1437.0, + 1350.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1424.0, + 1396.0, + 1424.0, + 1396.0, + 1435.0, + 1384.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 1428.0, + 1430.0, + 1428.0, + 1430.0, + 1435.0, + 1423.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1459.0, + 309.0, + 1459.0, + 309.0, + 1473.0, + 296.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1458.0, + 346.0, + 1458.0, + 346.0, + 1476.0, + 328.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1459.0, + 382.0, + 1459.0, + 382.0, + 1472.0, + 370.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1465.0, + 414.0, + 1465.0, + 414.0, + 1474.0, + 399.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1458.0, + 524.0, + 1458.0, + 524.0, + 1476.0, + 505.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1459.0, + 558.0, + 1459.0, + 558.0, + 1476.0, + 539.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1459.0, + 592.0, + 1459.0, + 592.0, + 1472.0, + 580.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1465.0, + 625.0, + 1465.0, + 625.0, + 1474.0, + 613.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1458.0, + 733.0, + 1458.0, + 733.0, + 1476.0, + 715.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1458.0, + 768.0, + 1458.0, + 768.0, + 1476.0, + 750.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1460.0, + 802.0, + 1460.0, + 802.0, + 1472.0, + 790.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1459.0, + 942.0, + 1459.0, + 942.0, + 1473.0, + 928.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1458.0, + 979.0, + 1458.0, + 979.0, + 1476.0, + 961.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1459.0, + 1014.0, + 1459.0, + 1014.0, + 1470.0, + 1002.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1458.0, + 1155.0, + 1458.0, + 1155.0, + 1476.0, + 1136.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1458.0, + 1189.0, + 1458.0, + 1189.0, + 1476.0, + 1171.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1459.0, + 1223.0, + 1459.0, + 1223.0, + 1472.0, + 1211.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1459.0, + 1364.0, + 1459.0, + 1364.0, + 1473.0, + 1350.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1460.0, + 1398.0, + 1460.0, + 1398.0, + 1473.0, + 1384.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1456.0, + 1465.0, + 1467.0, + 1465.0, + 1467.0, + 1476.0, + 1456.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1496.0, + 312.0, + 1496.0, + 312.0, + 1514.0, + 293.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1494.0, + 348.0, + 1494.0, + 348.0, + 1511.0, + 328.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1494.0, + 522.0, + 1494.0, + 522.0, + 1514.0, + 505.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1493.0, + 558.0, + 1493.0, + 558.0, + 1511.0, + 539.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1497.0, + 589.0, + 1497.0, + 589.0, + 1508.0, + 577.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1496.0, + 733.0, + 1496.0, + 733.0, + 1514.0, + 715.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1493.0, + 768.0, + 1493.0, + 768.0, + 1511.0, + 749.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1498.0, + 942.0, + 1498.0, + 942.0, + 1511.0, + 928.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1494.0, + 979.0, + 1494.0, + 979.0, + 1511.0, + 961.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1496.0, + 1153.0, + 1496.0, + 1153.0, + 1514.0, + 1136.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1494.0, + 1189.0, + 1494.0, + 1189.0, + 1511.0, + 1171.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1498.0, + 1362.0, + 1498.0, + 1362.0, + 1511.0, + 1350.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1497.0, + 1398.0, + 1497.0, + 1398.0, + 1510.0, + 1384.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1536.0, + 312.0, + 1536.0, + 312.0, + 1547.0, + 296.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1536.0, + 414.0, + 1536.0, + 414.0, + 1547.0, + 401.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1538.0, + 521.0, + 1538.0, + 521.0, + 1547.0, + 506.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1536.0, + 1155.0, + 1536.0, + 1155.0, + 1547.0, + 1140.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 184.0, + 1565.0, + 255.0, + 1565.0, + 255.0, + 1590.0, + 184.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1564.0, + 308.0, + 1564.0, + 308.0, + 1588.0, + 294.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1564.0, + 365.0, + 1564.0, + 365.0, + 1588.0, + 355.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1564.0, + 417.0, + 1564.0, + 417.0, + 1589.0, + 368.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1563.0, + 570.0, + 1563.0, + 570.0, + 1591.0, + 505.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1564.0, + 623.0, + 1564.0, + 623.0, + 1589.0, + 583.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1564.0, + 780.0, + 1564.0, + 780.0, + 1588.0, + 718.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1566.0, + 832.0, + 1566.0, + 832.0, + 1587.0, + 795.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1564.0, + 990.0, + 1564.0, + 990.0, + 1588.0, + 927.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1567.0, + 1033.0, + 1567.0, + 1033.0, + 1587.0, + 1014.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1564.0, + 1210.0, + 1564.0, + 1210.0, + 1588.0, + 1139.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1564.0, + 1254.0, + 1564.0, + 1254.0, + 1589.0, + 1214.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1563.0, + 1413.0, + 1563.0, + 1413.0, + 1591.0, + 1347.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1435.0, + 1568.0, + 1454.0, + 1568.0, + 1454.0, + 1585.0, + 1435.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 176.0, + 1599.0, + 276.0, + 1599.0, + 276.0, + 1626.0, + 176.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1417.0, + 1458.5, + 1437.0, + 1458.5, + 1437.0, + 1471.0, + 1417.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 170.0, + 957.0, + 433.0, + 957.0, + 433.0, + 990.0, + 170.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1063.0, + 309.0, + 1063.0, + 309.0, + 1076.0, + 295.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1060.0, + 346.0, + 1060.0, + 346.0, + 1078.0, + 329.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1061.0, + 381.0, + 1061.0, + 381.0, + 1079.0, + 364.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1061.0, + 522.0, + 1061.0, + 522.0, + 1079.0, + 504.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1060.0, + 556.0, + 1060.0, + 556.0, + 1078.0, + 540.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1061.0, + 592.0, + 1061.0, + 592.0, + 1079.0, + 574.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1063.0, + 731.0, + 1063.0, + 731.0, + 1077.0, + 716.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1062.0, + 765.0, + 1062.0, + 765.0, + 1076.0, + 752.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1063.0, + 801.0, + 1063.0, + 801.0, + 1077.0, + 788.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1063.0, + 941.0, + 1063.0, + 941.0, + 1076.0, + 928.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1062.0, + 976.0, + 1062.0, + 976.0, + 1076.0, + 964.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1063.0, + 1011.0, + 1063.0, + 1011.0, + 1077.0, + 998.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1097.0, + 311.0, + 1097.0, + 311.0, + 1114.0, + 294.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1098.0, + 346.0, + 1098.0, + 346.0, + 1114.0, + 329.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1098.0, + 381.0, + 1098.0, + 381.0, + 1112.0, + 368.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1097.0, + 522.0, + 1097.0, + 522.0, + 1114.0, + 504.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1098.0, + 557.0, + 1098.0, + 557.0, + 1114.0, + 540.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1096.0, + 593.0, + 1096.0, + 593.0, + 1114.0, + 578.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1097.0, + 733.0, + 1097.0, + 733.0, + 1114.0, + 715.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1098.0, + 767.0, + 1098.0, + 767.0, + 1114.0, + 750.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1100.0, + 800.0, + 1100.0, + 800.0, + 1109.0, + 792.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1099.0, + 941.0, + 1099.0, + 941.0, + 1112.0, + 928.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1100.0, + 976.0, + 1100.0, + 976.0, + 1113.0, + 963.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1134.0, + 313.0, + 1134.0, + 313.0, + 1154.0, + 293.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1133.0, + 347.0, + 1133.0, + 347.0, + 1150.0, + 326.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1134.0, + 522.0, + 1134.0, + 522.0, + 1152.0, + 505.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1133.0, + 556.0, + 1133.0, + 556.0, + 1151.0, + 540.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1134.0, + 589.0, + 1134.0, + 589.0, + 1148.0, + 577.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1134.0, + 735.0, + 1134.0, + 735.0, + 1154.0, + 715.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1133.0, + 769.0, + 1133.0, + 769.0, + 1150.0, + 748.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1136.0, + 941.0, + 1136.0, + 941.0, + 1151.0, + 928.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1134.0, + 976.0, + 1134.0, + 976.0, + 1149.0, + 963.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 187.0, + 1203.0, + 255.0, + 1203.0, + 255.0, + 1231.0, + 187.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1199.0, + 419.0, + 1199.0, + 419.0, + 1232.0, + 292.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1199.0, + 625.0, + 1199.0, + 625.0, + 1232.0, + 503.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1199.0, + 835.0, + 1199.0, + 835.0, + 1232.0, + 714.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1200.0, + 1000.0, + 1200.0, + 1000.0, + 1231.0, + 925.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1207.0, + 1031.0, + 1207.0, + 1031.0, + 1226.0, + 1013.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 176.0, + 1238.0, + 275.0, + 1238.0, + 275.0, + 1265.0, + 176.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 908.0, + 801.0, + 908.0, + 801.0, + 943.0, + 137.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 135.0, + 604.0, + 625.0, + 604.0, + 625.0, + 647.0, + 135.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 137.0, + 908.0, + 801.0, + 908.0, + 801.0, + 943.0, + 137.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 306.0, + 733.0, + 306.0, + 733.0, + 344.0, + 412.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 339.0, + 731.0, + 339.0, + 731.0, + 373.0, + 407.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 166.0, + 1313.0, + 430.0, + 1313.0, + 430.0, + 1352.0, + 166.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 166.0, + 1676.0, + 429.0, + 1676.0, + 429.0, + 1715.0, + 166.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 226.0, + 315.0, + 310.0, + 315.0, + 310.0, + 361.0, + 226.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 135.0, + 604.0, + 625.0, + 604.0, + 625.0, + 647.0, + 135.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 170.0, + 954.0, + 435.0, + 954.0, + 435.0, + 992.0, + 170.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 306.0, + 733.0, + 306.0, + 733.0, + 344.0, + 412.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 339.0, + 731.0, + 339.0, + 731.0, + 373.0, + 407.0, + 373.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 34, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 164, + 287, + 642, + 287, + 642, + 314, + 164, + 314 + ], + "score": 0.708 + }, + { + "category_id": 1, + "poly": [ + 197, + 1762, + 678, + 1762, + 678, + 1792, + 197, + 1792 + ], + "score": 0.574 + }, + { + "category_id": 1, + "poly": [ + 199, + 1697, + 683, + 1697, + 683, + 1727, + 199, + 1727 + ], + "score": 0.563 + }, + { + "category_id": 1, + "poly": [ + 194, + 1826, + 721, + 1826, + 721, + 1857, + 194, + 1857 + ], + "score": 0.504 + }, + { + "category_id": 4, + "poly": [ + 162, + 347, + 422, + 347, + 422, + 376, + 162, + 376 + ], + "score": 0.472 + }, + { + "category_id": 3, + "poly": [ + 161, + 394, + 1498, + 394, + 1498, + 1549, + 161, + 1549 + ], + "score": 0.47 + }, + { + "category_id": 1, + "poly": [ + 162, + 347, + 422, + 347, + 422, + 376, + 162, + 376 + ], + "score": 0.145 + }, + { + "category_id": 1, + "poly": [ + 164, + 287, + 642, + 287, + 642, + 314, + 164, + 314 + ], + "score": 0.129 + }, + { + "category_id": 13, + "poly": [ + 223, + 1525, + 278, + 1525, + 278, + 1546, + 223, + 1546 + ], + "score": 0.42, + "latex": "= 1 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 224, + 911, + 286, + 911, + 286, + 932, + 224, + 932 + ], + "score": 0.33, + "latex": "= 1 2 . 2 5 " + }, + { + "category_id": 15, + "poly": [ + 161.0, + 282.0, + 643.0, + 282.0, + 643.0, + 318.0, + 161.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 342.0, + 424.0, + 342.0, + 424.0, + 381.0, + 160.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 480.0, + 316.0, + 480.0, + 316.0, + 495.0, + 303.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 480.0, + 351.0, + 480.0, + 351.0, + 493.0, + 336.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 480.0, + 386.0, + 480.0, + 386.0, + 493.0, + 371.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 480.0, + 527.0, + 480.0, + 527.0, + 495.0, + 514.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 480.0, + 562.0, + 480.0, + 562.0, + 495.0, + 548.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 480.0, + 597.0, + 480.0, + 597.0, + 495.0, + 584.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 478.0, + 740.0, + 478.0, + 740.0, + 496.0, + 722.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 480.0, + 773.0, + 480.0, + 773.0, + 493.0, + 758.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 481.0, + 948.0, + 481.0, + 948.0, + 493.0, + 934.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 480.0, + 984.0, + 480.0, + 984.0, + 493.0, + 970.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 480.0, + 1018.0, + 480.0, + 1018.0, + 493.0, + 1006.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 481.0, + 1159.0, + 481.0, + 1159.0, + 493.0, + 1145.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 480.0, + 1193.0, + 480.0, + 1193.0, + 493.0, + 1180.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 481.0, + 1229.0, + 481.0, + 1229.0, + 493.0, + 1216.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 478.0, + 1373.0, + 478.0, + 1373.0, + 496.0, + 1355.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1388.0, + 477.0, + 1407.0, + 477.0, + 1407.0, + 495.0, + 1388.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 478.0, + 1442.0, + 478.0, + 1442.0, + 496.0, + 1424.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 513.0, + 318.0, + 513.0, + 318.0, + 533.0, + 300.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 516.0, + 351.0, + 516.0, + 351.0, + 530.0, + 336.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 516.0, + 386.0, + 516.0, + 386.0, + 528.0, + 376.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 516.0, + 527.0, + 516.0, + 527.0, + 530.0, + 512.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 516.0, + 562.0, + 516.0, + 562.0, + 530.0, + 548.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 516.0, + 597.0, + 516.0, + 597.0, + 528.0, + 587.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 513.0, + 740.0, + 513.0, + 740.0, + 531.0, + 722.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 516.0, + 771.0, + 516.0, + 771.0, + 531.0, + 758.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 516.0, + 809.0, + 516.0, + 809.0, + 528.0, + 797.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 516.0, + 949.0, + 516.0, + 949.0, + 530.0, + 934.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 516.0, + 982.0, + 516.0, + 982.0, + 530.0, + 969.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 516.0, + 1020.0, + 516.0, + 1020.0, + 528.0, + 1009.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 516.0, + 1159.0, + 516.0, + 1159.0, + 528.0, + 1145.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 517.0, + 1193.0, + 517.0, + 1193.0, + 528.0, + 1180.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 516.0, + 1232.0, + 516.0, + 1232.0, + 528.0, + 1219.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 516.0, + 1370.0, + 516.0, + 1370.0, + 530.0, + 1356.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1389.0, + 516.0, + 1404.0, + 516.0, + 1404.0, + 528.0, + 1389.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1428.0, + 516.0, + 1440.0, + 516.0, + 1440.0, + 528.0, + 1428.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 551.0, + 319.0, + 551.0, + 319.0, + 570.0, + 300.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 549.0, + 352.0, + 549.0, + 352.0, + 567.0, + 334.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 551.0, + 529.0, + 551.0, + 529.0, + 570.0, + 511.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 552.0, + 562.0, + 552.0, + 562.0, + 564.0, + 548.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 551.0, + 740.0, + 551.0, + 740.0, + 569.0, + 722.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 549.0, + 774.0, + 549.0, + 774.0, + 569.0, + 755.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 552.0, + 807.0, + 552.0, + 807.0, + 564.0, + 794.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 551.0, + 951.0, + 551.0, + 951.0, + 569.0, + 933.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 551.0, + 982.0, + 551.0, + 982.0, + 566.0, + 969.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 552.0, + 1159.0, + 552.0, + 1159.0, + 567.0, + 1145.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 549.0, + 1196.0, + 549.0, + 1196.0, + 567.0, + 1178.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 552.0, + 1370.0, + 552.0, + 1370.0, + 567.0, + 1356.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1389.0, + 552.0, + 1404.0, + 552.0, + 1404.0, + 564.0, + 1389.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 619.0, + 264.0, + 619.0, + 264.0, + 649.0, + 193.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 620.0, + 365.0, + 620.0, + 365.0, + 646.0, + 301.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 620.0, + 422.0, + 620.0, + 422.0, + 646.0, + 376.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 620.0, + 575.0, + 620.0, + 575.0, + 646.0, + 512.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 620.0, + 629.0, + 620.0, + 629.0, + 647.0, + 590.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 620.0, + 786.0, + 620.0, + 786.0, + 646.0, + 723.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 619.0, + 840.0, + 619.0, + 840.0, + 647.0, + 803.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 620.0, + 997.0, + 620.0, + 997.0, + 646.0, + 934.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 620.0, + 1050.0, + 620.0, + 1050.0, + 647.0, + 1012.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 620.0, + 1208.0, + 620.0, + 1208.0, + 646.0, + 1145.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 623.0, + 1249.0, + 623.0, + 1249.0, + 644.0, + 1231.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 620.0, + 1418.0, + 620.0, + 1418.0, + 646.0, + 1355.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1442.0, + 623.0, + 1460.0, + 623.0, + 1460.0, + 644.0, + 1442.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 733.0, + 316.0, + 733.0, + 316.0, + 747.0, + 301.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 730.0, + 351.0, + 730.0, + 351.0, + 745.0, + 337.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 733.0, + 386.0, + 733.0, + 386.0, + 747.0, + 371.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 736.0, + 419.0, + 736.0, + 419.0, + 747.0, + 407.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 733.0, + 527.0, + 733.0, + 527.0, + 747.0, + 512.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 732.0, + 562.0, + 732.0, + 562.0, + 745.0, + 548.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 733.0, + 597.0, + 733.0, + 597.0, + 745.0, + 584.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 733.0, + 737.0, + 733.0, + 737.0, + 747.0, + 723.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 729.0, + 774.0, + 729.0, + 774.0, + 747.0, + 756.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 730.0, + 810.0, + 730.0, + 810.0, + 748.0, + 792.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 765.0, + 319.0, + 765.0, + 319.0, + 783.0, + 300.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 769.0, + 351.0, + 769.0, + 351.0, + 781.0, + 336.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 766.0, + 386.0, + 766.0, + 386.0, + 781.0, + 374.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 765.0, + 529.0, + 765.0, + 529.0, + 783.0, + 511.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 769.0, + 562.0, + 769.0, + 562.0, + 781.0, + 548.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 768.0, + 597.0, + 768.0, + 597.0, + 780.0, + 587.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 765.0, + 740.0, + 765.0, + 740.0, + 783.0, + 722.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 769.0, + 771.0, + 769.0, + 771.0, + 781.0, + 759.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 769.0, + 806.0, + 769.0, + 806.0, + 777.0, + 798.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 804.0, + 318.0, + 804.0, + 318.0, + 824.0, + 300.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 803.0, + 352.0, + 803.0, + 352.0, + 821.0, + 334.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 804.0, + 529.0, + 804.0, + 529.0, + 824.0, + 511.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 806.0, + 562.0, + 806.0, + 562.0, + 819.0, + 548.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 804.0, + 740.0, + 804.0, + 740.0, + 824.0, + 722.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 806.0, + 771.0, + 806.0, + 771.0, + 819.0, + 758.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 846.0, + 737.0, + 846.0, + 737.0, + 857.0, + 725.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 872.0, + 264.0, + 872.0, + 264.0, + 902.0, + 193.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 870.0, + 370.0, + 870.0, + 370.0, + 901.0, + 300.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 872.0, + 418.0, + 872.0, + 418.0, + 899.0, + 377.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 873.0, + 574.0, + 873.0, + 574.0, + 899.0, + 512.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 876.0, + 617.0, + 876.0, + 617.0, + 896.0, + 597.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 870.0, + 786.0, + 870.0, + 786.0, + 901.0, + 722.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 876.0, + 828.0, + 876.0, + 828.0, + 896.0, + 810.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 181.0, + 908.0, + 223.0, + 908.0, + 223.0, + 933.0, + 181.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 908.0, + 290.0, + 908.0, + 290.0, + 933.0, + 287.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 157.0, + 973.0, + 421.0, + 973.0, + 421.0, + 1012.0, + 157.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1095.0, + 316.0, + 1095.0, + 316.0, + 1109.0, + 301.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1091.0, + 352.0, + 1091.0, + 352.0, + 1109.0, + 334.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1095.0, + 386.0, + 1095.0, + 386.0, + 1109.0, + 371.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1095.0, + 526.0, + 1095.0, + 526.0, + 1107.0, + 514.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1094.0, + 562.0, + 1094.0, + 562.0, + 1107.0, + 548.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1095.0, + 737.0, + 1095.0, + 737.0, + 1109.0, + 723.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1094.0, + 773.0, + 1094.0, + 773.0, + 1107.0, + 758.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1095.0, + 807.0, + 1095.0, + 807.0, + 1107.0, + 795.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1095.0, + 948.0, + 1095.0, + 948.0, + 1107.0, + 934.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1092.0, + 982.0, + 1092.0, + 982.0, + 1107.0, + 970.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1095.0, + 1018.0, + 1095.0, + 1018.0, + 1109.0, + 1003.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1095.0, + 1159.0, + 1095.0, + 1159.0, + 1107.0, + 1145.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1095.0, + 1193.0, + 1095.0, + 1193.0, + 1107.0, + 1180.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1095.0, + 1229.0, + 1095.0, + 1229.0, + 1107.0, + 1216.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1095.0, + 1334.0, + 1095.0, + 1334.0, + 1109.0, + 1320.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1094.0, + 1370.0, + 1094.0, + 1370.0, + 1107.0, + 1356.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 1095.0, + 1404.0, + 1095.0, + 1404.0, + 1107.0, + 1391.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1127.0, + 319.0, + 1127.0, + 319.0, + 1146.0, + 300.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1130.0, + 351.0, + 1130.0, + 351.0, + 1143.0, + 336.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1130.0, + 388.0, + 1130.0, + 388.0, + 1142.0, + 374.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1127.0, + 529.0, + 1127.0, + 529.0, + 1145.0, + 511.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1131.0, + 562.0, + 1131.0, + 562.0, + 1143.0, + 548.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1130.0, + 737.0, + 1130.0, + 737.0, + 1143.0, + 725.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1131.0, + 773.0, + 1131.0, + 773.0, + 1143.0, + 759.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1130.0, + 949.0, + 1130.0, + 949.0, + 1143.0, + 934.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1131.0, + 982.0, + 1131.0, + 982.0, + 1143.0, + 970.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1130.0, + 1159.0, + 1130.0, + 1159.0, + 1143.0, + 1144.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1131.0, + 1195.0, + 1131.0, + 1195.0, + 1143.0, + 1180.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 1133.0, + 1229.0, + 1133.0, + 1229.0, + 1140.0, + 1222.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1127.0, + 1337.0, + 1127.0, + 1337.0, + 1145.0, + 1319.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1131.0, + 1368.0, + 1131.0, + 1368.0, + 1143.0, + 1356.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1165.0, + 319.0, + 1165.0, + 319.0, + 1184.0, + 300.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1163.0, + 353.0, + 1163.0, + 353.0, + 1181.0, + 334.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1166.0, + 383.0, + 1166.0, + 383.0, + 1178.0, + 371.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1165.0, + 529.0, + 1165.0, + 529.0, + 1184.0, + 511.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1163.0, + 563.0, + 1163.0, + 563.0, + 1181.0, + 545.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1168.0, + 737.0, + 1168.0, + 737.0, + 1181.0, + 723.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1163.0, + 774.0, + 1163.0, + 774.0, + 1181.0, + 756.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1165.0, + 951.0, + 1165.0, + 951.0, + 1184.0, + 933.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 1163.0, + 985.0, + 1163.0, + 985.0, + 1181.0, + 967.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 1166.0, + 1017.0, + 1166.0, + 1017.0, + 1178.0, + 1006.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1165.0, + 1160.0, + 1165.0, + 1160.0, + 1184.0, + 1142.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1163.0, + 1196.0, + 1163.0, + 1196.0, + 1181.0, + 1178.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1165.0, + 1228.0, + 1165.0, + 1228.0, + 1178.0, + 1216.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1165.0, + 1337.0, + 1165.0, + 1337.0, + 1184.0, + 1317.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1166.0, + 1368.0, + 1166.0, + 1368.0, + 1180.0, + 1356.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 1232.0, + 264.0, + 1232.0, + 264.0, + 1263.0, + 193.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1234.0, + 364.0, + 1234.0, + 364.0, + 1260.0, + 301.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1235.0, + 416.0, + 1235.0, + 416.0, + 1258.0, + 380.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1234.0, + 574.0, + 1234.0, + 574.0, + 1260.0, + 512.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1237.0, + 617.0, + 1237.0, + 617.0, + 1258.0, + 597.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1234.0, + 785.0, + 1234.0, + 785.0, + 1260.0, + 722.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1237.0, + 828.0, + 1237.0, + 828.0, + 1258.0, + 809.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1234.0, + 996.0, + 1234.0, + 996.0, + 1260.0, + 934.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1235.0, + 1048.0, + 1235.0, + 1048.0, + 1258.0, + 1015.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1234.0, + 1207.0, + 1234.0, + 1207.0, + 1260.0, + 1144.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 1237.0, + 1250.0, + 1237.0, + 1250.0, + 1258.0, + 1231.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1234.0, + 1416.0, + 1234.0, + 1416.0, + 1260.0, + 1355.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1442.0, + 1237.0, + 1461.0, + 1237.0, + 1461.0, + 1258.0, + 1442.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1346.0, + 316.0, + 1346.0, + 316.0, + 1361.0, + 303.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1344.0, + 352.0, + 1344.0, + 352.0, + 1362.0, + 334.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1347.0, + 385.0, + 1347.0, + 385.0, + 1361.0, + 373.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1347.0, + 527.0, + 1347.0, + 527.0, + 1361.0, + 512.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1343.0, + 563.0, + 1343.0, + 563.0, + 1362.0, + 547.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1347.0, + 596.0, + 1347.0, + 596.0, + 1361.0, + 584.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1347.0, + 738.0, + 1347.0, + 738.0, + 1361.0, + 725.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1343.0, + 774.0, + 1343.0, + 774.0, + 1361.0, + 756.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1346.0, + 807.0, + 1346.0, + 807.0, + 1361.0, + 794.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1380.0, + 319.0, + 1380.0, + 319.0, + 1398.0, + 300.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1383.0, + 351.0, + 1383.0, + 351.0, + 1397.0, + 336.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1380.0, + 529.0, + 1380.0, + 529.0, + 1398.0, + 511.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1383.0, + 562.0, + 1383.0, + 562.0, + 1397.0, + 548.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1380.0, + 741.0, + 1380.0, + 741.0, + 1398.0, + 722.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1383.0, + 773.0, + 1383.0, + 773.0, + 1397.0, + 758.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 1382.0, + 809.0, + 1382.0, + 809.0, + 1394.0, + 797.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1418.0, + 319.0, + 1418.0, + 319.0, + 1438.0, + 300.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1416.0, + 352.0, + 1416.0, + 352.0, + 1435.0, + 334.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1418.0, + 529.0, + 1418.0, + 529.0, + 1436.0, + 511.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1419.0, + 562.0, + 1419.0, + 562.0, + 1433.0, + 548.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1418.0, + 740.0, + 1418.0, + 740.0, + 1436.0, + 722.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1416.0, + 774.0, + 1416.0, + 774.0, + 1435.0, + 756.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1459.0, + 318.0, + 1459.0, + 318.0, + 1469.0, + 304.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1460.0, + 844.0, + 1460.0, + 844.0, + 1471.0, + 829.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 193.0, + 1486.0, + 262.0, + 1486.0, + 262.0, + 1516.0, + 193.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1487.0, + 365.0, + 1487.0, + 365.0, + 1513.0, + 301.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1489.0, + 416.0, + 1489.0, + 416.0, + 1512.0, + 380.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1487.0, + 574.0, + 1487.0, + 574.0, + 1513.0, + 512.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1490.0, + 617.0, + 1490.0, + 617.0, + 1510.0, + 599.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1487.0, + 785.0, + 1487.0, + 785.0, + 1513.0, + 723.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1490.0, + 828.0, + 1490.0, + 828.0, + 1510.0, + 810.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 183.0, + 1524.0, + 222.0, + 1524.0, + 222.0, + 1548.0, + 183.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 1524.0, + 282.0, + 1524.0, + 282.0, + 1548.0, + 279.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.25, + 479.0, + 802.25, + 479.0, + 802.25, + 492.0, + 798.25, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 1759.0, + 678.0, + 1759.0, + 678.0, + 1795.0, + 196.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 197.0, + 1695.0, + 684.0, + 1695.0, + 684.0, + 1729.0, + 197.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 196.0, + 1826.0, + 722.0, + 1826.0, + 722.0, + 1858.0, + 196.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 160.0, + 342.0, + 424.0, + 342.0, + 424.0, + 381.0, + 160.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 161.0, + 282.0, + 643.0, + 282.0, + 643.0, + 318.0, + 161.0, + 318.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 35, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/mSAKhLYLSsl/mSAKhLYLSsl_origin.pdf b/parse/train/mSAKhLYLSsl/mSAKhLYLSsl_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3cd1b48a70e7c9b9a90698c1bf592b969944b234 --- /dev/null +++ b/parse/train/mSAKhLYLSsl/mSAKhLYLSsl_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8110e0f5cf1635f531e0b48f259ef432871c9659227f56ae800f172ba98b6d87 +size 2857557 diff --git a/parse/train/o81ZyBCojoA/o81ZyBCojoA_layout.pdf b/parse/train/o81ZyBCojoA/o81ZyBCojoA_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f301e97249cd9023fc12634a3b32daff6679cbde --- /dev/null +++ b/parse/train/o81ZyBCojoA/o81ZyBCojoA_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8077e5bc112f4a04da6b1ba8acae4d5b46055ec7c37c99a5acb07a48e4dfa9b +size 6221005 diff --git a/parse/train/o81ZyBCojoA/o81ZyBCojoA_origin.pdf b/parse/train/o81ZyBCojoA/o81ZyBCojoA_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b15309a598679a977bd6ed519bd22a722e1d1d13 --- /dev/null +++ b/parse/train/o81ZyBCojoA/o81ZyBCojoA_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96bcff6cd5b547f086c17ffcddf68666bb5d2f3b95ab7bb1c3ef06c1a8b69e6b +size 6031078 diff --git a/parse/train/o81ZyBCojoA/o81ZyBCojoA_span.pdf b/parse/train/o81ZyBCojoA/o81ZyBCojoA_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6094462b261e3fc0dafaf50575c23213d2b3e9b8 --- /dev/null +++ b/parse/train/o81ZyBCojoA/o81ZyBCojoA_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c10d749342e32f80bac89e8a45b64d874e98a3f649fef751b4314a03f18052d7 +size 6232951 diff --git a/parse/train/qZzy5urZw9/qZzy5urZw9_layout.pdf b/parse/train/qZzy5urZw9/qZzy5urZw9_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..758f036ae37593d9501dc0e49af8b2a892fd685c --- /dev/null +++ b/parse/train/qZzy5urZw9/qZzy5urZw9_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80cc51301ee4699ff1b7a7f52d23bd9b9200ee67e2fe9dc088d01ccec8276615 +size 1514862 diff --git a/parse/train/qZzy5urZw9/qZzy5urZw9_origin.pdf b/parse/train/qZzy5urZw9/qZzy5urZw9_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..48ebe6e7e793ace8959312337bf4b2c39f7b94c1 --- /dev/null +++ b/parse/train/qZzy5urZw9/qZzy5urZw9_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1da26bfdfca9f90833c5f946c3611a102b888582a98b306e7ad3b4deba57efa6 +size 1317779 diff --git a/parse/train/qZzy5urZw9/qZzy5urZw9_span.pdf b/parse/train/qZzy5urZw9/qZzy5urZw9_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a42a657a801556bb53b3fcfe6f09dd339fe465ff --- /dev/null +++ b/parse/train/qZzy5urZw9/qZzy5urZw9_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bfa0acddc7c2655604d3fc1ed7eb8bbcba237a3f43fe3c136475dc792211a61 +size 1517898 diff --git a/parse/train/r1lfF2NYvH/r1lfF2NYvH_layout.pdf b/parse/train/r1lfF2NYvH/r1lfF2NYvH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e248af2a98c080bd912bcf39b90f543560c9a607 --- /dev/null +++ b/parse/train/r1lfF2NYvH/r1lfF2NYvH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c792ecc04e2b0773ac0eaf6afe5157f57598de0e444d872907c9f7531648e44 +size 664658 diff --git a/parse/train/r1lfF2NYvH/r1lfF2NYvH_origin.pdf b/parse/train/r1lfF2NYvH/r1lfF2NYvH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5ec0afa203505a6490b4ba8aa26d1ded92112295 --- /dev/null +++ b/parse/train/r1lfF2NYvH/r1lfF2NYvH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ca7c3fc33bab55d238772276f9348cae79da09183d4518eef494f4a5b2005a6 +size 438709 diff --git a/parse/train/r1lfF2NYvH/r1lfF2NYvH_span.pdf b/parse/train/r1lfF2NYvH/r1lfF2NYvH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e30be2dbf5f8a81208d51a4b01d918f8b90744ff --- /dev/null +++ b/parse/train/r1lfF2NYvH/r1lfF2NYvH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fd742a91a6f3ce921021e068d2bb4f9be5d10fc5a3618aec8e00e830d8da93a +size 664776 diff --git a/parse/train/r1nmx5l0W/r1nmx5l0W_layout.pdf b/parse/train/r1nmx5l0W/r1nmx5l0W_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e74a6ca95c5c7d42dd35bd9e046dee053aea5aff --- /dev/null +++ b/parse/train/r1nmx5l0W/r1nmx5l0W_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5da26d0c0c3c534fc00e84ac3269834728cebd4303bd2d646930b01691f1928a +size 2842657 diff --git a/parse/train/r1nmx5l0W/r1nmx5l0W_origin.pdf b/parse/train/r1nmx5l0W/r1nmx5l0W_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6be145191929e9c226d265cdf83eee139d5ee01d --- /dev/null +++ b/parse/train/r1nmx5l0W/r1nmx5l0W_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88d8fa0f289a601e3d840ab5f60e94607bb8b91fc797cda2ad7ea036368de9e0 +size 2755145 diff --git a/parse/train/r1nmx5l0W/r1nmx5l0W_span.pdf b/parse/train/r1nmx5l0W/r1nmx5l0W_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f603b7b4f9f74b7be0802dcc7af3e17f2bd03ba5 --- /dev/null +++ b/parse/train/r1nmx5l0W/r1nmx5l0W_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a13fca34c27037e7f8b5f427490be173ddb73edae50b81c9ea668803f7b7326 +size 2842447 diff --git a/parse/train/r1xQNlBYPS/r1xQNlBYPS_layout.pdf b/parse/train/r1xQNlBYPS/r1xQNlBYPS_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b59d6f39e8b16c6de6837d91ce4ee6806086f008 --- /dev/null +++ b/parse/train/r1xQNlBYPS/r1xQNlBYPS_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f6901c39c1dadacece858cc6eec231fe46b626544d54ec4d92a8a11771fa431 +size 679347 diff --git a/parse/train/r1xQNlBYPS/r1xQNlBYPS_origin.pdf b/parse/train/r1xQNlBYPS/r1xQNlBYPS_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..926b89d373f08ac7f5134d0503cffbd9db2cf5e2 --- /dev/null +++ b/parse/train/r1xQNlBYPS/r1xQNlBYPS_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64fc749f0a0b0967303b940b30198b17ad720097877e3bb415751a4c1f6337ce +size 523958 diff --git a/parse/train/r1xQNlBYPS/r1xQNlBYPS_span.pdf b/parse/train/r1xQNlBYPS/r1xQNlBYPS_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c06e212af4ad45942a83acf7de6cd9e55b9fa8cc --- /dev/null +++ b/parse/train/r1xQNlBYPS/r1xQNlBYPS_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a1a91fa39f17a1f0dbbd0f4cffef9e01bf3917b9c3d69574456f32c65527b41 +size 679084 diff --git a/parse/train/rJgsskrFwH/rJgsskrFwH_layout.pdf b/parse/train/rJgsskrFwH/rJgsskrFwH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fbf6f363b851356aeeccf7eb97ca173bcf2c6a51 --- /dev/null +++ b/parse/train/rJgsskrFwH/rJgsskrFwH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3e706e74317cd8bd20d097c723c2ea2b50821c03316557f36c30141bb94ad03 +size 10686091 diff --git a/parse/train/rJgsskrFwH/rJgsskrFwH_origin.pdf b/parse/train/rJgsskrFwH/rJgsskrFwH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..83d9463a12d376abc8d0f7c854eddf2589c8a752 --- /dev/null +++ b/parse/train/rJgsskrFwH/rJgsskrFwH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9399a127bc961ac936e01f818cba3b40ed0c6bef9f10aaa7301a447ed47b4cdc +size 10529318 diff --git a/parse/train/rJgsskrFwH/rJgsskrFwH_span.pdf b/parse/train/rJgsskrFwH/rJgsskrFwH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..56ffe9b76f119f1738f8cb2163321e3fd3240e83 --- /dev/null +++ b/parse/train/rJgsskrFwH/rJgsskrFwH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f127d3888790c72fff9bfd0cf215b3903852d68f732a536bca8973d274571094 +size 10690513 diff --git a/parse/train/rkMW1hRqKX/rkMW1hRqKX_layout.pdf b/parse/train/rkMW1hRqKX/rkMW1hRqKX_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fd1810ed132d590bb41a4ab1ecb6b4d3238bc341 --- /dev/null +++ b/parse/train/rkMW1hRqKX/rkMW1hRqKX_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47af11967cc6d938af3fafa133acca70cc0a5b163c465b44fd050a8ff26835f5 +size 1334083 diff --git a/parse/train/rkMW1hRqKX/rkMW1hRqKX_origin.pdf b/parse/train/rkMW1hRqKX/rkMW1hRqKX_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..afa1ee792661a67b1ded6f17151f816f6f40d0d9 --- /dev/null +++ b/parse/train/rkMW1hRqKX/rkMW1hRqKX_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4908b55107ae02c588542f52a24238cf618c20906b178973c7c11fcec7c2819 +size 1107745 diff --git a/parse/train/rkMW1hRqKX/rkMW1hRqKX_span.pdf b/parse/train/rkMW1hRqKX/rkMW1hRqKX_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ff6b4c5ee0554d717a54deafcabbace944e0be13 --- /dev/null +++ b/parse/train/rkMW1hRqKX/rkMW1hRqKX_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7317faf1a3e97de077405e039e9cdb68f95fed8c9e02fbe6d76e2279bf5d588d +size 1339252 diff --git a/parse/train/rkgHY0NYwr/rkgHY0NYwr_layout.pdf b/parse/train/rkgHY0NYwr/rkgHY0NYwr_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d4ae29b35eb5c64c49f36688698930096bfe3234 --- /dev/null +++ b/parse/train/rkgHY0NYwr/rkgHY0NYwr_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57c25d271e44ce26a707327493a9bfc8d5764fcad01a5b4f47c2c178a476e708 +size 8683111 diff --git a/parse/train/rkgHY0NYwr/rkgHY0NYwr_origin.pdf b/parse/train/rkgHY0NYwr/rkgHY0NYwr_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bc9b31f4e3c386b373b3f972cc4229b4c59ddd84 --- /dev/null +++ b/parse/train/rkgHY0NYwr/rkgHY0NYwr_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c842767ef8324ad0f259b972dbf4879de31c7ca0a731dea7a56583065cdd211 +size 8573263 diff --git a/parse/train/rkgHY0NYwr/rkgHY0NYwr_span.pdf b/parse/train/rkgHY0NYwr/rkgHY0NYwr_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..18dae336c6f8f6106cf2cc660ad0d668324c3d25 --- /dev/null +++ b/parse/train/rkgHY0NYwr/rkgHY0NYwr_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2630ca928de086404c4ee20bba51fc4d6a6b155edbefd0b722693d8f795f734 +size 8684772 diff --git a/parse/train/ryQu7f-RZ/ryQu7f-RZ_layout.pdf b/parse/train/ryQu7f-RZ/ryQu7f-RZ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9d3f9cc9115bf9796ae9712203fbfb4bc9b1c880 --- /dev/null +++ b/parse/train/ryQu7f-RZ/ryQu7f-RZ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f328f3a6e48f8f0407fa856ce5fc124aa1b0748cbae810524a49fe643f25f21 +size 1029666 diff --git a/parse/train/ryQu7f-RZ/ryQu7f-RZ_origin.pdf b/parse/train/ryQu7f-RZ/ryQu7f-RZ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ccdb3bcec65eb7c3189e8bc128273fcd5a3b5a14 --- /dev/null +++ b/parse/train/ryQu7f-RZ/ryQu7f-RZ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2c3b620f99b18855e73d07aed54bb9554f122201c4e618cc0d08684f46c8736 +size 521420 diff --git a/parse/train/ryQu7f-RZ/ryQu7f-RZ_span.pdf b/parse/train/ryQu7f-RZ/ryQu7f-RZ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9119d876946e4af82b420b6a76d1a43755b4cf83 --- /dev/null +++ b/parse/train/ryQu7f-RZ/ryQu7f-RZ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:224a11251412d3b4e941d973608333ffdbd9cad849a4c5707bcd289eae493b40 +size 1045485 diff --git a/parse/train/rylWVnR5YQ/rylWVnR5YQ_layout.pdf b/parse/train/rylWVnR5YQ/rylWVnR5YQ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d82c338772690732aded44cd461ae5af8548053e --- /dev/null +++ b/parse/train/rylWVnR5YQ/rylWVnR5YQ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6e301e1e105b4d498e70a628b5e7d0b6c746999963f87f877bebbed40f03e0c +size 1473964 diff --git a/parse/train/rylWVnR5YQ/rylWVnR5YQ_origin.pdf b/parse/train/rylWVnR5YQ/rylWVnR5YQ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fdc55796a071171d66807ce3b934074bfb18fb3a --- /dev/null +++ b/parse/train/rylWVnR5YQ/rylWVnR5YQ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00e06a93b54c85e3b621bb70d07de85d327a52bef1a295a27c7f2528872de15c +size 1384334 diff --git a/parse/train/rylWVnR5YQ/rylWVnR5YQ_span.pdf b/parse/train/rylWVnR5YQ/rylWVnR5YQ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6578642f8378b9ded87d9a1e766aecabcc451103 --- /dev/null +++ b/parse/train/rylWVnR5YQ/rylWVnR5YQ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8e9d6c78fc536d165e0168c5c0c2193a6a58b0e46effc3b854865f8f64f83bb +size 1473945 diff --git a/parse/train/sUgpxb9QD/sUgpxb9QD_layout.pdf b/parse/train/sUgpxb9QD/sUgpxb9QD_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3533be7115020c6b01e683981788f1e32f3f578e --- /dev/null +++ b/parse/train/sUgpxb9QD/sUgpxb9QD_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a10cce1b3bb7d75992c1085b81415c0b7805aacf14b7ad8ba4428682e27376f +size 669933 diff --git a/parse/train/sUgpxb9QD/sUgpxb9QD_origin.pdf b/parse/train/sUgpxb9QD/sUgpxb9QD_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0645c56b8bad560a87582b403c51b69b4f1a03f1 --- /dev/null +++ b/parse/train/sUgpxb9QD/sUgpxb9QD_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e62080e2c21d0a7c34ac60f0007030367492bc111f74602d30052ab2ed5c3f1 +size 472697 diff --git a/parse/train/sUgpxb9QD/sUgpxb9QD_span.pdf b/parse/train/sUgpxb9QD/sUgpxb9QD_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bbf8f3106ff0cd3bf5c9b48e6cdc350cd5428f04 --- /dev/null +++ b/parse/train/sUgpxb9QD/sUgpxb9QD_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:623bffcfa60ccc5b605ae107553ece028c9c02e1456ac5588e4cabb5775623be +size 688695 diff --git a/parse/train/umIdUL8rMH/umIdUL8rMH_layout.pdf b/parse/train/umIdUL8rMH/umIdUL8rMH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3b32d91c870ff313f1124e5e496ba05dca25098f --- /dev/null +++ b/parse/train/umIdUL8rMH/umIdUL8rMH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82bfdf2ec2d1c3a2059fa79bce4ee3a7ae76fdd9461ead163879e96c4856d8c8 +size 4532072 diff --git a/parse/train/umIdUL8rMH/umIdUL8rMH_origin.pdf b/parse/train/umIdUL8rMH/umIdUL8rMH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ea2b352705cd59e9113538cec5924981f0579ab9 --- /dev/null +++ b/parse/train/umIdUL8rMH/umIdUL8rMH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95dd00d2ea0fc8849daa192dea72b6ef3c4d27db576a3590b51bbfee1692ec27 +size 4186583 diff --git a/parse/train/umIdUL8rMH/umIdUL8rMH_span.pdf b/parse/train/umIdUL8rMH/umIdUL8rMH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b516dce86e202d252c611d059db4ea878ab971d5 --- /dev/null +++ b/parse/train/umIdUL8rMH/umIdUL8rMH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:289573ee5e2cce68ef3715d9589855f7808a6d3bf3de37abf443327f86a082ee +size 4530333 diff --git a/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_layout.pdf b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..11e814996b0b9def2afbc4a60fd5f7ae305e9fc6 --- /dev/null +++ b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb1304898d1ca5bd32c3d1330e1426232f57a61f04d8a1323ab407f6359cde6f +size 643471 diff --git a/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_origin.pdf b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3264796e9c96a7da6c1e254e343d4a6674b5e812 --- /dev/null +++ b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c7c905c24992b983bbb2bb29a35bc5e3fc3a30b04ef9b3a9412c38267db669a4 +size 499393 diff --git a/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_span.pdf b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5f9d8e1b85d45a00dbb71d1c272658af7a73b9ce --- /dev/null +++ b/parse/train/xWq1MVj7YrE/xWq1MVj7YrE_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dad8f56923ba8375544e313333340b12c3f01bf1f2acf92f197813e5a7e1ffab +size 655198 diff --git a/parse/train/zdrls6LIX4W/zdrls6LIX4W_layout.pdf b/parse/train/zdrls6LIX4W/zdrls6LIX4W_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..11735f976d6806a6566100e40522dcf3cf2ad5f0 --- /dev/null +++ b/parse/train/zdrls6LIX4W/zdrls6LIX4W_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29cccbd99d15c79ee23102a4a55fca8a40fd4aaf14070334f8cda95bf1e900ea +size 2930271 diff --git a/parse/train/zdrls6LIX4W/zdrls6LIX4W_origin.pdf b/parse/train/zdrls6LIX4W/zdrls6LIX4W_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..247ce03aa75173628cd4b972a9c3c1470d771b02 --- /dev/null +++ b/parse/train/zdrls6LIX4W/zdrls6LIX4W_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc630a579a1ad6b19fed1532c066de527a7663056e65701d4e2a4a10e85cc728 +size 2604510 diff --git a/parse/train/zdrls6LIX4W/zdrls6LIX4W_span.pdf b/parse/train/zdrls6LIX4W/zdrls6LIX4W_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..73cbcdfcbc26c23412ec5f74e994be190c525601 --- /dev/null +++ b/parse/train/zdrls6LIX4W/zdrls6LIX4W_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e32dc056e802fff29654cd13ab8468e711cdcf2f01a546ab6f06e86eab76a040 +size 2937150